Started in January,1974(Monthly)
Supervised and Sponsored by Chongqing Southwest Information Co., Ltd.
ISSN 1002-137X
CN 50-1075/TP
CODEN JKIEBK
Editors
Current Issue
Volume 49 Issue 6A, 16 June 2022
  
Smart Healthcare
Survey on Finger Vein Recognition Research
LIU Wei-ye, LU Hui-min, LI Yu-peng, MA Ning
Computer Science. 2022, 49 (6A): 1-11.  doi:10.11896/jsjkx.210400056
Abstract PDF(2606KB) ( 1729 )   
References | Related Articles | Metrics
Finger vein recognition has become one of the most popular research hotpots in the field of biometrics because of its unique technical advantages such as living body recognition,high security and inner features.Firstly,this paper introduces the principle,merits,and current research status of finger vein recognition,then making the time as the clue,sorts out the development history of finger vein recognition technology,and discusses the classical and state-of-the-art recognition algorithms.Secondly,focusing on each process of finger vein recognition,this paper expounds on the critical techniques including image acquisition,image preprocessing,feature extraction and matching in traditional methods,and deep learning-based recognition.Besides,the commonly used public datasets and the related evaluation metrics in this field are introduced.Thirdly,this paper summarizes the existing research problems,proposes the corresponding feasible solutions,and predicts the future research direction of finger vein recognition.Some new ideas in the following studies for researchers are provided at the end.
Brain Tumor Segmentation Algorithm Based on Multi-scale Features
SUN Fu-quan, CUI Zhi-qing, ZOU Peng, ZHANG Kun
Computer Science. 2022, 49 (6A): 12-16.  doi:10.11896/jsjkx.210700217
Abstract PDF(3873KB) ( 785 )   
References | Related Articles | Metrics
Brain tumors are the most common diseases of nervous system except cerebrovascular disease,and their segmentation is also an important field in medical image processing.Accurately segmenting the tumor region is the first step in the treatment of brain tumors.Aiming at the problem of information loss caused by the weak multi-scale processing ability of traditional fully convolutional networks,a fully convolutional network based on multi-scale features is proposed.Using spatial pyramid pooling to obtain advanced features of multiple receptive fields,thereby capturing contextual multi-scale information and improving the adaptability to different scale features.Replacing the original convolution layer with the residual compact module can alleviate the degradation problem and extract more features.The data augmentation technology is combined to enhance the segmentation perfor-mance maximally while avoiding over fitting.Through a large number of contrastive ablation experiments on the public low-grade glioma MRI dataset,using Dice coefficient,Jaccard index and accuracy as the main evaluation criteria,91.8% Dice coefficient,85.0% Jaccard index and 99.5% accuracy are obtained.Experimental results show that the proposed method can effectively segment brain tumor regions and have certain generalization,and the segmentation effect is better than other networks.
Drug-Drug Interaction Prediction Based on Transformer and LSTM
KANG Yan, XU Yu-long, KOU Yong-qi, XIE Si-yu, YANG Xue-kun, LI Hao
Computer Science. 2022, 49 (6A): 17-21.  doi:10.11896/jsjkx.210400150
Abstract PDF(3998KB) ( 1061 )   
References | Related Articles | Metrics
The adverse reactions of drug-drug interactions have become one of the important reasons for the increase in the incidence of diseases such as digestive system diseases and cardiovascular diseases,and leads to the withdrawal of drugs from the market.Therefore,accurate prediction of drug interactions attracte widespread attention.Aiming at the problem that the traditional Encoder-Decoder model cannot capture the dependence between drug substructures,this paper proposes a TransDDI(TransformerDDI) based on Transformer and LSTM drug interaction prediction model.TransDDI includes three parts:data preprocessing module,latent feature extraction module and mapping module.The data preprocessing module uses the SPM algorithm to extract the frequent substructures that characterize the drug from the SMILES format input of the drug to form the drug feature vector,and then generate the feature vector of the drug pair.The latent feature extraction module uses Transformer to fully mine the information contained in the substructures of the feature vector,highlight the different important roles of different substructures,and generate potential feature vectors.The mapping module mainly forms a dictionary representation of the potential feature vector of the drug pair and the vector of the frequent substructure in the database,and uses the neural network fused with LSTM to make predictions.Onreal data sets BIOSNAP and DrugBank,the proposed method is compared with other 6 machine learning and deep learning methods by experiments.The results show that TransDDI has a higher accuracy rate and is convenient for drug interaction prediction.
Multi Model Algorithm for Intelligent Diagnosis of Melanoma Based on Deep Learning
CHANG Bing-guo, SHI Hua-long, CHANG Yu-xin
Computer Science. 2022, 49 (6A): 22-26.  doi:10.11896/jsjkx.210500197
Abstract PDF(3517KB) ( 1002 )   
References | Related Articles | Metrics
Skin melanoma is a kind of disease that can be cured by early detection.The main diagnosis method is based on the manual visual observation of dermatoscope.Affected by the doctor's medical skill and experience,the diagnostic accuracy is 75%~80% and the diagnostic efficiency is low.In this paper,a multi-modal neural network algorithm based on metadata and image data is proposed.Metadata is the feature vector that extracts the basic information of patients,the location of lesions,the resolution and quantity of images through perceptual machine learning model.The image data is extracted from the feature vectors of CNN model,and the two feature vectors are fused and mapped to obtain the disease classification results,which can be used for early auxiliary diagnosis of melanoma.A total of 58 457 samples are collected from ISIC 2019 and ISIC 2020 mixed data sets.The training samples and test samples are divided according to the ratio of 4∶1.The multi-modal algorithm and convolutional neural network method proposed in this paper are used for comparative experimental research.The results show that the AUC value of the melanoma auxiliary diagnosis classification model constructed by this algorithm can be improved by about 1%,which has certain use value.
Visual Analysis of Multiple Probability Features of Bluetongue Virus Genome Sequence
CHEN Hui-pin, WANG Kun, YANG Heng, ZHENG Zhi-jie
Computer Science. 2022, 49 (6A): 27-31.  doi:10.11896/jsjkx.210300129
Abstract PDF(4283KB) ( 511 )   
References | Related Articles | Metrics
The sequence of a gene determines its structure,and the structure of a gene reflects biological functional traits.Therefore,scientific data visualization of viral gene sequences has become one of the widely used methods.There is an increasing demand for visual manipulation of biological gene sequences.Therefore,based on the most advanced bioinformatics analysis and hierarchical structured bioinformatics knowledge model,the method of multiple probability measures is proposed to statistically analyze the bluetongue virus gene sequence,and combined with computer visualization methods,the characteristics of the bluetongue virus under different projections are presented.Compared with traditional virus research methods,this method is intuitive and concise,and it is easy to use.This method provides rich visualization under different measurement coordinates,reflecting the classification characteristics of bluetongue virus.Results generated by this method are compared with the phylogenetic tree generated by traditional biological analysis methods,which could provide references for homology analysis and the study of the evolutionary relationship of bluetongue virus.It is conducive to in-depth study of bluetongue virus from various angles.
New Text Retrieval Model of Chinese Electronic Medical Records
YU Jia-qi, KANG Xiao-dong, BAI Cheng-cheng, LIU Han-qing
Computer Science. 2022, 49 (6A): 32-38.  doi:10.11896/jsjkx.210400198
Abstract PDF(2605KB) ( 818 )   
References | Related Articles | Metrics
The growth of electronic medical records forms the basis of user health big data,which can improve the quality of medi-cal services and reduce medical costs.Therefore,the rapid and effective retrieval of cases has practical significance in clinical medi-cine.Electronic medical records have strong professionalism and unique text characteristics.However,traditional text retrieval methods have the disadvantages of inaccurate text entity semantic expression and low retrieval accuracy.In view of the above characteristics and problems,this paper proposes a fusion BERT-BiLSTM model structure to fully express the semantic information of the electronic medical record text and improve the accuracy of retrieval.This research is based on public data.First,correlation extension retrieval keywords prerpocessing is carried on the open standard Chinese EMR data according to clinical diagnosis rules.Secondly,the BERT model is used to dynamically obtain the word granularity vector matrix according to the context of the medical record text,then the generated word vector is used as the input of the bidirectional long and short-term memory network model(BiLSTM) to extract the global semantic features of the context information.Finally,the feature vector of the retrieved document is mapped to the Euclidean space,and the medical record text closest to the retrieved document is found to realize the text retrieval of unstructured clinical data.Simulation results show that this method can dig out multi-level and multi-angle text semantic features from the medical record text,the F1 value obtained on the electronic medical record data set is 0.94,which can significantly improve the accuracy of text semantic retrieval.
Control Strategy Optimization of Medical CPS Cooperative Network
LIU Li, LI Ren-fa
Computer Science. 2022, 49 (6A): 39-43.  doi:10.11896/jsjkx.210300230
Abstract PDF(2750KB) ( 431 )   
References | Related Articles | Metrics
The information construction of hospitals has entered the intelligent era,more and more medical cyber physical systems(CPS) have been applied in hospitals.However,in the complication disease treatment scenario,medical CPS are not reliable enough because of the specialization of medical disciplines and the lack of medical knowledge base.In this paper,a collaborative architecture of medical CPS is proposed to improve the decision reliability of medical CPS.On the collaboration platform,CPS send cooperative tasks to intelligent units on the network,and the intelligent units assist CPS to make medical decisions together.In this paper,the control strategy of the cooperative network is optimized to improve the network communication efficiency because the physiological data of patients are continuous dynamic data and medical CPS have a high requirement on the timeliness of response.CCD and HCD algorithms are proposed respectively for the deployment of high-level controller and low-level controller.Finally,two algorithms are simulated and compared with K-means algorithm.The results show that HCD algorithm greatly improves the load balancing of low-level controllers at the expense of less average communication delay.CCD algorithm is more suita-ble for advanced controller deployment with fewer cluster nodes,and its optimization effect on objective function is obviously better than that of HCD algorithm and K-means algorithm.
Property Analysis Model of Pleural Effusion Based on Standardization of Pleural Effusion Ultrasonic Image
FENG Yi-fan, XU Qi, ZENG Wei-ming
Computer Science. 2022, 49 (6A): 44-53.  doi:10.11896/jsjkx.210700196
Abstract PDF(5466KB) ( 679 )   
References | Related Articles | Metrics
Pleural effusion is a complication of many major diseases.Invasive puncture and biochemical tests are the gold standard for diagnoising the property of pleural effusion.Therefore,a non-invasive pleural effusion analysis method is of great significance.A model based on standardization of pleural effusion ultrasonic image—Property analysis method of pleural effusion(PAMPE) is proposed.PAMPE can quickly and noninvasively classify three laboratory indexes:effusion color,effusion turbidity and Rivalta test.The construction of PAMPE is mainly divided into three steps:image standardization,construction of feature engineering and using v-SVM to build PAMPE after feature selection.In the image standardization step,a new standardization method—Standardi-zation of Pleural Effusion Ultrasonic Image(SOPEU) is also proposed.SOPEU suppresses the differences in the grayscale and scale of the images in the image set caused by the different parameters of ultrasound equipment,the different degree of obesity of patients,and the different degree to which pleural effusion is shielded by the bones and diaphragm.Experiment results illustrate that,PAMPE behaves well in a variety of evaluation indicators:accuracy,precision,recall,F1-score,confusion matrix,receiver operating characteristic(ROC) curve and area under ROC curve(AUC).Specifically,for the three classification problems,the accuracy can reach 0.800,0.743 and 0.719,the precision can reach 0.806,0.779 and 0.741,the recall can reach 0.921,0.815 and 0.893,the F1-score can reach 0.860,0.796 and 0.809 and the AUC can reach 0.820,0.700 and 0.709,which proves the effectiveness of PAMPE from different aspects.Comparative results shows that for the three classification problems,PAMPE has increased the accuracy of 0.090,0.048 and 0.086 respectively compared with the model constructed without SOPEU.The experimental results show that the normalized images effectively reduce the classification errors caused by the different quality of data sources.
Automatic Detection of Pulmonary Nodules in Low-dose CT Images Based on Improved CNN
YUE Qing, YIN Jian-yu, WANG Sheng-sheng
Computer Science. 2022, 49 (6A): 54-59.  doi:10.11896/jsjkx.210400211
Abstract PDF(2471KB) ( 747 )   
References | Related Articles | Metrics
With air pollution getting worse and worse,lung cancer has become one of the malignant tumors with the fastest increasing morbidity and mortality rate,which seriously endangers people's life and health.The early stage of lung cancer is mainly in the form of pulmonary nodules.If the early stage of lung cancer can be detected and treated in time,the treatment effect of lung cancer will be improved.Low-dose spiral CT is widely used in the diagnosis of pulmonary nodules because of its characteristics of fast acquisition speed,low cost and low radiation.At present,CT image diagnosis mostly adopts the traditional manual diagnosis and CAD system diagnosis,but these two methods have the disadvantages of low accuracy and poor generalization.In view of the above problems,this paper takes the detection of pulmonary nodules in the field of medical assisted diagnosis as the research object,and proposes an improved low-dose CT image automatic detection algorithm for pulmonary nodules based on CNN.Firstly,the CT images are preprocessed to extract the lung parenchyma.Secondly,the cascade-rcnn candidate nodule screening network is improved to extract higher quality targets.Thirdly,an improved 3D CNN false positive reduction network is proposed to improve the accuracy of nodular classification.Finally,experiments are carried out on Luna16 dataset.Compared with existing algorithms,the detection accuracy of the proposed algorithm is improved.
Alzheimer's Disease Classification Method Based on Attention Mechanism and Multi-task Learning
DU Li-jun, TANG Xi-lu, ZHOU Jiao, CHEN Yu-lan, CHENG Jian
Computer Science. 2022, 49 (6A): 60-65.  doi:10.11896/jsjkx.201200072
Abstract PDF(2784KB) ( 801 )   
References | Related Articles | Metrics
In recent years,using deep learning to classify Alzheimer's disease has become one of the hotspots in medical image research.But the existing models are difficult to extract medical image features effectively,what's more,the auxiliary information resources of disease classification are wasted.To solve these problems,a classification method of Alzheimer's disease with attention mechanism and multi-task learning based on the deep 3D convolution neural network is proposed.Firstly,using the improved traditional C3D network,a rough low-level feature map is generated.Secondly,this feature map is input into a convolution block with attention mechanism and a common convolution block respectively.The former focuses on the structural characteristics of MRI images,and can obtain the attention weight of different pixel in the feature map,which is multiplied by the output feature map of the latter.Finally,multi-task learning is used to obtain three kinds of outputs by adding different full connected layer.The other two outputs optimize the main classification task through back propagation in the training process.Experimental results show that,compared with the existing classification methods of Alzheimer's disease,the classification accuracy and other indicators of the proposed method on ADNI data set have been improved,which is helpful to promote the follow-up disease classification research.
Intelligent Computing
Survey of the Application of Natural Language Processing for Resume Analysis
LI Xiao-wei, SHU Hui, GUANG Yan, ZHAI Yi, YANG Zi-ji
Computer Science. 2022, 49 (6A): 66-73.  doi:10.11896/jsjkx.210600134
Abstract PDF(2488KB) ( 1577 )   
References | Related Articles | Metrics
With the rapid development of information technology and the dramatic growth of digital resources,enormous resumes is generated in the Internet.It is a concern of scholars to analyze the resumes of job seekers to obtain the information of various personnel of candidates,industry categories and job recommendations.The inefficiency of manual resume analysis has promoted the wide application of natural language processing(NLP) technology in resume analysis.NLP can realize automated analysis of resumes by using artificial intelligence and computer technology to analyze,understand and process natural language.This paper systematically reviews the relevant literature in the past ten years.Firstly,the natural language processing is introduced.Then based on the principal line of resume analysis in NLP,the recent works in three aspects:resume information extraction,resume classification and resume recommendation are generalized.Finally,discussing the future development trend in this research area and summarizing the paper.
Review of Reasoning on Knowledge Graph
MA Rui-xin, LI Ze-yang, CHEN Zhi-kui, ZHAO Liang
Computer Science. 2022, 49 (6A): 74-85.  doi:10.11896/jsjkx.210100122
Abstract PDF(2519KB) ( 3052 )   
References | Related Articles | Metrics
In recent years,the rapid development of Internet technology and reference models has led to an exponential growth in the scale of computer world data,which contains a lot of valuable information.How to select knowledge from it,and organize and express this knowledge effectively attract wide attention.Knowledge graphs are also born from this.Knowledge reasoning for knowledge graphs is one of the hotspots of knowledge graph research,and important achievements are obtained in the fields of semantic search and intelligent question answering.However,due to various defects in the sample data,such as the lack of head and tail entities in sample data,the long query path,as well as the wrong sample data.In the face of the above characteristics,the knowledge graph reasoning of zero-shot,one-shot,few-shot and multi-shot get more attention.Based on the basic concepts and basic knowledge of knowledge graph,this paper introduces the latest research progress of knowledge graph reasoning methods in recent years.Specifically,according to the size of sample data,the knowledge graph reasoning method is divided into multi-shot reasoning,few-shot reasoning,zero-shot and single-shot reasoning.Models that use more than five instances for reasoning are multi-sample reasoning,models that use two to five instances for reasoning are few-shot reasoning,and those use zero or one instance number for reasoning are zero-shpt and one-shot reasoning.The multi-shot knowledge graph reasoning is subdivided into rule-based reasoning,distributed-based reasoning,neural network-based reasoning,and other reasoning.The few-shot knowledge graph reasoning is subdivided into meta-learning-based reasoning and neighboring entity information-based reasoning.And these methods are analyzed and summarized.In addition,this paper further describes the typical application of knowledge graph reaso-ning,and discusses the existing problems,future research directions and prospects of knowledge graph reasoning.
Survey on Bayesian Optimization Methods for Hyper-parameter Tuning
LI Ya-ru, ZHANG Yu-lai, WANG Jia-chen
Computer Science. 2022, 49 (6A): 86-92.  doi:10.11896/jsjkx.210300208
Abstract PDF(1927KB) ( 1646 )   
References | Related Articles | Metrics
For most machine learning models,hyper-parameter selection plays an important role in obtaining high quality models.In the current practice,most of the hyper-parameters are given manually.So the selection or estimation of hyper-parameters is an key issue in machine learning.The mapping from hyper-parameter set to the modeĹs generalization can be regarded as a complex black box function.The general optimization method is difficult to apply.Bayesian optimization is a very effective global optimization algorithm,which is suitable for solving optimization problems in which their objective functions could not be expressed,or the functions are non-convex,computational expensive.The ideal solution can be obtained with a few function evaluations.This paper summarizes the basics of the Bayesian optimization based on hyper-parameter estimation methods,and summarizes the research hot spots and the latest developments in the recent years,including the researches in agent model,acquisition function,algorithm implementation and so on.And the problems to be solved in existing research are summarized.It is expected to help beginners quickly understand Bayesian optimization algorithms,understand typical algorithm ideas,and play a guiding role in future researches.
