Computer Science

Survey of Knowledge-enhanced Natural Language Generation Research

LIANG Mingxuan, WANG Shi, ZHU Junwu, LI Yang, GAO Xiang, JIAO Zhixiang

Computer Science. 2023, 50 (6A): 220200120-8. doi:10.11896/jsjkx.220200120

Abstract

PDF(2006KB) ( 363 )

References | Related Articles | Metrics

Natural language generation(NLG) task is a subclass of natural language processing(NLP) tasks and is a challenging task.With the massive application of deep learning in natural language processing,it has become the main method for handling various tasks in natural language generation.The main natural language generation tasks are question and answer tasks,summary generation tasks,comment generation tasks,machine translation tasks,generative dialogue tasks,etc.Traditional generative mo-dels rely on input text to generate text based on limited knowledge,and knowledge enhancement methods are introduced to solve this problem.Firstly,the research background and important models of natural language generation are introduced.Then,methods to improve model performance are introduced for natural language processing induction,and the methods and architectures based on the integration of internal knowledge(such as extracting keywords to enhance generation,surrounding subject words,etc.) and external knowledge(such as enhanced generation with the help of external knowledge graph) into the text generation process are introduced..Finally,the future challenges and research directions are discussed by analyzing some problems faced by the generation task.

Explainability of Artificial Intelligence:Development and Application

WANG Dongli, YANG Shan, OUYANG Wanli, LI Baopu, ZHOU Yan

Computer Science. 2023, 50 (6A): 220600212-7. doi:10.11896/jsjkx.220600212

Abstract

PDF(3193KB) ( 338 )

References | Related Articles | Metrics

In recent years,the extensive application of artificial intelligence in many fields and disciplines has shown its excellent performance.The improvement of this performance usually needs to sacrifice the transparency of the model.However,the complexity and black box nature of artificial intelligence models have become the main bottleneck in its application in high-risk fields,which seriously hinders the further application of artificial intelligence in specific fields.Therefore,it is urgent to improve the interpretability of the model to prove its reliability.Therefore,this paper introduces the typical models and methods of AI interpretability research from three aspects:machine learning model interpretability,deep learning model interpretability,and hybrid model interpretability,further describes the application of interpretable AI in teaching analysis,judicial judgment,and medical diagnosis,and summarizes and analyzes the shortcomings of existing interpretable methods,puts forward the development trend of the future research direction of AI interpretability,and hope to further promote the development and application of interpretability research.

Overview of Named Entity Recognition Tasks

GAO Xiang, WANG Shi, ZHU Junwu, LIANG Mingxuan, LI Yang, JIAO Zhixiang

Computer Science. 2023, 50 (6A): 220200119-8. doi:10.11896/jsjkx.220200119

Abstract

PDF(1989KB) ( 295 )

References | Related Articles | Metrics

Named entity recognition,as a very basic task in natural language processing,lays the foundation for the efficient completion of many other downstream tasks.Its purpose is to identify the corresponding entity from a text described in natural language and label its type,so as to make preparations for data labeling for other related tasks.This paper first introduces the deve-lopment process of named entity recognition tasks and the key methods used in related research in the corresponding context,including the rule-based and dictionary-based methods used in the early days of the birth,and the statistics and deep learning derived from the later development.Secondly,it summarizes some of the more mainstream research directions in this field,including named entity recognition under low-resource conditions,nested named entity recognition,and cross-language named entity recognition.These directions are the hot research trends of this task recently,including the most popular research method of this task at present.Finally,the relevant experience in the research is summarized,and the future development direction and difficulties of the task are prospected.

Summarization of Aspect-level Sentiment Analysis

LI Yang, WANG Shi, ZHU Junwu, LIANG Mingxuan, GAO Xiang, JIAO Zhixiang

Computer Science. 2023, 50 (6A): 220400077-7. doi:10.11896/jsjkx.220400077

Abstract

PDF(1868KB) ( 376 )

References | Related Articles | Metrics

Sentiment analysis is one of the important branches of natural language processing.With the development of the times,in order to extract more sentiment information from text,aspect-level sentiment analysis is paying more and more attention in sentiment analysis.Firstly,this paper introduces the background knowledge and related concepts of aspect-level sentiment analysis,and explains it from the perspective of two subtasks of aspect extraction and aspect sentiment classification.In terms of aspect extraction,related methods based on similarity algorithms,topic models and sequence labeling are introduced.In terms of aspect sentiment classification,related methods based on sentiment lexicon and rules,machine learning and deep learning are introduced,and the Chinese and English data sets and sentiment lexicon commonly used in aspect-level sentiment analysis are sorted out.Finally,making a summary and outlook for the current challenges and future development directions of aspect-level sentiment analysis.

Review on Causality Detection Based on Empirical Dynamic Modeling

CAO Zhihao, MU Shaomin, QU Hongchun

Computer Science. 2023, 50 (6A): 220600194-12. doi:10.11896/jsjkx.220600194

Abstract

PDF(2678KB) ( 355 )

References | Related Articles | Metrics

Correlation is an important analysis standard in curent scientific research,but it does not mean causality.As people rea-lize the universality of no nlinear dynamics,it is very likely to lead to wrong conclusions only by relying on corelation.At present,various correlation research algorithms,including machine learning,are developing rapidly,while the research of mining causal correlation between variables is still under exploration.Empirical dynamic modeling theory is a data-driven dynamic system mo-deling framework.Its biggest feature is to abandon the formulaic method in traditional data analysis and reconstruct the behavior of dynamic system only from time series.The core idea is that a dynamic system can be described as a process in which a group of states evolve over time driven by certain rules in high-dimensional space.The dynamic system can be modeled by reconstructing the states that evolve over time.Based on empirical dynamic modeling theory,the causal relationship betwen variables can be detected through the time series data of variables in dynamic system.If variable X is the cause of variable Y(X⇒Y),the information of variable X must be implicit in variable Y and can be recovered from variable Y.This paper first analyzes the dialectical relationship between correlation and causality.Correlation does not mean causality,and lack of correlation does not mean no causality.Then it comprehensively introduces the core idea of causality detection based on empirical dynamic modeling theory,andsummarizes the historical development of Takens embedding theorem,simplex projection algorithm and convergent cross mapping algorithm.It introduces some improved methods of empirical dynamic modeling theory and practical application of causal detection,and finaly looks forward to the future development trend of causal detection based on empirical dynamic modeling.

L-type Redundancy Property in Propositional Logic

LIU Lingrong, CHEN Shuwei, JIANG Shipan

Computer Science. 2023, 50 (6A): 220600013-5. doi:10.11896/jsjkx.220600013

Abstract

PDF(1984KB) ( 222 )

References | Related Articles | Metrics

The technique of clause set simplification is a crucial part in the process of propositional logic SAT solving.The clause elimination methods corresponding to the redundancy property can accurately identify and delete redundant clauses.Either in the pre-processing stage or in the process of SAT solving,the embedded clause elimination methods in the SAT solver accelerate the solving efficiency of the solver. A large number of efficient clause elimination methods are extended based on the redundancy properties of blocked clauses and implication module resolution clauses.In order to check whether clause C is redundant,we need only to consider whether clause C satisfies the redundancy conditions.The proposed L-type redundancy is an extension of the blocked redundancy property,subsumption redundancy property and implication module resolution redundancy. It extends the judgement condition of redundant clause from the resolution of single literal to the combination of the set of literals.According to the properties of L-type redundancy,this paper analyzes the properties of L-type redundancy clause,and applies comparison analy-sis on the efficiency of L-type redundancy clauses and other existed redundant clauses,so as to illustrate the efficiency of L-type redundancy clauses.

Entity Relation Extraction Method in Weapon Field Based on DCNN and GLU

LI Han, HOU Shoulu, TONG Qiang, CHEN Tongtong, YANG Qimin, LIU Xiulei

Computer Science. 2023, 50 (6A): 220200112-7. doi:10.11896/jsjkx.220200112

Abstract

PDF(2529KB) ( 282 )

References | Related Articles | Metrics

Unstructured text data in the field of weapons is usually very complex.In a single sentence,one weapon may be associated with multiple weapons or there may be multiple relations between two weapons.An entity relation extraction method based on dilated convolutional neural network and gated linear unit is proposed to solve the problem of overlapping relation in this type of data.This method introduces the sentence coding vector into the dilated convolutional neural network model with gated linear unit,which combines word vector and position vector.And it introduces the self-attention mechanism to extract the feature information of entities in sentences quickly.Through hierarchical sequence annotation,this model identifies all entities in the sentence and all relations and object entities corresponding to each subject entity,and generates the entity relation triplet in the field of weapons.The F1 value of this method on the self-labeled weapon field data set is 81.1%,and it has a certain entity relation extraction ability,according to the experimental results.The F1 value for various overlap types is greater than 78%,which solves the problem of unstructured data relation overlap.At the same time,it performs admirably on the NYT public data set.

UAV Dynamic Route Planning Algorithm Based on RRT

GU Zilyu, LIU Yu, SUN Wenbang, YUE Guang, SUN Shangwen

Computer Science. 2023, 50 (6A): 220700127-5. doi:10.11896/jsjkx.220700127

Abstract

PDF(3183KB) ( 269 )

References | Related Articles | Metrics

Aiming at the problems of slow speed,poor flight ability and insufficient dynamic adjustment ability in traditional route planning algorithms,a dynamic route planning algorithm for UAV based on improved rapidly exploring random tree(RRT) is proposed.Firstly,when the RRT method is introduced for global route planning,in order to accelerate the convergence speed of the algorithm,target heuristic information is introduced in the selection of random tree nodes to be expanded,and UAV dynamic constraints are incorporated in the generation and addition of new nodes to ensure that the generated route has realistic flight abi-lity.Secondly,considering the emergent threat,a method of dynamically expanding random tree is proposed to prune and reconstruct the original random tree,so as to avoid the threat quickly and generate a safe route.Experimental results show that compared with the traditional RRT algorithm,the improved algorithm can improve the planning speed by about 20% and reduce the number of nodes by 32%,and the planned route conforms to the basic dynamics constraints of UAV.In addition,when facing emergent threats,the route can be dynamically adjusted quickly to achieve route re-planning.

New Global Optimization Algorithm:Carbon Cycle Algorithm

YANG Da, LUO Liang, ZHENG Long

Computer Science. 2023, 50 (6A): 220300131-7. doi:10.11896/jsjkx.220300131

Abstract

PDF(3622KB) ( 258 )

References | Related Articles | Metrics

With the rapid development of human science and technology,in the fields of applied research and engineering design,there are problems of large dimensions,high order,many objective functions,and complex constraints,which are difficult to solve by traditional algorithms and need to be optimized and resolved.Based on the continuous development of computer operation and problem solving level,metaheuristic optimization algorithms have been proposed and proved to be superior to traditional optimization methods in solving the above categories of problems.As a complement to the metaheuristic optimization algorithm,this paper proposes a new metaheuristic algorithm,called the carbon cycle algorithm(CCA),for continuous global optimization.This algorithm simulates the carbon element cycle in nature(mainly the biosphere).Plant respiration,animal respiration,animal predation,plant death process,animal death process,decomposer’s de-composition and plant photosynthesis process are simulated by this algorithm which uses these as search strategies to explore and search space.The computational convergence procedure of the proposed algorithm is dissected by comparing the result of some well-known optimization algorithms on the 13 benchmark functions.The test results of benchmark functions reveal that the proposed algorithm can provide an excellent solution which proves CCA can solve the challenging problem and is a competitive algorithm.CCA provides better solution accuracy on most benchmark functions.

Aspect-based Sentiment Analysis Based on Prompt and Knowledge Enhancement

LI Yang, TANG Jiqiang, ZHU Junwu, LIANG Mingxuan, GAO Xiang

Computer Science. 2023, 50 (6A): 220300279-7. doi:10.11896/jsjkx.220300279

Abstract

PDF(2456KB) ( 368 )

References | Related Articles | Metrics

sentiment analysis is an emerging fine-grained sentiment analysis task that aims to judge sentiment polarity based on given sentences and aspect words.Currently widely used pre-trained language models are different due to their training objectives and those of aspect-based sentiment analysis,resulting in poor analysis results.In order to alleviate the difference between the pre-trained language model and the sentiment analysis target,prompt is introduced into aspect-based sentiment analysis,using pseudo-labels plus aspect words and opinion words to create prompt continuous templates,and using prompt-encoder to train pseudo-labels to have Semantic information;then,use the topic graph attention mechanism to fuse external knowledge about aspect words and opinion words,and predict candidate label words composed of sentiment dictionaries according to the hidden vector fused with external knowledge;The probabilities of label words are mapped onto the sentiment polarity distribution space.Experiments show that the model improves the accuracy by 1.53% and 3.5% on the Laptops dataset and Restaurants dataset of the SemEval 2014 task.

Building Natural Language Interfaces for Distributed SCADA Systems Using Semantic Parsing

WANG Tao, GUO Wushi, DENG Jian, CHEN Liang

Computer Science. 2023, 50 (6A): 220300141-9. doi:10.11896/jsjkx.220300141

Abstract

PDF(2751KB) ( 269 )

References | Related Articles | Metrics

Due to the traditional program fixed window interface human-computer interaction,large distributed industrial process SCADA systems are mainly operated in the central control room and maintained by professional staff,so the system construction and operation and maintenance costs are very high,and it is significant to explore the natural human-computer interaction interface and guide the system adaptive services.Taking a distributed SCADA system for various professional fields as the background,this paper analyzes the core requirements of natural human-computer interaction from the perspective of actual operation.Different semantic parsing algorithms are recommended according to the complexity of natural language instructions.For basic natural language instructions,TF-IDF keyword extraction algorithm is used and combined with cosine similarity for structured extraction,which is parsed into SCADA manipulation intermediate language and converted into actual manipulation instructions by formalization.For complex natural language instructions,a structured instruction parsing algorithm based on dependency syntax analysis is used to realize the real-time control interface.Experimental results show that the proposed natural language interface can better solve the human-computer natural language interaction problem of SCADA system.The average accuracy,recall and F-value of instruction parsing is 89.27%,89.28% and 89.27%,respectively.The average response time is 1.593s,which provides a more convenient means of interaction,especially for industrial and agricultural information control.

Text Classification Based on Weakened Graph Convolutional Networks

HUANG Yujiao, CHEN Mingkai, ZHENG Yuan, FAN Xinggang, XIAO Jie, LONG Haixia

Computer Science. 2023, 50 (6A): 220700039-5. doi:10.11896/jsjkx.220700039

Abstract

PDF(1892KB) ( 240 )

References | Related Articles | Metrics

Text classification is a classic problem in the field of natural language processing.The traditional text classification model needs to extract features manually,the classification accuracy is not high,and it is difficult to deal with non-European spatial data.In order to solve the above problems and further improve the accuracy of text classification,the W-GCN model is proposed.This model is improved on the basis of the Text-GCN model,and a new weakened structure model is established to replace the text-GCN model.The dropout operation of neurons,and by weakening the weight,accurately control the weakening strength,and on the basis of retaining the dropout to a certain extent to prevent overfitting,it avoids the loss of features caused by directly discarding neurons,thus improving the accuracy of model classification..Compared with the Text-GCN model,the W-GCN model based on the weakened graph convolutional network improves the accuracy by 0.38% on the R8 dataset and 0.62% on the R52 dataset.The experimental results prove that the model Improve and weaken the effectiveness of the structure.

Study on Intelligent Decision Making of Aerial Interception Combat of UAV Group Based onMADDPG

LIN Xiangyang, XING Qinghua, XING Huaixi

Computer Science. 2023, 50 (6A): 220700031-7. doi:10.11896/jsjkx.220700031

Abstract

PDF(4332KB) ( 357 )

References | Related Articles | Metrics

Based on the requirements of future modern operations,a combat scenario is built.Under this scenario,reinforcement learning is used to solve the multi-target intelligent decision-making problem about aerial interception mission of UAVs.The multi-agent reinforcement learning algorithm is selected according to the operational mode and application requirements,and the algorithm principle and process are briefly introduced.The simulated combat system is developed.Design network model,network parameters and training methods.After training,the expected results have been achieved.The effectiveness of the experiment is proved,which not only provides technical support for practical application of this kind of problem,but also provides theoretical basis and experimental reference for the study of intelligent decision making in more complex combat scenarios and combat mission conditions.

Path Planning of Mobile Robot Based on Improved B-RRT^* Algorithm

YU Jiuyang, ZHANG Dean, DAI Yaonan, HU Tianhao, XIA Wenfeng

Computer Science. 2023, 50 (6A): 220500038-7. doi:10.11896/jsjkx.220500038

Abstract

PDF(5356KB) ( 291 )

References | Related Articles | Metrics

In the field of mobile robot motion path planning,the asymptotically optimal bidirectional rapid exploration random tree(B-RRT^*) algorithm has good obstacle avoidance and path search capabilities,but B-RRT^* has the shortcomings of many iterations and long planning time.As an efficient branch of the B-RRT^* algorithm,the kinematic constraint-based bidirectional rapid exploration random tree(KB-RRT) algorithm can effectively reduce the expansion of invalid trees and speed up the search for the optimal path,but the number of iterations of the algorithm is too large.For the improvement of the B-RRT^* algorithm,the latest B-RRT^* improved algorithm is the kinematically constrained B-RRT^*(KB-RRT^*) algorithm with efficient branches.Although the KB-RRT^* algorithm can effectively reduce the expansion of invalid trees,to speed up the search for the optimal path,but the number of iterations of the algorithm is still too large.Therefore,this paper proposes an improved B-RRT^* algorithm(AFB-RRT^*) based on adaptive sampling and fast search,which sets a safe area for obstacles and determines the search direction of a random tree according to the proposed adaptive sampling and fast search,reduces redundant sampling points,that is,AFB-RRT^* can achieve fast convergence in path planning.Simulation and experiments show that,compared with KB-RRT^*,AFB-RRT^* reduces the planning time and the number of convergence iterations under the premise that the planned path length is basically the same.

Study on Named Entity Recognition Method Based on Knowledge Graph Enhancement

GAO Xiang, TANG Jiqiang, ZHU Junwu, LIANG Mingxuan, LI Yang

Computer Science. 2023, 50 (6A): 220700153-6. doi:10.11896/jsjkx.220700153

Abstract

PDF(2090KB) ( 296 )

References | Related Articles | Metrics

Named entity recognition is a very basic task in natural language processing,and its purpose is to identify the corresponding entities and types from a text described in natural language.As external knowledge in the form of triples,knowledge graphs have been applied in many natural language processing tasks and achieved good results.This paper proposes an attention-aligned named entity recognition method based on knowledge graph information enhancement.Firstly,the knowledge graph information is embedded through the embedding layer and attention mechanism to obtain the representation of the knowledge graph triple information.Secondly,the sentence is obtained through BERT-BiLSTM.Then,an attention alignment module is used to assign triple weights to fuse the representation of knowledge graph information and sentence information.Finally,the prediction output of the fused representation vector is controlled by softmax,and the label of the entity is obtained.This method effectively avoids changing the semantic information of the original sentence due to the fusion of knowledge graphs,and also enables the word vectors in the sentence to have rich external knowledge.The proposedmethod achieves F1 values of 95.73% and 93.80% on the Chinese general data set MSRA and the medical domain specific data set Medicine,respectively,achieving advanced perfor-mance.

Observability of Probabilistic Boolean Control Networks

FAN Zhuoyou

Computer Science. 2023, 50 (6A): 220200068-6. doi:10.11896/jsjkx.220200068

Abstract

PDF(1825KB) ( 247 )

References | Related Articles | Metrics

The uncertainty of the transfer matrix brings difficulties to the observability and controllability analysis for probabilistic Boolean control networks(PBCNs).This paper mainly studies the observability of PBCNs,and the conditions of observability are also developed for PBCNs.On this basis,the method for calculating the initial state vector of the system is given.Firstly,according to the reachable state set of PBCNs,the distinguishable and indistinguishable states of the system are defined,and the concept of $d$-step distinguishability and the necessary and sufficient conditions for its judgment are given.Secondly,based on the output and state model of PBCNs,the probabilistic initial state set of the system is also obtained.Then,the definition of strong observability and weak observability of PBCNs are given.Meanwhile,the methods of calculating the initial state vector and determining whether a given PBCN is observable are obtained.Finally,an example is given to illustrate the effectiveness of the proposed methods.

Study on Diagonal Method Based on Indefinite Extensibility Concept

DUAN Tianlong

Computer Science. 2023, 50 (6A): 211100070-5. doi:10.11896/jsjkx.211100070

Abstract

PDF(1669KB) ( 264 )

References | Related Articles | Metrics

Using diagonal method yields not only basic theorems for mathematical logic,but also leads to logical paradoxes.By exa-mining the concept of indefinite extensibility and its association with the diagonal method,this paper points out that:1)since Simmons does not associate the concept of indefinite extensibility with the diagonal method,its division of “diagonal argument types” is only superficial;2)Thomson diagonal lemma does not depict the essence of diagonal method; 3) diagonal method can be used for a concept if and only if the concept is indefinite extensibility;4)the use of diagonal method can help us to find the concept of indefinite extensibility and provide ideas for describing the dynamic semantic model of natural language.

