Computer Science
Started in January,1974（Monthly）
Supervised and Sponsored by Chongqing Southwest Information Co., Ltd.
ISSN 1002-137X
CN 50-1075/TP
CODEN JKIEBK
Authors
Reviewers
Editors

Featured ArticlesMore...

Current Issue
  • Volume 51 Issue 8, 15 August 2024
      
      Discipline Frontier
      Driving Towards Intelligent Future:The Application of Deep Learning in Rail Transit Innovation
      SUN Yumo, LI Xinhang, ZHAO Wenjie, ZHU Li, LIANG Ya’nan
      Computer Science. 2024, 51 (8): 1-10.  doi:10.11896/jsjkx.240300099
      Abstract ( 147 )   PDF(3613KB) ( 323 )   
      References | Related Articles | Metrics
      Nowadays,rail transit plays a crucial role in urban transportation due to its convenience and efficiency.However,the operation of existing rail transit systems faces complex challenges.Processes such as passenger flow prediction and train scheduling still rely on manual methods,leading to low efficiency and accuracy,which has a certain impact on the system’s performance.In recent years,with the flourishing development of deep learning,its powerful feature extraction and image recognition capabilities provide more possibilities for the automation and intelligence of rail transit.This paper first outlines the challenges faced by current rail transit in various real-life scenarios,and then analyzes the main applications of deep learning in the rail transit field,including perception tasks,prediction tasks,optimization tasks,etc.Finally,the future development direction of deep learning in rail transit is prospected from four aspects:high-precision and robust safety detection,lightweight rail transit models,fully automated intelligent operation of rail transit,and efficient information processing through cloud computing and big data.
      Advancements and Prospects in Dysarthria Speaker Adaptation
      KANG Xinchen, DONG Xueyan, YAO Dengfeng, ZHONG Jinghua
      Computer Science. 2024, 51 (8): 11-19.  doi:10.11896/jsjkx.230700161
      Abstract ( 79 )   PDF(2194KB) ( 209 )   
      References | Related Articles | Metrics
      Automatic speech recognition tools make communication between dysarthria and normal individuals smoother,therefore,dysarthric speech recognition has become a hot research topic in recent years.The research on dysarthric speech recognition includes:collecting pronunciation data from dysarthria and normal individuals,representing acoustic features of dysarthria speech and normal speech,comparing and recognizing the content of pronunciation by machine learning model,and locating differences,so as to help dysarthria to improve their pronunciation.However,due to the significant difficulties in collecting a large amount of speech data from dysarthria,and the strong variability of their pronunciation,the performance of universal speech recognition models is often poor.To address this issue,many studies have proposed to introduce speaker adaptation methods into dysarthric speech recognition.Through extensive research on relevant literature,it has been found that current research mainly focuses on analyzing dysarthria speech in the feature domain and model domain.This paper focuses on analyzing how feature transformation and auxiliary features solve the differential representation of speech features,how linear transformation of acoustic models,fine-tuning of acoustic model parameters,and domain adaptation methods based on data selection improve the accuracy of model recognition.Finally,the current problems encountered in the research of dysarthria speaker adaptation are summarized,and it is pointed out that future research can improve the effectiveness of dysarthric speech recognition models from the perspectives of analyzing speech variability,fusing multi-feature and multi-modal data,and using a small number of speaker adaptation methods.
      Database & Big Data & Data Science
      Review of Outlier Detection Algorithms
      KONG Lingchao, LIU Guozhu
      Computer Science. 2024, 51 (8): 20-33.  doi:10.11896/jsjkx.230600052
      Abstract ( 85 )   PDF(3902KB) ( 255 )   
      References | Related Articles | Metrics
      Outlier detection,as an important research direction in the field of data mining,aims to discover data points in a dataset that are different from the majority and have potential analytical value,assistresearchers in identifying potential issues in the data source.Currently,outlier detection has been widely applied in various domains such as fraud detection,smart healthcare,intrusion detection,and fault diagnosis.This study,based on summarizing previous experiences,first discusses the definition of outliers,their causes,and typical application domains.It reviews the advantages and limitations of classical outlier detection algorithms such as DBSCAN and LOF,as well as their improved algorithms.Additionally,it analyzes the advantages of deep learning me-thods in the field of outlier detection.Secondly,considering the requirements for processing massive,high-dimensional,and temporal data in the current internet context,further research is conducted on the development status of outlier detection algorithms in new environments.Finally,the evaluation indicators of outlier detection algorithms,the role of cost factors in outlier detection evaluation,as well as commonly used toolkits and datasets,are introduced.The challenges and future development directions of outlier detection are summarized and prospected.