Review of Multi-instance Learning Algorithms
ZHAO Lu, YUAN Li-ming, HAO Kun
Computer Science. 2022, 49 (6A): 93-99.  doi:10.11896/jsjkx.210500047
Abstract PDF(3279KB) ( 770 )   
References | Related Articles | Metrics
Multi-instance learning(MIL) is a typical weakly supervised learning framework,where every training example,called bag,is a set of instances.Since the learning process of an MIL algorithm depends on only the labels of bags rather than those of any individual instances,MIL can fit well with applications in which instance labels are difficult to get.Recently,deep multi-instance learning methods attract widespread attention,so deep MIL has become a major research focus.This paper reviews some research progress of MIL.Firstly,MIL algorithms are divided into shallow and deep models according to their hierarchical structure.Secondly,various algorithms are reviewed and summarized in these two categories,and then different pooling methods of deep MIL models are analyzed.Moreover,the fundamental theorem of symmetric functions for models with set-type data as training samples and its application in deep MIL are expounded.Finally,the performance of different algorithms is compared and analyzed through experiments,and their interpretability is analyzed thoroughly.After that,problems to be further investigated are discussed.
Fast and Transmissible Domain Knowledge Graph Construction Method
DENG Kai, YANG Pin, LI Yi-zhou, YANG Xing, ZENG Fan-rui, ZHANG Zhen-yu
Computer Science. 2022, 49 (6A): 100-108.  doi:10.11896/jsjkx.210900018
Abstract PDF(5171KB) ( 1042 )   
References | Related Articles | Metrics
Domain knowledge graph can clearly and visually represent domain entity relations,acquire knowledge efficiently and accurately.The construction of domain knowledge graph is helpful to promote the development of information technology in rela-ted fields,but the construction of domain knowledge graph requires huge manpower and time costs of experts,and it is difficult to migrate to other fields.In order to reduce the manpower cost and improve the versatility of knowledge graph construction me-thod,this paper proposes a general construction method of domain knowledge graph,which does not rely on a large of artificial ontology construction and data markup.The domain knowledge graph is constructed through four steps:domain dictionary construction,data acquisition and cleaning,entity linking and maintenance,and graph updating and visualization.This paper takes the domain of network security as an example to construct the knowledge graph and details the build process.At the same time,in order to improve the domain correlation of entities in the knowledge graph,a fusion model based on BERT(Bidirectional Encoder Representations from Transformers) and attention mechanism model is proposed in this paper.The F-score of this model in text classification is 87.14%,and the accuracy is 93.51%.
Redundant Literals of Ternary Clause Sets in Propositional Logic
LI Jie, ZHONG Xiao-mei
Computer Science. 2022, 49 (6A): 109-112.  doi:10.11896/jsjkx.210700036
Abstract PDF(2901KB) ( 463 )   
References | Related Articles | Metrics
Automatic reasoning is one of the core issues in the field of artificial intelligence.Since a large number of redundant li-terals and redundant clauses are generated in the process of automatic reasoning based on resolution,the resolution efficiency will be affected.It is of great significance to eliminate redundant literals and redundant clauses in the clause set.In propositional logic,according to the related concepts and properties of necessary literals,useful literals and useless literals,this paper classifies and gives the judgment methods of redundant literals in some ternary clause sets,and explains these judgment methods through specific examples.
Active Metric Learning Based on Support Vector Machines
HOU Xia-ye, CHEN Hai-yan, ZHANG Bing, YUAN Li-gang, JIA Yi-zhen
Computer Science. 2022, 49 (6A): 113-118.  doi:10.11896/jsjkx.210500034
Abstract PDF(3241KB) ( 591 )   
References | Related Articles | Metrics
Metric learning is an important issue in machine learning.The measuring results will significantly affect the perfor-mance of machine learning algorithms.Current researches on metric learning mainly focus on supervised learning problems.How-ever,in real world applications,there is a large amount of data that has no label or needs to pay a high price to get labels.To handle this problem,this paper proposes an active metric learning algorithm based on support vector machines(ASVM2L),which can be used for semi-supervised learning.Firstly,a small size of samples randomly selected from the unlabeled dataset are labeled by oracles,and then these samples are used to train the support vector machine metric learner(SVM2L).According to the output measuring result,the rest unlabeled samples are classified by K-NN classifiers with different values of K,and the sample with the largest voting differences is selected and submitted to the oracle to get a label.Then,the sample is added to the training set to retrain the ASVM2L model.Repeating the above steps until the termination condition is met,then the best metric matrix can be obtained from the limited labeled samples.Comparative experiments on the standard datasets verify that the proposed ASVM2L algorithm can obtain more information with the least labeled samples without affecting the classification accuracy,and therefore has better measuring performance.
Relation Classification Method Based on Cross-sentence Contextual Information for Neural Network
HUANG Shao-bin, SUN Xue-wei, LI Rong-sheng
Computer Science. 2022, 49 (6A): 119-124.  doi:10.11896/jsjkx.210600150
Abstract PDF(2150KB) ( 649 )   
References | Related Articles | Metrics
Information extraction is a technique of extracting specific information from textual data.It has been widely used in knowledge graph,information retrieval,question answering system,sentiment analysis and text mining.As the core task and important part of information extraction,relation classification can realize the recognition of semantic relations between entities.In recent years,deep learning has made remarkable achievements in relation extraction tasks.So far,researchers have focused their efforts on improving neural network models,but there is still a lack of effective methods to obtain cross-sentence semantic information from paragraphs or discourse level texts with close semantic relationships between different sentences.However,semantic relationships between sentences for relation extraction tasks are of great use.In this paper,for such paragraphs or discourse level relation extraction datasets,a method to combine sentences with their cross-contextual information as the input of the neural network model is proposed,so that the model can learn more semantic information from paragraphs or discourse level texts.Cross-sentence contex tual information is introduced into different neural network models,and experiments are carried out on two relation classification datasets in different fields including San Wen dataset and Policy dataset.The effects of cross-sentence contex-tual information on model accuracy are compared.The experiment show that the proposed method can effectively improve the performance of relation classification models including Convolutional neural network,bidirectional long short-term memory network,attention-based bidirectional long short-term memory network and convolutional recurrent neural network.In addition,this paper proposes a relation classification dataset named Policy based on the texts of policies and regulations in the field of four social insurance and one housing fund,which is used to verify the necessity of introducing cross-sentence contextual information into the relation classification tasks in some practical fields.
Hybrid Improved Flower Pollination Algorithm and Gray Wolf Algorithm for Feature Selection
KANG Yan, WANG Hai-ning, TAO Liu, YANG Hai-xiao, YANG Xue-kun, WANG Fei, LI Hao
Computer Science. 2022, 49 (6A): 125-132.  doi:10.11896/jsjkx.210600135
Abstract PDF(2655KB) ( 575 )   
References | Related Articles | Metrics
Feature selection is very important in the stage of data preprocessing.The quality of feature selection not only affects the training time of the neural network but also affects the performance of the neural network.Grey Wolf improved Flower pollination algorithm(Grey Wolf improved Flower pollination algorithm,GIFPA) is a hybrid algorithm based on the fusion of flower pollination algorithm framework and gray wolf optimization algorithm.When it is applied to feature selection,it can not only retain the connotation information of the original features but also maximize the accuracy of classification features.The GIFPA algorithm adds the worst individual information to the FPA algorithm,uses the cross-pollination stage of the FPA algorithm as the global search,uses the hunting process of the gray wolf optimization algorithm as the local search,and adjusts the search process of the two through the conversion coefficient.At the same time,to overcome the problem that swarms intelligence algorithm is easy to fall into local optimization,this paper uses the RelifF algorithm in the field of data mining to improve this problem and uses the RelifF algorithm to filter out high weight features and improve the best individual information.To verify the performance of the algorithm,21 classical data sets in the UCI database are selected for testing,k-nearest neighbor(KNN) classifier is used for classification and evaluation,fitness value and accuracy are used as evaluation criteria,and K-fold crossover verification is used to overcome the over-fitting problem.In the experiment,a variety of classical algorithms and advanced algorithms,including the FPA algorithm,are compared.The experimental results show that the GIFPA algorithm has strong competitiveness in feature selection.
Construction of Named Entity Recognition Corpus in Field of Military Command and Control Support
DU Xiao-ming, YUAN Qing-bo, YANG Fan, YAO Yi, JIANG Xiang
Computer Science. 2022, 49 (6A): 133-139.  doi:10.11896/jsjkx.210400132
Abstract PDF(3935KB) ( 815 )   
References | Related Articles | Metrics
The construction of the knowledge graph in the field of military command and control support is an important research direction in the process of the military information equipment support.Aiming at the current situation that the named entity re-cognition model lacks the corresponding basic training corpus in the construction of the guarantee domain knowledge graph,based on the analysis of the relevant research status,this paper designs and implements a GUI named entity recognition corpus construction system based on the basic framework of the PyQt5 application program.First,it briefly describes the overall system architecture and corpus processing technical process.Secondly,it introduces the system's data preprocessing,labeling system,automatic labeling,labeling analysis and coding conversion related content in five major functional modules.Among them,the automatic labeling function module is automatic.The implementation of automatic labeling and the realization of automatic de-duplication algorithm is the most important and difficult point,and also is the core of the entire system.Finally,the graphical user interface of each functional module is implemented through the basic framework of the PyQt5 application program and various functional components.The design and implementation of this system can automatically process various original equipment manuals on military computers,and quickly generate the corpus required for named entity recognition model training,so as to provide effective technical support for the subsequent construction of the corresponding domain knowledge graph.
Topological Properties of Fuzzy Rough Sets Based on Residuated Lattices
XU Si-yu, QIN Ke-yun
Computer Science. 2022, 49 (6A): 140-143.  doi:10.11896/jsjkx.210200123
Abstract PDF(3009KB) ( 440 )   
References | Related Articles | Metrics
This paper is devoted to the study of the topological structure of L-fuzzy rough sets based on residuated lattices.The L-fuzzy topologies induced by the lower approximation operators determined by fuzzy implication operators are presented and its basic properties being discussed.The knowledge of the L-fuzzy approximation space is a general L-fuzzy relation,and there is no need to assume its reflexivity and strong seriality.Based on the transitive closures of the L-fuzzy relations,the interior operators and closure operators of the corresponding L-fuzzy topologies are constructed.The relationships among L-fuzzy topologies induced by lower approximation operators corresponding to different L-fuzzy relations are investigated,and a classification method for L-fuzzy relations is presented by using related topologies.
Aspect-level Sentiment Classification Based on Imbalanced Data and Ensemble Learning
LIN Xi, CHEN Zi-zhuo, WANG Zhong-qing
Computer Science. 2022, 49 (6A): 144-149.  doi:10.11896/jsjkx.210500205
Abstract PDF(2680KB) ( 601 )   
References | Related Articles | Metrics
Sentiment classification remains an important part of the field of natural language processing.The general task is to classify the emotional data into two categories,which is positive and negative.In many models,it is assumed that the positive and negative data are balanced.Contrarily,the two class of data are always imbalanced in reality.This paper proposes an ensemble learning model based on aspect-levelLSTM to process aspect-level problem.Firstly,the data sets are under-sampled and divided into multiple groups.Secondly,a classification algorithm is assigned to each group of data for training.Finally,it yields the classification result through joining all models.The experimental results show that the ensemble learning model based on aspect-level LSTM significantly improves the accuracy of classification,and its performance is better than the traditional LSTM model.
Deep Integrated Learning Software Requirement Classification Fusing Bert and Graph Convolution
KANG Yan, WU Zhi-wei, KOU Yong-qi, ZHANG Lan, XIE Si-yu, LI Hao
Computer Science. 2022, 49 (6A): 150-158.  doi:10.11896/jsjkx.210500065
Abstract PDF(3362KB) ( 655 )   
References | Related Articles | Metrics
With the rapid growth of software quantity and types,effectively mine the textual features of software requirements and classify the textual features of software functional requirements becomes a major challenge in the field of software enginee-ring.The classification of software functional requirements provides a reliable guarantee for the whole software development process and reduces the potential risks and negative effects in the requirements analysis stage.However,the validity of software requirement analysis is limited by the high dispersion,high noise and sparse data of software requirement text.In this paper,a two-layer lexical graph convolutional network model(TVGCCN) is proposed to model the graph of software requirement text innovatively,build the graph neural network of software requirement,and effectively capture the knowledge edge of words and the relationship between words and text.A deep integrated learning model is proposed,which integrates several deep learning classification models to classify software requirement text.In experiments of data set Wiodows_A and data Wiodows_B,the accuracy of deep ensemble learning model integrating Bert and graph convolution reaches 96.73% and 95.60% respectively,which is ob-viously better than that of other text classification models.It is fully proved that the deep ensemble learning model integrating Bert and graph convolution can effectively distinguish the functional characteristics of software requirement text and improve the accuracy of software requirement text classification.
Solve Data Envelopment Analysis Problems with Particle Filter
HUANG Guo-xing, YANG Ze-ming, LU Wei-dang, PENG Hong, WANG Jing-wen
Computer Science. 2022, 49 (6A): 159-164.  doi:10.11896/jsjkx.210600110
Abstract PDF(2084KB) ( 441 )   
References | Related Articles | Metrics
Data envelopment analysis is a method to evaluate the production efficiency of multi-input&multi-output decision ma-king units.The data envelopment analysis method is widely used to solve efficiency analysis problems in various fields.However,the current methods for solving data envelopment analysis problems mainly use some specialized software to solve the problem,and the entire process requires a high specialization.In order to solve the data envelopment analysis problem conveniently,the optimization philosophy is used to solve the data envelopment analysis problem.In this paper,an optimization method based on particle filter is proposed for solving the data envelopment analysis problem.Firstly,the basic principles of the particle filter method are systematically interpreted.Then the optimization problem of the data envelopment analysis is transformed into the minimum variance estimate problem of particle filter.Therefore,the basic principles of particle filter can be used to solve the optimization problem of data envelopment analysis to obtain a global optimal solution.Finally,several simulation examples are conducted to verify the effectiveness of the proposed method.The simulation results show that the optimization method based on particle filter can accurately and effectively solve the problem of data envelopment analysis.
TS-AC-EWM Online Product Ranking Method Based on Multi-level Emotion and Topic Information
YU Ben-gong, ZHANG Zi-wei, WANG Hui-ling
Computer Science. 2022, 49 (6A): 165-171.  doi:10.11896/jsjkx.210400238
Abstract PDF(2255KB) ( 434 )   
References | Related Articles | Metrics
The information of e-commerce platforms has a significant impact on consumers' purchase decisions.It is of great research value to integrate the information of large-scale stores,commodity information and online review information and get online commodity ranking to assist consumers in purchasing decisions.To solve the problems,this paper proposes an online product ranking method TS-AC-EWM,which integrates multi-level emotion and topic information,and makes full use of scoring information and review content information.Firstly,the online commodity ranking evaluation system is designed from two dimensions of measurement and content,including four measurement indexes and three content indexes.Secondly,we crawl the measurement indexes and online review content of each candidate commodity.Thirdly,three content indexes are calculated by TS method,which combines topic and affective information,and AC method,which is based on appending comments.Finally,using the entropy weight method to calculate the index weight,commodity grading and sorting.Experiments on Jingdong microwave oven dataset prove the feasibility and effectiveness of the proposed method,so the ranking method has a practical significance.
Automatic Generation of Patent Summarization Based on Graph Convolution Network
LI Jian-zhi, WANG Hong-ling, WANG Zhong-qing
Computer Science. 2022, 49 (6A): 172-177.  doi:10.11896/jsjkx.210400117
Abstract PDF(2170KB) ( 507 )   
References | Related Articles | Metrics
The patent specification contains much useful information.However,due to the long space,it is difficult to obtain effective information quickly.Patent summarization is a summary of a complete patent specification.The right-claiming document determines the scope of protection of the patent application documents.It found that there is a special structure in the right-clai-ming document.Therefore,this paper proposes a method of automatic generation of patent summarization based on graph convolution network.The patent summarization is generated through the patent right-claiming document and its structural information.Firstly,this model obtains patent structural information,and the graph convolution neural network is introduced in the encoder to fuse the serialization information and structural information,to improve the quality of summarization.Experimental results show that this method has a significant improvement in ROUGE evaluation compared with the current main stream extractive summarization method and the traditional encoder-decoder abstractive summarization.
Projected Gradient Descent Algorithm with Momentum
WU Zi-bin, YAN Qiao
Computer Science. 2022, 49 (6A): 178-183.  doi:10.11896/jsjkx.210500039
Abstract PDF(2413KB) ( 533 )   
References | Related Articles | Metrics
In recent years,deep learning is widely used in the field of computer vision and has achieved outstanding success.However,the researchers found that the neural network is easily disturbed by adding subtle perturbations in the dataset,that can cause the model to give incorrect outputs.Such input examples are called “adversarial examples”.At present,a series of algorithms for generating adversarial examples have emerged.Based on the existing adversarial sample generation algorithm-projected gradient descent(PGD),this paper proposes an improved method-MPGDCW algorithm,which combines momentum and adopts a new loss function to ensure the stability of the update direction and avoid bad local maximums.At the same time,it can avoid the disappearance of the gradient by replacing the cross-entropy loss function.Experiments on 4 robust models containing 3 architecturesconfirm that the proposed MPGDCW algorithm has better attack effect and stronger transfer attack capacity.
Data Debiasing Method Based on Constrained Optimized Generative Adversarial Networks
XU Guo-ning, CHEN Yi-peng, CHEN Yi-ming, CHEN Jin-yin, WEN Hao
Computer Science. 2022, 49 (6A): 184-190.  doi:10.11896/jsjkx.210400234
Abstract PDF(3438KB) ( 589 )   
References | Related Articles | Metrics
With the wide application of deep learning technology in image recognition,natural language processing and financial predicting,once there is bias in analysis results,it will cause negative impacts both on individuals and groups,thus any effects on its performance it is vital to enhance the fairness of the model without affecting the perfomance of deep learning model.Biased information about data is not only sensitive attributes,and non-sensitive attributes will also contain bias due to the correlation among attributes,therefore,the bias cannot be eliminated when debiasing algorithms only consider sensitive attributes.In order to eliminate the bias in the classification results of the deep learning model caused by the correlated sensitive attributions in the data,this paper proposes a data debiasing method based on the generative adversarial network.The loss function of the model combines the fairness constraints and the accuracy loss,and the model utilizes adversarial code to eliminate bias to generate debiased dataset,then with the alternating gaming training of the generator and the discriminator to reduce the loss of the no-bias information in the dataset,and the classification accuracy is ensured while the bias in the data is eliminated to improve the fairness of the subsequent classification tasks.Finally,data debiasing experiments are carried out on several real-world dataset to verify the effectiveness of the proposed algorithm.The results show that the proposed method can effectively decrease the bias information in datasets and generate datasets with less bias.
Vehicle Routing Problem with Time Window of Takeaway Food ConsideringOne-order-multi-product Order Delivery
YANG Hao-xiong, GAO Jing, SHAO En-lu
Computer Science. 2022, 49 (6A): 191-198.  doi:10.11896/jsjkx.210400005
Abstract PDF(2375KB) ( 665 )   
References | Related Articles | Metrics
The rapid growth of take-out food transaction makes take-out food develop fast and becomes a kind of new demand in consumers' market.With more and more transactions in take-out food order volumes,consumers require more on the basis of fundamental take-out food delivery service.The demand of consumers for take-out food is becoming increasingly various,which captures the structural characteristic that one take-out food order can be composed of different kinds of food provided by two or more different food merchants.Under the background of one-order-multi-product for take-out food delivery,aiming at the problem of takeaway order delivery with time window,this paper studies vehicle routing planning for delivery.This application can improve the performance of merchant service level and efficiency of delivery vehicles.Food merchants accept orders from consumers via the online food-selling platform,then prepare food.The delivery vehicle will come and pick up the food in the specific time window and send to consumers.Then this paper constructs the objective function for the mathematical model considering the lowest delivery cost during the whole delivery process,and set the time window limits of entity merchants and consumer.The genetic algorithm is used to solve the problem of take-out order delivery.Finally,the validity and feasibility of the mathematical model are verified by an example experiment.At last,suggestions on practical management and enlightenments on vehicle path planning problem are given from the perspective of practice.