Transfer Learning Based Cross-object Sign Language Gesture Recognition Method

WANG Tianran, WANG Qi, WANG Qingshan

Computer Science. 2023, 50 (6A): 220300232-5. doi:10.11896/jsjkx.220300232

Abstract

PDF(2583KB) ( 331 )

References | Related Articles | Metrics

Sign language is an important communication tool for hearing impaired people,and accurate recognition of sign language can reduce the communication barrier between able-bodied and hearing impaired people.The performance of general deep learning recognition models is highly dependent on the collected data,which leads to poor cross-object generalization ability of the models.Therefore,this paper designs a sign language gesture recognition model with cross-object generalization capability through a transfer learning approach.Firstly,a feature extractor is used to fuse the surface electromyography (sEMG) signal and the inertial measurement unit (IMU) signal.Then,a domain adversarial training method is proposed,which can complete the adversarial training of the feature extractor and domain classifier by relying on the source domain data only,and realize the migration of feature extraction from the source domain to the target domain.Finally,domain-invariant features are used in the gesture classifier to achieve sign language gesture cross-object recognition,which improves the generalization ability of the model in this paper.Experiments show that the proposed model can improve the accuracy of sign language cross-object recognition to 85.1% on a dataset containing 200 sign language gestures with a total of 60 000 sign language samples.

Study on Improvement of Deep Point Cloud Network Based on Multiple Emphasis Mechanisms

LIU Hui, TIAN Shuaihua

Computer Science. 2023, 50 (6A): 220400164-7. doi:10.11896/jsjkx.220400164

Abstract

PDF(3027KB) ( 243 )

References | Related Articles | Metrics

Machine vision is a key technology for robots to identify working objects from complex spatial environments.Kinect depth cameras or laser scanning sensors commonly used in robotic systems are capable of acquiring three-dimensional information about the target,which makes it possible for robots to perform more complex work tasks such as assembly,disassembly,and grasping.However,this also places higher demands on the robot system’s ability to process 3D information such as 3D localization,work object size measurement,and estimation.We analyze the main feature emphasis mechanisms of soft threshold squeeze-and-excitation,channel-wise gated,and attention mechanisms based on PointNet networks,and improve PointNet networks by using soft threshold squeeze-and-excitation,channel-wise gated,and attention networks,respectively,and experimentally validate them on the publicly available ShapeNet dataset from Stanford University.Experimental results show that the improvement of original network by the three emphasis mechanisms improves segmentation accuracy(mean intersection and merge ratio) of 3D point clouds by 0.24%,0.68%,and 0.93%,respectively,in comparison with original PointNet network.The improved method lays foundation for the subsequent solution of accurate estimation for the size of working objects in tasks such as assembly,disassembly and grasping by robots.

Study on Satire Detection Based on Sentiment-Topic-Satire Hybrid Model

FU Yue, SHI We

Computer Science. 2023, 50 (6A): 220300018-6. doi:10.11896/jsjkx.220300018

Abstract

PDF(2079KB) ( 260 )

References | Related Articles | Metrics

Satire detection is a subtask of opinion mining.Its main purpose is to identify the opinion or emotions expressed by users in written texts.Satirical sentences in texts often have mixed sentiment polarity.Correctly identifying satirical sentences and non satirical sentences plays an important role in sentiment analysis.Various satire detection methods are based on machine lear-ning classifiers,in which the training of classifiers is mainly based on simple words or dictionary features.The purpose of this studyis to establish an unsupervised probabilistic relationship model to identify satirical themes according to the sentiment distribution of words in microblog.The model estimates the related sentiment based on the topic level distribution,evaluates the sentiment related words appearing in the short text,and gives the sentiment related labels.Experimental results show that the model is superior to other latest baseline models in satire detection,and is very suitable for satire prediction of short text.

New Cost Sensitive SVDD Binary Classification Method

WU Chongming, WANG Xiaodan, ZHAO Zhenchong

Computer Science. 2023, 50 (6A): 220300202-5. doi:10.11896/jsjkx.220300202

Abstract

PDF(1760KB) ( 209 )

References | Related Articles | Metrics

In order to improve the performance of cost-sensitive classification,this paper improves the learning accuracy of higher misclassification cost categories to reduce the total misclassification cost,uses support vector domain description(SVDD) to reali-ze cost sensitive classification,and proposes a cost sensitive SVDD two-class classification method,CS-SVDD.This method first expands single class SVDD to two class classification SVDD,and constructs SVDD hyperspheres for different categories.by adjusting the classification accuracy of SVDD classifier for different class samples through the misclassification cost,the class with high misclassification cost can be more accurately learned,so as to reduce the total misclassification cost.For the samples with ambiguous category attributes outside the two hyperspheres or in the coverage area,cost sensitive decision rules are defined based on the principle of minimum misclassification cost.Experimental results on artificial data sets and UCI data sets show the effectiveness of the proposed method.

Intelligent Winding Motion Control Method Based on Bionic Snake-like Robot

DU Hongjian, GUO Peng, XIAO Wenyu, YIN Jun

Computer Science. 2023, 50 (6A): 220700060-5. doi:10.11896/jsjkx.220700060

Abstract

PDF(2608KB) ( 267 )

References | Related Articles | Metrics

Winding gait control is one of the basic motion control of special bionic snake-like robots.The tracking of snake-like robots in winding motion has the problems of target loss,head shaking and unstable direction.Combining sliding mode variable structure control and the Canny multi-level edge detection algorithm,this paper proposes an intelligent winding motion control method.First,a linear tracking controller is designed based on sliding mode control theory.Second,a new winding gait control function is proposed to solve the problem of head shaking of the snake-like robot.Finally,combined with the Canny algorithm,linear tracking controller and the winding gait control function,a complete control loop of the snake-shaped robot is established to realize the directional control of the robot.Experimental results show that the proposed intelligent winding gait control method is more stable and robust than the traditional winding motion.

Graph Attention Networks Based on Causal Inference

ZHANG Tao, CHENG Yifei, SUN Xinxu

Computer Science. 2023, 50 (6A): 220600230-9. doi:10.11896/jsjkx.220600230

Abstract

PDF(3074KB) ( 308 )

References | Related Articles | Metrics

Graph attention network(GAT) is an important graph neural network with a wide range of applications in classification tasks.However,when the neighborhood nodes in the graph are disturbed,the model classification accuracy will be affected and degraded.In response,a graph attention network based on causal inference named causal graph attention network(C-GAT) is proposed to improve the robustness of the network.The model first calculates the causal weights between the neighborhood of the target node and its label and uses them to sample the neighborhood.Then the attention coefficient between the sampled neighborhood and the target node is calculated.Finally,the embedding features of the target nodes are obtained by weighted aggregation of the neighborhood information based on the attention coefficients.Experimental results on the Cora,Pubmed and Citeseer datasets show that the classification performance of C-GAT is on par with the classical model in the case of no perturbation.In the presence of perturbations,the classification accuracy of C-GAT improves by at least 6% compared to the classical model,with a better balance of robustness and time cost.

Speech Emotion Recognition Based on Improved MFCC and Parallel Hybrid Model

CUI Lin, CUI Chenlu, LIU Zhengwei, XUE Kai

Computer Science. 2023, 50 (6A): 220800211-7. doi:10.11896/jsjkx.220800211

Abstract

PDF(3781KB) ( 199 )

References | Related Articles | Metrics

The traditional MFCC not only ignores the influence of the pitch frequency in the voiced signal,but also cannot characterize the dynamic characteristics of the speech.Therefore,a moving average filter is proposed to filter out the pitch frequency of the voiced signal.After extracting the static MFCC features,the dynamic features are obtained by extracting their first-order difference and second-order difference.The obtained features are sent to the model for training.To construct a more efficient speech emotion recognition model,a parallel hybrid model integrating a multi-head attention mechanism is built.The multi-head attention mechanism can not only effectively prevent the gradient disappearance phenomenon from constructing a deeper network,but also perform different tasks to improve the accuracy.Finally,when classifying emotional features,the traditional softmax may increase the intra-class distance during classification,resulting in poor confidence in the model.Therefore,the center loss function is introduced to combine the two for classification.Experimental results show that the accuracy of the proposed method can reach 98.15 % and 96.26 % on the RAVDESS dataset and EMO-DB dataset,respectively.

EEG Emotion Recognition Based on Multiple Directed Weighted Graph and ConvolutionalNeural Network

LUO Ruiqi, YAN Jinlin, HU Xinrong, DING Lei

Computer Science. 2023, 50 (6A): 220600128-8. doi:10.11896/jsjkx.220600128

Abstract

PDF(3432KB) ( 309 )

References | Related Articles | Metrics

In recent years,research on EEG signals applied to emotion recognition has received extensive attention,and mapping time series into visibility graph representation can effectively perform EEG emotion recognition through the metric of edges and nodes of the visibility graph.Traditional visibility graph algorithms ignore the correlation between multi-channel EEG signals,and it is difficult to retain the complex feature information on the time series.Therefore,this paper proposes a method to extract EEG signal features from multiple directed weighted visibility graphs and use EEG signal features for emotion recognition.Firstly,the EEG signal is converted into a directed weighted network graph to enhance the feature representation of the signal,the complex network structure is characterized using weighted clustering coefficients,the EEG connection matrix is established with multiple complex networks,and finally the convolutional neural network is used for feature learning,and the emotion recognition results are obtained through learning.The complex network model constructed by multiple directed weighted visibility maps achieves 93.85% accuracy in public dataset validation,which is better than the existing traditional visibility graph methods,and the multiple weighted visibility graph improves the emotion recognition accuracy by 9.4% compared with univariate visibility graph.Experimental results show that the method is also applicable to cross subject data and has good robustness.

Study of Multi-task Learning with Joint Semantic Segmentation and Depth Estimation

LUO Huilan, YE Ju

Computer Science. 2023, 50 (6A): 220100111-10. doi:10.11896/jsjkx.220100111

Abstract

PDF(3680KB) ( 244 )

References | Related Articles | Metrics

Semantic segmentation and depth estimation are two highly related tasks of image pixel-level classification.This paper proposes two different multi-task learning architectures from the perspectives of both shared feature extraction and feature interaction fusion:multi-task learning with SE and pyramid pooling (MTL_SPP) based on the squeeze and excitation (SE) and pyramid pooling,and multi-task learning network (MTL_SSW) based on se and selective weights (SW) to jointly learn semantic segmentation and depth estimation.The MTL_SPP architecture consists of shared backbone feature network and task-specific sub-networks,using the SE module to construct task-specific sub-networks and pyramid pooling to enhance feature extraction.Based on MTL_SPP,MTL_SSW adds SW modules which allows the semantic segmentation features and depth estimation features from task-specific sub-networks to guide and optimize each other, o it can learn more discriminative features.Experimental results show that the two proposed methods obtain better results than the state-of-the-art methods on both NYUD_v2 and SUNRGBD datasets.

Few-shot Learning Method Based on Multi-graph Feature Aggregation

ZENG Wu, MAO Guojun

Computer Science. 2023, 50 (6A): 220400029-10. doi:10.11896/jsjkx.220400029

Abstract

PDF(3224KB) ( 229 )

References | Related Articles | Metrics

Few-shot learning can learn the characteristics of various samples from fewer samples,but due to the problem of low data,that is,the number of samples is small,how to more accurately extract the important feature information in the image,and how to better learn from the image.The characteristics of the target object and the more accurate judgment of the similarity between the unlabeled samples and the support set category become the key.A few-shot learning method MGFAN based on multi-graph feature aggregation is proposed.Specifically,the model expands the original image through various data enhancement me-thods,and then uses a self-attention module to obtain important feature information between the original image and different expanded images,so as to obtain more accurate features vector about the image.Secondly,the self-supervised learning task of predicting different augmentation methods of images is introduced into the model as an auxiliary task to promote the feature learning ability of the model.Finally,multiple distance functions are used to calculate the similarity between samples more accurately.Experiments in 3 standard datasets miniImageNet,tieredImageNet and Stanford Dogs using 5-way 1-shot and 5-way 5-shot experimental settings show that the MGFAN method can significantly improve the classification performance of the classifier.

Document-level Relation Extraction of Graph Attention Convolutional Network Based onInter-sentence Information

DUAN Jianyong, YANG Xiao, WANG Hao, HE Li, LI Xin

Computer Science. 2023, 50 (6A): 220800189-6. doi:10.11896/jsjkx.220800189

Abstract

PDF(2042KB) ( 301 )

References | Related Articles | Metrics

In order to solve the problem of insufficient structural information mining for documents in existing models,a graph attention convolution network model based on inter-sentence information is proposed.In this model,a document level encoder is improved,which uses a new attention mechanism,inter-sentence attention mechanism,so that the final representation of a sentence pays more attention to the important information in the previous sentence and the previous document,which is more conducive to mine the structural information of the document.Experiments show that the F₁ evaluation index of this model on DocRED data set reaches 56.3%,and its performance is better than that of the baseline model.When integrating the inter sentence attention mechanism,the model needs to perform inter-sentence attention operations for each sentence,so it needs to consume more memory and time when training the model.The graph attention convolution network model based on inter-sentence information can effectively aggregate the relevant information in the document,and enhance the mining ability of the document structure information,so that the model can improve the effect of the document level relationship extraction task.

Few-shot Segmentation Based on Multi-scale Prototype Hierarchical Matching

SUN Kaiwei, LIU Hu, RAN Xue, GUO Hao

Computer Science. 2023, 50 (6A): 220300275-7. doi:10.11896/jsjkx.220300275

Abstract

PDF(2340KB) ( 207 )

References | Related Articles | Metrics

Traditional semantic segmentation tasks usually need a lot of labeled data for training,and it is difficult to generalize to new categories.Few-shot segmentation aims to segment new categories of target objects from query images using a small number of annotated supporting images.Due to the limited supporting image data,how to extract representative guidance information from limited support images is an important challenge for few-shot segmentation task.In order to solve this problem,a few-shot segmentation method based on multi-scale prototype hierarchical matching is proposed in this paper.Firstly,the middle-level and high-level features of the query image and the support image are obtained through the residual network ResNet.In order to further extract the rich context feature information of the target object,the extracted middle-level features are fed into the pyramid pooling module for multi-scale feature extraction.Based on the idea of prototype learning,middle-level features and high-level features are layered to generate prototypes and matched to obtain the final predicted segmentation mask.Experiments are carried out on the PASCAL-5ⁱ dataset and experimental results show that the mIoU of the proposed method achieves 66.7% in 1-way 5-shot setting,which is 11% and 4.8% higher than the current mainstream PANet and PFENet models,respectively,demonstrating the effectiveness and advanced nature of the method.

Extractive Automatic Summarization Model Based on Knowledge Distillation

ZHAO Jiangjiang, WANG Yang, XU Yingying, GAO Yang

Computer Science. 2023, 50 (6A): 210300179-7. doi:10.11896/jsjkx.210300179

Abstract

PDF(2502KB) ( 262 )

References | Related Articles | Metrics

The objective of the extractive summarization is to extract the important sentences from the original text to form a short summary while retaining the important content of the original text.Query-focused extractive summary model can further satisfy users’ different needs for summary content.Extractive summary model has the natural advantage of ensuring the correctness of summary and the readability of sentences.On this basis,ensuring the relevance and importance of the summary content is the key to the goal of the model.In order to satisfy the relevance of query and ensure the importance of summary content,this paper uses query information as a model study target,creates an extended summary data set based on the title and picture information,an extractive summary model based on knowledge distillation is proposed.In experiments,the pre-training language model BERT is adopted as the encoder,two model training strategies based on knowledge distillation theory are proposed:guided trai-ning and distillation training.Experimental results on CNN/DailyMail,a publicly available data set of news summaries,show that both training methods have achieved significant effects.It is also found that the model based on guiding training could effectively improve the significance of the summary content,while the model based on distillation training achieves the best effect in improving the relevance and significance of the summary.

Study on Long Text Topic Clustering Based on Doc2Vec Enhanced Features

CHEN Jie

Computer Science. 2023, 50 (6A): 220800192-6. doi:10.11896/jsjkx.220800192

Abstract

PDF(3686KB) ( 252 )

References | Related Articles | Metrics

Aimed at the difficulties of semantic representation of long news text,an enhanced document feature representation is constructed based on Doc2Vec embedding and word vector weighting.Enhanced features from the specific parts-of-speech contents on the head and tail of the document are extracted by the method of DV-sim or DV-tfidf.These features are then combined with doc2vec to form a new global representation.DV-sim uses the similarity between feature words and doc2vec vectors to obtain word weight from the semantic point of view,and DV-tfidf uses term frequency inverse document frequency to obtain word weight from the word frequency statistics point of view.Then the HDBSCAN algorithm is applied to cluster topics on the Thucnews and Sogou datasets.Compared with the Doc2Vec vector,the noise number on the two datasets reduces by 60.82% and 60.63%,the accuracy improves by 12.14% and 20.58%,and the F1-score increases by 15.61% and 11.58%,respectively,with DV-sim.The noise number on the two datasets reduces by 15.20% and 59.55%,the accuracy improves by 10.85% and 17.93%,and the F1-score increases by 15.60% and 9.21%,respectively,with DV-tfidf.Experiments show that both DV-sim and DV-tfidf can improve the performance of topic clustering,and the enhancement feature based on semantics is better than that based on word frequency.DV-sim has also been effectively applied in topic clustering of excellent female character reports.

FIR Low Pass Digital Filter Based on Multi-strategy Discrete Artificial Bee Colony Algorithm

SHAO Peng

Computer Science. 2023, 50 (6A): 220700026-5. doi:10.11896/jsjkx.220700026

Abstract

PDF(1848KB) ( 263 )

References | Related Articles | Metrics

Aiming at the shortcomings of artificial bee colony(ABC) algorithm in solving complex problems,such as lower accuracy and slower convergence speed,a discrete artificial bee colony fusing the refraction learning and Lévy flight(DABC-RL) algorithm is proposed to design finite impulse response(FIR) low-pass digital filter,so as to further improve its filtering performance.In the DABC-RL algorithm,on the one hand,Lévy flight strategy is used to enhance its local search ability,and refraction learning is employed to enhance its global search ability.On the other hand,the candidate solution in the DABC-RL algorithm is discretized by designing an appropriate discrete coding scheme,which makes it suitable for designing FIR low pass digital filter.In order to test the performance of the FIR low-pass digital filter designed by the proposed DABC-RL algorithm,the FIR low-pass digital filters designed by ABC algorithm and the refrPSO algorithm based on the refraction learning is selected as two comparative algorithms.Experimental results and analysis show that compared with other algorithms,the FIR low-pass digital filter designed by DABC-RL algorithm has the best performance,and obtains the fastest convergence accuracy and convergence speed.

LDM-EEG:A Lightweight EEG Emotion Recognition Method Based on Dual-stream Structure Scaling and Multiple Attention Mechanisms

LEI Ying, LIU Feng

Computer Science. 2023, 50 (6A): 220300262-9. doi:10.11896/jsjkx.220300262

Abstract

PDF(3257KB) ( 308 )

References | Related Articles | Metrics

EEG emotion recognition is a multi-channel time-series signal classification problem with high complexity,high information density and massive data.In order to achieve optimal accuracy and performance of EEG emotion recognition with fewer computational parameters while maintaining the existing classification accuracy,this paper proposes a lightweight network(LDM-EEG) based on dual-stream structural scaling and multiple attention mechanisms.The network takes the time-space and frequency-space maps constructed based on the differential entropy features of EEG signals as the input,processes the two features separately using a symmetric dual-stream structure,achieves lightweighting through a novel parameter-saving residual module and a network scaling mechanism,and enhances the model feature aggregation capability using a novel channel-time/frequency-space multiple attention mechanism and a post-attention mechanism.Experimental results show that the accuracy of the model is 95.18% with significantly reduced number of parameters,which achieves the optimal result in the domain.Further,about 98% reduction in the number of parameters has been achieved with slightly lower accuracy than the existing models.

Path Planning of Hydrographic Mapping UAV Based on Multi-constraint Petri Net

YAO Xi, CHEN Yande

Computer Science. 2023, 50 (6A): 220700079-7. doi:10.11896/jsjkx.220700079

Abstract

PDF(2381KB) ( 252 )

References | Related Articles | Metrics

With the development of surveying and mapping technology,the application of unmanned aerial vehicle(UAV) in hydraulic engineering has been deepened.The use of UAV has revolutionized the working mode of surveying and improved working efficiency.Due to the reasons such as unmanned aerial vehicle,limited duration of flight and restriction of aerial survey picture splice,it is necessary to carry out scientific route planning.It can meet the requirements of flight safety,validity of survey data and operation efficiency.In view of this,a path planning method based on multi-constraint Petri net is proposed.The problem scene is described.The multi-constraint Petri net is defined and the method of reachability analysis is given.The multi-constraint Petri net model for path planning is constructed.The optimal route planning scheme is obtained based on the reachability marking diagram.Experimental results show that this method has superiority in UAV path planning scheme optimization.

Review of Research on Denoising Algorithms of ECG Signal

HOU Yanrong, LIU Ruixia, SHU Minglei, CHEN Changfang, SHAN Ke

Computer Science. 2023, 50 (6A): 220300094-11. doi:10.11896/jsjkx.220300094

Abstract

PDF(2774KB) ( 643 )

References | Related Articles | Metrics

One of the most common signal processing problems with the electrocardiogram(ECG),an important indicator for identifying cardiac abnormalities in humans,is the elimination of unwanted noise.These noises can distort the clean signal,which can affect the diagnosis and analysis of the human heart.This paper reviews five different frameworks of ECG signal denoising techniques and the latest research results within these frameworks,and finally summarizes the best noise reduction models in last five years and compares them by performance evaluation criteria such as signal-to-noise ratio.The comparison shows that the deep learning models show good performance in ECG denoising,whether based on single noise or composite noise.Finally,the problems with the current denoising model are discussed and an outlook on the next step of the research is given.

Review of 3D Target Detection Methods Based on LiDAR Point Clouds

QIN Jing, WANG Weibin, ZOU Qijie, WANG Zumin, JI Changqing

Computer Science. 2023, 50 (6A): 220400214-7. doi:10.11896/jsjkx.220400214

Abstract

PDF(2652KB) ( 411 )

References | Related Articles | Metrics

In recent years,3D target detection using LiDAR point cloud is a research hotspot in the field of computer vision and has attracted much attention in the field of autonomous driving.Compared with 2D,3D combines depth information to better reflect the characteristics of the real world,to effectively solve practical problems such as path planning,motion prediction,target detection,and other aspects.This paper introduces the development background of 3D target detection,summarizes the flow of 3D target detection framework based on LiDAR point cloud data,compares several common data sets containing point cloud information,and classifies the main research methods.The performance and limitations of different methods are analyzed and compared.Finally,the current technical difficulties are summarized and the future development prospects of this field are forecasted.