      Study on Classification and Grading Allocation of Data Property Rights Protection Rules UnderC&M Framework
      CONG Yingnan, PENG You, ZHU Jinqing
      Computer Science. 2024, 51 (8): 34-44.  doi:10.11896/jsjkx.240100030
      Abstract ( 57 )   PDF(1758KB) ( 209 )   
      References | Related Articles | Metrics
      In the critical period of digital transformation of society and economy,the establishment of an efficient data factor market is an important foundation and basic prerequisite for the sustainable and rapid development of the digital economy,and is also a multidisciplinary issue of the times.The property rights protection system of data is the basic system of data factor market,and there are hundreds of theoretical discussions on it,which together with legal regulations and judges’ opinions constitute a “list of rules” to choose from.The C&M framework provides a method of rule selection based on economic efficiency,which is in line with the policy objective of building an efficient data factor market.This paper compares and selects rules for the protection of property rights of personal data,enterprise data and public data from two perspectives:ex ante efficiency and ex post efficiency,and concludes that for personal data and enterprise data,liability rules providing ex post remedies are more efficient than property rules granting absolute property rights,while for public data,each has its own advantages.Based on this,this paper further proposes a legislative proposal to build a “three-tier structure” protection model and a data anonymization system for personal data protection,the direction to build non-absolute property rights for enterprise data protection,and a proposal to establish a classified and graded openness pattern with three types of rules for public data.
      Interpretable Credit Evaluation Model for Delayed Label Scenarios
      XIN Bo, DING Zhijun
      Computer Science. 2024, 51 (8): 45-55.  doi:10.11896/jsjkx.230900107
      Abstract ( 47 )   PDF(3048KB) ( 168 )   
      References | Related Articles | Metrics
      With the rapid development of social economy,credit business plays an increasingly important role in the financial field,and using machine learning algorithms for credit evaluation has become the mainstream method.However,there are still some problems to be solved,such as the inadequacy of labeled data and model lag caused by delayed labels,and the lack of interpretability in dynamic credit evaluation models.To address these problems,this paper proposes an interpretable credit evaluation model for delayed label scenarios.Built upon the foundation of dynamic model trees,the model incorporates weighted enhancements.It combines delayed label update algorithms and a pseudo-label selection strategy with adaptive thresholds,treating delayed label data as both feedback data and pseudo-label data,effectively mitigating the impacts of insufficient labeled data and model lag.Moreover,the model achieves interpretability.It is finally tested on some synthetic and real credit evaluation datasets,demonstrating superior balance between predictive performance and interpretability compared to other mainstream algorithms.
      Power-PCSR:An Efficient Dynamic Graph Storage Structure for Power-law Graphs
      MAO Zhixiong, LIU Zhinan, GAO Xuning, WANG Mengxiang, GONG Shufeng, ZHANG Yanfeng
      Computer Science. 2024, 51 (8): 56-62.  doi:10.11896/jsjkx.231000155
      Abstract ( 41 )   PDF(2377KB) ( 153 )   
      References | Related Articles | Metrics
      Graph data is widespread in real life and changes over time.The traditional efficient static graph storage structure,compressed sparse row/column(CSR/CSC) requires a large amount of data migration when inserting/deleting edges to/from graphs,which is not suitable for dynamic graphs.Although the adjacency list(AL) is able to update graphs efficiently,it is inefficient in reading and analyzing graphs since it has a large number of pointers,which results in random memory access.PCSR is a novel dynamic graph storage structure based on CSR.It employs a packed memory arrays(PMA) to store the edges rather than a continuous array.Because there are empty slots in PMA,it is easier to insert/delete edges.Thus,packed compressed sparse row(PCSR) is efficient in both graph updating and analysis.However,we find that the performance of PCSR suffers from large degree vertices when storing power-law graphs.For this,this paper proposes a new graph storage structure based on PCSR,Power-PCSR,which supports efficient updating and analysis of dynamic power-law graphs.In Power-PCSR,each large-degree vertex is stored in an independent PMA separately,and other vertices with small degrees are stored in a PMA.The data migration caused by the small-degree vertices will not lead to the migration of large-degree vertices,thus greatly reducing the amount of data migration.Similarly,the data migration caused by the update of large-degree vertices is only limited to its PMA,and will not involve the data migration of other large-degree vertices and small-degree vertices.Experiments show that Power-PCSR has similar performance to PCSR when analyzing graphs,and is 2 times faster than PCSR when updating graph data.