Cutting Edge Method for Traveling Salesman Problem Based on the Shortest Paths in Optimal Cycles of Quadrilaterals
WANG Yong, CUI Yuan
Computer Science. 2022, 49 (6A): 199-205.  doi:10.11896/jsjkx.210400065
Abstract PDF(2048KB) ( 440 )   
References | Related Articles | Metrics
With the expansion of traveling salesman problem,the search space for optimal solution on the complete graph increases exponentially.The cutting edge algorithm is proposed to reduce the search space of the optimal solution for the traveling salesman problem.It proves that the probability that an optimal Hamiltonian cycle edge contained in the shortest paths in the optimal cycles of quadrilaterals is different from that for a common edge.A number of shortest paths in the optimal cycles of quadrilaterals are selected to compute the edge frequency of edges.As the edges are cut according to the average edge frequency of all edges,the retention probability that an optimal Hamiltonian cycle edge is derived based on the constructed binomial distributions.Given Knof a traveling salesman problem,there are four steps for eliminating edges.Firstly,a finite number of quadrilaterals containing each edge are chosen.Secondly,the shortest path in the optimal cycle of every selected quadrilateral is used to compute the edge frequency.Thirdly,5/6 of edges having the smallest edge frequencies are cut.In the last step,some edges are added for vertices of degree below 2 according to the edge frequency.Experiments illustrate that the preserved edges occupy 1/6 of the total edges.Moreover,the computation time of the exact algorithms for the traveling salesman problem on the remaining graphs is reduced to some extent.
TI-FastText Automatic Goods Classification Algorithm
SHAO Xin-xin
Computer Science. 2022, 49 (6A): 206-210.  doi:10.11896/jsjkx.210500089
Abstract PDF(2547KB) ( 820 )   
References | Related Articles | Metrics
In order to achieve automatic classification of goods according to title information,a Chinese words goods classification algorithm based on TF-IDF(term frequency-inverse document frequency) and FastText is proposed.In this algorithm,the lexicon is represented as a prefix tree by FastText.The TF-IDF filting is performed on the dictionary processed by n-grammar model.Thus,the high group degree of the entries is biased in the process of computing the mean value of input word sequence vectors,making them more suitable for the Chinese short text classification environment.This paper uses Anaconda platform to implement and optimize the product classification algorithm based on FastText.After evaluation,the algorithm has a high accuracy rate and can meet the needs of goods classification on e-commerce platforms.
Fishing Type Identification of Marine Fishing Vessels Based on Support Vector Machine Optimized by Improved Sparrow Search Algorithm
SHAN Xiao-ying, REN Ying-chun
Computer Science. 2022, 49 (6A): 211-216.  doi:10.11896/jsjkx.220300216
Abstract PDF(3833KB) ( 529 )   
References | Related Articles | Metrics
The identification of fishing type has significance for monitoring the fishing activities of motor vessels and maintaining the marine ecological balance.To protect the marine environment and improve the supervision efficiency of fishing vessels,a fi-shing type identification algorithm based on support vector machine optimized by the improved sparrow search algorithm (ISSA-SVM) is proposed.First,the t-distribution mutation operator is introduced to optimize the population selection,which improves the global search ability and local development ability of the original SSA.Second,the position update formula of the spectators of SSA is modified to further improve the convergence speed of the algorithm.Finally,the fishing type identification model ISSA-SVM is constructed by using ISSA to optimize the parameters of SVM.The experimental results on 3 546 fishing vessels show that compared with SVM,PSO-SVM,GWO-SVM and SSA-SVM,the fishing type identification model of ISSA-SVM proposed in this paper has higher accuracy and faster convergence speed.
Improved Sparrow Search Algorithm Based on A Variety of Improved Strategies
LI Dan-dan, WU Yu-xiang, ZHU Cong-cong, LI Zhong-kang
Computer Science. 2022, 49 (6A): 217-222.  doi:10.11896/jsjkx.210700032
Abstract PDF(4076KB) ( 897 )   
References | Related Articles | Metrics
To solve the shortcomings of sparrow search algorithm,such as slow convergence speed,easy to fall into local optimal value and low optimization precision,an improved sparrow search algorithm(IM-SSA) based on Various improvement strategies is proposed.Firstly,the initial population of the sparrow search algorithm is enriched by Tent chaotic sequence,which expand the search area.Then,the adaptive crossover and mutation operator is introduced into the finders to enrich the diversity of the producers population and balance the global and local search ability of the algorithm.Secondly,the t-distribution perturbation or differential mutation is used to perturbate the population after each iteration according to the individual characteristics,which can avoid the population singularity in the later stage of the algorithm and enhance the ability of jump out of the local optimal value of the algorithm.Finally,the IM-SSA algorithm proposed in this paper,gray Wolf algorithm,particle swarm optimization algorithm,whale algorithm and classical sparrow search algorithm are used to simulate the eight test functions,respectively.Through the comparative analysis of simulation results,it can be concluded that the IM-SSA algorithm has faster convergence speed,stronger ability to get out of local optimal value and higher optimization precision than the other four algorithms.Compared the simulation results of the IM-SSA algorithm with the ones of the existing improved sparrow search algorithm,it is found that the strategy of IM-SSA algorithm proposed in this paper is better.
Study on Computing Capacity of Novel Numerical Spiking Neural P Systems with MultipleSynaptic Channels
YIN Xiu, LIU Xi-lin, LIU Xi-yu
Computer Science. 2022, 49 (6A): 223-231.  doi:10.11896/jsjkx.210200171
Abstract PDF(3646KB) ( 429 )   
References | Related Articles | Metrics
Membrane system,also known as P system,are a distributed parallel computing model.The P systems can be roughly divided into three types:cell-like,tissue-like and neural-like.Numerical spiking neural P systems(NSN P systems) gains the ability to process numerical information by introducing numerical variables and production functions in Numerical P systems(NP systems).Based on NSN P systems,this paper proposes novel numerical spiking neural P systems with multiple synaptic channels(MNSN P systems).In MNSN P systems,each production function is assigned a threshold to control firing and each neuron has one or more synaptic channels to transmit the production value.This paper mainly studies the computing power of MNSN P systems,i.e.,through the simulation of register machines,it is proved that MNSN P systems are Turing universal as a number gene-rating/accepting device and construct a universal MNSN P system containing 70 neurons to compute functions.
Big Data & Data Science
Clustered Federated Learning Methods Based on DBSCAN Clustering
LU Chen-yang, DENG Su, MA Wu-bin, WU Ya-hui, ZHOU Hao-hao
Computer Science. 2022, 49 (6A): 232-237.  doi:10.11896/jsjkx.211100059
Abstract PDF(2918KB) ( 1081 )   
References | Related Articles | Metrics
Federated learning is to solve the problem of data fragmentation and isolation in machine learning based on privacy protection.Each client node trains the data locally and uploads the model parameter information to the central server,which aggregates the parameter information to achieve the purpose of common training.In the real environment,the distribution of data among nodes is often inconsistent.By analyzing the influence of independent identically distributed data on the accuracy of federated learning,it is proved that the accuracy of the model obtained by the traditional federated learning method is low.Therefore,a diversified sampling strategy is adopted to simulate the data inclination distribution,and a Clustered Federated Learning Methods algorithm based on DBSCAN clustering(DCFL) is proposed,which solves the problem that the learning accuracy is reduced when the data of different nodes are not independently and identically distributed in federated learning.Through the experimental comparison of Mnist and Cifar-10 standard data sets,compared with the traditional federated learning algorithm,DCFL can greatly improve the accuracy of the model.
Improved Collaborative Filtering Algorithm Combining Similarity and Trust
CAI Xiao-juan, TAN Wen-an
Computer Science. 2022, 49 (6A): 238-241.  doi:10.11896/jsjkx.210400088
Abstract PDF(1993KB) ( 558 )   
References | Related Articles | Metrics
The rapid development of e-commerce not only gives consumers more choice but has also causes information overload.As an indispensable method in information filtering technology,recommendation system has been widely concerned by the society.Collaborative filtering algorithm is the most widely used technology in recommendation systems,but it faces problems such as data sparsity,cold start and data scalability.This paper proposes an improved collaborative filtering algorithm model based on the fusion of trusted values and user similarity.This algorithm comprises three steps:first,we calculate the trust values between users;then we calculate the similarity between users;at last,we integrate the trust and the similarity to re-calculate the trust value between users and get the final rating prediction equation.Experimental results show that for different neighborhood sets,the performance of the proposed algorithm is better than that of traditional collaborative filtering algorithms.
Adaptive Ensemble Ordering Algorithm
WANG Wen-qiang, JIA Xing-xing, LI Peng
Computer Science. 2022, 49 (6A): 242-246.  doi:10.11896/jsjkx.210200108
Abstract PDF(3354KB) ( 495 )   
References | Related Articles | Metrics
Ordinal variables are used to express people's attitudes and preferences towards things.For example,in recommendation system,consumers' grades for goods are ordinal variables,and the emotion in sentiment analysis of NLP is also ordinal variables.At present,the ordered Logit model is adoptedto deal with the ordinal variables.However,the ordered Logit regression mo-del requires that theordinal variables generally follow uniform distribution.When theordinal variables do not conform to uniform distribution,the prediction result of the ordered Logit regression is not ideal.Based on this,this paper proposes an adaptive ensemble ordering algorithm.Firstly,this paper proposes a boosting-like algorithm with the aid of the idea of boosting.According to the concept of the ordered Logit regression model,the ordered multi-layer perceptron model and the ordered random fo-rest model are constructed.The two models,combined with the Softmax multi classification model and the ordered Logit model,constitute a boosting-like algorithm.In data processing,when the prediction values of the four models are not identical,the sample enters the boosting-like model and continues to train until the number of training rounds exceeds a certain threshold.Then,the random fo-rest model is adopted to construct the mapping function from all the predicted values of the training set to the real values.The proposed algorithm has a high prediction accuracy when the ordered variables are arbitrarily distributed,which greatly improves the application scope of the ordered Logit regression model.When the proposed algorithm is applied to the Baijiu quality datasets and the red wine quality datasets,its prediction accuracy is superior to that of the ordered Logit model and Softmax algorithm,Multi-layer Perceptron and KNN.
Relation Prediction for Railway Travelling Group Based on Hidden Markov Model
WANG Xin, XIANG Ming-yue, LI Si-ying, ZHAO Ruo-cheng
Computer Science. 2022, 49 (6A): 247-255.  doi:10.11896/jsjkx.210500001
Abstract PDF(2440KB) ( 463 )   
References | Related Articles | Metrics
In recent years,with the continuous development of transportation network as well as technology in high-speed railway,the speed and comfort of railway travel have been greatly improved,more and more people choose to travel by railway.As a result,co-travel behaviors have become even common in rail trips.The travel behavior of passengers can be influenced by their peers,and different travel groups will present different travel preferences.For example,for a travelling group with family mem-bers,the elderly and children will be taken good care of,hence group members are more inclined to pursue comfort during the trip.When a few young people who are mutual friends form a travelling group,they care more about the sense of experience and freshness.Therefore,predicting the type of a travel group will be beneficial for learning travel preference of this group,e.g.,not only help transportation,tourism and other related industries to define their products and services that travel groups interest in,but also provide support for market decision-making in the railway transportation industry.Based onthis,this paper proposes a methodology for analyzing railway passengers'travelling behavioral using ticket booking data.Firstly,based on ticket booking data,it proposes the quantitative method of co-travel times of a travelling group.Secondly,it formalizes the prediction problem by incorporating Hidden Markov Model.Lastly,the accuracy and consistency of the model are verified with real-life data and experiment results show that the accuracy of our model can even reach 96.38%,in the meanwhile,the consistency is as high as 95%.Thus,we conclude that the proposed method can effectively and accurately predict the relationship of railway travel groups.
SDFA:Study on Ship Trajectory Clustering Method Based on Multi-feature Fusion
YU Shu-hao, ZHOU Hui, YE Chun-yang, WANG Tai-zheng
Computer Science. 2022, 49 (6A): 256-260.  doi:10.11896/jsjkx.211100253
Abstract PDF(2569KB) ( 876 )   
References | Related Articles | Metrics
With the rapid development of ocean transportation,the technology of vessel trajectory mining and analysis has become more and more important.Trajectory clustering has many practical applications in the ship field,such as anomaly detection,position prediction,ship collision avoidance and so on.Traditional trajectory similarity calculation methods are relatively low in accuracy and efficiency,and most existing deep learning methods only extract features of static ones,ignoring the multi-feature combination of dynamic and static features.In order to solve the problem,a static-dynamic-feature fusion model based on convolutional auto-encoder is proposed,which can extract more perfect trajectory features.It makes up for the deficiency of multi-feature fusion technique in vessel trajectory clustering.Experiments on real datasets have demonstrated that compared with traditional methods such as LCSS,DTW and multi-feature extraction model based on deep learning,the proposed model has at least 5%~10% improvement in metrics such as precision,accuracy and so on.
Study on Prediction of Educational Statistical Data Based on DE-LSTM Model
LIU Bao-bao, YANG Jing-jing, TAO Lu, WANG He-ying
Computer Science. 2022, 49 (6A): 261-266.  doi:10.11896/jsjkx.220300120
Abstract PDF(2735KB) ( 614 )   
References | Related Articles | Metrics
At present,educational data presents the characteristics of large amount of data and diverse data types.Accurate and effective analysis and prediction of educational statistical data,which has important reference value for the formulation of relevant policies in education sector and social development.In this paper,DE-LSTM model is proposed,which takes the annual enrollment of a city as the data basis.The proposed model uses differential evolution algorithm to optimize the hidden layer nodes and lear-ning rate in the long-term and short-term memory neural network and has the better prediction performance in compared with the LSTM and BP models.Furthermore,effectiveness of the proposed DE-LSTM model is verified by a large number of simulation experiments.
Recommendation of Android Application Services via User Scenarios
WANG Yi, LI Zheng-hao, CHEN Xing
Computer Science. 2022, 49 (6A): 267-271.  doi:10.11896/jsjkx.210700123
Abstract PDF(3370KB) ( 421 )   
References | Related Articles | Metrics
With the development of mobile hardware and 5G communication technologies,smart applications are booming,which has penetrated into all the aspects of our life and work.A large number of Android applications not only meet the needs of people's daily life,but also make people need to spend more time to find the applications they want to start.In order to let users quickly find the application they want to start and perform the target function,this paper proposes a method of Android application service recommendation based on user scenarios.Specifically,this paper first analyzes the user scenarios,and extracts the text information in the user scenarios through the Accessibility API.Then,the label corresponding to the text information is calculated based on the method of knowledge base.Finally,through similarity calculation,the services related to user scenarios in the service library are searched,and the most relevant similar services and complementary services are recommended to users.This paper evaluates 300 Android application services of 10 popular apps in Android App store Wandoujia,and verifies the feasibility and effectiveness of the method.
Matrix Transformation and Factorization Based on Graph Partitioning by Vertex Separator for Recommendation
HE Yi-chen, MAO Yi-jun, XIE Xian-fen, GU Wan-rong
Computer Science. 2022, 49 (6A): 272-279.  doi:10.11896/jsjkx.210600159
Abstract PDF(3239KB) ( 446 )   
References | Related Articles | Metrics
Model-based collaborative filtering algorithms usually express user's preferences and item's attributes by latent factors through matrix factorization,but the traditional matrix factorization algorithm is difficult to deal with the serious data sparsity and data variability problems in the recommendation system.To solve the above problem,a matrix factorization algorithm based on bordered block diagonal matrix is proposed.Firstly,the original sparse matrix is transformed into bordered block diagonal matrix by a graph partitioning algorithm based on community discovery,which merges users with the same preference and items with similar characteristics into the same diagonal block,and then splices the diagonal blocks and the bordered into several sub-diagonal matrices which have higher densities.The experimental results show that,by decomposing the sub-diagonal matrices in parallel can not only improve the precision of prediction,but also improve the interpretability of the recommendation results.At the same time,each sub-diagonal matrix can be decomposed independently and in parallel,which can improve the efficiency of the algorithm.
Nonlinear Dynamics Information Dissemination Model Based on Network Media
DU Hong-yi, YANG Hua, LIU Yan-hong, YANG Hong-peng
Computer Science. 2022, 49 (6A): 280-284.  doi:10.11896/jsjkx.210500043
Abstract PDF(2359KB) ( 514 )   
References | Related Articles | Metrics
Existing network information propagation models that suppose all infected nodes are capable of infecting susceptible nodes,it cannot reflect objectively the fact that information propagation is time-efficient.For this problem,based on mean field theory,from the macro perspective of information dissemination,a novel dynamical model of network media information dissemination is proposed in this paper.According to real situation,the model assumes that only newly infected nodes will infect susceptible nodes in the social network,and there are two ways for susceptible nodes to become infected nodes,one is the spread of infection in social networks,the other is random browsing.Furthermore,based on graph theory,the states of nodes in the model and their transition relations are abstracted as weighted directed graphs.Based on Bayes' theorem,the state transitions between nodes are expressed as probability events,and the probability expression of event occurrence is given.Then the state transformation relationship matrix is determined.Finally,the model is solved using Gauss-Seidel iterative method.Numerical simulation results show that the dissemination of online media information is time-sensitive.Hot events will reach a peak of diffusion in a day,and then the range of diffusion will decline rapidly.Then the statistical data of the hot event from Baidu index is used to verify the validity of the model.The results show that the proposed model can reflect the spreading trend of network information more accurately than existing models.
FCM Algorithm Based on Density Sensitive Distance and Fuzzy Partition
MAO Sen-lin, XIA Zhen, GENG Xin-yu, CHEN Jian-hui, JIANG Hong-xia
Computer Science. 2022, 49 (6A): 285-290.  doi:10.11896/jsjkx.210700042
Abstract PDF(3386KB) ( 485 )   
References | Related Articles | Metrics
The traditional fuzzy C-means(FCM) algorithm is sensitive to noise data and only considers the distance factor in the iterative process.Therefore,the use of Euclidean distance for distance measurement will result in only considering the local consistency feature between sample points,while ignoring the global consistency feature.To solve these problems,an improved FCM algorithm based on density sensitive distance and fuzzy partition is proposed.First,the density sensitive distance is used to replace the Euclidean distance in the calculation of the similarity matrix,and then fuzzy entropy is introduced as a constraint condition in the clustering process to derive the new clustering center and the membership calculation formula with Gaussian distribution characteristics.In addition,in view of the problem that the traditional FCM algorithm randomly selects the initial clustering center may cause the clustering result to be unstable,according to the two principles of denser sample points around the cluster center point and longer distance between the cluster center points,combined with the density sensitive distance to select the initial cluster center point.Finally,the experimental comparison proves that the improved algorithm not only has higher clustering perfor-mance and anti-noise,but also significantly improves the convergence speed compared with the traditional FCM clustering algorithm and its derivative algorithm.