Semi-supervised Semantic Segmentation for High-resolution Remote Sensing Images Based on DataFusion

GU Yuhang, HAO Jie, CHEN Bing

Computer Science. 2023, 50 (6A): 220500001-6. doi:10.11896/jsjkx.220500001

Abstract

PDF(3623KB) ( 285 )

References | Related Articles | Metrics

Due to the need for pixel-wise annotation,semantic segmentation usually requires higher labor costs than tasks such as classification and object recognition.Especially in land classification based on high-resolution remote sensing images,complex backgrounds and dense targets make semantic annotation intolerably expensive,which seriously limits the practicability of semantic segmentation algorithms.In addition,although traditional semi/weak supervised learning methods can effectively reduce trai-ning costs,it is difficult to have high application value for the low quality of the segmentation results.In order to solve the above two pain points,this paper proposes a new semi-supervised semantic segmentation model using a self-correcting fusion strategy.By introducing data fusion technology and self-correction mechanism,the dependence of the segmentation model on pixel-wise annotation can be effectively reduced.Our method obtains mean F1-scores of 86.5% and 81.7% on Potsdam and Vaihingen datasets with only 15% pixel-wise annotation.Experimental results show that the proposed model can greatly reduce the cost of training process,and achieve high-quality segmentation results comparable to fully-supervised prediction.

Overview of Application of Virtual Reality in Sports Simulation:New Developments Since 2003

JI Qingge, CHEN Haodong, HE Suishen, ZHU Yonglin, ZHU Jiefu, ZHANG Huankai

Computer Science. 2023, 50 (6A): 220500168-10. doi:10.11896/jsjkx.220500168

Abstract

PDF(2747KB) ( 341 )

References | Related Articles | Metrics

Many sports are difficult to carry out due to the limitations of venues,weather,economy and other factors.With the rapid development of computer technology,virtual reality technology is widely used in sports,aiming to break through the above limitations.This paper introduces the technology of the combination of virtual reality and sports simulation from 2003 to now,and classifies it into competitive sports simulation,entertainment sports simulation,medical sports simulation and sports related scene simulation from the point of view of sports type.From the examples of sports simulation in recent years,it can be found that the sports simulation based on VR technology is not limited to the field of competitive sports,and is constantly developing in the direction of mass entertainment and interdisciplinary.VR based entertainment sports simulation and medical sports simulation are more mature,which can bring the public a VR experience closer to daily life.Although sports related scene simulation is a relatively new field,due to its commercial characteristics,the applications of virtual advertising and event system develop rapidly.Finally,a new outlook on the future of virtual reality technology in sports simulation is presented.

Remote Sensing Image Classification Based on Improved ResNeXt Network Structure

YANG Xing, SONG Lingling, WANG Shihui

Computer Science. 2023, 50 (6A): 220100158-6. doi:10.11896/jsjkx.220100158

Abstract

PDF(3294KB) ( 325 )

References | Related Articles | Metrics

Remote sensing image classification is one of the key directions of remote sensing image information processing,and its classification accuracy greatly limits the overall development of remote sensing technology.Traditional machine learning algorithms and model structures have the disadvantages that they cannot quickly extract feature maps from remote sensing images,and the classification results are not accurate enough.Aiming at this problem,an improved model based on the ResNeXt network model combined with the attention mechanism is proposed to replace the fully connected layer model with the optimized SVM(support vector machine) algorithm.Firstly,it introduces the attention mechanism in computer vision,assigns different weights to different features,improves the ability to extract effective information for the classification part of the image,then combines the ResNeXt network,and finally replaces the end of the convolutional neural network with the optimized SVM algorithm.The fully connected layer is used to improve the classification effect,and at the same time optimize the network performance without increasing the hyperparameters of the model as a whole.Experimental results of the network model on the data set AID show that the improved network model has a significant improvement in the ability to extract deep features,and the optimized network mo-del has a better classification effect for multi-classification tasks.

Remote Sensing Image Change Detection of Construction Land Based on Siamese AttentionNetwork

LI Tao, WANG Hairui

Computer Science. 2023, 50 (6A): 220500040-5. doi:10.11896/jsjkx.220500040

Abstract

PDF(2730KB) ( 262 )

References | Related Articles | Metrics

Aiming at the problems of under segmentation or over segmentation and rough edge segmentation in the process of urban construction land change detection using traditional semantic segmentation network,this paper proposes a high-resolution remote sensing image change detection method based on twin attention network.In the coding part,twin neural network is used for feature acquisition to retain more image features of different phases.In the deep coding stage,the hole convolution feature pyramid is introduced to realize the extraction and fusion of multi-scale features and increase the receptive field of the network.In the decoding part,the attention mechanism CBAM is used to highlight the useful features and enhance the useful information to improve the accuracy of edge segmentation.Finally,experiment is carried out on the data set of land use change in Loudi City.Experiment shows that the accuracy rate of this method is 92.56%,the accuracy rate is 89.15%,the recall rate is 85.61%,the IOU is 77.53%,the Miou is 83.76%,the F1 score is 87.34%,and the kappa coefficient is 31.42% on the land use change detection data set of Loudi city.The performance index is better than FCN network,u-net network and CBAM u-net network.Experimental results show that this method can effectively solve the problems of under segmentation or over segmentation of change detection results and rough edge segmentation.

Robot Visual Inertial Optical Measurement Method Based on Improved PL-VIO

WANG Haifang, LI Mingfei, LI Guangyu, CUI Yangyang

Computer Science. 2023, 50 (6A): 220400171-5. doi:10.11896/jsjkx.220400171

Abstract

PDF(3220KB) ( 225 )

References | Related Articles | Metrics

An improved point-line vision inertial measurement algorithm(PL-VIO) is proposed to solve the problem of numerous inertial measurement and visual track identification and imprecise image pose and edge accuracy in map object pose recognition.In the front end of vision,the sub-pixel edge extraction method is used to iterate and improve the accuracy of image edge corners,and the edge constraints are applied to the extracted corners to prevent sub-pixel edge detection from crossing the boundary.In order to improve the extraction accuracy and reduce the repeated extraction of line features at the visual backend,the line features and point features extracted by LSD are extracted and optimized.After SFM,the extracted line features are combined and redundant lines are deleted.Experiments are carried out using EuRoc data set based on ROS platform,and the obtained experimental data are imported into Evo.Evo is used to analyze and plot the experimental data,and the error parameters are evaluated.The overall reduction of error parameters in the experimental results verified the superiority and accuracy of the improved PL-VIO algorithm.

Endoscopic Image Enhancement Algorithm Based on Luminance Correction and Fusion Channel Prior

AN Ziheng, XU Chao, FENG Bo, HAN Jubao

Computer Science. 2023, 50 (6A): 220300265-7. doi:10.11896/jsjkx.220300265

Abstract

PDF(3367KB) ( 302 )

References | Related Articles | Metrics

In order to solve the problems of uneven illumination,blurred blood vessels in submucosal tissue,and low contrast in medical endoscopic images,a novel endoscopic image enhancement algorithm is proposed in this paper.The method is divided into two parts.The first part uses a method based on quadrant clipping histogram gamma correction to achieve brightness enhancement.In this part,the histogram of the brightness channel is first divided into quadrant clipping to obtain a smooth cumulative distribution function(CDF),and then use the truncated CDF way to control the size of the gamma parameter.The second part enhances the contrast and sharpness of the image based on the fusion channel prior.This part first uses discrete wavelet transform to fuse the green channel and red channel of the image to obtain a layer with rich details,which is used to generate the initial transmission map of the Image Formation Model(IFM).After that,the initial transmission image is corrected by the proposed ideal function model,and a clear image is obtained.finally,the contrast enhancement of tissue and blood vessels is realized by combining with CLAHE.The method and several other existing methods are analyzed subjectively and objectively on the MEDS dataset built by the laboratory.The results show that the proposed method can improve the contrast of blood vessels and tissues while avoiding excessive image enhancement.

Pathological Image Super-resolution Reconstruction Based on Sparse Coding Non-local AttentionDual Network

LIANG Meiyan, ZHANG Yu, LIANG Jianan, CHEN Qinghui, WANG Ru, WANG Lin

Computer Science. 2023, 50 (6A): 220700016-8. doi:10.11896/jsjkx.220700016

Abstract

PDF(3820KB) ( 261 )

References | Related Articles | Metrics

High-resolution pathological images are the objective criteria for high-precision disease diagnosis,which have great significance in the field of precision medicine.However,it is difficult to obtain high-resolution pathological images in real time,due to the limited resolution and constrained scanning time of hardware devices.Classical image super-resolution reconstruction algorithm is not suitable for pathological images because the parameters of the model are difficult to estimate,resulting in blurred and unrealistic image details after super-resolution reconstruction.Therefore,sparse-coding non-local attention dual network(SNADN) is proposed,which uses Gaussian constraints,hash coding and parameter sharing strategy in the dual branches,to achieve high accuracy and high efficiency super-resolution reconstruction of pathological images.The PSNR and SSIM of the reconstructed pathological images can reach 30.84dB and 0.914,respectively.The results show that SNADN can not only achieve accurate reconstruction of high-frequency details in pathological images,but also the lightweight sparse coding non-local attention mechanism can effectively improve the modeling efficiency.It is an effective method for super-resolution reconstruction of pathological images.

Cardiac MRI Image Segmentation Based on Faster R-CNN and U-net

HAN Junling, LI Bo, KANG Xiaodong, YANG Jingyi, LIU Hanqing, WANG Xiaotian

Computer Science. 2023, 50 (6A): 220600047-9. doi:10.11896/jsjkx.220600047

Abstract

PDF(3612KB) ( 319 )

References | Related Articles | Metrics

In order to solve the problem that the segmentation accuracy of the existing MRI neural network is reduced due to the diversity of input image information.An MRI image segmentation method using Faster R-CNN and U-net mechanism is proposed.Selecting the public cardiac MRI segmentation challenge datasets ACDC and SCD,cleaning and modifing the format of the dataset and sending them to the subsequent neural network.First,Faster R-CNN is applied to target image detection to preprocess the original input image and remove redundant background information.Second,performing U-net segmentation on the preprocessed images.At the same time,in order to test whether the performance and accuracy of the segmentation network are improved after the introduction of Faster R-CNN,ablation experiments and comparison experiments are conducted.In the ablation experiment,the detection and cropping module in the U-net segmentation network is removed,and the U-net and its improved network are selected to do a set of ablation experiments respectively.Experiments show that the average intersection ratio and Dice coefficient of the new method is 0.89 and 0.94 on the ACDC dataset,respectively,which is 7.3% and 5% higher.On the SCD dataset,it is 0.96 and 0.98,which is 5% and 3% higher,respectively.Automatic preprocessing and segmentation of MRI images is achieved.

Fabric Defect Detection Algorithm Based on Improved Cascade R-CNN

BAI Mingli, WANG Mingwen

Computer Science. 2023, 50 (6A): 220300224-6. doi:10.11896/jsjkx.220300224

Abstract

PDF(3656KB) ( 277 )

References | Related Articles | Metrics

Automatic detection of fabric defects is a difficult problem in textile industry.To solve the problem that the current fabric defect detection algorithms have unsatisfactory detection effect on samples with large scale and aspect ratio changes and numerous small targets,a fabric defect detection algorithm based on improved Cascade R-CNN network is proposed.The main improvements are as follows.Firstly,deformable convolution is incorporated into the feature extraction network ResNet-50 to extract more shape and scale features of defects adaptively.Secondly,balanced feature pyramid is introduced in the feature pyramid network before sampling to narrow the semantic gap between each feature layer before feature fusion and get more expressive multi-scale features.Then,more suitable initial anchor boxes are redesigned according to the scale and aspect ratio of defects.Finally,GIoU Loss with scale invariance is used as the regression loss of cascade detector to obtain more accurate defect prediction boundary boxes.Experimental results show that compared with the algorithm based on Cascade R-CNN,the improved Cascade R-CNN algorithm significantly improves the average precision of fabric defect detection.

Image Retrieval Based on Independent Attention Mechanism

ZHANG Shunyao, LI Huawang, ZHANG Yonghe, WANG Xinyu, DING Guopeng

Computer Science. 2023, 50 (6A): 220300092-6. doi:10.11896/jsjkx.220300092

Abstract

PDF(3935KB) ( 363 )

References | Related Articles | Metrics

In recent years,deep learning methods has taken a dominant position in the field of content-based image retrieval.To improve features extracted by off-the-shelf backbones and enable the network produce more discriminative image descriptors,the attention module ICSA(independent channel-wise and spatial attention),which is independent with features input into the mo-dule,is proposed.Attention weights of the proposed module keeps the same when input features change,while attention weights are usually computed with input features in other attention mechanisms,which is a main difference between ICSA and other attention modules.This feature also enables the module to be quite small(only 6.7kB,5.2% the size of SENet,2.6% of the size of CBAM) and relatively fast(similar with SENet in speed and 14.9% the time of CBAM).The attention of ICSA is divided as two parts:channel-wise and spatial attention,and they store the weights along orthogonal directions.Experiments on Pittsburgh shows that ICSA made improvement from 0.1% to 2.4% at Recall@1 when with different backbones.

Two-stage Method for Restoration of Heritage Images Based on Muti-scale Attention Mechanism

LIU Haowei, YAO Jingchi, LIU Bo, BI Xiuli, XIAO Bin

Computer Science. 2023, 50 (6A): 220600129-8. doi:10.11896/jsjkx.220600129

Abstract

PDF(5256KB) ( 360 )

References | Related Articles | Metrics

The use of virtual technology is important for the restoration of relics,which are often damaged by improper preservation or physical restoration methods.Existing traditional image restoration techniques and deep learning-based restoration methods are mainly suitable for images with simple structural textures,small damaged areas,or natural images with regular damage,and cannot be directly applied to heritage images.Using landscape painting image restoration as an example,a two-stage method for restoration of heritage images based on a multi-scale attention mechanism is proposed in this paper to address the problems of complex structural textures,discreet colouring and small size of existing datasets of heritage images.The method firstly performs coarse restoration of the overall structure and base tones of the image based on the global attention mechanism,then performs local fine restoration of small structures and fine textures of the image using the local attention mechanism and the residual module,as well as global fine restoration of large structures and textures using the contextual attention mechanism on the result of coarse restoration to borrow information accurately at a distance.Finally,the local and global fine restoration results are fused to achieve the restoration of heritage images.The proposed method has the advantage of improving the peak signal-to-noise ratio by 3.76 dB and the structural similarity by 0.034 compared with the comparative methods on average.Both the subjective and objective analysis of the experimental results show that the method has some advantages in semantic rationality,information accuracy and visual naturalness compared with the existing methods,and has a high potential for application in the field of heritage restoration.

Study on BGA Packaging Void Rate Detection Based on Active Learning and U-Net++ Segmentation

QI Xuanlong, CHEN Hongyang, ZHAO Wenbing, ZHAO Di, GAO Jingyang

Computer Science. 2023, 50 (6A): 220200092-6. doi:10.11896/jsjkx.220200092

Abstract

PDF(2066KB) ( 245 )

References | Related Articles | Metrics

Bump void is one of the most common physical defects in BGA packaging,which may lead to electrical failures and shortened lifetime.At present,the commonly used quality inspection is based on manual check on X-ray images,which has low accuracy and high time consumption.Therefore,automated chip detection methods based on deep learning draws increasing attention in industry.This paper proposes an active learning and U-Net+＋based void rate detection network.Based on active lear-ning,we apply equidistant partition for the whole dataset.For each sub-dataset,we take training-prediction-labeling-extension as pattern to optimize U-Net+＋network.The average dice coefficient on separated model sets reaches 80.99% on test set,while the overall accuracy rate reaches 94.89%.We innovatively apply active learning in in-line defect detection,and the result shows that,it can effectively enhance the labeling standard of data and model’s division precision.

Combining Multi-focus Fusion and DSGEF Two-stage Network to Reconstruct Solar Speckle Image

JIN Yahui, JIANG Murong, LI Fuhai, YANG Lei, CHEN Junyi

Computer Science. 2023, 50 (6A): 220600182-6. doi:10.11896/jsjkx.220600182

Abstract

PDF(2718KB) ( 196 )

References | Related Articles | Metrics

Because the solar speckle image has the characteristics of low contrast,similar structure of rice grains and small diffe-rence between frames,there are some problems such as insufficient high-frequency features and unrecoverable local details when using the existing reconstruction network for single frame deblurring.In this paper,a high-resolution reconstruction method of solar speckle image is proposed by combining multi-focus fusion and building gradient enhancement and FPN two-stage network.Firstly,the block-focused image fusion algorithm is performed to compensate for high-frequency details lost in the images by utilizing the complementary characteristics of similar information between sequence images.Secondly,a two-stage reconstruction network DSGEF is constructed based on the generative adversarial network(GAN),which combines gradient branches and structural feature branches to enhance high-frequency details,uses FPN network for multi-scale feature reconstruction to improve the definition of rice grain edges.Finally,a joint training loss including adversarial loss,pixel loss and perceptual loss is introduced to guide the network to implement high-resolution reconstruction of solar speckle images.Experimental results show that,compared with existing deep learning methods,the proposed method can significantly improve the image peak signal-to-noise ratio(PSNR) and structural similarity(SSIM) indicators,and can meet the requirements of high-resolution reconstruction of solar observation images.

Metric Regularized Infrared and Visible Cross-modal Person Re-identification

WU Hanxiao, ZHAO Qianqian, ZHU Jianqing, ZENG Huanqiang, DU Jixiang, LIAO Yun

Computer Science. 2023, 50 (6A): 221100046-8. doi:10.11896/jsjkx.221100046

Abstract

PDF(3914KB) ( 266 )

References | Related Articles | Metrics

Infrared and visible cross-modal person re-identification plays an important role in improving the all-day combat capability of intelligent video surveillance systems.Existing methods usually focus on the alignment of cross-modal features,and neglect the metric alignment among multiple modalities,resulting in re-identification lacking robustness to modal changes.For that,this paper proposes a metric regularized infrared and visible cross-modal person re-identification.First,this paper designs a metric regularized loss function to constrain the difference among matching behaviors under different modal retrieval modes to improve the robustness.Second,considering that the number of infrared images is less than that of visible images in actual surveillance scenes,this paper applies the modal data proportion to modify the cross-entropy function to reduce the adverse effect of the imba-lance between different modalities.Experimental results show the superiority of the proposed method,e.g.,using visible images to retrieval infrared images,the rank-1 identification rate reaches 89.52% on the RegDB dataset.

Image Super-resolution Reconstruction Based on Structured Fusion Attention Network

YU Jiuyang, ZHANG Dean, DAI Yaonan, HU Tianhao, XIA Wenfeng

Computer Science. 2023, 50 (6A): 220600240-5. doi:10.11896/jsjkx.220600240

Abstract

PDF(2850KB) ( 276 )

References | Related Articles | Metrics

Aiming at the problems of weak feature extraction ability and complex model parameters in existing image super-resolution models,an image super-resolution reconstruction model based on structured hybrid attention network is proposed.This model can reduce the number of parameter while improving the super-resolution reconstruction effect.First,the encoder is structured to extract more image features through the difference in the number of channels.Second,the attention network hybrid reorganization is performed on the output features of the encoder to enhance the feature characteristics of the image.Finally,a residual method is used to directly mix the input shallow image features with the enhanced features to reduce the amount of network parameters.Experimental results show that under the premise of public data sets and different magnifications,the PSNR value and SSIM value of the proposed model are basically optimal,and the parameter amount of the network structure is low,which better balances the relationship between performance and parameter complexity in the process of image super-resolution reconstruction.

Target Detection Algorithm Based on Compound Scaling Deep Iterative CNN by RegressionConverging and Scaling Mixture

WANG Guogang, WU Yan, LIU Yibo

Computer Science. 2023, 50 (6A): 220500230-9. doi:10.11896/jsjkx.220500230

Abstract

PDF(4141KB) ( 215 )

References | Related Articles | Metrics

A novel algorithm named as target detection algorithm based on compound scaling deep iterative CNN by regression converging and scaling mixture is proposed to avoid the disadvantages of low robustness,label marginalization and poor convergence performance of the regression loss function in the EfficientDet algorithm.After utilizing the 2×2 scaling mixture regularization strategy to enhance the training samples,the proposed method avoids the over fitting and improves the generalization ability of the model.The convergence speed,the positioning accuracy and the CNN regression accuracy are improved,since the aspect ratio and the center distance are taken into account in the penalty items of the CIOU loss function that can predict the bounding frame coordinate and suppress the redundant boxes.The proposed method improves the label fault tolerance rate because the cross entropy loss with label smoothing for class is established after generating the label smoothing regularization distribution,which is a weighted sum of the marginal label distribution and the uniform distribution by setting the smoothing parameter.Experiments are performed on the PASCAL VOC 2007 and 2012 datasets,and the results show that while the number of the network model parameters remain unchanged,the mean average precision of the proposed algorithm reaches 88.31 %,which is 3.29% higher than that of the original network(EfficientDet-D2,84.12%).Compared with YOLOv4,YOLOv3,SSD,Faster R-CNN and Fast R-CNN,the mean average precision increases by 5.2%,10.71 %,14.01%,15.11% and 18.30 %,respectively,and the number of network model parameters is reduced by 55.94×10⁶,52.91×10⁶,16.09×10⁶,55.18×10⁶ and 53.11×10⁶,respectively.Not only the algorithm improves the detection accuracy and the F1 score,but also it takes 0.73 s to detect each test image,which meets the real-time requirements during the detecting phase.