      Out-of-Distribution Hard Disk Failure Prediction with Affinity Propagation Clustering and Broad Learning Systems
      WANG Yiyang, LIU Fagui, PENG Lingxia, ZHONG Guoxiang
      Computer Science. 2024, 51 (8): 63-74.  doi:10.11896/jsjkx.230600103
      Abstract ( 44 )   PDF(3516KB) ( 180 )   
      References | Related Articles | Metrics
      Hard disk is the primary storage device in cloud data centers,and hard disk failure prediction is crucial for ensuring data security.However,there is a significant imbalance between failure and healthy SMART samples,which can lead to model bias.Moreover,hard disk models exhibit varying data distributions.Prediction models trained on specific hard disk data may not be suitable for other hard disks.To address these issues,this paper proposes a method for out-of-distribution hard disk failure prediction by combining the AP clustering algorithm and the broad learning system.To tackle the sample imbalance problem,this paper uses the AP clustering algorithm to cluster samples close to failure and treats all samples in the cluster containing determined failure instances as additional failure samples.To address the distribution differences of hard disk models,this paper combines the manifold regularization framework and the broad learning system to learn the low-dimensional structure of hard disk data,thereby improving the model’s generalization ability to unknown data.Experimental results show that,on the dataset resampled by the AP clustering algorithm,the F1_Score of multiple methods increases by an average of 0.2 compared to the datasets resampled by comparative methods.Additionally,in the task of predicting out-of-distribution hard disk failures,the F1_Score of the proposed model increases by 0.1~0.2 compared to other methods.
      Knowledge Compatibility Representation and Reasoning in Incomplete Formal Contexts from Logical Perspective
      ZHANG Shaoxia, LI Deyu, ZHAI Yanhui
      Computer Science. 2024, 51 (8): 75-82.  doi:10.11896/jsjkx.240400104
      Abstract ( 35 )   PDF(1530KB) ( 156 )   
      References | Related Articles | Metrics
      The incomplete information in formal contexts leads to the incompatibility of knowledge,that is,implications cannot hold simultaneously in any completion of an incomplete formal context.Logical description is a methodology for representing knowledge from a semantic aspect and establishing inference rules with semantic coordination from a syntactic aspect.This paper firstly studies the compatibility semantic representation within incomplete data from a logical perspective,characterizes the soundness and compatibility of knowledge via incomplete instances,and constructs the most compact compatible set(namely compatible canonical basis).Secondly,this paper establishes inference rules with semantic soundness,compatibility,and completeness to avoid incompatible knowledge and invalid knowledge in knowledge reasoning.Finally,this paper applies the logical research results to incomplete formal contexts by introducing two types of implication forms,namely ↓↓-type implication and ↑↑-type implication,which are both compatible and more stringent than acceptable implication.The compatible canonical bases of the two types of implications are constructed and their completeness and non-redundancy are verified.
      Multi-granularity Intuitive Fuzzy Rough Set Model Based on θ Operator
      ZHENG Yu, XUE Zhan’ao, LYU Mingming, XU Jiucheng
      Computer Science. 2024, 51 (8): 83-96.  doi:10.11896/jsjkx.230600185
      Abstract ( 41 )   PDF(2223KB) ( 155 )   
      References | Related Articles | Metrics
      In order to solve the problem that it is difficult for decision makers to make accurate judgment when multiple attributes conflict with each other in the multi-attribute decision making.In the intuitive fuzzy approximation space,this paper firstly uses the membership degree,non-membership degree and fuzzy implication operator of intuitive fuzzy set,and proposes the concepts of membership degree and non-membership degree based on θ operator and θ* operator,and proves a series of properties of them.Then,in the intuitive fuzzy set and the multi-granularity rough set,the pessimistic and optimistic models of theintuitive fuzzy rough set based on θ operator are defined,and the related properties of the two models are discussed.Finally,a multi-attribute decision algorithm based on the multi-granularity intuitive fuzzy rough set model based on θ operator is presented.The evaluation of talents introduced by universities and the evaluation of businesses in the green economy supply chain of enterprises are analyzed as examples.The correctness of the proposed method is proved by comparing the results of the optimistic and pessimistic models with those of the existing methods.The effectiveness of the model algorithm is also verified.