Microblog Popular Information Detection Based on Hidden Semi-Markov Model
XIE Bai-lin, LI Qi, KUANG Jiang
Computer Science. 2022, 49 (6A): 291-296.  doi:10.11896/jsjkx.210800011
Abstract PDF(2256KB) ( 464 )   
References | Related Articles | Metrics
In recent years,microblog has become great places for people to communicate with each other and share knowledge.However,microblog has also become the main grounds for rumors' transmission.If we can identify popular information in early stage,then we can identify and quell rumors early,we can also identify hot topics early in microblog.Therefore,the research on popular information detection is important.In this paper a new method is presented for identifying popular information based on hidden semi-Markov model(HSMM),from the perspective of the transmission processes of popular information in microblog.In this method,the observation value is constructed based on the influence level of the information forwarder and the time interval between two adjacent forwarders,and the influence level of the forwarder is automatically obtained by using the random forest classification algorithm.The proposed method includes a training phase and an identification phase.In the identification phase,the average log likelihood of every observation sequence is calculated,and the popularity of information is updated in real time.So this method can identify the popular information in early stage.An experiment based on real datasets of Sina Weibo and Twitter is conducted to evaluate this method.The experiment results validate the effectiveness of this method.
Acceleration of SVM for Multi-class Classification
CHEN Jing-nian
Computer Science. 2022, 49 (6A): 297-300.  doi:10.11896/jsjkx.210400149
Abstract PDF(1811KB) ( 481 )   
References | Related Articles | Metrics
With excellent classification effect and solid theoretical foundation,support vector machines have become one of the most important classification method in the field of pattern recognition,machine learning and data mining in recent years.How-ever,their training time becomes much longer with the increase of training instances.In the case of multi-class classification,the training process will become even more complex.To deal with above problems,a fast data reduction method named as MOIS is proposed for multi-class classification.With cluster centers being used as reference points,redundent instances can be deleted,bound instances crucial for the trainning can be selected,and the distribution imbalance between classes can also be relieved by the proposed method.Experiments show that MOIS can enormously improve the training efficiency while keeping or even improving the classification accuracy.For example,on Optdigit dataset,the classification accuracy is increased from 98.94% to 99.05%,while the training time is reduced to 0.15% of the original.What's more,on the dataset formed by the first 100 classes of HCL2000,the training time of the proposed method is reduced to less than 6% of original,while the accuracy is improved slightly from 99.29% to 99.30%.Furthermore,MIOS is highly efficient.
Image Processing & Multimedia Technology
Research Progress on Speech Style Transfer
LIU Chang, WEI Wei-min, MENG Fan-xing, CAI Zhi
Computer Science. 2022, 49 (6A): 301-308.  doi:10.11896/jsjkx.210300134
Abstract PDF(3132KB) ( 917 )   
References | Related Articles | Metrics
Speech style transfer technology refers to the conversion of timbre or speech style of the source speaker to timbre or speech style of the target speaker without changing the speech content.With people's urgent need for social media privacy protection and the rapid development of neural network tampering technology,speech style transfer technology has been deeply stu-died in the field.On the basis of introducing the basic principle of speech style transfer,this paper analyzes the research status from the perspective of three important factors vocoder,corpus alignment and transfer model,mainly including traditional vocoder and WaveNet vocoder,parallel corpus,unparallel corpus,conventional migration model and neural network model.It summarizes current problems of speech style transfer technology and challenges,and prospects the future development direction.
Identification of 6mA Sites in Rice Genome Based on XGBoost Algorithm
SUN Fu-quan, LIANG Ying
Computer Science. 2022, 49 (6A): 309-313.  doi:10.11896/jsjkx.210700262
Abstract PDF(2288KB) ( 457 )   
References | Related Articles | Metrics
N6-methyladenine(6mA) sites plays an important role in regulating gene expression of eukaryotes organisms.Accurate identification of 6mA sites may helpful to understand genome 6mA distributions and biological functions.At present,various experimental methods have been used to identify 6mA sites in different species,but they are too expensive and time-consuming.In this paper,a novel XGBoost-based method,P6mA-Rice,is proposed for identifying 6mA sites in the rice genome.Firstly,DNA sequence coding method based on sequence,which introduces and emphasizes the position specificity information,is first employed to represent the given sequences.Effective feature extraction criteria is proposed from seven aspects to make the expression of DNA information more comprehensive.Then,the selected feature set PS6mA based on the XGBoost feature importance is put into the integrated tree boosting algorithm XGBoost to construct the proposed model P6mA-Rice.The jackknife test on a benchmark dataset demonstrates that P6mA-Rice could obtain 90.55% sensitivity,88.48% specificity,79.00% Mathews correlation coefficient,and a 89.49% accuracy.Extensive experiments validate the effectiveness of P6mA-Rice.
Real-time Extend Depth of Field Algorithm for Video Processing
LAI Teng-fei, ZHOU Hai-yang, YU Fei-hong
Computer Science. 2022, 49 (6A): 314-318.  doi:10.11896/jsjkx.201100187
Abstract PDF(3153KB) ( 632 )   
References | Related Articles | Metrics
The extend depth of field algorithm(EDF) refers to extracting the clear parts of the image focusing on different areas of the sample and fusing them into an image to make each area of the sample in the fused image clear.The article proposes aEDF algorithm for video.First,the difference image is used to filter out the images that are considered to be the key frame whose focal depth changes.Then image registration is used to reduce the fusion error.Finally,the Laplacian pyramid based fusion algorithm is used to fuse the frame and previous fusion result and generate new EDF result.By repeating this process,the real-time dynamic EDF of the video is realized.The article designs experiments,using the spatial domain-based and the wavelet transform-based image fusion algorithm as a reference,comparing the operating efficiency and fusion quality in the scene of video from subjective and objective perspectives,and proves that the algorithm based on the Laplacian pyramid has good real-time performance,and is robust to out-of-focus blurred image.
Dam Crack Detection Based on Multi-source Transfer Learning
WANG Jun-feng, LIU Fan, YANG Sai, LYU Tan-yue, CHEN Zhi-yu, XU Feng
Computer Science. 2022, 49 (6A): 319-324.  doi:10.11896/jsjkx.210500124
Abstract PDF(3667KB) ( 674 )   
References | Related Articles | Metrics
The existing deep models will encounter overfitting and low computational efficiency when they are directly used for dam crack detection.This paper proposes a new dam crack detection algorithm based on multi-source transfer learning,which aims to improve the accuracy,reduce the model calculation and speed up the detection speed.Firstly,this method combines MobileNet with SSD object detection algorithm to construct a MobileNet-SSD network,which effectively reduces model parameters and computational complexity.Then,the proposed deep network is trained by using multi-source data sets such as road cracks,wall cracks and bridge cracks.Based on the transfer learning idea,the learned knowledge is transferred to the target domain model of dam crack to further improve the detection accuracy.Finally,a multi-model fusion method is proposed to integrate the detection results of different models obtained through transfer learning,which can effectively enhance the location of output boxes.
Traffic Sign Detection Based on MSERs and SVM
HU Cong, HE Xiao-hui, SHAO Fa-ming, ZHANG Yan-wu, LU Guan-lin, WANG Jin-kang
Computer Science. 2022, 49 (6A): 325-330.  doi:10.11896/jsjkx.210300117
Abstract PDF(3427KB) ( 520 )   
References | Related Articles | Metrics
Traffic sign detection is an important research content in the field of vehicle assistant driving system and automatic driving.It can instantly assist drivers or automatic driving systems to detect and identify traffic signs effectively.Based on this requirement,a traffic sign detection method based on real traffic scene is proposed.Firstly,the appropriate database is selected to convert the road scene image in the database into gray-scale image,and the gray-scale image is processed by simplified Gabor filtering to enhance the edge information of traffic signs.Secondly,the region recommendation algorithm MSERs is used to process the Gabor filtered feature map to form the proposal region of traffic signs.Finally,by extracting hog features,SVM is used for classification.Through experiments,the feature extraction performance of simplified Gabor filter,the performance of SG-MSERs region recommendation and filtering are analyzed,and the classification accuracy and processing time of the algorithm are obtained.The results show that the algorithm achieves good detection performance on both GTSDB and CSTD datasets,and basically meets the needs of real-time processing.
Analysis and Trend Research of End-to-End Framework Model of Intelligent Speech Technology
LI Sun, CAO Feng
Computer Science. 2022, 49 (6A): 331-336.  doi:10.11896/jsjkx.210500180
Abstract PDF(2473KB) ( 1158 )   
References | Related Articles | Metrics
The end-to-end framework is a probability model based on the depth neural network which can directly predict the speech signal and the target language character.From the original data input to the result output,the intermediate processing process and neural network are integrated,which can be separated from human subjective bias,directly extract the features,fully mine the data information,and simplify the task processing steps.In recent years,with the introduction of attention mechanism,the auxiliary end-to-end architecture realizes the mutual mapping between multimode,further improving the overall performance of the technology.Through the research on the technology and application of end-to-end technology in the field of intelligent speech in recent years,the end-to-end architecture provides a new idea and method for speech model algorithm,but there are also problems such as the mixed framework can not effectively balance and take into account the single technical characteristics,the complexity of the internal logic of the model makes it difficult for human intervention debugging,and the customization scalability is weakened.In the future,there will be further development in the application of the end-to-end integrated model in the field of speech.On the one hand,the front-end to back-end modules ignore the multiple input assumptions in front-end speech enhancement and back-end speech recognition to integrate speech enhancement and acoustic modeling.On the other hand,the end-to-end interactive information carrier focuses on the information extraction and processing of speech signal data itself the human-compu-ter interaction is closer to the real human language communication.
Study on Knowledge Distillation of Target Detection Algorithm Based on YOLOv4
CHU Yu-chun, GONG Hang, Wang Xue-fang, LIU Pei-shun
Computer Science. 2022, 49 (6A): 337-344.  doi:10.11896/jsjkx.210600204
Abstract PDF(4670KB) ( 883 )   
References | Related Articles | Metrics
Knowledge distillation,as a training method based on the teacher-student network,guides the relatively simple student network to be trained through the complex teacher network,so that the student network can obtain the same precision as the teacher network.It has been widely studied in the field of natural language processing and image classification,while the research in the field of object detection is relatively less,and the experimental effect needs to be improved.The Distillation Algorithm of object detection is mainly carried out in the feature extraction layer,and the distillation method of single feature extraction layer will cause students can't learn the teacher's network knowledge fully,which makes the accuracy of the model poorly.In view of the above problem,this paper uses the “knowledge” in feature extraction,target classification and border prediction of teacher network to guide student network to be trained,and proposes a multi-scale attention Distillation Algorithm to make the know-ledge of teacher network influence student network.Experimental results show that the distillation algorithm proposed in this paper based on YOLOv4 can effectively improve the detection accuracy of the original student network.
Image Arbitrary Style Transfer via Criss-cross Attention
YANG Yue, FENG Tao, LIANG Hong, YANG Yang
Computer Science. 2022, 49 (6A): 345-352.  doi:10.11896/jsjkx.210700236
Abstract PDF(8424KB) ( 770 )   
References | Related Articles | Metrics
Arbitrary style transfer is a technique for transferring an ordinary photo to an image with another artistic style.With the development of deep learning,some image arbitrary style transfer algorithms have emerged to generate stylized images with arbitrary styles.To solve the problems in adapting to both global and local styles,maintaining spatial consistency,this paper proposes an arbitrary style transfer via criss-cross attention network,which can efficiently generate stylized images with coordinated global and local styles by capturing long-range dependencies.To address the problem of the distorted content structure of stylized images,a group of the parallel channel and spatial attention networks are added before style transfer,which can further emphasize key features and retain key information.In addition,a new loss function is proposed to eliminate artifacts while preserving the structural information of the content images.This algorithm can match the closest semantic style feature to the content feature,and adjust the local style efficiently and flexibly according to the semantic spatial distribution of the content image.Moreover,it can retain more original information about the structure.The experimental results show that the proposed method can transfer the image into different styles with higher quality and better visual effects.
Classification Method for Small Crops Combining Dual Attention Mechanisms and Hierarchical Network Structure
YANG Jian-nan, ZHANG Fan
Computer Science. 2022, 49 (6A): 353-357.  doi:10.11896/jsjkx.210200169
Abstract PDF(5392KB) ( 465 )   
References | Related Articles | Metrics
The image recognition of small crops is very challenging for several reasons.First,the crop is small in size and a single sample is not representative of a collection.Second,different categories or different grades of the same crop may look very similar in shapes and colors.At present,there is a lack of research on image classification methods for small crops such as dried tea,rice and soybean,and most of the research datasets are taken in the laboratory environment with professional equipment,which brings difficulties to the practical application.For this,a method for image acquisition and processing of small crop samples using mobile phones is proposed.By taking tea and rice as a case study,we design a hierarchical network structure combined with two attention mechanisms.Through the coarse-grained to fine-grained classification process,coarse-grained classification is made first,namely different categories of samples,and then combined with two attention mechanisms,the network pays more attention to the diffe-rences between similar samples of different grades under the same category,so that they can be more accurate to classification of samples.Finally,the proposed method achieves the accuracy of 93.9% on the collected datasets.
Spatial Non-cooperative Target Components Recognition Algorithm Based on Improved YOLOv3
HAO Qiang, LI Jie, ZHANG Man, WANG Lu
Computer Science. 2022, 49 (6A): 358-362.  doi:10.11896/jsjkx.210700048
Abstract PDF(2705KB) ( 623 )   
References | Related Articles | Metrics
Due to the large amount of neural network parameters and insufficient computing power of embedded devices,it is difficult to effectively deploy neural networks on embedded platforms when using deep learning methods to identify spatial non-cooperative target components.Aiming at this problem,an improved lightweight target detection network is proposed in this paper.On the basis of YOLOv3,a new lightweight feature extraction backbone network Res2-MobileNet is designed,drawing on the ideaof Depth Separable Convolution,introducing the Bottleneck module to reduce the amount of model parameters to improve the detection speed,and introducing the Res2Net residual module to increase the sensitivity of network to small targets by increasing the modeĹs receptive field scale richness and structural depth,and combines multi-scale detection methods to recognize spatial non-cooperative target components.The experimental results show that compared with the YOLOv3 model,the size of this model is reduced by 55.5%,the detection speed is increased from 34fps to 65 fps,and the detection effect for small targets is also significantly improved.
Multi-tree Network Multi-crop Early Disease Recognition Method Based on Improved Attention Mechanism
GAO Rong-hua, BAI Qiang, WANG Rong, WU Hua-rui, SUN Xiang
Computer Science. 2022, 49 (6A): 363-369.  doi:10.11896/jsjkx.210500044
Abstract PDF(3241KB) ( 567 )   
References | Related Articles | Metrics
In the early stage of crop infection,timely acquisition of crop disease information,identification of the cause and severity of disease,and the right remedy,can prevent and control the decline in crop yield caused by the spread of the disease in time.In view of the low accuracy of traditional deep learning network for early crop disease recognition,based on the difference in information contained in each channel of the disease feature image and the characteristics of multilayer prceptron(MLP) that can approximate any function,a multi-tree network crop early disease identification method based on improved attention mechanism is proposed.It combines the attention mechanism with residual network to recalibrate disease features(SMLP_Res).At the same time,combined with the multi-tree structure,the SMLP_ResNet network with high feature extraction ability is expanded,and the constructed multi-tree SMLP_ResNet network model can simplify the task of early disease recognition of multiple crops and effectively extract early disease features.In experiments,Plant Village and AI Challenger 2018 are used to train18-layer model ResNet,SE_ResNet,SMLP_ResNet,as well as the multi-tree structure model with the same structure,to verify the influence of SMLP_Res and multi-tree structure on crop disease recognition models.Experimental analysis shows that,the disease recognition accuracy of the three network models on Plant Village dataset with obvious disease features is more than 99%,but their accuracy on the early disease data set AI Challenger 2018 is not more than 87%.SMLP_ResNet has sufficient feature extraction of crop disease data due to the addition of SMLP_Res module,and the detection results are better.The three early disease recognition models of the multi-tree structure significantly improves the recognition accuracy on AI Challenger 2018 dataset.The multi-tree SMLP_ResNe has better performance than the other two models,and the early disease recognition accuracy of cherry is 99.13%,the detection result is the best.The proposed multi-tree SMLP_ResNet crop early disease recognition model can simplify the recognition task,suppress noise transmission,and achieve a higher accuracy rate.
Lightweight Micro-expression Recognition Architecture Based on Bottleneck Transformer
ZHANG Jia-hao, LIU Feng, QI Jia-yin
Computer Science. 2022, 49 (6A): 370-377.  doi:10.11896/jsjkx.210500023
Abstract PDF(3567KB) ( 873 )   
References | Related Articles | Metrics
Micro-expressions are spontaneous facial movements at a marginal spatiotemporal scale,which reveal one's true fee-lings.Its duration is short,the amplitude of the movement is slight,and it is difficult to recognize,but it has important research value.In order to solve the micro-expression recognition problem,a novel extremely lightweight micro-expression recognition neural architecture is proposed.The neural network which takes apex-onset optical-flow features as the input and integrates approaches in residual convolutional networks and visual Transformers,could effectively solve the micro-expression sentiment classification problem.This architecture containsnovel parameter-saving residual blocks,and a bottleneck Transformer block which replace the convolution operators in residual blocks with self-attention mechanism.The model evaluation experiments are conducted with a LOSO cross-validation strategy on a combined database con-sists of the 3 CASME datasets.With obviously fewer total parameters(39 685),the model achieves an average recall of 73.09% and an average F1-Score of 72.25%,exceeding those mainstream architectures in this domain.A series ablation experiments are also conducted to ensure the superiority of the optical strain strength,self-attention mechanism and relativeposition encoding.
Remote Sensing Aircraft Target Detection Based on Improved Faster R-CNN
ZHU Wen-tao, LAN Xian-chao, LUO Huan-lin, YUE Bing, WANG Yang
Computer Science. 2022, 49 (6A): 378-383.  doi:10.11896/jsjkx.210300121
Abstract PDF(4953KB) ( 695 )   
References | Related Articles | Metrics
Optical remote sensing image airplane target detection technology has been widely used in the fields of air traffic and military reconnaissance.Aiming at the problem of low detection accuracy and high false alarm rate when detecting multiple aircraft of different sizes,this paper proposes an improved ROI pooling method based on bilinear interpolation for aircraft target detection,which solves the problem of mis-alignment.Experimental results show that the improved method achieves a detection performance of 95.35% AP(IOU≤0.5).In the task of multi-size airplane detection,the target positioning accuracy and average accuracy are improved.
Single Backlit Image Enhancement Based on Virtual Exposure Method
ZHAO Ming-hua, ZHOU Tong-tong, DU Shuang-li, SHI Zheng-hao
Computer Science. 2022, 49 (6A): 384-389.  doi:10.11896/jsjkx.210400243
Abstract PDF(5989KB) ( 609 )   
References | Related Articles | Metrics
The visual quality of backlit image in target area is low and its background area is overexposed,which are important factors affecting image quality.Aiming at the problem that the existing backlit image enhancement methods cannot suppress the excessive enhancement of bright areas while enhancing the detail information of dark areas well,a single backlit image enhancement method based on virtual exposure is proposed in this paper.First,the virtual exposure image is introduced,and the best low-exposure image and high-exposure image are determined according to parameters.Then,the dark and bright areas are processed by nonlinear enhancement method and the neighborhood correlation method respectively.Finally,the details and features of the dark and bright areas are fused by Laplacian pyramid fusion method.Experiments results based on natural images and synthetic images show that the proposed method has less color and brightness distortion,and the visual effect is more natural.