Graph Neural Network Few Shot Image Classification Network Based on Residual and Self-attention Mechanism

LI Fan, JIA Dongli, YAO Yumin, TU Jun

Computer Science. 2023, 50 (6A): 220500104-5. doi:10.11896/jsjkx.220500104

Abstract

PDF(2357KB) ( 270 )

References | Related Articles | Metrics

Few shot learning is proposed to solve the problem of small size of data set required for model learning or high cost of data annotation in deep learning.Image classification has always been an important research content in the research field,and there may be insufficient annotation data.In view of the lack of image annotation data,researchers have put forward many solutions,one of which is to classify small sample images by using graph neural network.In order to better play the role of graph neural network in the field of small sample learning,aiming at the unstable situation of graph neural network convolution operation,residual graph convolution network is used to improve the graph neural network,and residual graph convolution network is designed to improve the stability of graph neural network.Based on the convolutional network of residual graph,the self-attention mechanism of residual graph is designed in combination with the self-attention mechanism,and the relationship between nodes is deeply mined to improve the efficiency of information transmission and improve the classification accuracy of the classification model.After testing,the training efficiency of the improved Res-GNN is improved.The classification accuracy in 5way-1shot task is 1.1% higher than that of GNN model,and 1.42% higher than that of GNN model in 5way-5shot task.In the 5way-1shot task,the classification accuracy of ResAT-GNN is 1.62% higher than that of GNN model.

Superpixel Segmentation Iterative Algorithm Based on Ball-k-means Clustering

LIU Yao, GUAN Lihe

Computer Science. 2023, 50 (6A): 220600114-7. doi:10.11896/jsjkx.220600114

Abstract

PDF(2730KB) ( 282 )

References | Related Articles | Metrics

Considering the problem of superpixel segmentation,this paper propose an iterative algorithm of superpixel segmentation based on Ball-k-means clustering to further improve the edge fit of superpixels.Firstly,the superpixels are regarded as five-dimensional hyperspheres,and the image is evenly segmented to obtain the initial superpixels.Secondly,the neighbor superpixels are searched according to the radius and distance between the centers of adjacent superpixels.Then,using the distances between the superpixels and their neighbor superpixel centers,the superpixels are divided into a stable region and multiple ring active regions.Finally,the pixels in each annular active area are divided into the nearest neighbor superpixel only according to their distance from the center of some neighbor superpixels,so as to realize the superpixel segmentation iteratively.In order to reduce the distance calculation and speed up the convergence,a judgment theorem of the relation between the nearest neighbor superpixels is given,and an adaptive partition updating strategy is designed for the superpixel class labels of pixels.Experimental comparison and analysis on BSD500 data set show that the proposed algorithm has better segmentation effect on different types of images,with higher edge fitting degree,less influence by parameters,and more stable segmentation results.

COVID-19 Instance Segmentation and Classification Network Based on CT Image Semantics

BAI Zhengyao, FAN Shenglan, LU Qianjie, ZHOU Xue

Computer Science. 2023, 50 (6A): 220600142-9. doi:10.11896/jsjkx.220600142

Abstract

PDF(3722KB) ( 256 )

References | Related Articles | Metrics

To assist clinicians in the diagnosis and treatment of COVID-19 patients,a computer-aided diagnosis network AIS-Net is proposed to classify,detect and segment COVID-19 lesions in CT images.First,the network integrates semantic and instance segmentation to improve the accuracy.Then,the two modules are designed,the information enhanced attention module(IEAM) for weighing input features and the instance segmentation monitoring module focusing on the lesions at different scales.Furthermore,the classification module with the main header and the auxiliary header discerns COVID-19 pneumonia,common pneumonia,and non-pneumonia.Finally,the Swin Transformer is introduced into the auxiliary classification to distinguish the lesions of common pneumonia and COVID-19.On the CC-CCII dataset,the mean average precision(mAP) of instance segmentation is 56.53%,which is 11.77% higher than the state-of-the art(SOTA).Dice coefficient,sensitivity and specificity is 80%,85.1% and 99.3% respectively,which is 4.7%,3.7% and 1.2% higher than the SOTA.The overall classification accuracy is 99.07% on the COVIDX-CT dataset,0.92% higher than the SOTA.AIS-Net can effectively diagnose COVID-19 patients through CT images,and segment and detect the lesion sites.

Maximum Overlap Single Target Tracking Algorithm Based on Attention Mechanism

SUN Kaiwei, WANG Zhihao, LIU Hu, RAN Xue

Computer Science. 2023, 50 (6A): 220400023-5. doi:10.11896/jsjkx.220400023

Abstract

PDF(3025KB) ( 204 )

References | Related Articles | Metrics

With the development of artificial intelligence,deep learning has attracted extensive attention in the research of computer vision.In the field of single target tracking,the single target tracking algorithm based on deep learning has been studied.The algorithm complexity of deep learning algorithm is relatively high.The complete segmentation of target classification and target state estimation is conducive to the in-depth discussion of each task.However,the current single target tracking algorithm can not deal with the complex tracking environment well.When the model encounters the complex tracking environment,it often tracks a certain area of the background or tracks the surrounding similar targets.In order to solve the above problems.In this paper,a method is proposed:different attention mechanisms are added to the task of target classification and target state estimation respectively,so that the model can better deal with background confusion and occlusion of similar targets.In order to verify the effectiveness of the above methods,this paper has done a lot of comparative experiments on multiple datasets,and compared with the previous single target tracking algorithm based on deep learning.The proposed algorithm improves 3.1% in the EAO index and 2.3% in the Robustness index.It shows the effectiveness and progressiveness of this method.

Pavement Crack Detection Based on Attention Mechanism and Deformable Convolution

LONG Tao, DONG Anguo, LIU Laijun

Computer Science. 2023, 50 (6A): 220300214-6. doi:10.11896/jsjkx.220300214

Abstract

PDF(4060KB) ( 298 )

References | Related Articles | Metrics

Aiming at the pavement crack detection problem under complex background,due to the unsatisfactory detection effect of image segmentation algorithm based on deep learning,and the imbalance of pixel categories in the crack image itself,this paper proposes a pavement crack detection network based on attention mechanism and deformable convolution,which is constructed based on encoder-decoder structure.In order to solve the problem of difficult crack detection in complex background,firstly,deformable convolutional is used to improve the learning ability of linear features of cracks with different shapes.Secondly,the dense connection mechanism is used to strengthen the feature information.Then,in the decoder stage,the feature fusion of transpose convolution and bridge are adopted,and the multi-stage feature fusion is combined to improve the detection accuracy of the network.Finally,the attention module(SimAM) is introduced to pay more attention to the extraction of target features and suppress background features without increasing network parameters.Experiments are carried out on two open crack datasets to ve-rify the effectiveness of the algorithm.The experimental results show that the performance evaluation criteria of the algorithm are better than the comparison algorithms.The mean pixel accuracy and mean intersection over union of the BCrack dataset reached 92.12% and 84.79%,respectively.The mean pixel accuracy and mean intersection over union of the CFD dataset reached 91.02% and 74.75%,respectively.The average accuracy and average intersection ratio of CFD data set is 91.02% and 74.75%,respectively.The algorithm performs well in crack detection under complex background,and can be applied to pavement maintenance engineering.

Font Transfer Based on Glyph Perception and Attentive Normalization

LYU Wenrui, PU Yuanyuan, ZHAO Zhengpeng, XU Dan, QIAN Wenhua

Computer Science. 2023, 50 (6A): 220100205-6. doi:10.11896/jsjkx.220100205

Abstract

PDF(4438KB) ( 277 )

References | Related Articles | Metrics

The style transfer of font is a very challenging task,and its aim is to transfer the target font to the source font through a certain mapping method,so that it can realize the conversion of fonts.Existing methods in glyph transfer are limited in robustness,it highlights the poor maintenance of the structural integrity of the generated fonts.None of these methods can get satisfactory results,especially with the presence of a huge difference among different glyph styles.To address this problem,an end-to-end font transfer network framework model is proposed,and the attentive normalization is introduced in the model to better extract the high-level semantic features of the font images,thus improving the quality of the generated images.Additionally feature fusion is performed using adaptive instance normalization for font transformation.In terms of maintaining the integrity of the glyph structure,the perception loss and context loss are designed to constrain the generation of the glyph structure.A regularization term is added to the design of the adversarial loss function to stabilize the training of GAN.To verify the validity of the model,experiment is trained and tested in multiple sets using publicly available datasets in FET-GAN,and compared with the latest methods in FET-GAN,CycleGAN and StarGANv2.It is experimentally verified that the model is able to achieve mutual transfer of fonts between a given number of font domains,and both its transfer effect and model generalization ability have some advantages compared with the latest work.

Ultrasonic Image Segmentation Based on SegFormer

YANG Jingyi, LI Fang, KANG Xiaodong, WANG Xiaotian, LIU Hanqing, HAN Junling

Computer Science. 2023, 50 (6A): 220400273-6. doi:10.11896/jsjkx.220400273

Abstract

PDF(3444KB) ( 337 )

References | Related Articles | Metrics

Ultrasonic image segmentation is not only an important part of medical image processing,but also a common technical means of clinical diagnosis.In this paper,the SegFormer network model is proposed to realize the accurate segmentation of medical ultrasound images.On the one hand,the ultrasonic label image is transformed into a single channel and processed by binarization to complete the preprocessing of the data set image;on the other hand,the pre-training model is loaded into the pre-training model to fine-tune the trained model parameters,and a random gradient descent optimizer with momentum is selected to accelerate the convergence speed and reduce the oscillation.Experimental results show that,compared with FCN,UNet and DeepLabV3,all the evaluation indexes of the proposed model are the best in the breast nodule ultrasound image data set,and the evaluation indexes of mIoU,Acc,DSC and Kappa is 81.32%,96.22%,88.91% and 77.85% respectively.The experimental results also show that the model is robust in different ultrasonic image data sets.

Electiric Bike Helment Wearing Detection Alogrithm Based on Improved YOLOv5

XIE Puxuan, CUI Jinrong, ZHAO Min

Computer Science. 2023, 50 (6A): 220500005-6. doi:10.11896/jsjkx.220500005

Abstract

PDF(2835KB) ( 549 )

References | Related Articles | Metrics

In electric vehicle traffic accidents,craniocerebral injury is the main cause of death of electric vehicle riders,and most electric vehicle riders rarely wear helmets.Therefore,it is of strong practical significance to supervise the helmet wearing situation of electric vehicle riders by combining the target detection algorithm with road cameras.For the current problems of electric vehicle helmet wearing detection:the high leakage rate of targets blocking each other,and the high leakage rate of smaller targets,this paper proposes an improved YOLOv5 target detection algorithm to achieve the detection of electric vehicle helmet wearing.The method first adds the channel attention mechanism ECA-Net to the YOLOv5 network,so that the model can detect the target features,thus improving the model detection performance;the Bi-FPN weighted bidirectional feature pyramid module is used toachieve a balance of the importance of features at different levels,which is conducive to improving the small target miss detection problem;the loss function of Alpha-CIoU Loss is used to improve the accuracy of model localization.Experimental results show that the detection accuracy of the method is higher than other models for the helmet wearing situation of electric vehicle riders in all three scenarios,with an average accuracy of 95.8%,which is higher than the original network detection accuracy,and achieves high accuracy detection of electric vehicle helmet wearing situation.

Low-resource Thai Speech Synthesis Based on Alternate Training and Pre-training

CAI Haoran, YANG Jian, YANG Lin, LIU Cong

Computer Science. 2023, 50 (6A): 220800127-5. doi:10.11896/jsjkx.220800127

Abstract

PDF(2183KB) ( 276 )

References | Related Articles | Metrics

As a language spoken by tens of millions of people,Thai is widely used.In the late 1990s,some scholars carried out research on Thai speech synthesis.In recent years,end-to-end speech synthesis systems based on deep neural networks and trained with large-scale high-quality “text-audio” data have been able to synthesize high-quality speech.At present,Chinese,English and other common languages have massive speech synthesis databases.However,the “text-audio” database available for Thai as a non-common language is often small in scale.Under the condition of low resources,this paper aims to improve the quality of Thai speech synthesis,selects the end-to-end speech synthesis model Tacotorn2 as the baseline model,studies the alternate training method and pre-training method,and studies the effect of different text embedding methods on the effect of Thai speech synthesis.Then,the speech synthesized by the six models designed in this paper is evaluated from the attention alignment map and the MOS score.Experimental results show that the system using the method of “vowel consonant embedding+pre-training+alternate training” has the best speech synthesis quality,and the MOS score of the synthesized speech can reach 3.95,which is significantly better than the baseline system’s 1.71.

Defect Detection of Transmission Line Bolt Based on Region Attention Mechanism andMulti-scale Feature Fusion

WU Liuchen, ZHANG Hui, LIU Jiaxuan, ZHAO Chenyang

Computer Science. 2023, 50 (6A): 220200096-7. doi:10.11896/jsjkx.220200096

Abstract

PDF(5560KB) ( 296 )

References | Related Articles | Metrics

Bolts play a role in fixing the connection between lines in transmission lines.Once loose or detached,it may cause po-wer transmission failures and cause large-scale power outages.Obviously,regular inspection of bolts in transmission lines is essential to ensure the safety and stability of the entire power system.Most of the existing detection methods are based on deep convolutional neural networks.However,the unobvious features and small size of the bolts pose a challenge to the detection work.Aiming at the above problems,this paper proposes a bolt defect detection method for transmission lines based on region attention mechanism and multi-scale feature fusion.Firstly,a region attention module suitable for object detection is proposed,which is embedded in the residual block of ResNet50 to enhance the network’s feature extraction for bolts.Secondly,based on the feature pyramid networks(FPN),a bottom-up path is extended,and shallow features are fully utilized to improve the detection accuracy of small objects.Finally,in order to alleviate the imbalance between samples,the PrIme Sample Attention(PISA) soft sample sampling strategy is introduced.Experimental results show that the proposed method achieves a mean average precision(mAP) of 74.3% and an average recall(AR) of 86.4% with a detection speed of 8.2 FPS when detecting transmission line bolts.Compared with other detection networks,the proposed method improves the detection accuracy of bolt defects without sacrificing too much detection speed.

Motion Contrast Enhancement-based Crowd Motion Segmentation Method

ZHANG Xinfeng, NI Qili, CHEN Shuhan, YANG Baoqing, LI Bin

Computer Science. 2023, 50 (6A): 211200205-7. doi:10.11896/jsjkx.211200205

Abstract

PDF(4363KB) ( 266 )

References | Related Articles | Metrics

In surveillance videos of public places,the movement states of the crowds are various and complex,and it is difficult to analyze the movement state of the whole crowd through detecting or segmenting every individual.Therefore,it is an effective way to understand and analyze the movement state of the crowd by dividing the crowd into areas with basically the same movement state.Supervised crowd motion segmentation methods require pixel-level training sets with high labeling costs,and thus unsupervised clustering methods are more promising for crowd motion segmentation.However,since the local features describing crowd movements usually change gradually,leading to the unsupervised methods based on clustering algorithm need to choose different parameters for different crowd scenarios,it is difficult to adapt to a variety of different application scenarios.To this end,this paper proposes a motion contrast improvement-based crowd motion segmentation method.The method is an unsupervised model that first enhances the contrast of different motion states based on the distribution law of movement and noise in the motion field,and then combines the adaptive threshold segmentation algorithm and the marker watershed algorithm to extract the essentially consistent region for each motion state,avoiding the difficulty of parameter selection for unsupervised clustering methods.Based on the results of crowd motion segmentation,this paper presents an energy model to describe the stability of crowd movement.The energy model can enable early warning of abnormal crowd motion state by deducing the change process of the whole crowd motion state.Experiments are conducted on crowd motion segmentation in different types of complex crowd motion state scenes.Experimental results verify the effectiveness and segmentation accuracy of the motion contrast enhancement-based crowd motion segmentation method and the validity of the proposed energy model.

Multi-path Semantic Segmentation Based on Edge Optimization and Global Modeling

CHEN Qiaosong, ZHANG Yu, PU Liu, TAN Chongchong, DENG Xin, WANG Jin, SUN Kaiwei, OUYANG Weihua

Computer Science. 2023, 50 (6A): 220700137-7. doi:10.11896/jsjkx.220700137

Abstract

PDF(3549KB) ( 242 )

References | Related Articles | Metrics

In the current semantic segmentation convolutional network,the spatial and detail information is gradually lost with the deepening of the convolutional layer,resulting in inaccurate segmentation of boundary parts and small objects.Meanwhile,the local feature capability of convolution restricts the network's ability to obtain effective global modeling,resulting in confusion of internal segmentation of objects.Aiming at these problems,a multi-path semantic segmentation algorithm based on edge optimization and global modeling is designed.The algorithm proposes a multi-path adjacent dislocation fusion network.Four branches of different resolutions are interlaced and fused adjacently.In order to reduce the loss of spatial information and detail information,the detail information between the adjacent four different resolution paths is fused,and the semantic information is fused between the tail of the high-resolution path and the header of the low-resolution path.The adaptive edge feature module is proposed to obtain edge features which are integrated into the middle layer and depth supervision layer of the network to enhance the expressive ability of edge features and the segmentation effect of small objects.The Transformer global feature module is proposed,which uses different convolutions for downsampling operations to reduce the length of self-attention sequences and fuse channel information and self-attention information to obtain effective high-level semantic global information.Experimental results show that the mIoU value on the CamVid test set reaches 76.2%,and the mIoU value on the Cityscapes validation set reaches 79.1%.

Person Re-identification Method Based on Progressive Attention Pyramid

ZHANG Shuaiyu, PENG Li, DAI Feifei

Computer Science. 2023, 50 (6A): 220200084-8. doi:10.11896/jsjkx.220200084

Abstract

PDF(3632KB) ( 231 )

References | Related Articles | Metrics

Aiming at the problem that the existing person re-identification algorithms do not fully extract person features,resulting in low accuracy of the algorithm in scenes such as person occlusion and posture change,a person re-identification method based on progressive attention pyramid is proposed.This method designs a progressive feature pyramid structure based on the attention mechanism,embeds the channel and spatial attention modules into the feature pyramid structure,and applies them to the channel and spatial dimensions of the feature.Channel attention pyramid aggregates the noteworthy features in different channel dimensions at each level of the backbone network,and the spatial attention pyramid extracts the noteworthy features in different spatial dimensions.Each level of the pyramid follows the principle of “split-attend-concat”,and continuously learns the person feature map under different segmentation levels from the bottom up.Attention allows the network to fully mine key features from different channel dimensions and different spatial dimensions.At the same time,the multi-level feature alignment is realized through the cascade structure and deformable convolution,which further improves the re-identification accuracy of the model.In this paper,the method is tested on two mainstream datasets,Market-1501 and DukeMTMC-reID,respectively.Experimental results show that this method can allow the model to focus on richer person features.Compared with the baseline network,the Rank-1 index of the model increases by 3.2% and 5.8%,and the mAP index increases by 6.8% and 6.6%,respectively.

Real-time Detection of Motorcycle Lanes Based on Deep Learning

WAN Haibo, JIANG Lei, WANG Xiao

Computer Science. 2023, 50 (6A): 220200066-5. doi:10.11896/jsjkx.220200066

Abstract

PDF(3255KB) ( 281 )

References | Related Articles | Metrics

Motorcycle driving is more dangerous than other driving styles but lacks effective driving assistance systems,such as lane assist systems,obstacle detection,pre-collision system,etc.The position of the lane line when driving is often used for determining whether the motorcycle has deviated.Therefore,lane line detection is very important for developing assisted driving systems,so this paper proposes a real-time detection algorithm for motorcycle lanes based on deep learning.This paper proposes three improvements based on the Lanenet architecture:1) using the absolute position of the lane coordinates as the input feature;2)using the K-means algorithm instead of the Mean-Shift algorithm;3) removing the H-net structure.Due to the lack of public motorcycle lane data sets,the collected motorcycle lane data will be used to fit the model in this paper.Experimental results prove the effectiveness of the proposed algorithm.The detection speed can reach 47.6fps,and the cross-combination ratio can reach 0.71560.Compared with the algorithm in reference [3],the accuracy improves by 15.5% and the speed improves by 53.3%.

Study on Phased Target Detection in CT Image

WANG Xiaotian, LI Bo, KANG Xiaodong, LIU Hanqing, HAN Junling, YANG Jingyi

Computer Science. 2023, 50 (6A): 220200063-10. doi:10.11896/jsjkx.220200063

Abstract

PDF(6295KB) ( 243 )

References | Related Articles | Metrics

CT is one of the most commonly used imaging examinations in clinic,and the computer-aided diagnosis of CT images has important clinical significance.In order to optimize target detection in CT images,eight different target detection algorithms are used to detect hepatic hemangioma enhanced CT images,cerebral artery stenosis CTA images and colonic polyp CT images,and the applicability of different algorithms are compared.Firstly,the enhanced CT images of hepatic hemangioma,CTA images of cerebral artery stenosis and CT images of colonic polyps are labeled and datasets are made.Secondly,different parameter optimization algorithms are used,and AP-epoch and AP-FPS curves are drawn to compare the detection performance of different algorithms.Experimental results show that the AP,AP50,AP75 and Recall of PPYOLOv2 are optimal in different data sets,the prediction boundary box is close to the target to be tested,the prediction confidence is high,and it has good generalization ability and robustness.