      Adaptive Density Peak Clustering Algorithm Based on Shared Nearest Neighbor
      WANG Xingeng, DU Tao, ZHOU Jin, CHEN Di, WU Yunzheng
      Computer Science. 2024, 51 (8): 97-105.  doi:10.11896/jsjkx.230500226
      Abstract ( 35 )   PDF(4182KB) ( 168 )   
      References | Related Articles | Metrics
      Density peak clustering algorithm(DPC) is a simple and efficient unsupervised clustering algorithm.Although the algorithm can automatically discover cluster centers and realize efficient clustering of arbitrary shape data,it still has some defects.Aiming at the three defects of density peak clustering algorithm,which does not consider the location information of data when defining the correlation value,the number of clustering centers needs to be set manually in advance,and the chain reaction is easy to occur when distributing sample points,an adaptive density peak clustering algorithm based on shared nearest neighbor is proposed.Firstly,the shared nearest neighbor is used to redefine the local density and other measures,and the local characteristics of data distribution are fully considered,so that the spatial distribution characteristics of sample points can be better reflected.Se-condly,by introducing the phenomenon of density attenuation,the sample points are automatically gathered into micro-clusters,which realizes the adaptive determination of cluster number and the adaptive selection of cluster center.Finally,a two-stage distribution method is proposed,in which the micro-clusters are merged to form the backbone of the cluster,and then the backbone of the cluster allocated in the previous step guides the distribution of the remaining points,avoiding the occurrence of chain reactions.The implementation on two dimensional composite datasets and UCI datasets shows that this algorithm has better perfor-mance in most cases than the classical density peak clustering algorithm and its improved algorithms in recent years.
      Measurable Shapelets Extraction Based on Symbolic Rrepresentation for Time Series Classification
      WANG Liqin, WAN Yuan, LUO Ying
      Computer Science. 2024, 51 (8): 106-116.  doi:10.11896/jsjkx.230500161
      Abstract ( 32 )   PDF(3160KB) ( 151 )   
      References | Related Articles | Metrics
      In the time series classification problems,shapelets extraction method based on symbol representation has good classification accuracy and efficiency,but the quality measurement of symbols,such as calculating TFIDF scores,is time-consuming and computatively heavy,leading to low classification efficiency.In addition,there are still a large number of shapelets candidates extracted,and the discriminating power needs to be improved.To solve these problems,this paper proposes a measurable shapelets extraction method based on symbolic representation,which includes three stages:time series data preprocessing,determining shapelets candidate set and learning shapelets,so that high-quality shapelets can be obtained quickly.In the data preprocessing stage,the time series is transformed into a symbolic aggregation approximation(SAX)representation to reduce the dimensions of the original time series.In the stage of determining the candidate set of shapelets,Bloom filters are used to filter repeated SAX words,and the filtered SAX words are stored in the hash table for quality measurement.Then,the similarity of SAX words is discriminated,and the final shapelets candidate set is determined based on the concepts of similarity and coverage.In the learning phase of shapelets,the logistic regression model is used to learn real shapelets for time series classification.In this paper,a large number of experiments are conducted on 32 datasets,and the experimental results show that the average classification accuracy and average classification efficiency of the proposed method rank second on 32 datasets.Compared with the existing time series classification methods based on shapelets,the proposed method can improve the classification efficiency while ensuring the accuracy,and has good interpretability.