Face Anti-spoofing Algorithm Based on Texture Feature Enhancement and Light Neural Network
SHEN Chao, HE Xi-ping
Computer Science. 2022, 49 (6A): 390-396.  doi:10.11896/jsjkx.210600217
Abstract PDF(4081KB) ( 673 )   
References | Related Articles | Metrics
Face anti-spoofing is an important part of face recognition,which is of great significance to the safety of related industries in reality,such as authentication,security key,financial payment and so on.The mainstream face anti-counterfeiting algorithm based on deep learning has achieved quite advanced results,but there are still some problems,such as too many model parameters increases the difficulty of actual deployment,and the generalization performance of light neural network structure is not good,etc.Aiming at the problems of poor generalization ability and too large parameters of the related face anti-spoofing algorithm.This paper proposes a texture feature enhancement method and a face anti-spoofing detection algorithm based on improved FeatherNet network.By enhancing the texture difference features of real and fake face information as the input of the backbone network.In the design of the backbone network,DropBlock module and multi-channel attention feature map branch are introduced.The generalization performance is enhanced while maintaining the speed.The designed algorithm shows good performance improvement in both data-set test and cross data-set test.
Study on Activity Recognition Based on Multi-source Data and Logical Reasoning
XIAO Zhi-hong, HAN Ye-tong, ZOU Yong-pan
Computer Science. 2022, 49 (6A): 397-406.  doi:10.11896/jsjkx.210300270
Abstract PDF(4476KB) ( 564 )   
References | Related Articles | Metrics
The use of smart terminal equipment to identify and record people's daily behaviors is of great significance for health monitoring,the disabled assistance and elderly care.Most existing related studies adopt machine learning-based ideas,but there are problems such as high consumption of computing resources and training.Due to the heavy burden of data collection and low scalability in different scenarios,this paper proposes a behavior recognition technology based on multi-source perception and logical reasoning.By determining the logical correlation between the actions of different limbs,the accurate description of basic behaviors of users' daily life is realized.Compared with existing work,this technical solution has the advantages of lightweight calculation,low training cost and strong expansion ability to the diversity of users and scenes.This paper realizes a behavior recognition system based on the above technology.A large number of experiments have been carried out to evaluate the performance of the system.The results show that the proposed method has a recognition accuracy of more than 90% for 11 daily behavior activities such as walking,running,and up and down stairs.At the same time,compared with the behavior recognition method based on machine learning,the proposed technique greatly reduces the amount of training data collected by users.
Face Recognition Based on Locality Regularized Double Linear Reconstruction Representation
HUANG Pu, DU Xu-ran, SHEN Yang-yang, YANG Zhang-jing
Computer Science. 2022, 49 (6A): 407-411.  doi:10.11896/jsjkx.210700018
Abstract PDF(2767KB) ( 455 )   
References | Related Articles | Metrics
The solving process of sparse representation classifier(SRC) is relatively complicated and costs a long time,collaborative representation classifier(CRC) treats all the training samples as the dictionary of unknown samples and the dictionary is large without considering the label information,linear regression classifier(LRC) does not take the differences between inter-class samples into account and ignores the distance information and the neighborhood relations between samples.To address the problems and shortcomings in these representation learning based classification algorithms,this paper proposes a locality regularized double linear reconstruction representation classification method(LRDLRRC) for face recognition.Firstly,LRDLRRC calculates the intra-class nearest neighbors of the query sample and uses the intra-class nearest neighbors to linearly reconstruct the query sample.Then the query sample is represented as a linear combination of all the intra-class reconstruction samples,and the representation coefficient is constrained by the reconstruction error between the query sample and the intra-class reconstruction samples.Finally,the Lagrange multiplier method is applied to solve the representation coefficient,and the classification result of the query sample is determined by the ratio between the reconstruction error and the representation coefficient.Experiments on AR,FRGC and FERET datasets show that the proposed algorithm has superior accuracy,time complexity and strong robustness.
High Performance Insulators Location Scheme Based on YOLOv4 with GDIoU Loss Function
MA Bin, FU Yong-kang, WANG Chun-peng, LI Jian, WANG Yu-li
Computer Science. 2022, 49 (6A): 412-417.  doi:10.11896/jsjkx.210600089
Abstract PDF(2968KB) ( 517 )   
References | Related Articles | Metrics
In this paper,a Gaussian Distance Intersection over Union(GDIoU) loss function based YOLOv4 deep learning method is proposed to solve the problem of low speed and low accuracy insulator positioning in the process of power line health inspection.In the scheme,a GDIoU loss function is designed to accelerate the convergence speed of the YOLOv4 deep learning network,and the two-dimensional Gaussian model is used to improve the convergence ability of the network,through which the perfor-mance of the YOLOv4 network is enhanced and the insulator's positioning accuracy is accordingly improved.At the same time,an adaptive tilt correction algorithm is proposed to improve the positioning accuracy of the insulators in different spatial angle states by rotating the image with only one insulator.The experimental results show that the average precision is increased by 7.37% compared with the naive YOLOv4 scheme.And the GDIoU based YOLOv4 deep learning network combined with the adaptive tile correction method accelerates the insulator positioning speed by three times compared with the other insulator positioning me-thods at the same level of accuracy.The proposed method makes a good balance between accuracy and speed,and its performance can meet the requirement of online insulator positioning adequately.
Influence of Different Data Augmentation Methods on Model Recognition Accuracy
WANG Jian-ming, CHEN Xiang-yu, YANG Zi-zhong, SHI Chen-yang, ZHANG Yu-hang, QIAN Zheng-kun
Computer Science. 2022, 49 (6A): 418-423.  doi:10.11896/jsjkx.210700210
Abstract PDF(3278KB) ( 784 )   
References | Related Articles | Metrics
The effect of deep learning depends heavily on the quantity and quality of data,and insufficient data will cause the mo-del to overfit.In practical application research,it is often difficult to obtain a large number of high-quality sample data,especially image data.In response to the above problems,this paper takes the ANIMAL-10 dataset as the research object,and designs a method based on foreground target extraction and using pure colors to replace the original background to achieve data augmentation.Combining traditional data augmentation methods to construct new datasets,four neural network models of AlexNet,Inception,ResNet and VGG-16 are used to analyze the impact of different color backgrounds and different data augmentation methods on the accuracy of model recognition.Experiments shows that different color backgrounds have no significant influence on the accuracy of model recognition.And on this basis,the green background is used for subsequent data augmentation operations,four datasets A,B,C,and D are designed and the above four models are compared.The test results show that the model have a significant impact on the recognition accuracy,while the data set has no significant impact on the recognition accuracy.However,for the AlexNet and Inception-v3 models,the recognition accuracy of the enhanced dataset including the prominent foreground data is increased by 3.78% and 4.55%,respectively,compared with the original image and traditional data augmentation methods.This shows that under small datasets,the data augmentation method that highlights the foreground could make the model easier to notice and learn the key features of the images,so that the performance of the model is better,and the recognition accuracy of the model is improved,which has certain practical value in actual engineering applications.
Real-time Helmet Detection Algorithm Based on Circumcircle Radius Difference Loss
CHEN Yong-ping, ZHU Jian-qing, XIE Yi, WU Han-xiao, ZENG Huan-qiang
Computer Science. 2022, 49 (6A): 424-428.  doi:10.11896/jsjkx.220100252
Abstract PDF(2985KB) ( 664 )   
References | Related Articles | Metrics
For large demands of fast and accurate helmet detection,this paper proposes a real-time helmet detection algorithm.Firstly,to solve the gradient vanish problem of using bounding box regression loss functions,this paper proposes the circumcircle radius difference (CRD) loss function.Secondly,to solve the problem of complex multi-scale feature fusion layers restricting detection speeds,this paper proposes a lightweight focus on small object (FSO) feature fusion layer.Finally,this paper combines the YOLO network,CRD,and FSO to form a YOLO-CRD-FSO (YCF) model for real-time helmet detection.On a Jetson Xavier NX device,experiments show that the detection speed of YCF reaches 43.4 frames per second for 640×640 sized videos,which is nearly 2 frames per second faster than the state-of-the-art YOLO-V5 model,and the mean average precision has been improved by nearly 1%.The proposed YCF detection model comprehensively optimizes boundary box regression loss functions and feature fusions,acquiring good helmet detection results.
Face Recognition Based on Locality Constrained Feature Line Representation
HUANG Pu, SHEN Yang-yang, DU Xu-ran, YANG Zhang-jing
Computer Science. 2022, 49 (6A): 429-433.  doi:10.11896/jsjkx.210300169
Abstract PDF(4265KB) ( 473 )   
References | Related Articles | Metrics
To solve the problem of low feature representation capacity and discriminality of collaborative representation based classifier(CRC) and related algorithms in face recognition,a locality constrained feature line representation based classifier(LCFLRC) is proposed for face recognition.At first,LCFLRC represents a test image as a linear combination of the projections of the test image on the overall feature lines,and a constraint with respect to the distance between the test image and each feature line is imposed.Then,the L2norm based optimization problem is solved by using the Lagrange multiplier method.At last,the label of the test image is decided according to the reconstruction residual between the test image and the projections of the test image on the feature lines of each class.LCFLRC could capture more variations iin facial images by using the feature lines to represent the test image,and contains more discriminant information by taking advantage of the distance information between the test image and feature line such that the reconstruction coefficient of the projection on the feature line nearer to the test image is larger.Experimental results on CMU PIE,Extended Yale-B and AR face databases demonstrate that the proposed method significantly outperform other classification methods with varying illumination,facial expressions and poses in images.
Dual-field Feature Fusion Deep Convolutional Neural Network Based on Discrete Wavelet Transformation
SUN Jie-qi, LI Ya-feng, ZHANG Wen-bo, LIU Peng-hui
Computer Science. 2022, 49 (6A): 434-440.  doi:10.11896/jsjkx.210900199
Abstract PDF(2941KB) ( 624 )   
References | Related Articles | Metrics
Pooling operation is an essential part of deep convolutional neural networks,and also one of the key factors for the success of deep convolutional neural network.However,in the process of image recognition,the traditional direct pooling operation will lead to the loss of feature information and affect the accuracy of recognition.In this paper,a dual-field feature fusion module based on discrete wavelet transform is proposed to overcome the disadvantage of the direct pooling operation.In this module,the dual-field feature fusion of spatial domain and channel domain is considered,and the pooling operation is embedded between spatial feature fusion module and channel feature fusion module,which effectively suppress the information loss of features caused by pooling directly.By replacing the existing pooling operation,the new dual-field feature fusion module can be easily embedded into the current popular deep neural network architectures.Extensive experimental results on CIFAR-10,CIFAR-100 and Mini-Imagenet datasets by using mainstream network architectures such as VGG,ResNet and DenseNet.The experimental results show that compared with the classical network,the popular network based on embedded attention mechanism or latest wavelet basis model,the proposed method can achieve higher classification accuracy.
Acceleration Algorithm of Multi-channel Video Image Stitching Based on CUDA Kernel Function
LIU Yun, DONG Shou-jie
Computer Science. 2022, 49 (6A): 441-446.  doi:10.11896/jsjkx.210600043
Abstract PDF(2214KB) ( 961 )   
References | Related Articles | Metrics
With the rapid growth of the number of general airports and the rapid development of remote towers,multi-channel real-time video stitching technology is constantly improved.Most of the video mosaic algorithms are based on the improvement of feature point extraction algorithm,feature point matching algorithm and image fusion algorithm.The improved feature point extraction algorithm makes the feature points more accurate.The improved image fusion algorithm makes the color of each image on the panorama consistent and eliminates the stitching seam.Finally,the GPU is used to accelerate the process.This paper focuses on the overall calculation process of image mosaic,from the mathematical point of view to calculate the pixel coordinate transformation matrix and pixel value transformation matrix from the original image to panoramic image.In video stitching,based on the pixel coordinate transformation matrix of pixel value transformation matrix,the kernel function is used to calculate the coordinates and pixel values after matrix transformation.Set the thread configuration when the kernel function is called.The parallel computing ability of the graphics processor is fully utilized to accelerate the image stitching.The experimental results show that the stitching time of 8-channel 1080 * 1920 videos is about 22 ms.The core technology of video stitching includes streaming,decoding,stitching,coding and transmission.At the end of the paper,some suggestions on streaming and decoding technology are put forward.
Information Security
Overview of Research and Development of Blockchain Technology
FU Li-yu, LU Ge-hao, WU Yi-ming, LUO Ya-ling
Computer Science. 2022, 49 (6A): 447-461.  doi:10.11896/jsjkx.210600214
Abstract PDF(4331KB) ( 1221 )   
References | Related Articles | Metrics
Blockchain is called the next-generation Internet of Value,which is a basic system architecture for emerging decentra-lized cryptocurrencies.Since Satoshi Nakamoto proposed the term blockchain in 2008,it has gradually received widespread attention due to its immutability,traceability,and decentralization features.Two of the representatives are the Bitcoin block Chain system and Ethereum blockchain system.However,in the current literatures,most of the existing blockchain technology is applied to real life while the introduction of the underlying implementation of the blockchain is relatively vague.To this end,the blockchain should be separated from the actual one,and the working of the blockchain can be understood through the design ideas and key technologies of the Bitcoin blockchain system and the Ethereum blockchain system.The article mainly introduces the infrastructure of blockchain technology including the cryptographic principles,consensus algorithms,data storage structure and other aspects.Further the supplements about ambiguous concepts in the Bitcoin white paper and the Ethereum yellow paper are presented as well,which can provide more deeply research for readers later.Finally,the current application status and prospects of blockchain are discussed.
RegLang:A Smart Contract Programming Language for Regulation
GAO Jian-bo, ZHANG Jia-shuo, LI Qing-shan, CHEN Zhong
Computer Science. 2022, 49 (6A): 462-468.  doi:10.11896/jsjkx.210700016
Abstract PDF(4023KB) ( 872 )   
References | Related Articles | Metrics
To prevent the systemic risks brought by financial technology innovation,regulators have put forward more and more regulatory requirements,which increases the compliance costs of financial institutions and regulatory frictions.Blockchain technology can help to promote RegTech innovation and build a new regulatory scheme,but how to realize the digitalization of regulatory rules on the blockchain is still an urgent problem.In response to the core requirements of RegTech,this paper proposes a regulatory-oriented smart contract programming language RegLang,which is interactivity,calculability and concurrency.The language is convenient for regulatory experts to write regulatory policies into digital rules and run them in the form of smart contracts on the blockchain.We evaluate RegLang based on the digitization of real-world financial regulatory policy,and the results show that compared with existing programming languages,RegLang is easy to write,expressive,highly readable,supports concurrent operation,and is more suitable for expressing regulatory rules on blockchain and realizing blockchain-based digital regulation.
Secret Reconstruction Protocol Based on Smart Contract
WEI Hong-ru, LI Si-yue, GUO Yong-hao
Computer Science. 2022, 49 (6A): 469-473.  doi:10.11896/jsjkx.210700033
Abstract PDF(2106KB) ( 557 )   
References | Related Articles | Metrics
The secret reconstruction protocol for a verifiable secret sharing scheme is designed.Under real-life conditions where the participants are all rational,the punishment mechanism and the method of deducting the deposit are used to conduct the behavior of malicious parties verified by the verifiable secret sharing scheme Constraint,and using the tool of blockchain smart contract,according to the independence and non-tampering of smart contract,the problem of trusted third parties that are difficult to solve in previous research is solved.The protocol is divided into two parts,local protocol and smart contract.While ensuring the security and confidentiality,the protocol can also utilize the design of smart contracts to achieve fairness.
Empirical Security Study of Native Code in Python Virtual Machines
JIANG Cheng-man, HUA Bao-jian, FAN Qi-liang, ZHU Hong-jun, XU Bo, PAN Zhi-zhong
Computer Science. 2022, 49 (6A): 474-479.  doi:10.11896/jsjkx.210600200
Abstract PDF(2006KB) ( 612 )   
References | Related Articles | Metrics
The Python programming language and its echo-systems continue to play important roles in modern artificial intelligent systems like machine learning or deep learning,and are among one of the most popular implementation languages in modern machine learning infrastructures like TensorFlow,PyTorch,Caffe or CNTK.The security of the Python virtual machines is critical to the security of these machine learning systems.However,due to the existence of huge native code base in Python's CPython virtual machine,it's a great research challenge to study the security vulnerability patterns in Python virtual machines and the techniques to fix these vulnerabilities.This paper presents a novel vulnerability analysis framework PyGuard,which makes use of the static program analysis techniques to analyze the security of native code in Python virtual machines.This paper also introduces a prototype implementation of this framework and reports the experimental results of an empirical security study of the CPython virtual machine (version 3.9):we have found 45 new security vulnerabilities which demonstrates the effectiveness of this system.We have conducted a thorough study of the vulnerability patterns and given a taxonomy.
Model for the Description of Trainee Behavior for Cyber Security Exercises Assessment
TAO Li-jing, QIU Han, ZHU Jun-hu, LI Hang-tian
Computer Science. 2022, 49 (6A): 480-484.  doi:10.11896/jsjkx.210800048
Abstract PDF(3874KB) ( 533 )   
References | Related Articles | Metrics
The evaluation of trainee performance is one of the key points to improve the effectiveness of cyber security exercises,and the study of evaluation method includes two stages:evaluation based on training results and evaluation based on training behavior modeling.The first one cannot figure out the training details,the other can only pre-model some training paths so that it can't determine the correctness of non-preset training path training behavior.In order to solve the problems,a two-layer cyber security exercises trainee behavior description model based on the orientation graph and finite state automatic machine is proposed,and the correctness determination and detail evaluation of non-preset training behavior are realized by combining the characteri-stics of training behavior and training results.An experiment on typical computer network security training scenario shows that,compared with the description model that focuses only on training behavior,the model improves the accuracy of training behavior determination,and realizes the determination of non-preset path training behavior correctness and training details.
Timing Attack Resilient Sampling Algorithms for Binary Gaussian Based on Knuth-Yao
LIANG Yi-wen, DU Yu-song
Computer Science. 2022, 49 (6A): 485-489.  doi:10.11896/jsjkx.210600017
Abstract PDF(2071KB) ( 977 )   
References | Related Articles | Metrics
Lattice-based cryptography is considered to be a class of new type of cryptosystems that can resist the attacks from quantum computers.Discrete Gaussian sampling over the integers plays an important role in lattice-based cryptography,but it is vulnerable to timing attacks.The Knuth-Yao method is a sampling method based on the binary probability matrix.The simple improvement of Knuth-Yao can achieve constant running time to resist timing attacks,but it greatly sacrifices the sampling speed.By using the cyclic shift property of the probability matrix of the binary discrete Gaussian distribution,the Knuth-Yao is simplified,and the storage space of the probability matrix is reduced.Based on the idea of splitting the probability matrix,a class of timing attack resilient sampling algorithms for the binary discrete Gaussian based on Knuth-Yao is effectively implemented,and the sampling speed can be guaranteed to a large extent.Experimental results show that the sampling speed of the improved algorithm is only 27.6% lower than that of the conventional algorithm based on Knuth-Yao,which achieves a better balance between sampling speed and security.