Attentional Feature Fusion Approach for Siamese Network Based Object Tracking

LUO Huilan, LONG Jun, LIANG Miaomiao

Computer Science. 2023, 50 (6A): 220300237-9. doi:10.11896/jsjkx.220300237

Abstract

PDF(4818KB) ( 208 )

References | Related Articles | Metrics

In order to solve the problem of tracking drift due to target occlusion and tracking failure due to background interfe-rence during target tracking,this paper proposes a siamese network-based object tracking method with multi-feature integration,where feature fusion and attention mechanism are introduced to build multiple region-proposal-network based tracking modules.Firstly,two adjacent residual block are squeeze-and-excitation and then effectively fused,as a way to strengthen the feature information.Secondly,the parallel convolution attention module is used to filter the interference information contained in the channel information and spatial information.Finally,an algorithm similar to ensemble learning is proposed by constructing two different trackers,which receive deep semantic features and the aforementioned fused features,respectively,and weight them and train for the final object tracking.In addition,to verify the effectiveness of the algorithm,this paper also investigates the effects of diverse fusion schemes,different training weights to each tracker and the combination ways of the modules in the proposed model.Experi-mental results on the VOT2016 and VOT2018 datasets show that the proposed multi-feature integration method can effectively improve the robustness of the object tracking compared with other siamese network-based object tracking algorithms,while ensuring high accuracy.

Lightweight Target Detection Algorithm Based on Improved Yolov4-tiny

DOU Zhi, HU Chenguang, LIANG Jingyi, ZHENG Liming, LIU Guoqi

Computer Science. 2023, 50 (6A): 220700006-7. doi:10.11896/jsjkx.220700006

Abstract

PDF(3541KB) ( 356 )

References | Related Articles | Metrics

Video-oriented deep learning algorithms have high computational complexity and are difficult to meet real-time requirements,which seriously affects their applications in edge computing and real-time systems.Lightweight networks have become one of the research hotspots.Lightweight networks for large networks significantly reduce the scale of the original network parameters and improve the detection speed,but the detection accuracy is had to meet industrial needs.In view of the above problems,this paper proposes an improved lightweight target detection network,which can effectively improve the detection performance while maintaining a small parameter scale.In this paper,the vision transformer(VIT) structure is added to the YOLOv4-tiny backbone network,and the multi-head self-attention mechanism enables the network to extract deeper object features.Using the simplified Bi-FPN,the two detection channels are changed to three detection channels,and the attention mechanism is introduced in the feature map fusion node to improve the model's utilization of image features and the network’s detection accuracy for objects of different sizes.Using Ghost convolution to replace traditional convolution operations,so as to reduce network computational complexity and network parameters.Experimental results on the COCO dataset show that the improved algorithm has significantly improved the detection accuracy of the original YOLOv4-tiny network while keeping the network scale unchanged,it can simultaneously meet the requirements of edge computing and real-time systems for the lightweight and accuracy of deep networks.

Dual Gating-Residual Feature Fusion for Image-Text Cross-modal Retrieval

ZHANG Changfan, MA Yuanyuan, LIU Jianhua, HE Jing

Computer Science. 2023, 50 (6A): 220700030-7. doi:10.11896/jsjkx.220700030

Abstract

PDF(2834KB) ( 207 )

References | Related Articles | Metrics

Due to the rapid development of the Internet and social media,cross-modal retrieval has attracted extensive attention.The purpose of cross-modal retrieval is to achieve flexible retrieval of different modalities.The heterogeneity gap between diffe-rent modal suggests that the similarity of different modal features cannot be calculated directly,making it difficult to improve the accuracy of cross-modal retrieval.This paper proposes an image-text cross-modal retrieval method for dual gating-residual feature fusion(DGRFF),to narrow the heterogeneity gap between the image and text.By designing gating features and residual features to fusion the features of image modality and text modality,this method can gain more effective feature information from the opposite modality,making semantic feature information more comprehensive.At the same time,the adversarial loss is adopted to align the feature distribution of the two modalities,to maintain the modality invariance of the fusion feature and obtain a more recogni-zable feature representation in the public potential space.Finally,the model is trained by combining label prediction loss,cross-modal similarity loss and adversarial loss.Experiments on Wikipedia and Pascal Sentence datasets show that DGRFF performs well on cross-modal retrieval tasks.

Study on Safety Warning Method of Driver’s Blind Area Based on Machine Vision

WANG Wei, BAI Long, MA Huanchang, LIU Yanheng

Computer Science. 2023, 50 (6A): 220700141-7. doi:10.11896/jsjkx.220700141

Abstract

PDF(3728KB) ( 252 )

References | Related Articles | Metrics

In order to reduce the energy consumption and safety warning cost of the driver’s observation and judgment of the left front,right front blind area and the surrounding during driving,this paper studies algorithms and technologies related to pedes-trian safety automatic detection and ranging in the driver’s blind area,and proposes a driver’s blind area safety warning scheme based on machine vision.Firstly,based on the driver’s actual driving perspective,through the study on the image pedestrian recognition features,the multifeature fusion blind area pedestrian satety detection method is designed,the feature histogram of the fusion scheme is obtained for the positive and negative classification of automatic pedestrian detection,and GPU is introduced to accelerate the processing efficiency of data sharing transactions.Secondly,based on the interpolation measurement method and monocular ranging principle,the laboratory ranging is carried out,some pixels are calibrated and different location points are calibrated in combination with the actual scene,so as to improve the measurement accuracy and calculation speed under the fixed angle scene,and optimize the monocular camera ranging method under the vehicle video scene.the pedestrians on the left and right sides of the car are automatically identified and ranging,and the optimal reminder distance is calculated according to the driver’s reaction time and the braking distance of the car,and the driver is reminded appropriately to reduce the probability of accidents.Experimental results show that the scheme can effectively identify pedestrians and measure the distance,ensure the real-time reminder to drivers,and has low economic cost and good practicability.

Fusion Multi-feature Fuzzy Model for Target Recognition and Its Application

RUAN Wang, HAO Guosheng, WANG Xia, HU Xiaoting, YANG Zihao

Computer Science. 2023, 50 (6A): 220100138-7. doi:10.11896/jsjkx.220100138

Abstract

PDF(3395KB) ( 232 )

References | Related Articles | Metrics

In natural recognition scenes,image features are often characterized by complexity,diversity and fuzziness,and lack of consideration of the relationship between features when using multiple features for image recognition,a target recognition fuzzy model integrating multiple image features is proposed.Firstly,the image feature is extracted,the value of the feature is taken as the fuzzy set of the model,and the corresponding membership function is given.Secondly,the evaluation index of the model is gi-ven,and the feasibility of the model is demonstrated according to the index.Thirdly,particle swarm optimization algorithm is used to optimize the parameters of membership function of image features.Finally,the target recognition algorithm based on feature fusion fuzzy model is proposed,which is applied to filling-mark recognition and the hot rolled strip surface defect recognition.Experimental results show that the designed model performs well under the evaluation index,and the algorithm significantly improves the accuracy and robustness of target recognition and the rationality of feature fusion.

Cross-dataset Learning Combining Multi-object Tracking and Human Pose Estimation

ZENG Zehua, LUO Huilan

Computer Science. 2023, 50 (6A): 220400199-7. doi:10.11896/jsjkx.220400199

Abstract

PDF(4068KB) ( 258 )

References | Related Articles | Metrics

In recent years,multi-object tracking has gained significant progress,especially for pedestrians.By performing joint pose estimation on pedestrians,it is possible to improve the motion prediction of pedestrians by multi-object tracking algorithms,while providing more information for higher-order tasks such as autonomous driving.However,in the current multi-object tra-cking dataset containing human pose estimation labels,the video length is short and the targets are sparse,limits the research of multi-object tracking.In the paper,cross-dataset learning is performed using the multi-object tracking dataset MOT17 and the multi-human pose estimation dataset COCO with more pedestrians.The performance of the multi-object tracking algorithm under joint human pose estimation is effectively improved based on a round-robin training strategy.The use of simultaneous polarized self-attention down-sampling and attention up-sampling enhances the human pose estimation performance of the algorithm while improving the algorithm training speed.

Multimodal MRI Brain Tumor Segmentation Based on Multi-encoder Architecture

DAI Tianhong, SONG Jieqi

Computer Science. 2023, 50 (6A): 220200108-6. doi:10.11896/jsjkx.220200108

Abstract

PDF(3214KB) ( 309 )

References | Related Articles | Metrics

Glioma is a primary tumor originating from glial cells in the brain,accounting for about 45% of all intracranial tumors.Accurate segmentation of brain tumor in magnetic resonance imaging(MRI) images is of great clinical significance.In this paper,an automatic brain tumor segmentation method based on multi-encoder architecture is proposed.The model adopts a U-shaped network structure which expands the single contracting path into multiple paths to deeply exploit semantic information of diffe-rent modalities.In order to obtain the multiscale features of images,an inception module combined with dilated convolution is designed as the basic convolutional layer;a lightweight attention mechanism known as efficient channel attention(ECA) block is then introduced into the bottleneck layer and the decoder,so that the model pays more attention to the segmentation-related information and ignores the redundancy of the channel dimension,thereby further improving the segmentation results.Using the Brain Tumor Segmentation Challenge 2018(BraTS 2018) dataset for verification,the proposed model gets average Dice coefficientvalues of 0.880,0.784,and 0.757 for the whole tumor,tumor core and enhancing tumor respectively.Experiment results show that the proposed method achieves accurate and effective multimodal MRI brain tumor segmentation.

Contrastive Learning for Low-light Image Enhancement

WU Jufeng, ZHAO Xungang, ZHOU Qiang, RAO Ning

Computer Science. 2023, 50 (6A): 220600171-6. doi:10.11896/jsjkx.220600171

Abstract

PDF(2715KB) ( 353 )

References | Related Articles | Metrics

Insufficient lighting in image capturing can significantly degrade the visibility and quality of images.To tackle this problem,this paper proposes a low-light image enhancement network based on contrastive learning.It is a challenging work to apply image-to-image translation task to image enhancement since the gap between low and normal light is too huge and complex for pixel-level restoration.Therefore,the proposed method takes in two steps.In order to build intermediate states that lie between the low and normal light,this paper first adopts a traditional method based on Retinex theory to initially enhance the low-light images.Second,in order to make the mappings between two domains,the subsequent enhancement is decomposed into two stages,content enhancement and degradation learning.This work is based on contrastive learning,which can enhance the representation ability of the networks,and achieves high-naturalness recovery.Extensive experimental results demonstrate the efficiency of proposed method,which can enhance the low-light image effectively with better image quality and detail restoration ability than the SOTA low-light image enhancement methods.

Improved Forest Optimization Feature Selection Algorithm for Credit Evaluation

HUANG Yuhang, SONG You, WANG Baohui

Computer Science. 2023, 50 (6A): 220600241-6. doi:10.11896/jsjkx.220600241

Abstract

PDF(1795KB) ( 192 )

References | Related Articles | Metrics

Credit evaluation is a key problem in finance,which predicts whether a user is at risk of defaulting and thus reduces bad debt losses.One of the key challenges in credit evaluation is the presence of a large number of invalid or redundant features in the dataset.To solve this problem,an improved feature selection using forest optimization algorithm(IFSFOA) is proposed.It addresses the shortcomings of the original algorithm FSFOA by using a cardinality check-based initialization strategy instead of randomized initialization in the initialization phase to improve the algorithm’s search capability;using a multi-level variation strategy in the local seeding phase to optimize the local search capability and solve the problems of restricted search space and localization of FSFOA;using a greedy selection strategy to select high-quality trees and eliminate low-quality trees when updating the candidate forest.In updating the candidate forest,we use the greedy selection strategy to select high-quality trees and eliminate low-quality trees,and converge the search dispersion process.Finally,the results show that IFSFOA outperforms FSFOA and more efficient feature selection algorithms proposed in recent years in terms of classification ability and dimension reduction ability,and validates the effectiveness of IFSFOA by setting up comparison experiments on public credit evaluation datasets covering low,medium and high dimensions.

GDLIN:A Learned Index By Gradient Descent

CHEN Shanshan, GAO Jun, MA Zhenyu

Computer Science. 2023, 50 (6A): 220600256-6. doi:10.11896/jsjkx.220600256

Abstract

PDF(2402KB) ( 230 )

References | Related Articles | Metrics

In the era of big data,data access speed is an important indicator to measure the performance of large-scale storage systems.Index is one of the main technologies to improve data access performance in database system.In recent years,learned index(LI) is proposed,which uses machine learning models instead of traditional B+-tree indexes,leverages pattern about the under-lying data distribution to train the models and optimize the indirect search of data query into the direct search of function calculation,learned index can speed up queries and reduce the size of an index.However,the fitting effect of LI is general,and it assumes that the data is static and read-only,it does not support modification operations such as insertion.This paper presents GDLIN,a novel form of a learned index,which uses gradient descent algorithm to fit the data.Gradient descent algorithm can reduce the error between the predict position and the actual position,which can reduce the cost of local research.Besides,GDLIN recursive calls the construction algorithm until only one model is created,which makes full use of keys’ distribution,and avoids the increase of the size of index with the data volume.In addition,GDLIN uses the sorted linked list to address the problem of data insertion.Experiment results demonstrate GDLIN improves the lookup throughput by 2.1× compared with the traditional B＋-trees without insertion.Besides,GDLIN improves the lookup performance by 1.08× compared with the LI when the factor of insertion is 0.5.

Temporal Hierarchical Data Management Based on Nested Intervals Scheme in Relational Database

YANG Zhenkai, CAO Yibing, ZHAO Xinke, ZHENG Jingbiao

Computer Science. 2023, 50 (6A): 220500290-5. doi:10.11896/jsjkx.220500290

Abstract

PDF(2361KB) ( 199 )

References | Related Articles | Metrics

Temporal hierarchical data is a kind of hierarchical data characterized by time dimension description and is used to model the hierarchical structure that changes over time.Compared with management methods for common hierarchical data,there are still problems in temporal hierarchical data management such as the complexity of storage scheme design and inefficiency of query and update.To solve the above problems,a temporal hierarchical data management method based on nested intervals scheme is proposed.4 types of change in hierarchical data are firstly analyzed from the perspective of the node change,based on which the storage and query capabilities of multi-version nodes in a rational database are then realized by extending the time labels.Finally,the abundantly gapped nested intervals scheme(AGNIS) is put forward to solve the problem of data insertion inefficiency in common nested intervals scheme.Experiments based on the data of Chinese administrative division and its adjustment from 2021 to 2022 show that the proposed method can implement the storage of historical hierarchical data and the query of hie-rarchical snapshot at any time,with a high efficiency in data query and update operation.

Anomaly Detection of Time-series Based on Multi-modal Feature Fusion

ZHANG Guohua, YAN Xuefeng, GUAN Donghai

Computer Science. 2023, 50 (6A): 220700094-7. doi:10.11896/jsjkx.220700094

Abstract

PDF(2243KB) ( 364 )

References | Related Articles | Metrics

Effective anomaly detection of multivariate time series is important for data mining analysis.However,most of the exi-sting detection methods are based on single modality,they cannot effectively utilize the distribution information of time series in multi-modal space.For multi-modal features,there is no effective adaptive fusion method and extraction method of spatial-temporal dependence.In this paper,a time series anomaly detection method based on multi-modal feature fusion is proposed.The multi-modal feature adaptive fusion module is established,it can adaptively fuse the multi-modal features through convolution network and soft selection mode.The spatial-temporal attention module is proposed,it is composed of temporal attention and spatial attention.It extracts spatial-temporal dependence of the multi-modal features and outputs the spatial-temporal attention vector.Then the model prediction results are obtained based on the spatial-temporal attention vector.By learning the distribution of normal samples,anomaly detection result is obtained according to the error measure between the predicted values and the real values.The proposed method is compared with other state-of-the-art models on four public datasets,and results demonstrate its effectiveness.

Tripartite Evolutionary Game Analysis of Medical Data Sharing Under Blockchain Architecture

YANG Jian, WANG Kaixuan

Computer Science. 2023, 50 (6A): 221000080-7. doi:10.11896/jsjkx.221000080

Abstract

PDF(3024KB) ( 317 )

References | Related Articles | Metrics

To promote the development of health and medical big data and actively promote the safe sharing of medical data,this paper constructs a tripartite evolutionary game model of the system manager,data provider and data demander based on the blockchain architecture.Firstly,prospect theory is combined with evolutionary game,and the parameters of traditional evolutio-nary game are improved by the prospect value function.Secondly,the possibility of game equilibrium and its evolution trend are discussed.Finally,the influence of different factors on the decision-making of each participant in medical data sharing under blockchain architecture is discussed through numerical simulation.The results show that the choice of initial strategy has a signi-ficant influence on the stability of game strategy.The evolution of the system can be accelerated by improving the regulatory bene-fits of the system manager,reducing the perceived losses of the data provider,and improving the compensation of the data demander for actively reporting non-compliance behaviors,thus enhancing the trust of all participants and promoting the formation of trust relationships.

Local Community Detection Algorithm for Attribute Networks Based on Multi-objective Particle Swarm Optimization

ZHOU Zhiqiang, ZHU Yan

Computer Science. 2023, 50 (6A): 220200015-6. doi:10.11896/jsjkx.220200015

Abstract

PDF(2651KB) ( 223 )

References | Related Articles | Metrics

Community structure is an important feature in complex networks,and the goal of local community detection is to query a community subgraph containing a set of seed nodes.Traditional local community detection algorithms usually use the topology of the network for community query,ignoring the rich node attribute information in the network.A local community detection algorithm based on multi-objective particle swarm optimization is proposed for realistic and widespread attribute networks.Firstly,attribute relationship edges are constructed based on the attribute similarity between nodes and their multi-order neighbours,and topological relationship edges are obtained by weighting the network structure based on the motif information,followed by sampling the two relationship edges around the core nodes using a random walk algorithm to obtain alternative node sets.Based on this,the alternative node sets are iteratively filtered by a multi-objective particle swarm optimization algorithm to obtain a topologically tight and attribute-homogeneous community structure.Experimental results on real datasets show that the proposed method improves the performance of local community detection.

Spatial-Temporal Graph-CoordAttention Network for Traffic Forecasting

LIU Jiansong, KANG Yan, LI Hao, WANG Tao, WANG Hailing

Computer Science. 2023, 50 (6A): 220200042-7. doi:10.11896/jsjkx.220200042

Abstract

PDF(2713KB) ( 291 )

References | Related Articles | Metrics

Traffic prediction is an important research component of urban intelligent transportation systems to make our travel more efficient and safer.Accurately predicting traffic flow remains a huge challenge due to complex temporal and spatial depen-dencies.In recent years,graph convolutional network(GCN) has shown great potential for traffic prediction,but GCN-based mo-dels tend to focus on capturing temporal and spatial dependencies,ignoring the dynamic correlation between temporal and spatial dependencies and failing to integrate them well.In addition,previous approaches use real-world static traffic networks to construct spatial adjacency matrices,which may ignore the dynamic spatial dependencies.To overcome these limitations and improve the performance of the model,a novel spatial-temporal Graph-CoordAttention network(STGCA) is proposed.Specifically,the spatial-temporal synchronization module is proposed to model the spatial-temporal dependence of the crossing relations at different moments.Then,a dynamic graph learning scheme is proposed to mine potential graph information based on data correlation between traffic flows.Compared with the existing baseline models on four publicly available datasets,STGCA exhibits excellent perfor-mance.

Study on Multibeam Sonar Elevation Data Prediction Based on Improved CNN-BP

XIONG Haojie, WEI Yi

Computer Science. 2023, 50 (6A): 220100161-4. doi:10.11896/jsjkx.220100161

Abstract

PDF(2363KB) ( 226 )

References | Related Articles | Metrics

In order to establish an accurate multibeam sonar elevation data prediction model and solve the problem of the accuracy of air-squared prediction of artificial reefs,a multibeam sonar elevation data prediction method based on a combined model of improved convolutional neural network(CNN) and BP neural network is proposed.First,the improved CNN is used to extract topographic trend features by full convolutional operation of the elevation data,and then input to BP to further explore the internal topographic trend change pattern,so as to achieve the prediction of multibeam sonar elevation data.Experiments are conducted with multibeam sonar elevation data from a submarine ranch and cross-validated using the null square volume of artificial reefs.Finally,it is compared with the traditional kriging,BP,GA-BP,and PSO-BP models.The results show that the improved CNN-BP model performs the best prediction results on multibeam sonar elevation data and artificial reef air-square volume,which verifies the feasibility,reliability and high accuracy of the proposed method.

Data Augmentation for Cardiopulmonary Exercise Time Series of Young HypertensivePatients Based on Active Barycenter

HUANG Fangwan, LU Juhong, YU Zhiyong

Computer Science. 2023, 50 (6A): 211200233-11. doi:10.11896/jsjkx.211200233

Abstract

PDF(2816KB) ( 234 )

References | Related Articles | Metrics

The gradual rise of precision medicine,such as mining cardiopulmonary exercise time series of young hypertensive patients,can understand the response of different individuals to aerobic exercise training.This helps to improve the efficiency of hypertension management plan and achieve aerobic exercise intervention more effectively.One of the bottlenecks in this study is that it is difficult to obtain sufficient sample data.To solve the above problem,this paper adopts the weighted dynamic-time-warping barycenter averaging algorithm(WDBA) to realize data augmentation of time series,focusing on the barycenter selection and the weight assignment.In this paper,the concept of active barycenter is introduced for the first time,and the selection strategies of representative barycenter and diversity barycenter are proposed to improve the effect of data augmentation.Furthermore,aiming at the shortcomings of the existing weight assignment strategies,a random strategy with decreasing distance is proposed to further improve the generalization ability of the model by avoiding the synthesis of duplicate samples.Experimental results show that the accuracy of predicting the efficacy of aerobic exercise intervention in young hypertensive patients can be further improved by considering both the barycenter selection and the weight assignment for data augmentation in the background of this study.