      Hohai Graphic Protein Data Bank and Prediction Model
      WEI Xiangxiang, MENG Zhaohui
      Computer Science. 2024, 51 (8): 117-123.  doi:10.11896/jsjkx.231100014
      Abstract ( 42 )   PDF(1564KB) ( 133 )   
      References | Related Articles | Metrics
      Protein is a kind of substance with spatial structure.The main goal of protein structure prediction is to extract effective information from existing large-scale protein datasets,so as to predict the structure of proteins in nature.At present,one of the problems in protein structure prediction experiments is the lack of data sets that can further reflect the spatial structure of proteins.Although the current mainstream PDB(protein data bank) is experimentally measured,it does not utilize the spatial characteristics of proteins,and there are problems of doping nucleic acid data and partial data is incomplete.In view of the above pro-blems,this paper studies the prediction of protein from the perspective of spatial structure.Based on the original PDB,the Hohai graphic protein data bank is proposed.The dataset expresses the spatial structure characteristics of proteins based on the graph structure.Based on the traditional Transformer network model,relevant protein structure prediction experiments are carried out on the new dataset,and the prediction accuracy of HohaiGPDB could reach 59.38%,which proves the research value of Hohai-GPDB.The HohaiGPDB could be used as a general data set for protein-related studies.
      Intelligent Evidence Set Selection Method for Diverse Data Cleaning Tasks
      QIAN Zekai, DING Xiaoou, SUN Zhe, WANG Hongzhi, ZHANG Yan
      Computer Science. 2024, 51 (8): 124-132.  doi:10.11896/jsjkx.230900003
      Abstract ( 52 )   PDF(2188KB) ( 156 )   
      References | Related Articles | Metrics
      Due to the limitations of data cleaning algorithms designed specifically for individual data quality issues and their inability to effectively address multiple coexisting data quality enhancement requirements,a collaborative approach employing multiple data cleaning methods can be adopted to fulfill various data cleaning needs.This paper formulates the data cleaning problem as a task of evidence set generation and selection.By utilizing an incremental quality assessment scheme based on aggregate queries and an operator result selection scheme based on intermediate operator evidence sets,efficient data cleaning involving a combination of diverse cleaning methods is achieved across various cleaning tasks.In the proposed cleaning model,the operator repository yields data cleaning results and transforms them into intermediate operators.The sampler in the midstream module distributes and prunes the set of intermediate operators to provide the searcher with a high-quality candidate evidence set.The downstream searcher,guided by the quality evaluator,selects evidence sets.Upon completion of the search process,the upstream operator repository updates data and necessary parameters,facilitating the reiteration of intermediate operator generation.Finally,extensive experiments are conducted on three real-world datasets of varying scales.Performance verification across different data cleaning tasks demonstrates the feasibility of operator orchestration for any type of data cleaning requirement,underpinning the proposed method’s stable precision and recall in scenarios involving diverse data quality constraints,dynamics,and large-scale data clea-ning.Furthermore,a performance comparison with existing intelligent data cleaning systems reveals that the proposed method outperforms these systems by over 15% in tasks related to outlier detection,rule violations,and mixed errors,all within the same cleaning time.
      Computer Graphics & Multimedia
      Scene Segmentation Model Based on Dual Learning
      LIU Sichun, WANG Xiaoping, PEI Xilong, LUO Hangyu
      Computer Science. 2024, 51 (8): 133-142.  doi:10.11896/jsjkx.230700207
      Abstract ( 41 )   PDF(4682KB) ( 171 )   
      References | Related Articles | Metrics
      For complex tasks such as urban scene segmentation,there are problems such as low utilization of feature map space information,inaccurate segmentation boundaries,and excessive network parameters.To solve these problems,DualSeg,a scene segmentation model based on dual learning,is proposed.Firstly,depthwise separable convolution is used to significantly reduce the number of model parameters Secondly,accurate context information is obtained by fusing hollow pyramid pooling and double attention mechanism modules.Finally,dual learning is used to construct a closed-loop feedback network,and the mapping space is constrained by duality,while training the two tasks of “image scene segmentation” and “dual image reconstruction”,it can assist the training of the scene segmentation model,help the model to better perceive the category boundary and improve the recogni-tion ability.Experimental results show that the DualSeg model based on the Xception skeleton network achieves 81.3% mIoU and 95.1% global accuracy on natural scene segmentation dataset PASCAL VOC,respectively,and the mIoU reaches 77.4% on the CityScapes dataset,and the number of model parameters decreases by 18.45%,which verifies the effectiveness of the model.A more effective attention mechanism will be explored in the future to further improve the segmentation accuracy.