Design and Implementation of Cross-chain Trusted EMR Sharing System Based on Fabric
YUAN Hao-nan, WANG Rui-jin, ZHENG Bo-wen, WU Bang-yan
Computer Science. 2022, 49 (6A): 490-495.  doi:10.11896/jsjkx.210500063
Abstract PDF(4681KB) ( 1108 )   
References | Related Articles | Metrics
Electronic medical record(EMR) is a sensitive and important privacy data asset of patients.Its trusted sharing is significant to the development of medical informalization.Aiming at the problems of unsafe data storage,difficult cross-domain trusted sharing and long access period of EMRs,this paper integrates blockchain and edge computing,designs and implements a cross-chain trusted EMR sharing system based on Fabric alliance chain frame.The system is mainly divided into patient mobile application,hospital Web application and RFID tag bracelet,including medical record encryption and authentication,cross-chain trusted sharing,remote authorization and other functions.In addition,this paper designs encryption and authentication mechanism based on biometric key and national commercial cipher algorithm series,to control the flow of privacy data with patients as the main bodyand realize personalized privacy protection.It applies a master-slave multi-chain hierarchical cross-chain model on the Hyperledger Fabric to achieve reliable access and control.Experiments and comparative analysis show that the system has great advantages in data security and performance.
Training Method to Improve Robustness of Federated Learning
YAN Meng, LIN Ying, NIE Zhi-shen, CAO Yi-fan, PI Huan, ZHANG Lan
Computer Science. 2022, 49 (6A): 496-501.  doi:10.11896/jsjkx.210400298
Abstract PDF(3291KB) ( 1092 )   
References | Related Articles | Metrics
The federated learning method breaks the bottleneck of data barriers and has been widely used in finacial and medical aided diagnosis.However,the existence of counter attacks poses a potential threat to the security of federated learning models.In order to ensure the safety of the federated learning model in actual scenarios,improving the robustness of the model is a good solution.Adversarial training is a common method to enhance the robustness of the model,but the traditional adversarial training methods are all aimed at centralized machine learning,and usually can only defend against specific adversarial attacks.This paper proposes an adversarial training method to enhance the robustness of the model in a federated learning scenario.This method adds single-strength adversarial samples,iterative adversarial samplesand normal samples in the training process,and adjusts the weight of the loss function under each training sample to complete the local training and the update of global model.Experiments on Mnist and Fashion_Mnist datasetsshow that although the adversarial training is only based on FGSM and PGD,this adversa-rial training method greatly enhances the robustness of the federated model in different attack scenarios.The adversarial training is based on FGSM and PGD,and is also effective for other attck methods such as FFGSM,PGDDLR,BIM and so on.
Analysis of Bitcoin Entity Transaction Patterns
HE Xi, HE Ke-tai, WANG Jin-shan, LIN Shen-wen, YANG Jing-lin, FENG Yu-chao
Computer Science. 2022, 49 (6A): 502-507.  doi:10.11896/jsjkx.210600178
Abstract PDF(4055KB) ( 637 )   
References | Related Articles | Metrics
Since the Bitcoin system went online,people have conducted decentralized transfer transactions through Bitcoin addresses,which greatly increased the convenience of transactions,and the transaction records generated by peer-to-peer transactions have always been the focus of research.Due to the huge scale of the Bitcoin transaction network,it takes a long time and huge computing power to explore the entire network directly,and it is also not conducive to observing the internal transaction pattern of the entity.Bitcoin transaction records are permanently stored in the blockchain ledger,and the entity behavior and internal transaction pattern of Bitcoin entity service can be further explored by constructing and analyzing the transaction network.By improving the traditional label propagation algorithm,a label propagation algorithm based on central nodes is proposed to divide the communities of the Bitcoin entity transaction network,and the transaction patterns of the core communities are analyzed,such as Exchanges and mining pools.This paper summarizes two kinds of transaction patterns which are easy to understand and conform to reality.The experimental results prove the differences in transaction patterns within different services,and the graphical display improves the readability of the Bitcoin transaction network.
Android Malware Detection Method Based on Heterogeneous Model Fusion
YAO Ye, ZHU Yi-an, QIAN Liang, JIA Yao, ZHANG Li-xiang, LIU Rui-liang
Computer Science. 2022, 49 (6A): 508-515.  doi:10.11896/jsjkx.210700103
Abstract PDF(2977KB) ( 818 )   
References | Related Articles | Metrics
Aiming at the problem of limited detection accuracy of a single classification model,this paper proposes an Android malware detection method based on heterogeneous model fusion.Firstly,by identifying and collecting the mixed feature information of malicious software,the random forest algorithm based on CART decision tree and the Adaboost algorithm based on MLP are used to construct the integrated learning model respectively,and then the two classifiers are fused by Blending algorithm.Finally,a heterogeneous model fusion classifier is obtained.On this basis,the mobile terminal malware detection is implemented.Experimental results show that the proposed method can effectively overcome the problem of insufficient accuracy of single classification model.
Network Attack Path Discovery Method Based on Bidirectional Ant Colony Algorithm
GAO Wen-long, ZHOU Tian-yang, ZHU Jun-hu, ZHAO Zi-heng
Computer Science. 2022, 49 (6A): 516-522.  doi:10.11896/jsjkx.210500072
Abstract PDF(3858KB) ( 838 )   
References | Related Articles | Metrics
In the field of penetration testing,the discovery of attack paths is of great significance to the realization of attack automation.Most of the existing attack path discovery algorithms are suitable for static global environments,and there is a problem that the solution fails due to the explosion of the state space.To solve the problem of attack path discovery under dynamic network environment and improve the efficiency of path discovery,a method of network attack path discovery based on bidirectional ant co-lony algorithm is proposed.First,model the network information and define the attack cost.Then,a new two-way ant colony algorithm is proposed for attack path discovery.The main improvements include different search strategies,cross-optimization operations and new pheromone update methods,etc.Simulation experiments verify the improved quality and efficiency.At the same time,compared with other path discovery methods,it has a certain time or space advantages in large network scale.When the attack path host fails,the re-planning mechanism is used to realize the attack path discovery in the local area,which is more suitable for attack path discovery under actual automated penetration testing.
Study on Automatic Synthetic News Detection Method Complying with Regulatory Compliance
MAO Dian-hui, HUANG Hui-yu, ZHAO Shuang
Computer Science. 2022, 49 (6A): 523-530.  doi:10.11896/jsjkx.210300083
Abstract PDF(3446KB) ( 474 )   
References | Related Articles | Metrics
Automatic Synthetic news has been widely used in formatted news information such as financial market analysis and event reports,which has great social impact.However,when centralized storage is adopted for regulatory,it is easy for regulators or third parties to steal and tamper with the information.Therefore,under the premise of ensuring detection efficiency and accur-acy,it is particularly important to protect private information from being leaked.In this paper,an automaticsynthetic news detection method is proposed,which meets the requirements of regulations.The goal is to record data activities in distributed ledger while ensuring that only regulatory agencies can process news information by data access token.This method designs two types of distributed ledgers and calls them through intelligent contracts to realize authorization mechanism and log recording.Only honest participation can be recognized by the blockchain and prove compliance with regulation.Furthermore,the method endows edge nodes with computing power by adopting lightweight detection algorithms IDF-FastText,prevents the proliferation of various synthetic news from the source,and realizes the timeliness of regulation.The GPT-2 detection algorithm based on general adversarial networks(GAN) is deployed on the server for the regulator to verify the detection results.Finally,the feasibility of the proposed design concept is proved by experiments.
Robust Speaker Verification with Spoofing Attack Detection
GUO Xing-chen, YU Yi-biao
Computer Science. 2022, 49 (6A): 531-536.  doi:10.11896/jsjkx.210500147
Abstract PDF(3295KB) ( 568 )   
References | Related Articles | Metrics
Spoofing attacks seriously affect the security application of speaker verification system.This paper proposes a speaker verification system with replay attack detection capability,which has a series connection structure of front-end attack detection and back-end speaker verification.In addition,this paper proposes a channel frequency response difference enhancement cepstral coefficient(CDECC) through channel frequency response analysis and speaker personality analysis.The CDECC enhances the low and high frequency bands of the speech signal spectrum by the third-order polynomial nonlinear frequency transform,which can effectively reflect the channel frequency response difference of different input channels and the speech spectrum difference of different speakers.The speaker and text independent replay attack detection experiment based on ASVspoof 2017 2.0 dataset shows that the equal error rate(EER) of CDECC based replay attack detection is 25.03%,which is 10.00% lower than the baseline system.By embedding the replay attack detection module at the front end of the speaker verification,the speaker verification system's false acceptance rate(FAR) is significantly reduced,the system's EER is reduced from 3.32% to 1.01%,and the robustness is effectively improved.
Design of Cross-domain Authentication Scheme Based on Medical Consortium Chain
CHEN Yan-bing, ZHONG Chao-ran, ZHOU Chao-ran, XUE Ling-yan, HUANG Hai-ping
Computer Science. 2022, 49 (6A): 537-543.  doi:10.11896/jsjkx.220200139
Abstract PDF(3090KB) ( 859 )   
References | Related Articles | Metrics
Secure cross-domain authentication is the key to ensure the interconnection of medical data.Most of the existing cross-domain authentication models rely on trusted third parties,and there are heavy key management overhead and private key escrow problems.By introducing blockchain and certificateless authentication technology,a safe and efficient cross-domain authentication scheme based on medical consortium chain is proposed.Using hash function,digital signature and other cryptography technology to achieve safe and reliable authentication of foreign users,and using improved practical Byzantine mechanism to ensure that medical institutions in the alliance can quickly agree on the verification results without central nodes.The analysis shows that in terms of security,the scheme has security properties such as resistance to distributed attacks;in terms of efficiency,compared with the existing cross-domain authentication scheme,the scheme has advantages in computational overhead and communication overhead.
Survey of Network Traffic Analysis Based on Semi Supervised Learning
PANG Xing-long, ZHU Guo-sheng
Computer Science. 2022, 49 (6A): 544-554.  doi:10.11896/jsjkx.210600131
Abstract PDF(3586KB) ( 763 )   
References | Related Articles | Metrics
Semi supervised learning is a new machine learning method.It combines supervised learning with unsupervised lear-ning,and uses a small number of tags to analyze a large number of unlabeled data sets.In recent years,semi supervised learning has become one of the research hotspots of scholars at home and abroad,and has been widely used in various fields.With the rise of 5G and other technologies,the complexity and diversification of network traffic data flow have brought new difficulties to the field of network security.Therefore,applying semi supervised technology to the analysis of network traffic data has become one of the main methods.This paper introduces the characteristics and processing methods of current network traffic data,expounds the advantages of semi supervised learning in processing network traffic,summarizes the research progress of semi supervised learning in processing traffic analysis,and expounds the practical application of semi supervised learning in network traffic analysis from the aspects of semi supervised classification,semi supervised clustering and semi supervised dimensionality reduction.Finally,the challenges and new research directions of the current semi supervised network traffic analysis methods in the future are pointed out.
DDoS Attack Detection Method in SDN Environment Based on Renyi Entropy and BiGRU Algorithm
YANG Ya-hong, WANG Hai-rui
Computer Science. 2022, 49 (6A): 555-561.  doi:10.11896/jsjkx.210800095
Abstract PDF(3211KB) ( 727 )   
References | Related Articles | Metrics
Based on the bidirectional gated recurrent unit,BiGRU network can solve the gradient disappearance or gradient explosion problem of the traditional RNN model,a DDoS attack detection method in SDN environment based on Renyi entropy and bigru algorithm is proposed.First of all,the abnormal flow detection is carried out by Renyi entropy,and the detection is divided into normal and abnormal results.Traffic detected as abnormal will be detected using the BiGRU algorithm.Then,the switch is used to collect flow meter information,6 feature vectors are extracted as the characteristic vectors of attack detection.Finally,the network topology of the SDN is simulated by Minet,which is based on the controller OpenDaylight.The experimental results show that compared with SVM and BPNN neural network detection algorithm,the proposed detection scheme has improved detection accuracy,higher recognition rate and better comprehensive detection capability.
SMOTE-SDSAE-SVM Based Vehicle CAN Bus Intrusion Detection Algorithm
ZHOU Zhi-hao, CHEN Lei, WU Xiang, QIU Dong-liang, LIANG Guang-sheng, ZENG Fan-qiao
Computer Science. 2022, 49 (6A): 562-570.  doi:10.11896/jsjkx.210700106
Abstract PDF(6362KB) ( 686 )   
References | Related Articles | Metrics
With the rapid development of in-vehicle equipment intelligence on the Internet of Vehicles,due to its increasingly deepened connection with the Internet,the number of network attacks on the vehicle CAN bus has been increased,the attack methods have become more complex and the attack characteristics have become more concealed.At present,the intrusion detection of the Internet of Vehicles has just started.Traditional detection models based on firewall or rule bases are unable to obtain the hidden deep features of network attacks,but the intelligent detection models based on deep learning present problems such as “over-fitting” or “under-fitting” due to too many training parameters and unbalanced training datasets.To solve the above problems,an SMOTE-SDSAE-SVM based intrusion detection algorithm for CAN bus of vehicles is proposed in this paper,which is simply called 3S.This algorithm tries to combine deep learning and machine learning techniques to extract deep features of network attacks and ensure the efficiency of model training.The main contributions are as follows.Firstly,to balance the training samples of different categories,SMOTE method is used to generate more similar samples through the nearest neighbor sampling strategy.Secondly,sparse autoencoder and denoising autoencoder are combined to increase the speed of feature extraction and eliminate noise effects.And the deep feature of the CAN message is eventually extracted by stacking multi-layer sparse denoising autoencoder.Finally,SVM is used to accurately classify the extracted deep features of CAN messages,thereby discovering network attacks.According to the extensive experiments on the Volvo CAN dataset and the CAR-HACKING dataset,the proposed 3S algorithm is proved to have better accuracy and lower false alarm rate than other algorithms.
Security Analysis of A Key Exchange Protocol Based on Tropical Semi-ring
HUANG Hua-wei, LI Chun-hua
Computer Science. 2022, 49 (6A): 571-574.  doi:10.11896/jsjkx.210700046
Abstract PDF(1846KB) ( 419 )   
References | Related Articles | Metrics
This paper analyses the security of a key exchange protocol based on tropical semi-ring proposed by Grigoriev et al,and provides a method of algebraic cryptanalysis.Some tropical matrix equations are constructed according to the public information of the protocol.And the shared key of the protocol is obtained by solving the linear system of equations over tropical semi-ring.The parameters of the protocol should be increased appropriately for resisting the algebraic cryptanalysis.
Back-propagation Neural Network Learning Algorithm Based on Privacy Preserving
WANG Jian
Computer Science. 2022, 49 (6A): 575-580.  doi:10.11896/jsjkx.211100155
Abstract PDF(1942KB) ( 640 )   
References | Related Articles | Metrics
Back-propagation neural network learning algorithms based on privacy preserving are widely used in medical diagnosis,bioinformatics,intrusion detection,homeland security and other fields.The common of these applications is that all of them need to extract patterns and predict trends from a large number of complex data.In these applications,how to protect the privacy of sensitive data and personal information from disclosure is an important issue.At present,the vast majority of existing back-propagation neural network learning algorithms don't consider how to protect the data privacy in the process of learning.This paper proposes a back propagation neural network algorithm based on privacy-preserving,which is suitable for horizontally partitioned data.In the construction process of neural networks,it is need to compute network weight vector for the training sample set.To ensure the private information of neural network learning model can not be leaked,the weight vector will be assigned to all participants,so that each participant owns a part of private values of weight vector.In the calculation of neurons,we use secure multiparty computation,thus ensuring the middle and final values of the neural network weight vector are secure and will not be leaked.Finally,the constructed learning model will be securely shared by all participants,and each participant can use the model to predict the corresponding output for their respective target data.Experimental results show that the proposed algorithm has advantages over the traditional non-privacy protection algorithm in execution time and accuracy error.
Study on Key Technologies of Unknown Network Attack Identification
CAO Yang-chen, ZHU Guo-sheng, SUN Wen-he, WU Shan-chao
Computer Science. 2022, 49 (6A): 581-587.  doi:10.11896/jsjkx.210400044
Abstract PDF(3465KB) ( 855 )   
References | Related Articles | Metrics
Intrusion detection is a technology that proactively defends against attacks in the network and plays a vital role in network management.Traditional intrusion detection technology cannot identify unknown attacks,which is also a problem that has plagued this field for a long time.Aiming at unknown types of intrusion attacks,an unknown attack recognition model combining K-Means and FP-Growth algorithms is proposed to extract the rules of unknown attacks.First,for the data of a mixture of multiple unknown attacks,cluster analysis is performed with K-Means based on the similarity between samples,and the silhouette coefficient is introduced to evaluate the effect of clustering.After the clustering is completed,the same unknown attacks are classified into the same cluster,the feature of unknown attack is manually extracted,the feature data is preprocessed,the continuous feature is discretized,and then the frequent item sets and association rules of the unknown attack data are mined by the FP-Growth algorithm,and finally the rule unknown attack is obtained by analyzing it.The rules of attack are used to detect this type of unknown attack.The results show that the accuracy rate can reach 98.74%,which is higher than that of the related algorithms.
Quantitative Method of Power Information Network Security Situation Based on Evolutionary Neural Network
LYU Peng-peng, WANG Shao-ying, ZHOU Wen-fang, LIAN Yang-yang, GAO Li-fang
Computer Science. 2022, 49 (6A): 588-593.  doi:10.11896/jsjkx.210200151
Abstract PDF(3632KB) ( 519 )   
References | Related Articles | Metrics
Electric power information networks are facing increasingly severe threats of cyber attacks.Traditional quantification methods of network security situation only analyze from the perspective of network performance,ignoring the impact of the importance of various power application services on the security situation,making it difficult for the quantitative results to fully reflect the power information Cyber risk status.This paper proposes a power information network security situation quantification method based on evolutionary neural networks.First,by analyzing the characteristics of power communication network applications,a power communication network-oriented security situation architecture(PIN-NSSQ)is designed.Secondly,Starting from the three dimensions ofnetwork reliability,threat and vulnerability,combined with the importance of power business,a coupled and interconnected spatial element index system is established,to realize the mathematical representation of key element indicators.Then,integrating the BP neural network optimized by genetic evolution algorithm into the calculation process of element indicators,a power information network security situation quantification model based on evolutionary neural network is constructed to effectively realize the efficient calculation and precise quantification of the process of comprehensive perception of the power information network security situation.Finally,a simulation experiment environment is built according to a certain power department network topology to verify The effectiveness and robustness of the method proposed in the article.
Computer Network
Strategies for Improving Δ-stepping Algorithm on Scale-free Graphs
CHEN Jun-wu, YU Hua-shan
Computer Science. 2022, 49 (6A): 594-600.  doi:10.11896/jsjkx.210400062
Abstract PDF(2317KB) ( 574 )   
References | Related Articles | Metrics
The single-source shortest path problem is an important graph primitive.It computes the shortest paths in a weighted graph from a source vertex to every other vertex.Combining the classical Dijkstra's algorithm and Bellman-Ford's algorithm,Δ-stepping algorithm is used widely to complete SSSP computations in parallel settings.Due to the generic mechanism of preferential attachment,most large-scale networks are significantly skewed in the vertex degree distributions.This paper presents two strategies that utilize the skewness to improve the efficiency and scalability of Δ-stepping algorithm on large graphs.A preprocessing is executed on the input graph to compute an upper bound for the distance between every pair of vertices.The preprocessing results are used to reduce the redundant edge relaxations in Δ-stepping algorithm,so to improve the algorithm's efficiency and scalability in parallels settings.First,these upper bounds are used to reduce the relaxed edges.An edge is skippable in the SSSP computation when its weight is greater than upper bound of the distance between the connected vertices.Second,they are used to reduce relaxations repeated on the same edges.During the computation,a vertex is forbidden to be relaxed before its tentative distance to the source vertex has been updated to be less than the upper bound.Experimental results show that the improved algorithm on the Graph500 benchmark graphs provides nearly 10 ×performance improvement over its implementation on Graph500,and the performance improvement is between 2.68 and 5.58 on some real graphs.