Review on Methods and Applications of Text Fine-grained Emotion Recognition

WANG Xiya, ZHANG Ning, CHENG Xin

Computer Science. 2023, 50 (6A): 220900137-7. doi:10.11896/jsjkx.220900137

Abstract

PDF(1927KB) ( 321 )

References | Related Articles | Metrics

Emotional information contained in massive texts on the Internet expresses public views and attitudes.How to identify and utilize emotional resources has become the focus of research in various fields.By combing the relevant theories and literature on fine-grained emotion recognition,this paper summarizes the classification methods and application scenarios,and discusses the technical challenges and practical gaps.Through analysis,it is found that fine-grained emotion recognition methods mainly include emotion lexicon,traditional machine learning and neural network learning,which are mostly used in business analysis and public opinion management.In view of the future research trend,firstly,the real-time updating of online emotion words,domain lexicon construction and semantic analysis technology can be studied.Secondly,how to improve the automatic classification of training data and build a semi-supervised learning model need to be further discussed.In addition,the research of business analysis and public opinion management can explore the integration of aspect extraction and emotion recognition.This paper summarizes and comments on emotion recognition technology and its application,which can provide a reference for the subsequent research.

Analysis of Academic Network Based on Graph OLAP

YANG Heng, ZHU Yan

Computer Science. 2023, 50 (6A): 220100237-5. doi:10.11896/jsjkx.220100237

Abstract

PDF(2894KB) ( 232 )

References | Related Articles | Metrics

In recent years,academia has gradually accumulated a large amount of data.As an effective method for representing and analyzing big data,network structure has rich dimensions and can model a large amount of data in real life.Graph online analytic processing(Graph OLAP) technology inherits the related ideas of traditional OLAP technology,allowing users to analyze multi-dimensional network data from different angles and granularities.However,most of the existing graph OLAP technologies revolve around the construction of data cubes,and most of the related operations are simple extensions of traditional OLAP technologies on graph data,and the built models have weak ability to mine the topology of the network itself.To this end,the aca-demic network constellation model and related graph OLAP analysis algorithms are firstly designed,which more clearly highlights the topological structure information of academic networks and improves the analysis ability of graph OLAP.Secondly,the corresponding materialization strategy is proposed,which effectively improves the efficiency of graph OLAP analysis.

City Traffic Flow Prediction Method Based on Dynamic Spatio-Temporal Neural Network

MENG Xiangfu, XU Ruihang

Computer Science. 2023, 50 (6A): 220600266-7. doi:10.11896/jsjkx.220600266

Abstract

PDF(2489KB) ( 245 )

References | Related Articles | Metrics

Traffic flow forecasting is of great importance to urban road planning,traffic safety issues and building smart cities.However,most existing traffic prediction models cannot capture the dynamic spatio-temporal correlation of traffic data well enough to obtain satisfactory prediction results.To address this problem,a dynamic spatio-temporal neural network-based city traffic flow prediction method is proposed to solve the traffic flow prediction problem.First,by modelling the nearest cycle dependence,daily cycle dependence and weekly cycle dependence of the traffic data,a 3D convolutional neural network is used on each component to extract the high-dimensional features of urban traffic.Then,an improved residual structure is used to capture the correlation between remote area pairs and the prediction area,and a fusion of spatial attention and temporal attention mechanisms is used to capture the dynamic correlation between traffic flows in different time periods in different areas.Finally,the outputs of the three components are weighted and fused using a parameter matrix-based approach to obtain the prediction results.Experiments on two publicly available datasets,TaxiBJ and BikeNYC,show that the proposed model outperforms the mainstream traffic forecasting models.

Dynamic Neighborhood Density Clustering Algorithm Based on DBSCAN

ZHANG Peng, LI Xiaolin, WANG Liyan

Computer Science. 2023, 50 (6A): 220400127-7. doi:10.11896/jsjkx.220400127

Abstract

PDF(3072KB) ( 265 )

References | Related Articles | Metrics

The traditional density clustering algorithms do not consider the attribute difference between data points in the clustering process,but treat all data points as homogenous points.Based on the traditional DBSCAN algorithm,a dynamic neighborhood－－density based spatial clustering of applications with noise(DN-DBSCAN) is proposed.When it is working,each point’s neighborhood radius is determined by the properties of itself,so the neighborhood radius is dynamic changing.Thus,different influences on datasets produced by points with different properties is reflected in the clustering results,making the density clustering algorithm has more practical meaning and can be more reasonable to solve practical problems.On the basis of example analysis,the DN-DBSCAN algorithm is applied to solve the urban agglomeration division problem in the Yangtze river delta,and the results of DBSCAN algorithm,OPTICS algorithm and DPC algorithm are compared and analyzed.The results show that DN-DBSCAN algorithm can reasonably classify urban agglomerations in the Yangtze river delta according to the different attributes of each city with an accuracy of 95%,which is much higher than the accuracy of 85%,85% and 88% of the other three algorithms respectively,indicating that it has a better ability to solve practical problems.

Explainable Constraint Mechanism for Modeling Temporal Sentiment Memory in Sequential Recommendation

ZHENG Lin, LIN Yixuan, ZHOU Donglin, ZHU Fuxi

Computer Science. 2023, 50 (6A): 220100066-8. doi:10.11896/jsjkx.220100066

Abstract

PDF(3165KB) ( 334 )

References | Related Articles | Metrics

In recent years,the research of sequential recommendation has developed rapidly in the recommendation field,existing methods are good at capturing users’ sequential behavior to achieve preference prediction.Among them,some advanced methods integrate users’ sentiment information to guide behavior mining.However,the advanced sentiment-based models do not consider mining relations between multi-category user sentiment sequences.Moreover,such methods cannot intuitively explain the contribution of temporal sentiments to user preferences.To make up for the above shortcomings,this paper first attempts to store temporal sentiments in the form of memory and impose constraints on them.Specifically,this research proposes two mechanisms including sentiment self-constraint and sentiment mutual-constraint to explore the associations between multiple categories of sentiments and assist user behaviors in completing sequential recommendations.Furthermore,the proposed memory framework is able to record users’ temporal sentiment attention,so that it can provide a certain degree of intuitive explanation on the basis of accurately predicting users’ temporal preference.Experimental results show that our approach outperforms existing state-of-the-art sequential methods,and it has better explainable effects than the sentiment-based sequential recommendation models.

Recommendation Model Based on Decision Tree and Improved Deep & Cross Network

KE Haiping, MAO Yijun, GU Wanrong

Computer Science. 2023, 50 (6A): 220300084-7. doi:10.11896/jsjkx.220300084

Abstract

PDF(2920KB) ( 271 )

References | Related Articles | Metrics

Feature mining is a key step to learn the interaction between users and items in the recommendation algorithm model,which is of great significance to improve the accuracy of the recommendation model.Among the existing feature mining models,although the linear logistic regression model is simple and can achieve good fitting effect,its generalization ability is weak,and the model has a large demand for feature parameters.Deep & Cross network can effectively realize the cross extraction of features,but its representation ability of data features is still insufficient.Therefore,by introducing the idea of multiple residual structure and cross coding,an improved recommendation model of Deep & Cross network based on decision tree is proposed.Firstly,it designs a tree structure based on GBDT algorithm to construct enhanced features,which strengthens the deep mining of the model on potential features.Secondly,the input parameter dimension of the embedded layer of the model is amplified and optimized.Finally,the improved Deep & Cross network recommendation model is used for recommendation prediction.This design can not only break the limitations of existing models in generalization ability,but also keep the feature parameters simple and strengthen their representation ability,so as to effectively mine the hidden associations of users and improve the accuracy of recommendation.Experimental results based on the public test data set show that the prediction effect of the proposed model is better than the exis-ting feature interaction methods.

Edge Server Placement for Energy Consumption and Load Balancing

FU Xiong, FANG Lei, WANG Junchang

Computer Science. 2023, 50 (6A): 220300088-5. doi:10.11896/jsjkx.220300088

Abstract

PDF(2456KB) ( 255 )

References | Related Articles | Metrics

At present,the traditional cloud computing mode can not meet the needs of users in low latency scenarios,so mobile edge computing comes into being.In order to make the edge servers placed in the same area have lower total energy consumption and balanced workload,and an ant colony optimization energy consumption load balancing placement algorithm ACO-ELP(ant colony optimization energy-consumption load-balancing placement) for energy consumption optimization and load balancing is proposed.Firstly,by constructing the power consumption model and load balancing model,the problem is defined,and the actual parameters are matched with the algorithm variables.In the iterative process,the ant colony algorithm is optimized.By dynamically controlling the volatilization and retention rate of pheromone,the iterative speed of the algorithm is accelerated,and the maximum and minimum value of pheromone is controlled to ensure that the algorithm can search the global optimal solution as much as possible and will not fall into the local optimal solution.Finally,the algorithm is simulated and evaluated with the data of Telecom base stations in Shanghai.The results show that compared with the basic placement algorithm,the algorithm not only reduces the number of servers and energy consumption,but also significantly reduces the load deviation.

MEC Offloading Model Based on Linear Programming Relaxation

LEI Xuemei, LIU Li, WANG Qian

Computer Science. 2023, 50 (6A): 211200229-5. doi:10.11896/jsjkx.211200229

Abstract

PDF(2473KB) ( 213 )

References | Related Articles | Metrics

In the mobile edge computing(MEC),the local device can offload tasks to the edge node near the network for computation processing,thereby reducing the delay,power consumption and overload of the client,also the computing loading core network.For the complex MEC environment of multi-type edge nodes,a three-stage computing offloading decision is modeled based on linear programming relaxation,that is CART-CRITIC-LR(CCLR) algorithm.First,the classification and regression decision tree algorithm(CART) is used to screen out the locally executed calculation tasks.Secondly,the multi-attribute decision-making algorithm(CRITIC) is used to determine the weight of the three performance indicators respectively.Then the calculation offloa-ding problem is modeled as a linear programming relaxation(LR ) to optimize the equilibrium solutions among the total delay,total energy consumption and total cost.Each offloading strategy is analyzed by comprehensively comparing the energy consumption,cost,delay.experimental results show that the CCLR algorithm achieves the shortest total delay while ensuring the multi-objective global optimization,which illustrates the effectiveness and applicability of the algorithm.

Controlled Short-distance Quantum Teleportation for Arbitrary Two-particles State in Pauli Noise Environment

XIANG Shengjian

Computer Science. 2023, 50 (6A): 220700024-4. doi:10.11896/jsjkx.220700024

Abstract

PDF(1725KB) ( 252 )

References | Related Articles | Metrics

The quantum teleportation is one of the hot topics in the quantum communication.The short-distance teleportation,different from the traditional teleportation,can further save costly quantum entanglement resource based on the restriction in the distance.However,this also increases the probability in terms of cheating for the participants.Therefore,this paper proposes another short-distance quantum teleportation for arbitrary two-particles state scheme with a controller in order to enhance the safety.At the same time,it is impossible for quantum teleportation in an ideal environment,due to a fact that the particle will be ine-vitably affected by the noise channel during the distributing period.This paper also analyzes the influence of Pauli noise,which is a widely used noise channel model,on the fidelity of a two-particles state.As a result,the different concurrence in the two-particles state can generate different fidelity in some typical Pauli noise channel.This research can provide some theoretical value in the aspect of quantum communication network and the experiment research.

Energy Efficiency Planning with SWIPT-MISO Dynamic Energy Consumption Model

XU Chenyang, XUE Liang, WANG Jinlong, ZHU Long

Computer Science. 2023, 50 (6A): 220400185-7. doi:10.11896/jsjkx.220400185

Abstract

PDF(2683KB) ( 201 )

References | Related Articles | Metrics

In simultaneous wireless information and power transfer networks,multiple antennas are usually equipped at the transmitter,which is able to serve all sensors in one-time transmission over the same frequency band.However,collecting channel state information from all sensors may cause a colossal waste of time and frequency resources.Therefore,the energy-saving beamfor-ming design with only channel distribution information at the transmitter is studied in multi-user multi-input single-output network.Under the constraints of information interruption probability,total available power and available power of authorized users,the network energy efficiency is maximized by the improved teaching-learning-based optimization algorithm.In addition,for the proposed power consumption scheme,the nonlinear energy receiving mechanism is considered,and the power-splitting energy harvesting receiver architecture is proposed to prevent the receiver from entering the saturation region,so as to improve the power receiving efficiency.The improved teaching-learning-based optimization algorithm has the advantages of whale algorithm,solves the constructed nonconvex optimization problem,and improves the convergence speed.Simulation experiments analyze the effects of outage probability,dynamic power consumption coefficient and available power at the transmitter on the system energy efficiency in the dynamic energy allocation scenario,and verify the effectiveness of the proposed algorithm.

Cloud Computing Load Prediction Method Based on Hybrid Model of CEEMDAN-ConvLSTM

ZAHO Peng, ZHOU Jiantao, ZHAO Daming

Computer Science. 2023, 50 (6A): 220300272-9. doi:10.11896/jsjkx.220300272

Abstract

PDF(3784KB) ( 227 )

References | Related Articles | Metrics

With the rapid development of cloud computing technology,more and more users choose to use cloud services,and the problem of mismatch between load requests and resource supply becomes increasingly prominent.As a result,user requests cannot be timely responded,which greatly affects the cloud service quality.Real-time prediction of load requests will help the timely supply of resources.To solve the problem of low performance of load prediction methods in the cloud computing environment,a cloud computing load prediction method based on hybrid model of complete ensemble empirical mode decomposition with adaptive noise and convolutional long short-term memory(CEEMDAN-ConvLSTM) is proposed.To begin with,the data sequence is decomposed into several sub-sequences which are easy to analyze and model.Then the convolutional long short-term memory(ConvLSTM) prediction model is used to predict the series of sub-sequences.The research idea based on multi-process parallel computation is adopted to realize multi-sequence parallel prediction and Bayesian optimization parameter tuning.Finally,the prediction values are integrated and superimposed to obtain the prediction output of the whole model,to achieve the goal of high-precision prediction of the original complex sequence data.The CEEMDAN-ConvLSTM hybrid model is verified by using the Google cluster workload data set.Experiment results show that the CEEMDAN-ConvLSTM hybrid model had a good prediction effect.Compared with the autoregressive differential moving average model(ARIMA),long short-term memory network(LSTM) and the convolutional long short-term memory(ConvLSTM),the Root Mean Square Error(RMSE) increases by 30.9%,30.1% and 22.5%,respectively.

Study on Performance of Wireless Train Communication Network Based on Wi-Fi 6

YANG Shaolong, ZHU Guosheng, PANG Xinglong, LI Xiuyuan, PAN Deng

Computer Science. 2023, 50 (6A): 220600179-5. doi:10.11896/jsjkx.220600179

Abstract

PDF(2011KB) ( 323 )

References | Related Articles | Metrics

The normal operation of modern trains is inseparable from the cooperation of mechanical and electronic systems,especially the train control and management system(TCMS) plays a key role in it.TCMS-related applications and services run on the train communication network(TCN),which is wired and often redundant,resulting in a large number of wired links interconnection,making network deployment and maintenance difficulties,and poor flexibility.This paper proposes a Wi-Fi 6-based train wireless communication networking scheme,which applies Wi-Fi 6 technology to the vehicle-level ECN network.The work includes the design of network architecture,the selection of communication data and the experimental verification in simulation environment.Experimental results show that the proposed Wi-Fi 6-based train communication QoS in terms of delay,jitter and packet loss rate meets the IEC 61375-3-4 standard,and have advantages over long term evolution(LTE).

Tag Identification for UHF RFID Systems Based on Deep Learning

YU Jiabao, YAO Junmei, XIE Ruitao, WU Kaishun, MA Junchao

Computer Science. 2023, 50 (6A): 220200151-6. doi:10.11896/jsjkx.220200151

Abstract

PDF(3305KB) ( 244 )

References | Related Articles | Metrics

The most basic function of radio frequency identification(RFID) system is tag identification.However,the current authentication system cannot detect forged or cloned tags,which leads to potential security and privacy issues.At present,there are encryption based authentication protocols and feature extraction based solutions,among which encryption based authentication protocol is incompatible with existing protocols and feature extraction based authentication protocol has limitations such as difficulty in feature extraction or short recognition distance.This paper proposes a tag identification method for UHF RFID systems to overcome the two shortenings.The core idea is to first extract signals irrelevant to the logical information of tags from the backscattered RFID signals,and then send them to the convolutional neural network for similarity matching.According to the score of similarity matching and a given threshold,the authenticity of the tag is finally recognized.In this paper,we establish an experimental system which contains an USRP N210 used as the reader of the RFID system,and contains 150 UHF commercial tags to backscatter signals from the reader.We then collects the RFID signals based on this experiment.Experimental results show that the tag recognition accuracy based on deep learning can reach more than 94%,and its equal error ratio(EER) is 0.034 when the recognition distance is up to 2m.

Cluster Head Selection Algorithm Based on Improved Butterfly Optimization Algorithm in WSN

YANG Shiyu, ZHAO Bing, PENG Yue

Computer Science. 2023, 50 (6A): 220100166-5. doi:10.11896/jsjkx.220100166

Abstract

PDF(2114KB) ( 256 )

References | Related Articles | Metrics

Aiming at the problem that the cluster head selection of clustering routing protocol in wireless sensor networks is unreasonable,resulting in uneven network load and shortened network life cycle,a cluster head selection algorithm CIBOA based on improved butterfly optimization algorithm IBOA is proposed.Firstly,based on the butterfly optimization algorithm BOA,the Circle chaotic map and nonlinear dynamic convergence factor are introduced to control the parameter,which improves the search speed and convergence accuracy of butterfly optimization algorithm,and makes the search ability stronger..In the process of cluster head selection,a new fitness function is built on the basis of the residual energy,distance among the nodes and BS and average distance between neighbor nodes.The IBOA is used for improving the random problem of cluster head selection and comprehensively select better cluster heads.Simulation results show that the cluster head selection algorithm CIBOA based on the improved butterfly optimization algorithm can comprehensively consider the factors such as node energy and distance and prolong the network lifetime.

Missing Localization Characteristic Estimation Algorithm for Passive UHF RFID Tag

ZHAO Yang, LI Lingyun, ZHAO Xiaoxia, LIU Xianhui, ZHANG Liang

Computer Science. 2023, 50 (6A): 220500055-6. doi:10.11896/jsjkx.220500055

Abstract

PDF(2732KB) ( 243 )

References | Related Articles | Metrics

For the problem of missing localization characteristics caused by the activation failure of passive UHF RFID tags,given the significant challenges in precisely modeling the channel,this paper proposes a missing localization characteristic estimation algorithm based on the linear model of signal strength Euclidean distance-space Euclidean distance to improve the localization accuracy of the scene analysis algorithms by increasing the number of characteristic dimensions.To increase the completeness of the scene matching data for nonactivated reference tags,the missing localization characteristics could be calculated directly by using the linear model.For the nonactivated target tags,the linear model is used to estimate the distance between the target tag and multiple benchmark reference tags,the least squares algorithm is used to estimate the preliminary location information of the target tag,and again the missing localization features are estimated using the linear model in reverse to complete the localization characteristics of the target tags.Experiments show that the proposed algorithm can not only effectively improve the localization accuracy of all missing target tags,but also the target tags around the missing reference tags.In addition,there is no additional hardware equipment included for this algorithm,which meets the application requirements of low-cost and high-precision.

Dynamic Energy Optimization Strategy Based on Relay Selection and Queue Stability

CHEN Che, ZHENG Yifeng, YANG Jingmin, YANG Liwei, ZHANG Wenjie

Computer Science. 2023, 50 (6A): 220100082-8. doi:10.11896/jsjkx.220100082

Abstract

PDF(2236KB) ( 280 )

References | Related Articles | Metrics

Relay-assisted mobile edge computing(MEC) has recently emerged as a promising paradigm to enhance resource utilization and data processing capability of low-power networks,such as 5G networks and Internet of things (IoT).Nevertheless,the design of relay selection and computation offloading policies to improve the energy efficiency for queue stability system remains challenging.In order to solve the energy consumption optimization problem in relay-assisted MEC system,a mixed integer nonli-near stochastic optimization model is established,with the objective of minimizing the long-term average energy consumption,subject to a task buffer stability constraint.The problem is solved by decomposing into two stages:relay selection and relay offloa-ding decision.In relay selection stage,the relay node is determined by setting a weighted parameter V₁ to minimize the weighted sum of transmission energy consumption and buffer queue length.In offloading decision stage,the stochastic optimization is converted to a deterministic optimization problem based on Lyapunov optimization method.Specifically,at each time slot,the theore-tical expressions of optimal relay calculation frequency,relay transmission power and remote calculation frequency are obtained under the constraint of task buffer queue stability.Simulation results show that the energy optimization strategy can effectively reduce the long-term average energy consumption under the constraint of buffer queue stability,and converge to the optimal solution obtained by exhaustive searching.Besides,the weight of energy consumption and waiting time can be changed by adjusting the values of parameters V₁ and V₂ in algorithm.

Overview of Blockchain Consensus Algorithms

TAN Pengliu, WANG Runshu, ZENG Wenhao, WANG Shikun, ZOU Wenshi

Computer Science. 2023, 50 (6A): 220400200-12. doi:10.11896/jsjkx.220400200

Abstract

PDF(2934KB) ( 342 )

References | Related Articles | Metrics

Consensus algorithm not only maintains the stability and security of distributed system,but also is the key technology in the development direction of blockchain.With the rapid development of blockchain technology,the research of consensus algorithm has attracted more and more attention and favor of researchers.Nowadays,choosing an appropriate consensus algorithm in different application scenarios is a selective problem that researchers have to face.Starting from the types of service object nodes,this paper classifies the consensus algorithm into three categories:public chain,alliance chain and private chain.Based on these three categories,this paper expounds the basic principles of some mainstream and some new blockchain consensus algorithms,a total of 9 consensus algorithms,and evaluates the performance of these 9 consensus algorithms from three aspects:decentralization,security and scalability.This paper also analyzes and summarizes the advantages and disadvantages of the relevant algorithms,and gives the relevant directions to optimize the blockchain consensus algorithm for researchers’ research and reference,so as to promote the steady development of blockchain consensus algorithm.