      Super-resolution Reconstruction for Low-dose CT Based on Guidance of Gradient
      XU Ying, ZHANG Daoqiang, GE Rongjun
      Computer Science. 2024, 51 (8): 143-151.  doi:10.11896/jsjkx.230700162
      Abstract ( 38 )   PDF(5756KB) ( 124 )   
      References | Related Articles | Metrics
      Low-dose CT(LDCT) scan plays a pivotal role in clinical practice,effectively decreasing cancer risks for radiologists and patients.However,the utilization of low-dose radiation introduces notable noise into the resulting CT images,highlighting the necessity of low-dose CT reconstruction.Another important task in the field of image reconstruction is super-resolution(SR),with the aim of achieving high-resolution CT imaging while minimizing computational expenses.High resolution CT images afford the capacity to capture intricate anatomical details in greater fidelity.Although significant progress has been made in their respective domains,there is still a lack of effective methodologies that can effectively harness the inherent correlation between these tasks and handle them concurrently.We employ edge information as a link between the two tasks,and utilize gradients to extract shared features from both tasks.This allows the LDCT reconstruction process to assist the SR reconstruction process and gene-rate resulting images with sharp edges.Our work consists of three components:1)Edge-enhanced framework.The framework fully exploits the correlation between the two tasks by extracting relevant features using gradient information,enabling the denoising(DN)task to assist the SR task in achieving superior performance.2)Gradient guided fusion block(GGFB),which enhances the highly correlated edge features while suppressing irrelevant features,thereby enabling effective reconstruction in edge regions.3)Gradient loss,which introduces richer gradient features into the model and guides the network to prioritize the reconstruction of edge regions.Extensive experiment demonstrates that our noise reduction and super resolution reconstruction network(NRSR-Net)achieves promising PSNR,SSIM,and LPIPS in quantitative evaluations,as well as gains high-quality readable visualizations.All of these advantages demonstrate the great potential of NRSR-Net in clinical CT imaging.
      Few-shot Image Classification Based on Pseudo-label Dependence Enhancement and NoiseInterferenceReduction
      TANG Ruiqi, XIAO Ting, CHI Ziqiu, WANG Zhe
      Computer Science. 2024, 51 (8): 152-159.  doi:10.11896/jsjkx.230500066
      Abstract ( 36 )   PDF(3172KB) ( 146 )   
      References | Related Articles | Metrics
      The success of deep learning in image classification relies heavily on large-scale data.However,in many application scenarios,it is difficult to collect enough data for model training.Therefore,few-shot learning aimed at obtaining high-performance models with limited data becomes a hot research direction.In the field of few-shot image classification,using unlabeled data to augment the training datasets is a common method,but it faces two urgent problems:how to obtain pseudo-labels of unlabeled data and how to mitigate the negative impact of accumulated noise labels？ Firstly,in order to obtain high-quality pseudo-labels,it is necessary to solve the problem of noise labels caused by the distribution shift of the source domain and the target domain.A dependence enhancement method based on Hilbert-Schmidt independent criterion is proposed to improve the prediction reliability of pseudo-labels by maximizing the correlation between image feature representation and labels.Secondly,in order to overcome the problem of label prediction error that accumulates over time,a noise label interference reduction method is proposed to ensure that the gradient of samples with correct labels always dominates the training dynamics,so as to lead the model to the optimal solution.The above methods are evaluated on the benchmark datasets for few-shot image classification,namely mini-ImageNet and tiered-ImageNet.The results demonstrate that the proposed approach effectively utilizes unlabeled data to improve classification accuracy and achieves impressive classification performance.
      Task-aware Few-shot SAR Image Classification Method Based on Multi-scale Attention Mechanism
      ZHANG Rui, WANG Ziqi, LI Yang, WANG Jiabao, CHEN Yao
      Computer Science. 2024, 51 (8): 160-167.  doi:10.11896/jsjkx.230500171
      Abstract ( 39 )   PDF(2289KB) ( 150 )   
      References | Related Articles | Metrics
      Aiming at the problem of the lack of labeled samples in SAR image classification,this paper proposes a task-aware few-shot SAR image classification method based on multi-scale attention mechanism.In order to fully mine local features and focus on the key local semantic patches under specific tasks,this paper introduces two effective attention mechanisms to obtain more efficient and rich feature representation.First,in the feature extraction stage,the complemented squeeze-and-excitation attention block(CSE Block) is used to focus on the salient features of different semantic parts of the original features.It can extract secon-dary salient features from the suppressed features and merge them with the main salient features,which can obtain more efficient and rich feature representation.Subsequently,an adaptive episodic attention block(AEA Block) is used to obtain key semantic patches in the entire task,which can enhance the differentiated information between tasks and improve the accuracy of SAR image classification tasks.The results show that the classification accuracy of the 5-way 1-shot task is 2.9% higher than that of the sub-optimal task on the SAR image classification standard MSTAR dataset.In the two tasks,the runtime of the proposed method is the same as other metric-learning methods,without additional excessive computing resources,which verifies its effectiveness.