Optimization of Offloading Decisions in D2D-assisted MEC Networks
FANG Tao, YANG Yang, CHEN Jia-xin
Computer Science. 2022, 49 (6A): 601-605.  doi:10.11896/jsjkx.210200114
Abstract PDF(2567KB) ( 649 )   
References | Related Articles | Metrics
Mobile edge computing(MEC) can provide convenient service for users due to the properties of resource subsidence.To further reduce the offloading pressure caused by massive users and smart applications,this paper utilizes device-to-device(D2D) technology to achieve reasonable utilization of idle local resources.That is,the compute-intensive users can offload their complex tasks to their idle neighbors in the proximity by D2D communications besides the local computing.First,the problem is formulated to minimize the aggregate delay of all users.Then,to analyze the resource competition among users and reduce the complexity,game theory is introduced and the multi-user cooperation offloading game is proposed.The proposed game is proved to be an exact potential game with at least one pure-strategy nash equilibrium(NE).Meanwhile,this paper proposes a better reply based distributed offloading algorithm to obtain the desired solution.Finally,simulation results show that the formulated game model and the proposed algorithm can decrease the network delay and average delay effectively,which validates the feasibility and effectiveness of our work.
WiFi-PDR Fusion Indoor Positioning Technology Based on Unscented Particle Filter
ZHOU Chu-lin, CHEN Jing-dong, HUANG Fan
Computer Science. 2022, 49 (6A): 606-611.  doi:10.11896/jsjkx.210700108
Abstract PDF(2658KB) ( 545 )   
References | Related Articles | Metrics
In order to improve the accuracy and stability of indoor positioning,this paper proposes an indoor positioning method based on WiFi-PDR fusion without trace particle filter.In order to reduce the influence of indoor complex environment on WiFi positioning,the weighted path loss algorithm is used to improve WiFi positioning.To reduce the cumulative effect of pedestrian track estimation errors,the walking period is divided by setting reference values and the acceleration data is smoothed and noise-reduced to improve the accuracy of step measurement.On the basis of improving WiFi and PDR positioning,a fusion positioning method using unscented particle filter is proposed,and the particle filter is optimized for robustness and adaptive to improve its robustness.Experimental simulation results show that this method can effectively improve the accuracy and stability of indoor positioning.
Strong Barrier Construction Algorithm Based on Adjustment of Directional Sensing Area
WANG Fang-hong, FAN Xing-gang, YANG Jing-jing, ZHOU Jie, WANG De-en
Computer Science. 2022, 49 (6A): 612-618.  doi:10.11896/jsjkx.210300291
Abstract PDF(3287KB) ( 549 )   
References | Related Articles | Metrics
K-barrier coverage is one of the hotspots in directional sensor networks.Traditional barrier construction algorithm consumes a lot of node energy and reduces the network lifetime.This paper innovatively exploits the adjustable characteristics of the direcitonal sensing area to efficiently construct directional barrier without consuming node energy.It firstly creates the adjustment of directional sensing area to reveal the regulation of sensing region adjustment.So that two nodes far from each other form continuous sensing regions without locomotivity.Then,it proposes a barrier construction scheme based on the adjustment of sensing area,optimizes and adjusts the directional sensing area,and selects optimal node to form barrier in distributed manner.Simulation results show that,compared with other methods using actuating capability,the proposed method could form barrier with less network resources,and achieve longer service lifetime.This research has important theoretical and practical significance.
Computation Offloading and Deployment Optimization in Multi-UAV-Enabled Mobile Edge Computing Systems
LIU Zhang-hui, ZHENG Hong-qiang, ZHANG Jian-shan, CHEN Zhe-yi
Computer Science. 2022, 49 (6A): 619-627.  doi:10.11896/jsjkx.210600165
Abstract PDF(2868KB) ( 604 )   
References | Related Articles | Metrics
The combination of unmanned aerial vehicles(UAVs) and mobile edge computing(MEC) technology breaks the limitations of traditional terrestrial communications.The effective line-of-sight channel provided by UAVs can greatly improve the communication quality between edge servers and mobile devices(MDs).To further enhance the quality-of-service(QoS) of MEC systems,a multi-UAV-enabled MEC system model is designed.In the proposed model,UAVs are regarded as edge servers to offer computing services for MDs,aiming to minimize the average task response time by jointly optimizing UAV deployment and computation offloading.Based on the problem definition,a two-layer joint optimization method(PSO-GA-G) is proposed.On one hand,the outer layer of the proposed method utilizes a discrete particle swarm optimization algorithm combined with genetic algorithm operators(PSO-GA) to optimize the UAV deployment.On the other hand,the inner layer of the proposed method adopts a greedy algorithm to optimize the computation offloading.Extensive simulation experiments verify the feasibility and effectiveness of the proposed method.The results show that the proposed method can achieve shorter average task response time,compared to other baseline methods.
Coverage Optimization of WSN Based on Improved Grey Wolf Optimizer
FAN Xing-ze, YU Mei
Computer Science. 2022, 49 (6A): 628-631.  doi:10.11896/jsjkx.210500037
Abstract PDF(2161KB) ( 529 )   
References | Related Articles | Metrics
How to use mobile nodes to maximize coverage and reduce energy consumption is an important direction in the research of wireless sensor networks.A grey wolf optimization(GWO) algorithm is proposed to solve the coverage problem of wireless sensor network by using the improved Levy flight strategy and the energy position fusion mechanism based on the circle mapping.Simulation results show that the improved GWO without considering energy has higher convergence speed and bigger coverage rate than the basic GWO and other related algorithms.After considering the energy,the coverage can still be guaranteed and the node life can be extended.
Enhanced ELM-based Superimposed CSI Feedback Method with CSI Estimation Errors
QING Chao-jin, DU Yan-hong, YE Qing, YANG Na, ZHANG Min-tao
Computer Science. 2022, 49 (6A): 632-638.  doi:10.11896/jsjkx.210800036
Abstract PDF(3587KB) ( 416 )   
References | Related Articles | Metrics
In massive multiple-input multiple-output(mMIMO) systems,the superimposed channel state information(CSI) feedback avoids the occupation of uplink bandwidth resources,while causing high calculation complexity and low feedback accuracy due to the superimposed interference,yet the actual application scenarios with CSI estimation errors are not considered.For these reasons,aiming at the superimposed CSI feedback in the scenario of CSI estimation errors and based on improving the extreme learning machine(ELM),this paper proposes enhanced ELM-based superimposed CSI feedback.First,the base station performs pre-equalization processing on the received signal to initially eliminate uplink channel interference.Then,the traditionalsuper imposed CSI feedback is iteratively unfolded by constructing an enhanced ELM network.This operation enhances the ability of the network to learn data distribution by standardizing the hidden layer output of each ELM network,thereby improving the accuracy of recoveries for downlink CSI and uplink user data sequences(UL-US).Experimental simulations show that compared with the classic and novel superimposed CSI feedback methods,the proposed method can obtain similar or better recovery accuracies for the downlink CSI and UL-US,while retaining the improvement robustness against the influence of different parameters.
Load Balancing Optimization Scheduling Algorithm Based on Server Cluster
TIAN Zhen-zhen, JIANG Wei, ZHENG Bing-xu, MENG Li-min
Computer Science. 2022, 49 (6A): 639-644.  doi:10.11896/jsjkx.210800071
Abstract PDF(2363KB) ( 737 )   
References | Related Articles | Metrics
In order to solve the problems of unbalanced request allocation and long task completion time when server clusters process concurrent task requests,a multi-objective optimal scheduling algorithm based on cuckoo search is proposed about server cluster load balancing.Firstly,according to the task request allocation characteristics of the server cluster,an optimization model related to the server real-time load information is constructed,which takes minimizing the task completion time and enhancing the effectiveness of load balancing as the objective function,by monitoring and recording the server real-time load information.And the decision variable is determined as the matching set between the task request and the server.Then,the non-dominated sorting cuckoo search algorithm with elite strategy is introduced to iteratively optimize the decision variables.Under the selection of fitness function,the Pareto solution set conforming to the global optimization is found.And the scheduling mechanism adjusts and forwards the tasks according to the determined optimal matching set.Simulation results show that the proposed scheduling algorithm can not only ensure the balance of server cluster but also reduce the task completion time as much as possible.Compared with other algorithm models,the scalability of the proposed algorithm is better.
Interdiscipline & Application
Application of Early Quantum Algorithms in Quantum Communication,Error Correction and Other Fields
Renata WONG
Computer Science. 2022, 49 (6A): 645-648.  doi:10.11896/jsjkx.210400214
Abstract PDF(1811KB) ( 402 )   
References | Related Articles | Metrics
At present,a development direction of quantum algorithm is to rethink the early quantum algorithms.Each of them involves an important,groundbreaking concept in quantum computing.They are generally considered to only belong to the theoretical category due to the fact that the problems they solve are of little practical value.However,theyare still important as they can solve a problem exponentially faster than a classical algorithm.Here,this paper elaborates on some recent developments in repurposing the early quantum algorithms for quantum key distribution and other fields.It especially focuses on Deutsch-Jozsa algorithm,Bernstein-Vazirani algorithm and Simon's algorithm.The Deutsch-Jozsa algorithm is used to determine whether a multi-argument function is balanced or constant.As recent research shows,it can be extended to application in the field of quantum communication and formal languages.The Bernstein-Vazirani algorithm finds a string encoded in a function.Its application can be extended to quantum key distribution and error correction.Simon's algorithm tackles the problem of identifying a string with a particular property.Its modern applications include quantum communication and error correction.
Optimization for Shor's Integer Factorization Algorithm Circuit
LIU Jian-mei, WANG Hong, MA Zhi
Computer Science. 2022, 49 (6A): 649-653.  doi:10.11896/jsjkx.210600149
Abstract PDF(2652KB) ( 630 )   
References | Related Articles | Metrics
With the help of techniques such as windowed arithmetic and the coset representation of modular integers,the overall optimization and resource estimation for the quantum circuit of Shor's algorithm has been shown.What's more,the simulation experiment of the designed quantum circuit has been carried out.The Toffoli gate and the depth of the overall circuit can be reduced by techniques such as windowed arithmetic and the coset representation of modular integers.The Toffoli count is 0.18n3+0.000 465n3log n and the measurement depth is 0.3n3+0.000 465n3log n.Due to the windowed semiclassical Fourier transform,the space usage includes 3n+O(log n) logical qubits.A tradeoff for resources consume between the time and the space has been made at the cost of adding some approximation errors.
Study on Improved BP Wavelet Neural Network for Supply Chain Risk Assessment
XU Jia-nan, ZHANG Tian-rui, ZHAO Wei-bo, JIA Ze-xuan
Computer Science. 2022, 49 (6A): 654-660.  doi:10.11896/jsjkx.210800049
Abstract PDF(2507KB) ( 442 )   
References | Related Articles | Metrics
In view of the impact of supply chain risks on upstream and downstream enterprises in the manufacturing industry,it is important to research the method of identification and evaluation for the supply chain risks.Firstly,based on supply chain operation reference model(SCOR) and taking automobile manufacturing enterprises as the research background,the identification process of supply chain risk indicators is studied by analyzing automobile supply chain risks and combining with the field survey results.An evaluation index system involving five risk categories,including strategic planning risk,procu-rement risk,manufacturing risk,distribution risk and return risk,is established.Secondly,considering that BP neural network model is prone to local optimal solution and other problems in the process of optimization evaluation,it is improved and optimized by increasing momentum,and the S-type function in the basic evaluation model is replaced by Morlet wavelet function to reconstruct the supply chain risk evaluation model.Finally,risk identification and assessment are studied with automobile enterprise of actual case,using the Matlab simulation to compare and analyze the improved BP wavelet neural network and fuzzy comprehensive evaluation,BP neural network,increased momentum of BP neural network.The results show that the improved BP wavelet neural network model has the better practicability and reliability.
Construction of Ontology Library for Machining Process of Mechanical Parts
WANG Yu-jue, LIANG Yu-hao, WANG Su-qin, ZHU Deng-ming, SHI Min
Computer Science. 2022, 49 (6A): 661-666.  doi:10.11896/jsjkx.210800013
Abstract PDF(3458KB) ( 762 )   
References | Related Articles | Metrics
In view of the common problem that enterprises cannot deeply utilize the typical part machining data resources scattered in various application systems,this paper proposes to realize the comprehensive integration of typical mechanical part machining process data by establishing a typical part machining process ontology library.Firstly,the typical part machining process knowledge is categorized into two types:theoretical knowledge and process knowledge,and a complete typical part machining process ontology library is established by using the ontology modeling language OWL.The research results of this paper provide strong support for building an intelligent manufacturing of the whole elements of typical parts processing.
Model Based on Spirally Evolution Glowworm Swarm Optimization and Back Propagation Neural Network and Its Application in PPP Financing Risk Prediction
ZHU Xu-hui, SHEN Guo-jiao, XIA Ping-fan, NI Zhi-wei
Computer Science. 2022, 49 (6A): 667-674.  doi:10.11896/jsjkx.210800088
Abstract PDF(3275KB) ( 494 )   
References | Related Articles | Metrics
Public-private partnership(PPP) projects can improve infrastructure,ensure people's livelihood,and promote the economic development,but there may be a huge loss of capitals and a serious waste of resources among the parties involved because of the characteristics of difficulty in withdrawing funds,long construction cycle and large numbers of participants.Thus,it is important to predict the risks of PPP projects scientifically and accurately.A risk prediction model based on spirally evolution glowworm swarm optimization(SEGSO) and back propagation neural network(BPNN) is proposed in this paper,which is applied for risk prediction in PPP infrastructure projects.Firstly,several strategies such as good point set,communication behavior,elite group and spiral evolution are introduced into the basic GSO,and SEGSO is proposed.Secondly,SEGSO is used to capture better initial weights and thresholds of BPNN to build a SEGSO-BPNN prediction model.Finally,the SEGSO algorithm searching performance is verified on five test functions,and the significance and validity of SEGSO-BPNN model are verified on seven UCI standard datasets.The model is applied to the risk prediction of Chinese PPP projects,and it gains good results,which provides a novel technique for PPP financing risk prediction.
Study on Cloud Classification Method of Satellite Cloud Images Based on CNN-LSTM
WANG Shan, XU Chu-yi, SHI Chun-xiang, ZHANG Ying
Computer Science. 2022, 49 (6A): 675-679.  doi:10.11896/jsjkx.210300177
Abstract PDF(3545KB) ( 956 )   
References | Related Articles | Metrics
The classification of satellite cloud images has always been one of the research hotspots in the field of meteorology.But there are some problems,such as the same cloud type has different spectral features,different cloud types have the same spectral features,and mainly use the spectral features and ignore spatial features.To solve the above problems,this paper proposes a cloud classification method of satellite cloud image based on CNN-LSTM,which makes full use of spectral information and spatial information to improve the accuracy of cloud classification.Firstly,the spectral features are screened based on the physical characteristics of the cloud,and the square neighborhood of the point cloud is used as the spatial information.Then,the convolutional neural network(CNN) is used to automatically extract the spatial features,which solves the problem of difficult classification with spectral feature alone.Finally,on this basis,combined with the spatial local difference features extracted by the long short-term memory(LSTM) network,it provides multi-view features for the classification of satellite cloud images,and solves the problem of misjudgment caused by the similarity of cloud spatial structure.Experimental results show that the overall classification accuracy of the proposed method for satellite cloud images reaches 93.4%,which is 2.7% higher than that of the single CNN method.
Complex Network Analysis on Curriculum System of Data Science and Big Data Technology
YANG Bo, LI Yuan-biao
Computer Science. 2022, 49 (6A): 680-685.  doi:10.11896/jsjkx.210800123
Abstract PDF(6398KB) ( 817 )   
References | Related Articles | Metrics
In recent years,more and more universities have begun to offer majors in data science and big data technology.As an emerging and popular multi-disciplinary major with wide caliber,its curriculum system is still being furthered improved.In this paper,we use complex network methods to analyze and visualize the course data set of 106 universities collected from the Internet.The course co-occurrence network and college relationship network are constructed respectively.For the highly coupled course co-occurrence network,a shell decomposition algorithm based on edge weights is proposed.The results are compared with the word frequency statistics and the frequent items obtained by the Apriori algorithm.Considering that this speciality can award a degree in science or engineering,the data set is divided into two sections science and engineering to analyze and visualize.This research can provide a certain reference to universities which are establishing or have established data science and big data technology speciality,and also provide an effective algorithm for the analysis of highly coupled networks.
Pop-up Obstacles Avoidance for UAV Formation Based on Improved Artificial Potential Field
CHEN Bo-chen, TANG Wen-bing, HUANG Hong-yun, DING Zuo-hua
Computer Science. 2022, 49 (6A): 686-693.  doi:10.11896/jsjkx.210500194
Abstract PDF(3556KB) ( 549 )   
References | Related Articles | Metrics
With the maturity of UAV-related technologies,the development prospects and potential application scenarios of UAVs are also recognized by more and more people.Among them,UAV formation can overcome the load,endurance and mission of a single UAV.Due to restrictions on types and other aspects,UAV formation flying is an important development direction in the future.During the flight,the UAV formation may be restricted by unknown obstacles such as new high-rise buildings and temporary no-fly zones.The main focus of current obstacle avoidance methods is to generate a reference flight path that does not intersect the obstacle for the UAV in the two-dimensional scene before the departure,with the obstacle information is known.However,this method is not flexible enough to meet the requirements of avoiding these unknown obstacles in the process of advancing in the actual three-dimensional environment.A formation collision avoidance system(FCAS) for collision risk perception is proposed.By analyzing the movement trend of UAVs,those UAVs within the formation that are most likely to collide are screened out,and the improved artificial potential field is used to unknown obstacles.Avoidance of such obstacles can effectively avoid collisions between UAVs within the formation during the obstacle avoidance process,effectively reduce the number of communication links within the formation,and minimize the impact of obstacles on the formation of UAVs.After the obstacle avoi-dance is completed,all UAVs will resume their original formation and return to the reference paths.Simulation results show that the system enables the UAV formation to deal with static unknown obstacles during the flight of the reference path,and finally reaches the destination without collision,thus verifying the feasibility of the strategy.
Evolutionary Game Analysis of WeChat Health Information Quality Optimization Based on Prospect Theory
WANG Xian-fang, ZHANG Liang, ZHANG Ning
Computer Science. 2022, 49 (6A): 694-704.  doi:10.11896/jsjkx.210900186
Abstract PDF(4001KB) ( 526 )   
References | Related Articles | Metrics
The quality of health information in WeChat varies from good to bad.Research on the evolutionary process of platform,official account and user behavioral decision-making,explore the key factors that prevent the official account from publi-shing false information,and provide a useful reference for optimizing the health ecological environment.By constructing a three-party game model,system equilibrium points and constraints are solved,then the influencing factors and optimal stable state of the system evolution are simulated and analyzed.Prospect theory is introduced to explore the influence of the subject's risk attitude and loss aversion on the optimal outcome.Simulation experiments show that the platform and users are more sensitive to risks than losses.Compared with the high cost of supervision,the platform pays more attention to the improvement of reputation.Compared with the misleading caused by false information,users focus on satisfying subjective needs.Official accounts are more sensitive to losses,and the sensitivity to platform penalties is greater than fan losses.When the initial willingness of the official account to release real information is low,although external factors such as platform punishment and media exposure can curb the spread of fake health information,the optimal system is difficult to achieve as soon as possible.