Research Progress of RSA Algorithm in Network Data Transmission

WANG Xinmiao, SUN Tingting, MA Jingjun

Computer Science. 2023, 50 (6A): 220300107-7. doi:10.11896/jsjkx.220300107

Abstract

PDF(1859KB) ( 260 )

References | Related Articles | Metrics

In the process of people’s “ face-to-face ” information exchange using various electronic devices,both sides do not want their information to be obtained by the third party,which leads to the problem of communication security.Data security is not enough,the data transmission process is vulnerable to external interference,resulting in information duplication,missing,packet loss or delay.These problems are still not completely avoided.One of the important reasons is that the security of data is not enough,and it is disturbed by the outside world in the transmission process.In response to this problem,the majority of scientific researchers have actively responded,invented data encryption technology and introduced cryptography,which uses cryptographic algorithms to encrypt the data to reduce the interference in the transmission process,thereby protecting the data.In essence,among many technologies that guarantee various functional characteristics of information security,data encryption technology is the core and key technology of information security.Data encryption technology encrypts data from plaintext to ciphertext and communicates through encryption algorithm,and then decrypts the plaintext through corresponding decryption algorithm,which can improve the security of data transmission and ensure the integrity of data transmission.In order to further understand the working principle of cryptographic algorithm in network data transmission,this paper selects the RSA algorithm in asymmetric cryptographic system as the research object,introduces the encryption and decryption process of this algorithm in detail,compares and analyzes the advantages and disadvantages of RSA algorithm and ECC algorithm,and summarizes the corresponding optimization measures and optimization results in view of the defects of RSA algorithm.Finally,the research progress and practical application of RSA algorithm in network data transmission are summarized,and the future of RSA algorithm is prospected,hoping to provide some reference for data protection.

Network Advanced Threat Detection System Based on Event Sequence Correlation Under ATT&CK Framework

ZHANG Yuxiang, HAN Jiujiang, LIU Jian, XIAN Ming, ZHANG Hongjiang, CHEN Yu, LI Ziyuan

Computer Science. 2023, 50 (6A): 220600176-7. doi:10.11896/jsjkx.220600176

Abstract

PDF(3178KB) ( 246 )

References | Related Articles | Metrics

With the rapid development of network technology,the network world is becoming more and more fierce in attack and defense confrontation,and advanced network threat behaviors are emerging,but there are still some differences in the process description of multi-step attack behaviors in the actual operation and maintenance of the current network security analysts,which causes huge semantic communication costs.In order to solve this pain point problem in network advanced threat detection,ATT&CK network adversarial behavior framework is adopted as the unified description language of multi-step attack behavior,and a network advanced threat detection system based on event sequence association is designed and implemented,which can achieve effective detection of multi-step attack behavior through event sequence association model and visualize the presentation through ATT&CK attack matrix,which helps analysts to clarify the means,strategies and purposes of malicious attacks,and analysts can reduce attacker’s attack effect by taking corresponding defense measures through the techniques and tactics presented by the detection system.Experimental results show that the detection rate of the detection system can reach 96.43%,which is of great practical significance for analysts to solve the “defense dilemma” in network attacks.

Formal Verification of Supply Chain Contract Based on Coloured Petri Nets

ZHENG Hong, QIAN Shihui, LIU Zerun, DU Wen

Computer Science. 2023, 50 (6A): 220300220-7. doi:10.11896/jsjkx.220300220

Abstract

PDF(3959KB) ( 255 )

References | Related Articles | Metrics

The security of smart contracts is particularly vital to the application of blockchain in the supply chain field.Currently,most of formal verification work on smart contracts focuses on vulnerability detection,and there is still relatively little attention to how to generate secure smart contracts before deploying them on chain,and there are difficulties in how to effectively and stan-dardly map the properties of specific fields to smart contracts.Therefore,this paper proposes formal specification of supply chain business logic based on coloured Petri Net(CPN) before writing contracts and constructing a two-layer simulation model with a graphical interface to describe transaction state changes for formal verification and state analysis,thus reducing logic vulnerabilities at the modeling stage.Finally,a conversion method from the CPN modeling language to contracts written in Solidity is provided to improve the security and reliability of smart contracts.

Image Compression and Encryption Based on Compressive Sensing and Hyperchaotic System

PAN Tao, TONG Xiaojun, ZHANG Miao, WANG Zhu

Computer Science. 2023, 50 (6A): 220200121-6. doi:10.11896/jsjkx.220200121

Abstract

PDF(4529KB) ( 273 )

References | Related Articles | Metrics

In medical,military,financial systems and other scenarios where important images need to be transmitted,image compression and encryption is a feasible and effective way to transmit images safely and efficiently.Image compression and transmission can reduce the transmission overhead.Compressed images can be encrypted to make images more secure,and ordinary people can not get key information from them.After encryption,it can also resist some attacks means to ensure the security of information.Based on the compression perception theory,sparse sampling can be completed,and images can be compressed to any scale.Hyperchaotic system can guarantee the security of the system.Chaotic characteristics such as Lyapunov exponents of hyperchao-tic system are also analyzed.It is proved that the system is chaotic and safe enough.Chaotic sequences generated by hyperchaotic system are also used to construct measurement matrix.This eliminates the need to transfer a matrix with large text during transmission,but only the key.On the basis of compression theory,scrambling diffusion operation is also used,and diffusion operation related to plain text is used,which greatly improves image security and ensures data security.Experiments show that the image is compressed and encrypted well,the key space is large,the key is sensitive enough,the cipher histogram is distributed evenly,the cipher information entropy is close to the theoretical value,and the correlation between cipher images is low,which shows that it can resist many common attacks such as violent attacks and statistical attacks.At the same time,the decrypted image restored under normal compression ratio has a small visual gap with the original image,even if the compression ratio is small,most of the information content of the image can be seen,which indicates that the algorithm has a good reconstruction quality and high security.

Cross-architecture Cryptographic Algorithm Recognition Based on IR2Vec

ZHAO Chenxia, SHU Hui, SHA Zihan

Computer Science. 2023, 50 (6A): 220100255-7. doi:10.11896/jsjkx.220100255

Abstract

PDF(4484KB) ( 241 )

References | Related Articles | Metrics

In the field of information security,encryption technology is used to ensure the security of information.Identifying cryptographic algorithm in executable file is of great significance to protect information security.Most of the existing cryptographic algorithm recognition technologies can only target a single architecture and have poor recognition ability in cross-architecture scenarios.Therefore,this paper proposes IR2Vec model to solve the problem of cryptographic algorithm recognition in cross-architecture.Firstly,the model solves the cross-architecture problem based on the characteristics of LLVM connecting different front-end and back-end.The executable file is decompiled into the intermediate representation language by LLVM-RetDec,and then the PV-DM model is improved to quantify the semantics of the intermediate representation language,and the semantic similarity is judged by calculating the cosine distance of the vector.Collecting a variety of cryptographic algorithms to establish the cryptographic algorithm library,comparing the executable files of the target to be detected with the files in the cryptographic algorithm library one by one,and taking the one with the highest similarity as the recognition result.Experimental results show that the technology can effectively identify the cryptographic algorithm in the executable file.The model can support the cross recognition of binary files of X86,ARM and MIPS,Clang and GCC compilers and O0,O1,O2 and O3 optimization options.

Practical Byzantine Consensus Algorithm Based on Verifiable Random Functions

HUANG Baohua, PENG Li, ZHAO Weihong, CHEN Ningjiang

Computer Science. 2023, 50 (6A): 220300064-6. doi:10.11896/jsjkx.220300064

Abstract

PDF(2238KB) ( 213 )

References | Related Articles | Metrics

To solve the problems of fixed primary node selection method and high communication cost of the practical Byzantine fault tolerance consensus algorithm widely used in alliance chain,this paper proposes a selective Byzantine fault tolerance consensus algorithm named SBFT based on verifiable random function.The first proposal is to dynamically calculate node contribution value by evaluating the node behavior after each round of consensus,and select the nodes participating in consensus based on the node contribution value.Next,a combination of node contribution value and verifiable random function is used for random selection of primary nodes by cryptographic sortation,which makes the selected primary node unpredictable while reducing the probability of non-honest nodes becoming primary node.Finally,the consistency protocol of PBFT is improved by changing the mesh communication network topology of PBFT into a star communication network topology and incorporating the view replacement process into the normal consensus process.Simulation experimental results show that the proposed SBFT algorithm has higher throughput,lower consensus latency and higher algorithmic efficiency compared with the PBFT algorithm.

Byzantine Fault Tolerant Consensus Algorithm Based on Traceable Ring Signature

TU Jun, JIA Dongli, WANG Jin

Computer Science. 2023, 50 (6A): 220300100-7. doi:10.11896/jsjkx.220300100

Abstract

PDF(2371KB) ( 196 )

References | Related Articles | Metrics

The practical Byzantine fault tolerance(PBFT) consensus algorithm of alliance chain has the problems of weak privacy protection between nodes,static network structure,unreliable selection of master node and high communication overhead.A Byzantine fault-tolerant consensus algorithm(tracePBFT) based on traceable ring signature is proposed.Firstly,the nodes are randomly divided into primary domain nodes and secondary domain nodes,and different weights are given,and the primary domain node with high weight is selected as the primary node.Then,the ring signature is introduced in the preparation stage to protect the privacy of the node,and the node can select the reliable node through the weight,verify the signature and track the Byzantine node in the confirmation stage,and finally appropriately punish the Byzantine node.In this way,the selected master node is more reliable and reduce the communication overhead caused by changing the view due to the error of the master node.Experiments show that the tracePBFT algorithm is better than the traditional PBFT algorithm in communication complexity,security,throughput and so on.

DGA Domain Name Detection Method Based on Similarity

SUN Haidong, LIU Wanping, HUANG Dong

Computer Science. 2023, 50 (6A): 220400122-6. doi:10.11896/jsjkx.220400122

Abstract

PDF(2593KB) ( 265 )

References | Related Articles | Metrics

Botnets expose the Internet to a huge threat.Malicious behaviors such as distributed denial of service attacks and spam relying on botnets can cause great losses to the attack targets.The communication of the botnet is mainly based on the DGA domain name,so the domain name needs to be detected.Existing detection methods are mainly based on character encoding to extract domain name features,and then use neural networks for classification.Since only character features are considered,the detection accuracy of malicious domain names is often not high.In order to accurately detect DGA domain names,a calculation method of domain name character similarity and domain name node similarity is proposed,and malicious domain names are detected according to the similarity.First,a model based on a bidirectional gated recurrent unit neural network is constructed to screen out the algorithm with obvious features in the data set to generate domain names.Then using the recurrent neural network to cluster the selected malicious domain names,and finally calculate the similarity between the domain name to be detected in the dataset and the domain names which are malicious,and classify the domain name with the similarity greater than the threshold as the malicious domain name.Experimental results show that the method has an accuracy of 99.03% in detecting datasets containing multi-category malicious names.

Ring Confidential Transaction Protocol Based on Multivariate Public-key Cryptosystem

HONG Xuan, YUAN Mengling

Computer Science. 2023, 50 (6A): 220100157-6. doi:10.11896/jsjkx.220100157

Abstract

PDF(1777KB) ( 232 )

References | Related Articles | Metrics

Similar to Bitcoin,Monero is also a cryptocurrency.The original Monero is based on the CryptoNote protocol,which uses ring signatures and one-time keys to hide the real identities of both parties to the transaction,but the specific transaction amount is exposed in the area.In the blockchain,there are certain security risks.To address this security hole,Shen Noether proposed ring confidential transactions(RingCT),which utilizes a random number to hide the real transaction amount.The ring confidential transaction protocol currently uses by the Monero community is based on the discrete logarithm problem.However,with the development of quantum computers,solutions based on traditional number theory problems will become no longer secure.Post-quantum solutions are a good alternative.Multivariate public key cryptography is one of the main research directions of post-quantum cryptography,and compared with other post-quantum cryptographic schemes,multivariate-based signature schemes tend to have faster computing speed and less computing resources in the process of signature and verification.It has good researchva-lue.Based on the multivariable ring signature scheme,this paper designs a multivariable ring confidential transaction protocol.The protocol uses the additive homomorphism of the public key of the multivariable signature scheme to realize the commitment to the transaction amount,and performs a ring signature on the commitment.By randomly selecting the user public key in the blockchain to form a ring,the identity of the actual transaction participants in the transaction is confused.At the same time,during the transaction generation process,the trader’s private key will be used to generate a unique key-image,and it will participate in the signature generation process and become a part of the signature.By comparing this part,the transaction double-spending can be effectively prevented.The security of the proposed scheme is proved in the random oracle model,and compared with the lattice-based post-quantum secure ring confidential transaction protocol,the proposed scheme has more advantages in signature efficiency and verification efficiency.

Rumor Propagation Model of Microblog Network with Attenuation Effect and Forgetting Mechanism

WANG Han, LIU Wanping, LU Ling

Computer Science. 2023, 50 (6A): 220100189-7. doi:10.11896/jsjkx.220100189

Abstract

PDF(4114KB) ( 228 )

References | Related Articles | Metrics

With the rapid growth of microblog users,the control of network rumors becomes more important.In order to quiet down quickly the microblog rumors,and reduce the propagation range of rumors on microblog network,the key factors affecting the propagation of rumors are studied.Firstly,the UNFR propagating model is innovatively proposed by combining the spreading scene of rumors on microblog network,the model divides users into four categories:unknown,neutral,forwarder and refuter.Node state transitions of model are redefined by considering attenuation effect and forgetting mechanism.Through dynamics ana-lysis of the model and numerical experiments,the propagation regulations of network rumors are analyzed.Then,the rationality of the model is verified by propagation simulation experiments on microblog network.The method of reducing the propagation range of rumors is obtained by analyzing the influences of model parameters on rumor propagation.Finally,the control effects of para-meters under different initial values of rumor are studied,and effective control strategies of microblog rumors are proposed according to the experimental results.

Network Security Situation Assessment for GA-LightGBM Based on PRF-RFECV Feature Optimization

REN Gaoke, MO Xiuliang

Computer Science. 2023, 50 (6A): 220400151-6. doi:10.11896/jsjkx.220400151

Abstract

PDF(2030KB) ( 253 )

References | Related Articles | Metrics

At present,in the field of cyber security,due to the shortcomings of long training time and high sensitivity to redundant features,traditional machine learning models have been unable to deal with the increasingly complex network space.To improve the accuracy and efficiency of network security situation awareness for massive and high-dimensional network security elements,a GA-LightGBM network security situation awareness model based on PRF-RFECV feature preference is proposed,which first uses parallel random forest to filter out feature importance,then combines recursive feature elimination with cross-validation to select the optimal feature set,and finally uses the global search property of genetic algorithm to select the optimal parameters of LightGBM model for classification.Experimental simulation shows that the model is more accurate and more efficient than the traditional network security situation awareness algorithm in terms of both accuracy and F1 score.

Blockchain-based Identity Authentication and Authorization Mechanism

LIN Feilong, YUE Yuedong, ZHENG Jianhui, CHEN Zhongyu, LI Minglu

Computer Science. 2023, 50 (6A): 220700158-9. doi:10.11896/jsjkx.220700158

Abstract

PDF(3026KB) ( 311 )

References | Related Articles | Metrics

The abuse of people’s identity information is a serious problem in nowadays society.In this paper,a blockchain-based identity authentication and authorization(BIAA) mechanism is proposed.BIAA requires users to provide the effective identity certificate and biological feature to authorize the business,to ensure that the business is authorized by the user.Then,the identity authorization together with the business contract will be written into the blockchain ledger with the secure and traceable manner.To fulfill BIAA,a stellate multi-blockchain structure is proposed for identity register and authorization.An identity register blockchain is built using consortium blockchain which is maintained by authorities to manage the identity registration.It also charges to identity authentication.Multiple identity authorization blockchains can be built with the permission from identity register blockchain.Each identity authorization blockchain can be maintained by a business sector and write the business contracts with identity authorizations into the blockchain ledger.For technical implementation,an identity register-authenticate-authorize(IRAA) terminal is designed.It transforms the identity and biological feature into ciphertext by hash function,thus to guarantee the identity information offline and secure.It is also embedded with the protocol to deal with the identity authentication in an encrypted way.IRAA terminal also charges to sign the business contract using digital signature and thus finish the identity authorization.Finally,a prototype system leveraging second-generation identity certificate and finger vein pattern as identity information is built,which verifies the security,feasibility,and effectiveness of BIAA mechanism and provides a valuable reference for solving the abuse of identity.

Network Reliability Analysis of Power Monitoring System Based on Improved Fuzzy ComprehensiveEvaluation Method

BING Ying’ao, WANG Wenting, SUN Shengze, LIU Xin, NIE Qigui, LIU Jing

Computer Science. 2023, 50 (6A): 220400293-7. doi:10.11896/jsjkx.220400293

Abstract

PDF(2446KB) ( 272 )

References | Related Articles | Metrics

Network attacks emerge one after another,and the importance of risk management and control in the power industry is increasing day by day.However,complex and new network attack methods,missing system loopholes in production equipment,and the complexity of the integration of physical and information networks will undoubtedly bring forward the risk management and control of power monitoring systems.Aiming at how to comprehensively,accurately and reasonably judge the network reliability of the power monitoring system under the comprehensive influence of multiple factors,a sound reliability evaluation system is established,and a reliability evaluation model of the power monitoring system based on fuzzy evaluation is proposed.This paper starts with the loopholes existing in the equipment nodes of the power monitoring system,and comprehensively analyzes the risks in the system and the environmental risks outside the system,so that the power monitoring system can be safely graded during the operation process to make security decisions.This method adopts network security classification.The standard is combined with the industrial system vulnerability database to establish an evaluation index system,and the reliability evaluation index is determined from the three aspects of network communication reliability,business reliability,and system reliability.The reliability analysis of the power monitoring and monitoring system is carried out,targeted maintenance is carried out according to the evaluation results,and more accurate quantitative indicators are used to achieve fine-grained risk assessment in the evaluation process.Finally,a set of semi-virtual system environment synthesis is built for power.The risk level and reliability of the power monitoring system are analyzed and evaluated to verify the validity of the reliability evaluation method.

Restart and Recovery Algorithm Based on Distributed Cluster Nodes

PAN Lu, LUO Tao, NIU Xinzheng

Computer Science. 2023, 50 (6A): 220300205-6. doi:10.11896/jsjkx.220300205

Abstract

PDF(2284KB) ( 199 )

References | Related Articles | Metrics

In view of the node downtime caused by malicious attack in a distributed cluster,the recovery efficiency of the snapshot mechanism of traditional node restart is insufficient.Based on the storage and transmission of snapshot files by cluster nodes,a Raft-based snapshot dual-trigger strategy is proposed to improve the rationality of snapshot triggering.Experiments show that the algorithm improves the time from node downtime to recovery compared to the original Raft algorithm,so as to avoid the shortcomings of traditional cluster node recovery,and can better adapt to the complex network conditions of the cluster.Cluster node failure recovery is of great reference significance.

Study on SQL Injection Detection Based on FlexUDA Model

WANG Qingyu, WANG Hairui, ZHU Guifu, MENG Shunjian

Computer Science. 2023, 50 (6A): 220600172-6. doi:10.11896/jsjkx.220600172

Abstract

PDF(2658KB) ( 191 )

References | Related Articles | Metrics

FlexUDA model based on semi-supervised learning is proposed to solve the problem that insufficient labeled data is easy to cause model over fitting when deep learning method detects SQL injection.Firstly,the collected data are preprocessed by decoding,generalization and word segmentation,and then the unlabeled data are augmented by calculating the TF-IDF value.The original data and augmented data are vectorized using TF-IDF and Word2Vec fusion algorithm.Finally,the FlexUDA model is used for training,and the trained model is compared with other models.Experimental results show the FlexUDA model only uses 1000 labeled data and 100000 unlabeled data for training,and achieves 99.42% accuracy and 99.23% recall.Compared with other supervised training models,it shows better generalization performance,and can well solve the over fitting problem caused by insufficient labeled data in SQL injection detection.

Image Recognition Method of Transmission Line Safety Risk Assessment Based on MultidimensionalData Coupling

XU Changqian, WANG Dong, SU Feng, ZHANG Jun, BIAN Haifeng, LI Long

Computer Science. 2023, 50 (6A): 220500032-6. doi:10.11896/jsjkx.220500032

Abstract

PDF(3282KB) ( 246 )

References | Related Articles | Metrics

Transmission lines located in high-altitude and high icing risk areas face the risk of large-area line breaking and tower falling in extreme climate.The traditional manual line patrol identification has slow speed and low accuracy,resulting in a lot of labor cost.A transmission line safety risk assessment method considering multi-dimensional image coupling driving is proposed.The icing image of key equipment is fused with the contour image of power grid operation state,so as to realize the rapid and accurate identification of relevant transmission line safety risks.Firstly,the transmission line electrical data and environmental data are coupled to generate a multi-dimensional thermal image,which can reflect the transmission line voltage offset,line load rate,ambient temperature and line icing degree in the whole system,and the line safety risk index is calculated according to the electrical data and environmental data.After that,the convolution neural network model based on MobileNet-V3 framework is built,and the generated multi-dimensional image data is used as the input of the model and the transmission line safety risk index is used as the output to train the model and generate the transmission line safety risk rapid assessment model.Finally,the model is tested on a 500kV transmission line in a province.The test results show that this method can realize the rapid and accurate assessment of transmission line safety risk.