      Road Extraction Algorithm for Remote Sensing Images Based on Improved DeepLabv3+
      WANG Qian, HE Lang, WANG Zhanqing, HUANG Kun
      Computer Science. 2024, 51 (8): 168-175.  doi:10.11896/jsjkx.230600118
      Abstract ( 36 )   PDF(3325KB) ( 143 )   
      References | Related Articles | Metrics
      Road extraction can help us better understand the urban environment and is an important part of urban transportation and planning.With the development of deep learning and computer vision,the use of deep learning-based semantic segmentation algorithm to extract roads from remote sensing images has become increasingly mature.However,existing deep learning road extraction algorithms suffer from slow extraction speed and susceptibility to background environmental factors,resulting in missed segmentation and discontinuity.To address these issues,a lightweight algorithm called CE-DeepLabv3+ based on ECANet attention mechanism and cascade atrous spatial pyramid pooling module is proposed.Firstly,the main feature extraction network is replaced with the lightweight MobileNetv2 to reduce parameter volume and improve model execution speed.Secondly,the convolution layers of the atrous spatial pyramid pooling module are further expanded to increase the receptive field,and different feature layers are cascaded to enhance semantic information reuse,thereby improving the ability to extract fine-grained features.Thirdly,the ECANet attention mechanism is added to suppress environmental interference and focus on road information.Finally,an improved loss function is used for training to address the impact of road and background sample imbalance on model performance.Experimental results show that the improved algorithm achieves excellent performance,with significant improvements in segmentation efficiency and accuracy compared to the original DeepLabv3+ algorithm.
      Color Transfer Method for Unpaired Medical Images Based on Color Flow Model
      WANG Xiaojie, LIU Jinhua, LU Shuyi, ZHOU Yuanfeng
      Computer Science. 2024, 51 (8): 176-182.  doi:10.11896/jsjkx.230700088
      Abstract ( 27 )   PDF(3774KB) ( 151 )   
      References | Related Articles | Metrics
      In clinical applications,CT image is a kind of image data that is relatively easy to obtain,but there is a large gap between them and the real human body color.The tomographic color image of the human body is the color response of the real human body,but it is a rare data.Combining the two,so that each case can get its own color CT data,which will have a effect on the doctor’s surgery and the patient’s understanding to the disease.Therefore,this paper proposes a medical image colorization framework based on a color flow model.It first inputs the CT and human color data into the color flow model and extracts the content and color features.Then,the color and texture transfer work is performed at the feature level.Finally,the processed feature information is re-input into the reversible color flow model for image reconstruction.After each flow module,we add a texture constraint loss to make the shaded image more textured.At the same time,we add edge constraints to ensure that the characteristics of small blood vessels and other tissues on the medical image are not lost.Qualitative and quantitative experiments prove that our method is more robust than other colorization methods,and the experimental results are more realistic.And we conduct extensive experiments on different data domains,proving that our method is not affected by domain shift and can obtain stable experimental results.At the same time,the proposed method can display a clear organizational structure without adjusting the window width/level.