Dynamic Customization Model of Business Processes Supporting Multi-tenant
ZHANG Ji-lin, SHAO Yu-cao, REN Yong-jian, YUAN Jun-feng, WAN Jian, ZHOU Li
Computer Science. 2022, 49 (6A): 705-713.  doi:10.11896/jsjkx.210200104
Abstract PDF(2815KB) ( 456 )   
References | Related Articles | Metrics
Process customization is an essential means to realize personal services of business processes.It provides differential business services by adjusting the internal structure of business process model while using a single software system.However,with the increasing scale and complexity of business processes,the existing process customization technology needs to reconstruct the process model when dealing with those complex and changeable business processes,which affects the development efficiency of process customization.Therefore,providing an efficient process customization method has always been a research hotspot in the field of business processes.From the perspective of multi-tenant application,this paper proposes a dyna-mic customization model of business processes supporting multi-tenant.Firstly,the business sub-process is constructed by means of assembling variable task nodes and then tenant identify identification and process instance derivation are realized by tenant sensor.Secondly,a dynamicprocess customization method is provided for the varying requirements of tenants.Finally,combined with case analysis,the validity of the model is verified.
Teleoperation Method for Hexapod Robot Based on Acceleration Fuzzy Control
YIN Hong-jun, DENG Nan, CHENG Ya-di
Computer Science. 2022, 49 (6A): 714-722.  doi:10.11896/jsjkx.210300076
Abstract PDF(5078KB) ( 503 )   
References | Related Articles | Metrics
In order to solve the problem that conventional speed-level controller is hard to guarantee the speed tracking capability of hexapod robot,this paper proposes a bilateral teleoperation method based on acceleration fuzzy control.Firstly,a semi-autonomous mapping scheme between the master's position and the slave's velocity is established.Then,the relationship between the acceleration of the body and the drive value of the leg joint is determined.Secondly,a fuzzy PD control algorithm is used to design the control law of teleoperation system.On this basis,for improving the operating performance of the system,the velocity or force information is fed back to the operator in the form of haptic force,after the stability range of these control law parameters is analyzed by Llewellyn criterion.Finally,a semi-physical simulation platform is developed for experiment.Experimental results show that the proposed method is feasible,and the speed-tracking and force transparency of teleoperation system are obviously improved.
Application Research of PBFT Optimization Algorithm for Food Traceability Scenarios
LI Bo, XIANG Hai-yun, ZHANG Yu-xiang, LIAO Hao-de
Computer Science. 2022, 49 (6A): 723-728.  doi:10.11896/jsjkx.210800018
Abstract PDF(2361KB) ( 577 )   
References | Related Articles | Metrics
The characteristics of blockchain such as immutability and traceability can better support the food traceability system,and there are problems such as long delay,many nodes and high system overhead in the application of food traceability combined with blockchain technology.To address the above problems,an optimized PBFT algorithm trace-PBFT(t-PBFT) is proposed for the food traceability scenario based on the practical Byzantine fault tolerance(PBFT) algorithm.Firstly,the nodes in the supply chain are divided into three classes,and the node status is dynamically updated according to the actual communication volume of the nodes in the consensus,which is used to evaluate the reliability of the nodes as the basis for electing the master node.Secon-dly,the consistency protocol in the original algorithm is optimized to reduce the number of node communications by combining the characteristics of the food supply chain.Experimental results show that the t-PBFT algorithm performs better than the PBFT algorithm in terms of communication overhead,request delay and throughput.Finally,based on the t-PBFT algorithm and combined with the consortium chain,an architectural model to meet the demand of food traceability is proposed.It can record the data of each link in the food supply chain,ensure data traceability and the safety of food circulation process.
Diagnosis Strategy Optimization Method Based on Improved Quasi Depth Algorithm
ZHANG Zhi-long, SHI Xian-jun, QIN Yu-feng
Computer Science. 2022, 49 (6A): 729-732.  doi:10.11896/jsjkx.210700076
Abstract PDF(2111KB) ( 393 )   
References | Related Articles | Metrics
In the existing diagnostic strategy optimization methods,there are few researches on the unreliability test of multi-valued system,and it is difficult to fully consider the dual effects of multi-valued test and unreliability test on the optimization of diagnostic strategy.A quasi-depth algorithm based on tabu search is proposed.Firstly,the uncertain correlation matrix between fault and multi-valued test and the multi-valued unreliable diagnosis strategy are described.Then,aiming at the problem,the steps of the improved quasi-depth algorithm for tabu search are described.Finally,an example is given to verify the proposed algorithm.Experimental results show that the algorithm can reduce the algorithm complexity while ensuring the fault detection and isolation effect,and make the optimization process of diagnosis strategy more accurate and efficient.
Hybrid Housing Resource Recommendation Based on Combined User and Location Characteristics
PIAO Yong, ZHU Si-yuan, LI Yang
Computer Science. 2022, 49 (6A): 733-737.  doi:10.11896/jsjkx.210800062
Abstract PDF(2339KB) ( 518 )   
References | Related Articles | Metrics
With the development of the times,the idea of users to purchase houses has also changed,paying more attention to the location resources in their decision-making process.This paper proposes a hybrid recommendation method based on user and location resource characteristics to provide more accurate purchasing suggestions,where the content-based recommendation algorithm and user based collaborative filtering algorithm are combined in a cascade way.By integrating 170 000+ housing transaction with 1200+location resource data,experiment result shows that the proposed hybrid model has better recommendation effect than the traditional ones.
Dynamic Model and Analysis of Spreading of Botnet Viruses over Internet of Things
ZHANG Xi-ran, LIU Wan-ping, LONG Hua
Computer Science. 2022, 49 (6A): 738-743.  doi:10.11896/jsjkx.210300212
Abstract PDF(2736KB) ( 728 )   
References | Related Articles | Metrics
With the innovation and progress of imformation technology,Internet of things(IoT) technology grows explosively growth in various fields.However,devices over these networks are suffering the threat of hackers.The rapid growth of IoT-Botnets in recent years leads to many security occurrences including large-scale DDoS attacks,which brings IoT users severe damages.Therefore,it is significant to study the spread of a group of botnets represented by Mirai virus among IoT networks.In order to describe the formation process of IoT botnet precisely,this paper classifies the nodes of IoT devices into transmission devices and function devices,and then proposes SDIV-FB,a novel IoT virus dynamics model,through the analysis of Mirai virus propagation mechanism.The spreading threshold and equiliabrium of the model system are calculated,and the stability of the equiliabria are proved and analyzed.Moreover,the rationality of the derived theories are proved through the numerical simulation experiments,and the effectiveness of the model parameters are verified as well.Finally,decreasing the infection rate and increasing the recovery rate are proposed in this paper as two effective strategies for controlling the IoT botnets.
Study on Information Sharing and Channel Strategy of Platform in Consideration ofInformation Leakage and Information Investing Cost
XU Ming-yue
Computer Science. 2022, 49 (6A): 744-752.  doi:10.11896/jsjkx.211000055
Abstract PDF(2364KB) ( 489 )   
References | Related Articles | Metrics
A model in which a manufacturer sells products through an e-commerce platform and a traditional offline retailer is constructed.Specifically,the e-commerce platform selects online selling format and establishes information sharing strategy.This paper compares and analyses four situations where the online selling format is either reselling or agency selling with or without information sharing.Based on Bayesian game and information leakage effect,what is the e-commerce platform's choice,agency selling or reselling with the interaction of information investing cost and sharing strategy in dual-channel supply chain.Research shows that,firstly,unless the demand uncertainty is low and the platform revenue sharing ratio is high,the e-commerce platform selects the reselling format.Secondly,the e-commerce platform's incentive to share information strongly depends on its online selling format selection and demand uncertain degree.Under reselling case,the e-commerce platform does not share information voluntarily.While under agency selling case,the e-commerce platform shares information with manufacturer voluntarily when the demand uncertainty is high.Finally,under agency selling case,information sharing is beneficial for all members of supply chain.More specially,information sharing is not always beneficial for the entire dual-supply chain under reselling case.Only if the demand uncertainty is high and the cross-channel substitutability is small,information sharing can improve the performance of entire supply chain.
Development of Electric Vehicle Charging Station Distribution Model Based on Fuzzy Bi-objective Programming
QUE Hua-kun, FENG Xiao-feng, GUO Wen-chong, LI Jian, ZENG Wei-liang, FAN Jing-min
Computer Science. 2022, 49 (6A): 753-758.  doi:10.11896/jsjkx.210700225
Abstract PDF(2389KB) ( 458 )   
References | Related Articles | Metrics
With the popularization of electric vehicles,the number of public charging stations in cities cannot meet the growing demand for charging.Charging station construction usually requires multi-cycle and multi-level strategic planning,which is also affected by policies,economic environment and other factors.There are great uncertainties in the charging demand,the construction cost and operation cost in each charging station construction cycle.Considering the limited-service capacity of charging stations and the constraints of service radius,this paper develops a bi-objective fuzzy programming model that maximizes the charging satisfaction of electric vehicle users in the full cycle and minimizes the total cost of charging stations.Furthermore,a modified genetic algorithm based on adaptive and reverse search mechanisms is proposed to solve this problem.The results of the improved genetic algorithm and the standard genetic algorithm are compared in a case study.The performance of the model with different confidence levels and service radius of charging stations on the objective function are also verified.
Pedestrian Navigation Method Based on Virtual Inertial Measurement Unit Assisted by GaitClassification
YANG Han, WAN You, CAI Jie-xuan, FANG Ming-yu, WU Zhuo-chao, JIN Yang, QIAN Wei-xing
Computer Science. 2022, 49 (6A): 759-763.  doi:10.11896/jsjkx.211200148
Abstract PDF(3365KB) ( 661 )   
References | Related Articles | Metrics
Due to the degraded performance of pedestrian navigation system when foot-mounted IMU is out of range during vigo-rous activities or collisions,a novel pedestrian navigation method is proposed based on construction of virtual inertial measurement unit(VIMU) assisted by gait classification.Attention-based convolutional neural network(CNN) is introduced to classify the common gaits of pedestrian.Then the inertial data from pedestrian's thigh and foot is collected synchronously via actual IMUs as training and testing samples.For different gaits,the corresponding ResNet-gated recurrent unit(Resnet-GRU) hybrid neural network models are built.According to these models,virtual foot-mounted IMU is constructed for positioning in case of actual foot-mounted IMU overrange.Experiments show that,the proposed method brings enhanced performance of pedestrian navigation system based on zero velocity update when the foot motion of pedestrian is violent,which makes the navigation system more adaptable in complex and unknown terrains.The positioning error during comprehensive gait is about 1.43% of the total walking distance,which satisfies the accuracy requirement of military and civilian applications.
Model Medial Axis Generation Method Based on Normal Iteration
ZONG Di-di, XIE Yi-wu
Computer Science. 2022, 49 (6A): 764-770.  doi:10.11896/jsjkx.210400050
Abstract PDF(2564KB) ( 450 )   
References | Related Articles | Metrics
As the dimensionality reduction representation of model,the medial axis has been widely used in many engineering fields because of its good performance.At present,the method of generating the medial axis of the model is mainly based on the idea of approximating the medial axis,or the quality of the medial axis is not high,or the calculation time cost is high.As a result,a method of generating model medial axis based on normal iteration is proposed.The normal iteration method first discretizes the model into a triangular mesh model,and then performs GPU parallel tracking calculations based on the definition of the medial axis on the sample points and triangular faces.After multiple normal iterations,the medial axis points corresponding to all sample points are obtained.Finally,connecting the corresponding medial axis points according to the topological connectivity of the sample points to obtain the medial axis of the model.Experiment results show that the method can generate the model medial axis relatively quickly and accurately under different models,which verifies that the method improves the time efficiency and accuracy of the medial axis generation.
Study on Optimal Scheduling of Power Blockchain System for Consensus Transaction ofEach Unit
ZHOU Hang, JIANG He, ZHAO Yan, XIE Xiang-peng
Computer Science. 2022, 49 (6A): 771-776.  doi:10.11896/jsjkx.210600241
Abstract PDF(3687KB) ( 442 )   
References | Related Articles | Metrics
With the deepening of coupling of cold and heat electricity,distributed generation,medium-sized generators and their associated loads are gradually integrated into the grid.It brings new challenges to power grid operation.In order to solve the mutual trust problem between medium-sized users and large power grids,promote the local consumption of renewable energy and the coordinated complementarity of distributed energy,firstly,for the problems of opaque links in transactions between medium-sized users and large power grids,the consensus trading architecture between third-party operators and distributed multi-energy units is established by applying the consensus and mutual trust mechanism of blockchain.Secondly,the objective function is constructed based on the benefit cost of the operator and the operating cost of the distributed multi-energy units at different stages.The objective function is used to construct a mathematical model of its reasonable operation.Then,based on the encrypted broadcast mechanism of blockchain technologies,a consensus trading strategy for operators and distributed multi-energy units is proposed.It ensures that each unit achieves the operational objectives of mutual trust,multi-energy coordinated complementary and local consumption of renewable energy through the integration of operators.Finally,experimental simulation results verify the effectiveness and feasibility of the proposed model and method.
Study on Hybrid Resource Heuristic Loop Unrolling Factor Selection Method Based on Vector DSP
LU Hao-song, HU Yong-hua, WANG Shu-ying, ZHOU Xin-lian, LI Hui-xiang
Computer Science. 2022, 49 (6A): 777-783.  doi:10.11896/jsjkx.210400146
Abstract PDF(2260KB) ( 469 )   
References | Related Articles | Metrics
For modern microprocessors,the very long instruction word(VLIW) architecture integrating vector processing units has gradually become a typical representative of high-performance digital signal processor(DSP) architectures.This architecture is mainly characterized by rich register resources and many instruction execution units.Based on these characteristics,a selection method for the corresponding loop unrolling factor is proposed to improve the effect of loop unrolling optimization.This method takes into account the vector or scalar attribute of the code in a loop body,and the usage rules of base address registers and index registers.Moreover,another two heuristics,i.e.,the proportion of the times that the execution units are used and the power alignment of unrolling factor,are used in the loop unrolling factor selection algorithm.The ability of this method in developing more instruction level parallelism is proved by experiments performed on three commonly used digital signal processing algorithms.Experiment results show that the average performance of the algorithms improves by more than 10% compared with the existing methods.In particular,experiments on FFT algorithm show that the proposed method can analyze the usage of related hardware resources more accurately through the hybrid resource heuristics,and makes the judgment of unrolling and obtains the corresponding value of loop unrolling factor.
Study on Machine Learning Algorithms for Life Prediction of IGBT Devices Based on Stacking Multi-model Fusion
WANG Fei, HUANG Tao, YANG Ye
Computer Science. 2022, 49 (6A): 784-789.  doi:10.11896/jsjkx.210400030
Abstract PDF(2609KB) ( 893 )   
References | Related Articles | Metrics
Insulated gate bipolar transistor(IGBT) device is a kind of core technology component which is widely used in industry,communication,computer,automotive electronics,and other fields.It is very important to improve the safety of the device.Recently,the prediction of the life of IGBT devices by machine learning has become a hot research issue.However,the common neural network prediction still has problems of long training time and low accuracy.In order to solve these problems,a machine learning model based on Stacking multi-model fusion is proposed to realize IGBT life prediction.The model effectively improves the accuracy and efficiency of prediction.The algorithm consists of a two-layer structure,which incorporates four complementary machine learning algorithm models.In the first layer,the mild gradient lift tree model(LGBM),extreme gradient lift tree model(XGBoost),and ridge regression model are used to predict the IGBT life,and then the prediction results are input into the second layer for training.The second layer uses a linear regression model and the final IGBT life is predicted by the two-layer model training.Through the comparison of experimental data,it is confirmed that the machine learning model based on Stacking multi-model fusion is superior to the commonly used long and short-term memory neural network(LSTM) algorithm model,its MSE of IGBT life prediction is 93% lower than that of the LSTM algorithm model.What's more,the average time of model training is reduced to 13% of the LSTM algorithm model.
Application of Grassberger Entropy Random Forest to Power-stealing Behavior Detection
QUE Hua-kun, FENG Xiao-feng, LIU Pan-long, GUO Wen-chong, LI Jian, ZENG Wei-liang, FAN Jing-min
Computer Science. 2022, 49 (6A): 790-794.  doi:10.11896/jsjkx.210800032
Abstract PDF(2328KB) ( 414 )   
References | Related Articles | Metrics
Power stealing seriously endangers the grid security.In order to improve the efficiency of electricity theft detection,this paper proposes a novel method for electricity stealing detection based on Grassberger entropy random forest.First,KPCA is applied to reduce the dimensionality of the original power time series for extracting the user power consumption characteristics.Then,considering the unbalance of the number of theft samples and normal samples,the data under sampling method is used to establish multiple quantitatively balanced sample subsets.The random forest with improved Grassberger entropy is used tocompute informantion gain,so as to improve the accuracy of the model in power theft detection.Finally,the electricity consumption dataset of China Southern Power Grid is used to verify the power stealing detection effect of the proposed model.
Optimization and Simulation of General Operation and Maintenance Path Planning Model for Offshore Wind Farms
TAN Ren-shen, XU Long-bo, ZHOU Bing, JING Zhao-xia, HUANG Xiang-sheng
Computer Science. 2022, 49 (6A): 795-801.  doi:10.11896/jsjkx.210400300
Abstract PDF(2342KB) ( 628 )   
References | Related Articles | Metrics
The path planning of offshore wind farm operation and maintenance is a challenging and complex task,which needs to determine the resources and transport paths needed by the operation and maintenance,so as to minimize the total operation and maintenance cost.In this paper,the abstract class method is adopted in the modeling of offshore wind farm operation and maintenance planning,and a general operation and maintenance path planning model framework is established.This model is conducive to the compatibility of different offshore wind farm operation and maintenance path planning and scheduling decision-making tasks.improve the scalability of the model and the flexibility of multi-scenario application.In this paper,an improved adaptive large neighborhood search algorithm (ALNS),is proposed to solve the general operation and maintenance path planning model based on abstract class on the basis of the algorithm with multiple destroy and repair operators.Finally,the data of a domestic wind farm is selected for simulation experiment.By comparing the results of six operators within ALNS,and comparing the results of ALNS with the results of accurate algorithm,the results show that the algorithm optimization has better effect and reliability.
Vehicle Controller of Pure Electric Vehicles Based on CAN Bus
CHEN Long-hua, LI Hong-lin, YANG Han-li, XIAN Guang-mei, ZHANG Yan-qi
Computer Science. 2022, 49 (6A): 802-807.  doi:10.11896/jsjkx.220300133
Abstract PDF(5406KB) ( 702 )   
References | Related Articles | Metrics
With the gradual enhancement of people's awareness of environmental protection and harmonious coexistence with nature,more and more people begin to pay attention to environmental pollution,especially the exhaust emissions of diesel and gasoline vehicles,which has become an urgent problem to be solved in many countries.Focusing on the energy saving and zero emission of electrical cars,this paper puts forward an pure electric vehicle controller based on CAN bus,and studies the power charging and power changing of electrical vehicles.Under the premise of ensuring the safety of power charging and changing,this vehicle controller tests in advance whether the vehicle is ready to work,then controls the process of power charging and changing of the vehicle.At the same time,the vehicle controller has the online upgrade function,which can realize the remote upgrade of the vehicle,and further provides convenience to users.