Robust Federated Learning Algorithm Based on Adaptive Weighting

ZHANG Lianfu, TAN Zuowen

Computer Science. 2023, 50 (6A): 230200188-9. doi:10.11896/jsjkx.230200188

Abstract

PDF(3422KB) ( 390 )

References | Related Articles | Metrics

Federated learning allows multiple data owners to jointly train machine learning models without sharing private training data.However,studies have shown that FL is vulnerable to Byzantine attacks and privacy breaches,this problem has not been well addressed by existing studies.In the federated learning scenario,protecting FL from Byzantine attacks while considering performance,efficiency,privacy,number of attackers,simplicity and feasibility is a challenging problem.To solve this problem,a privacy preserving robust federal learning algorithm DP-FedAWA is proposed based on l₂-norm distance and quadratic normalization.The proposed algorithm does not require any assumptions outside the training process and can deal with a few or a lot of attackers adaptively.In no defense setting,DP-FedAvg is used as the comparison baseline,while Krum and Median are used as the comparison baseline in the defense setting.Extensive experiments on MedMNIST2D data set confirm that the proposed DP-FedAWA algorithm is safe and robust to malicious clients,and comprehensively outperforms the existing Krum and Median in Accuracy,Precision,Recall and F1-Score.

New Image Watermarking Algorithm Based on Quantum Wavelet Transform

SU Yonghong, XIA Ting, WANG Xumei, QIAN Xiaohong

Computer Science. 2023, 50 (6A): 220300034-8. doi:10.11896/jsjkx.220300034

Abstract

PDF(3836KB) ( 212 )

References | Related Articles | Metrics

Image watermarking is a technology that embeds specific information into the carrier image for the purpose of copyright protection.A new image watermarking scheme based on quantum wavelet transform is studied,including scrambling process,embedding process and extraction process.The improved quantum Arnold scrambling method is used to scramble the binary image.The scrambled watermark image is applied to the least effective block qubit of the carrier image.For the carrier gray image,the quantum Haar wavelet transform and quantum least significant bit(LSB) blocking technology are used to embed the scrambled watermark image into the quantum wavelet coefficient.In the extraction process,firstly,the scrambled watermark image is extracted from the embedded image,and then the improved quantum Arnold inverse scrambling method is used to obtain the original watermark image.Simulation technology verifies the invisibility and high robustness of the watermark based on the quantum image watermarking method.The invisibility of the scheme is proved by peak signal-to-noise ratio(PSNR) test.The high robustness of the scheme is tested by bit error rate(BER) test and normalized correlation coefficient(NC).Simulation results show that the watermarking scheme not only has acceptable visual quality,but also has good resistance to different types of attacks.

Empirical Study on Application and Maintenance of OSS Community Profile Documentation

ZHANG Yu, WANG Zhe, LI Zhixing, YU Yue, WANG Tao, CAI Mengluan

Computer Science. 2023, 50 (6A): 220600221-8. doi:10.11896/jsjkx.220600221

Abstract

PDF(2471KB) ( 182 )

References | Related Articles | Metrics

Community profile documentation is crucial for the establishment and management of open source software(OSS) communities.Although prior research has conducted content analysis of community profile documentation,little is known about how common it is in practice and how it is maintained by OSS practitioners.We aim at complementing the current understanding of community profile documentation by providing a quantitative description of its prevalence and maintenance.We randomly collect 2000 OSS projects from GitHub,based on which we study the documentation popularity by programming language,repository owner type,repository age,and community size,respectively.We also investigate the maintenance practice of community profile documentation in terms of location,creation latency,maintainers,update frequency and change-triggering events,respectively.The README and LICENSE documentation is far more popular and created earlier than the CONTRIBUTING,CONDUCT and TEMPLATE documentation in GitHub OSS projects.Community profile documentation is more likely to be found in repositories of TypeScript,repositories of larger community size,and repositories owned by organizations.Community profile documentation is mainly placed in the root directory and changed by a small group of developers with a low frequency of update,which is mostly driven by perfective and adaptive requirements.

Service Recommendation Algorithm Based on Multi-features Crossing

GAO Wenbin, WANG Rui, ZU Jiachen, DONG Chenchen, HU Guyu

Computer Science. 2023, 50 (6A): 210800242-7. doi:10.11896/jsjkx.210800242

Abstract

PDF(2901KB) ( 221 )

References | Related Articles | Metrics

With the rapid growth of the number of web services,the problem of service overload has gradually emerged.To relieve service overload,and help users position high-quality services rapidly,service recommendation has become a hot research topic in the field of service computing.Aiming at the difficulties of cold start and data sparseness in current service recommendation,this paper proposes a quality of service(QoS) prediction recommendation algorithm SRMFC based on the multi-features crossing,which implements multi-features through the “word embedding” method to improve the performance of the algorithm in dealing with the cold start.At the same time,a neural network is used to complete the automatic cross of multi-features.Compared with traditional collaborative filtering,factorization machine and other methods,the proposed algorithm can achieve in-depth exploration of the relationship between features,and improve the learning ability of the algorithm in dealing with extremely sparse data scenarios.Experiments on public data sets show that,under different data sparsity scenarios,the service quality prediction error of the SRMCF intersection decrease by at least 20% compared with the mainstream service recommendation algorithm in recent years.

Synchronizing Algorithms for Bounded Partially Ordered Automata

WANG Zhixi, JIANG Guide

Computer Science. 2023, 50 (6A): 220500099-5. doi:10.11896/jsjkx.220500099

Abstract

PDF(1718KB) ( 249 )

References | Related Articles | Metrics

Synchronizing automata are the automata having a synchronizing word.They have been applied in many fields such as system testing,encoding,industrial automation,robotics and biological computation.Bounded partially ordered automata are the automata whose state set is equipped with a partial order that is compatible to the input letters.This paper reveals some important characterizations of synchronizing bounded partially ordered automata,proposes the algorithms for testing the synchronization,finding a synchronizing word and finding a shortest synchronizing word of an arbitrary bounded partially ordered automaton,and then determines the least upper bound of all n-state synchronizing bounded partially ordered automata.In the field of bounded partially ordered automata,these works solve the main problems on synchronizing automata.

Key Risk Node Identification Methods in New Energy Vehicle Supply Chain

YANG Xiaobo, GAO Haiwei, LIU Tianyue, GUO Binghui

Computer Science. 2023, 50 (6A): 221100052-7. doi:10.11896/jsjkx.221100052

Abstract

PDF(2968KB) ( 454 )

References | Related Articles | Metrics

In this paper,the open data of Tesla and XPENG,which have typical analytical value in the new energy automobile industry,are respectively used to construct networksof supply chain.And the key node identification method of risk transmission based on multi-dimensional fusion is studied according to the structural correlation characteristics of the network of supply chain.Firstly,the network centrality characteristics are introduced to analyze and calculate the key node.At the same time,considering the characteristics of systemic risk transmission in supply chain of automotive,the risk immune transmission model is introduced to determine the key nodes.Finally,the cascading failure model of the two networks is analyzed respectively,and the key nodes that have strong impact on network failure are selected.Through multi-dimensional key node analysis,it is found that the key nodes with strong impact include not only core enterprises such as batteries in the traditional sense,but also accessory enterprises with invisible leading position.Therefore,through the comprehensive analysis method of structure and transmission attribute proposed in this paper,the potential hidden key risk control nodes in the network of supply chain of new energy vehicles can be well found,which has good practical application value.

Fast Calculation Method of High-altitude Electromagnetic Pulse Environment Based on Machine Learning

WANG Jinjin, CHENG Yinhui, NIE Xin, LIU Zheng

Computer Science. 2023, 50 (6A): 220500046-5. doi:10.11896/jsjkx.220500046

Abstract

PDF(2597KB) ( 255 )

References | Related Articles | Metrics

The calculation of the environment of high altitude electromagnetic pulse is the basis of the electromagnetic pulse effect technology and the protection.In order to quickly calculate the high altitude electromagnetic pulse environment parameters and their distribution,a new method combining traditional calculation method with neural network method is proposed.Firstly,it calculates the HEMP discrete data with different explosive height,different equivalent and different position in space using the exis-ting method.Second,it establishes the fast multi-parameter calculation model using neural network based on these discrete data.Finally,this paper calculates the HEMP environment parameters of different explosive height,different equivalent and different position in space using the established model in parallel batch.The distribution of the high altitude electromagnetic pulse is also can be calculated quickly.Experimental results show that HEMP environment parameters and their distribution are accelerated using the fast method.It can provide a large number of incident field parameters for the calculation of high altitude electromagne-tic pulse conduction environment.

Formalization of Inverse Matrix Operation Based on Coq

SHEN Nan, CHEN Gang

Computer Science. 2023, 50 (6A): 220400108-7. doi:10.11896/jsjkx.220400108

Abstract

PDF(2495KB) ( 220 )

References | Related Articles | Metrics

Matrix is a data structure widely used in computer science,and its correctness of operation is of great significance.In matrix formalization,there is no reasonable and practical formalization work.The reason lies in the difficulty in formalizing the two common inverse methods in engineering.The first method is based on the adjoint matrix solution method.The difficulty lies in that the submatrices of n*n matrix cannot be formally expressed,which makes it very difficult to construct the matrix composed of cosubformulas.Therefore,it is difficult to achieve the formalization of adjoint matrix inverse solution.The second method is called Gauss-Jordan elementary transformation method,the difficulty lies in the construction of elementary matrix and its operation function.If Coq is used to design the operation function of inductive structure,that is,the idea of filling two-dimensional table with rows first is adopted,the description information of two-dimensional table from column dimension will be discarded,so that the operation function branches too much and complex inductive structure needs to be designed,which leads to the failure of subsequent formal verification.In this paper,a record-based matrix function construction method is proposed,which describes the matrix in both column and column dimensions,making it possible to construct and prove the elementary matrix.On this basis,the formalization of matrix inversion in Coq system based on gaussian-Jordan elimination method is realized.And we implement the first software matrix inversion function library under formal verification in a way with lower cost and time complexity.

Visualization of Ocean Data Vector Field Based on Streamline Distance Clustering

WANG Zhen, YANG Zhengwei, GAO Shunqi, ZHANG Lei

Computer Science. 2023, 50 (6A): 220300284-7. doi:10.11896/jsjkx.220300284

Abstract

PDF(2957KB) ( 222 )

References | Related Articles | Metrics

Streamlines visualization is an important research object of ocean vector field visualization,in which the setting of number and location for streamlines seed points is the basis.The key to the problem,how to eliminate the visual confusion and occlusion caused by multiple streamlines,is the accurate clustering of generated streamlines and the selection of appropriate represen-tative streamlines within the clustering.In this paper,the PDM distance is proposed to be the similarity measurement,then the fine streamline clustering is implemented after once endpoints streamline clustering.The proposed method effectively solves the problem of inaccurate endpoint clustering results and improves the accuracy of streamline clustering.After sort of all the PDM distance within each clustering,we extract midline and two boundary lines to redraw the streamlines,so as to reduce the pheno-menon of visual confusion and occlusion caused by multiple streamlines.For the problem of huge amount of calculation,the MDS algorithm is proposed to reduce dimension and accelerate computing speed.In addition,in order to further accelerate the calculation speed,the critical point detection algorithm is adopted to reduce the time-consuming vortex generation during the process ofstreamline generation.The effectiveness and superiority of our proposed method are verified by using ocean flow field data from China coast,and the drawing effect of streamline is good.

Grid-based Tracking Method for Hydrographic Mapping UAV

YAO Xi

Computer Science. 2023, 50 (6A): 220300023-7. doi:10.11896/jsjkx.220300023

Abstract

PDF(2460KB) ( 214 )

References | Related Articles | Metrics

With the development of surveying and mapping technology,unmanned aerial vehicles(UAV) have been widely used in hydraulic engineering.It has improved the efficiency of water conservancy surveying and mapping work.It has also brought about problems such as searching for and monitoring UAV,which are caused by the loss of communication and forced landing.How to effectively track the water conservancy mapping UAV has become a research topic.In view of this,this paper proposes a grid-based tracking method.The three-dimensional grid of the space is divided.The target grid is locked based on the spatial information scanned by radar.Driving the camera to collect UAV images.Image recognition and radar information are fused to realize UAV recognition and tracking.In this paper,three-dimensional mesh space division algorithm,mesh mapping algorithm,UAV identification algorithm,tracking and monitoring algorithms are described in detail.Experimental results show that this technique has advantages in fast capture,effective lock,tracking and surveillance of water conservancy mapping UAV.

Design of Indoor Mapping and Navigation System Based on Multi-sensor

LIU Jiawei, DU Xin, FAN Fangzhao, XIE Chengbi

Computer Science. 2023, 50 (6A): 220300218-8. doi:10.11896/jsjkx.220300218

Abstract

PDF(5948KB) ( 296 )

References | Related Articles | Metrics

Aiming at the problems of low positioning accuracy and limited ability to describe the environment in the process of map building and navigation of mobile robot based on single sensor,a multi-sensor sensing map building and navigation system based on robot operating system is developed.Firstly,an omnidirectional four-wheel Mecanum mobile chassis is built.Secondly,the RTAB-MAP algorithm is analyzed,and based on this algorithm,the RGB-D camera,lidar and odometer information are fused to realize the simultaneous construction of two-dimensional and three-dimensional maps of indoor environm ent.Thirdly,the extended Kalman filter algorithm is proposed to fuse the odometer information generated by the encoder with IMU data to improve the estimation accuracy of pose.Finally,the traditional robot navigation framework is optimized according to the fused data,the design of autonomous navigation function is completed.The test results show that the system adopts the multi-sensor sensing scheme,which can complete the construction of two-dimensional and three-dimensional maps of the indoor scene at the same time,and improve the ability to describe the environment.By using the fused data of extended Kalman filter,the positioning accuracy of the robot is significantly improved and the accuracy of navigation is ensured.

Study on Android Fake Application Detection Method Based on Interface Similarity

FU Xiong, NIE Xiaohan, WANG Junchang

Computer Science. 2023, 50 (6A): 220300114-7. doi:10.11896/jsjkx.220300114

Abstract

PDF(2738KB) ( 340 )

References | Related Articles | Metrics

With the development of the Android,fake applications appear and become active on the Android platform.The popularity of obfuscation and other technologies makes it difficult for fake applications to be detected by traditional detection methods.In order to effectively resist the reinforcement technology,an Android fake application detection method(InfSimiDetec) based on interface similarity is proposed.Firstly,the layout information of the running interface is extracted by the automatic test tool.Next,the interface structural features are extracted based on the layout information.Then the interfaces with similar structural features are selected for interface similarity calculation.Finally,the application similarity calculation is carried out based on the ratio of similar interfaces.Experiments are carried out using a dataset containing multiple types of fake applications and compared with traditional detection methods.The results show that the precision rate of this method is 94.11% and the recall rate is 96.12%.Compared with traditional detection methods,this method shows better performance.

Diesel Engine Fault Diagnosis Based on Kernel Robust Manifold Nonnegative Matrix Factorizationand Fusion Features

LIU Hongyi, WANG Rui, WU Guanfeng, ZHANG Yang

Computer Science. 2023, 50 (6A): 220400128-8. doi:10.11896/jsjkx.220400128

Abstract

PDF(3887KB) ( 288 )

References | Related Articles | Metrics

The diesel engine is one of the important power sources in industrial production,its failure will cause a huge impact on the efficiency and safety of industrial production,it is of great significance to diagnose the fault of diesel engine.Aiming at the difficulty and low accuracy of feature extraction in diesel engine valve fault diagnosis,a diesel engine fault diagnosis method based on kernel robust manifold non-negative matrix factorization method and fusion feature is proposed.Firstly,the pressure signal is analyzed in the time domain to extract the pressure characteristics.Secondly,the time-frequency analysis of the vibration signal is carried out using the short-time flourier transform(STFT),and the features of the vibration signal are extracted by the kernel robust manifold nonnegative matrix factorization.Then the features of the pressure signal and vibration signal are fused.Finally,support vector machine is used to realize fault diagnosis.Compared with the traditional method,the fault diagnosis accuracy of this method can reach 100% on the collected data set,which proves that it can effectively extract features and significantly improve the diagnosis accuracy.

Multidimensional Evaluation Method for Domestic Building Information Modeling Software Based on Entropy-Weight-AHP and Cloud Model

ZHAO Xuefeng, HOU Xiao, SUN Zhe, LI Mengxuan

Computer Science. 2023, 50 (6A): 220400216-9. doi:10.11896/jsjkx.220400216

Abstract

PDF(2433KB) ( 225 )

References | Related Articles | Metrics

With the key research and wide application of BIM in China’s architectural field,the strategic significance of BIM mo-deling software has been continuously improved.Long-term dependence on foreign BIM modeling software makes China’s construction industry face the risk of “getting stuck”.China’s construction industry urgently needs the high-quality development of domestic BIM modeling software.Systematic,scientific,reasonable and objective software evaluation is an important means to promote the development of China’s domestic BIM modeling software,and it is also a key problem to be solved urgently in China’sconstruction industry at present.Firstly,the characteristics of foreign mainstream BIM modeling software are investigated and systematically analyzed,and a multi-dimensional evaluation model of domestic BIM modeling software including function evaluation dimension,quality evaluation dimension and level of BIM implementation dimension is established.Then,based on entropy weight-AHP and cloud model theory,a multi-dimensional evaluation process for domestic BIM modeling software is established.Finally,based on the multi-dimensional evaluation model,the case software PKPM-BIM and Glodon numerical dimensional architecture design software are evaluated in a multi-dimensional way,and the feasibility and applicability of the evaluation model are verified.

Container-based Scheduling Architecture for Mixed-Criticality Systems

DENG Guanghong, ZHANG Qiheng

Computer Science. 2023, 50 (6A): 220800215-5. doi:10.11896/jsjkx.220800215

Abstract

PDF(2125KB) ( 252 )

References | Related Articles | Metrics

The mixed-criticality systems composed of real-time containers and non-real-time containers have difficulties in ensuring the real time of scheduling and the allocation of cpu resources.In this paper,we present an architecture of scheduling RT containers and NRT containers for mixed-criticality systems,which is based on the hierachical scheduler to schedule the runqueues of the container control groups.By this means,our architecture ensures the real-time of RT containers by limiting system resources to NRT containers.We also add monitor and load balancer for workloads to ensure equitable allocation of CPU resources occupied by NRT containers.Experimental results show that the proposed architecture can improve the degradation of real-time in RT containers when RT containers coexist with NRT containers in mixed-criticality systems.

College Students Employment Dynamic Prediction of Multi-feature Fusion Based on GRU-LSTM

ZHANG Jian, ZHANG Ye

Computer Science. 2023, 50 (6A): 220500056-6. doi:10.11896/jsjkx.220500056

Abstract

PDF(2453KB) ( 220 )

References | Related Articles | Metrics

At present,the employment prediction system of colleges and universities mostly adopts single traditional feature mo-deling,which leads to problems such as poor employment prediction effect and weak employment accurate service.This paper proposes a multi-feature fusion based on GRU-LSTM employment prediction method.Firstly,students’ behavior features are added to the traditional prediction model,and the feature vector of multi-information fusion is constructed.Then,considering the different contribution of different influencing factors to college students employment,an optimal feature extraction method of employment prediction based on Pearson correlation coefficient is proposed to optimize the feature subset.Finally,a combined prediction model of GRU and LSTM is proposed,which combines the advantages of high prediction accuracy of LSTM and short prediction time of GRU to make efficient and accurate prediction of employment data.Experimental results show that compared with the traditional methods,the accuracy of employment prediction by this method increases by 4.2%,providing reliable data support for improving the employment of college students.

Design and Implementation of Natural Gas Intelligent Scheduling Computer Platform System

DENG Shengnan, LUO Taiyu, HUANG Jingcai, REN Yuqing, SONG Wei, SU Chang, LEI Lili, HU Guanghui, XU Hong

Computer Science. 2023, 50 (6A): 220700258-7. doi:10.11896/jsjkx.220700258

Abstract

PDF(3229KB) ( 222 )

References | Related Articles | Metrics

Natural gas is an efficient green energy.For a long time in the future,natural gas will be an important part of China’s energy consumption structure.Chongqing has been leading the country in the exploitation and utilization of natural gas.However,at present,it is facing increasing difficulties in monitoring and optimizing the operation of pipe networks.Therefore,combined with the background of the rapid development of information technology such as big data and simulation technology,this paper introduces a natural gas intelligent scheduling platform that is being developed.The platform adopts browser/server mode architecture,which mainly includes pipe network operation information monitoring module,working condition optimization scheduling module,planned maintenance scheduling module and scheme library management module.As for its core function,i.e.,the intelligent scheduling optimization,this paper uses a typical example to illustrate it.The platform is conducive to the realization of intelligent scheduling and control of natural gas pipeline network.It can not only maximize the capacity of natural gas pipeline network,but also save relevant manpower and energy consumption.The research work in the near future will focus on the optimization scheduling of the pipeline network under abnormal and fault conditions.

Study on Product Recovery Model of Remanufacturing Enterprises Based on Game Theory

CAI Ran, HUANG Pengpeng

Computer Science. 2023, 50 (6A): 220300113-6. doi:10.11896/jsjkx.220300113

Abstract

PDF(2139KB) ( 222 )

References | Related Articles | Metrics

Based on the research assumptions in the closed-loop supply chain of different recycling channels,a product recycling model is established for recyclers(manufacturers,retailers or third-party recyclers),and the Stackelberg game theory is applied to analyze the models,and three types of recycling are obtained.For the equilibrium solution of the model,Matlab is used to numerically simulate the solution of the optimal recovery model.Research results show that in the market led by manufacturers,the pricing of remanufactured products’ sales channels and recycling channels do not change due to the different recycling models,but the profit distribution of channel members varies with the different recycling models.In the models where recycling is carried out by the manufacturer,the entire closed-loop supply chain has the largest profit,and the other two recycling models have the same closed-loop supply chain profit.The manufacturer always has the highest profit in all recycling models.