      Parallel Multi-scale with Attention Mechanism for Point Cloud Upsampling
      XIAO Xiao, BAI Zhengyao, LI Zekai, LIU Xuheng, DU Jiajin
      Computer Science. 2024, 51 (8): 183-191.  doi:10.11896/jsjkx.230500094
      Abstract ( 38 )   PDF(5481KB) ( 141 )   
      References | Related Articles | Metrics
      The current deep learning-based point cloud upsampling method lacks the attention to a local area feature correlation and multi-scale extraction of global features,resulting in the dense output point cloud with too many outliers and low fine-grained granularity.To solve the above problem,a parallel multi-scale with attention mechanism for point cloud upsampling(PMA-PU) network is proposed,which consists of a feature extractor,a feature expander,a coordinate refiner and a coordinate reconstructor.Firstly,giving an N×3 sparse point cloud as input,a parallel multi-scale feature extraction module(PMA) with an embedded attention mechanism is designed to map the point cloud in 3D space to the high-dimensional feature space to obtain the global and local feature information of the point cloud.Secondly,the high-dimensional point cloud features are obtained after expanding the dimensionality of the point cloud features using the edge convolution feature expander to better preserve the edge information of the point cloud features,and the high-dimensional point cloud features are converted back to the 3D space by the coordinate reconstructors.Finally,the output point cloud details are fine-tuned by using the coordinate refiners.The results of the PMA-PU comparison experiments on the synthetic dataset PU1K show that the generated dense point cloud has significant improvement in the three evaluation metrics,Chamfer Distance(CD),Hausdorff Distance(HD),and P2F(point-to-surface),which are significantly better than the second highest performance.The network models with the second highest performance are optimized by 7.863%,21.631%,and 14.686%,respectively.The visualization results demonstrate that PMA-PU has a better performce feature extractor,which can generate dense point clouds with higher fine granularity and a shape closer to the true value.
      Monocular 3D Object Detection Based on Height-Depth Constraint and Edge Fusion
      PU Bin, LIANG Zhengyou, SUN Yu
      Computer Science. 2024, 51 (8): 192-199.  doi:10.11896/jsjkx.230500071
      Abstract ( 32 )   PDF(2973KB) ( 146 )   
      References | Related Articles | Metrics
      Monocular 3D object detection aims to complete 3D object detection using monocular images,and most existing monocular 3D object detection algorithms are based on classical 2D object detection algorithms.To address the issue of inaccurate instance depth estimation through direct regression in monocular 3D object detection algorithms,which leads to poor detection accuracy,a monocular 3D object detection algorithm based on height-depth constraint and edge feature fusion is proposed.In the instance depth estimation method,the height-depth constraint is calculated by the instance 3D height and 2D height under the geometric projection relationship,mainly converting the prediction of instance depth into the prediction of 2D height and 3D height of the object.To address the issue of object truncation at image edges in monocular images,an edge fusion module based on depth separable convolution is used to enhance the feature extraction of edge objects.For the multi-scale problem caused by the proximity and distance of objects in the image,a multi-scale mix attention module based on dilated convolution is designed to enhance the multi-scale feature extraction of the highest layer feature map.Experimental results demonstrate the effectiveness of the proposed method,as it achieves a 7.11% improvement in car category detection accuracy compared to the baseline model on the KITTI dataset,outperforming the current methods.
      Diversified Label Matrix Based Medical Image Report Generation
      ZHANG Junsan, CHENG Ming, SHEN Xiuxuan, LIU Yuxue, WANG Leiquan
      Computer Science. 2024, 51 (8): 200-208.  doi:10.11896/jsjkx.230600018
      Abstract ( 29 )   PDF(5253KB) ( 132 )   
      References | Related Articles | Metrics
      Medical images play a vital role in medical diagnosis.Accurately described text reports are essential for understanding images and subsequent disease diagnosis.In recent years,the generation of standardized reports based on modeling methods has become a research hotspot in the field of medical imaging report generation.However,due to the data deviation problem caused by the large gap between positive and negative samples,the content of the generated report generally tends to describe the normal situation.This limitation creates challenges in accurately capturing abnormal information.To address this issue,this paper proposes a novel approach based on diversified label matrix for medical report generation.This method utilizes a diverse label matrix to perform differential learning on different diseases and generate diverse medical reports.Additionally,a text-matrix feature loss function is designed to optimize the diverse label matrix,enhancing its effectiveness.Furthermore,the Transformer network is enhanced by incorporating a feature intersection module.This module strengthens the mapping between images and text,and improves accuracy in disease description.Experimental results on the two datasets of IU-X-Ray and MIMIC-CXR show that,the proposed method achieves the best results in multiple indicators,such as BLEU and METEOR,compared with the current mainstream methods.
      A Robust Method for Range Grating Lobe Suppression in Stepped Frequency SAR
      GAO Wenbin
      Computer Science. 2024, 51 (8): 209-216.  doi:10.11896/jsjkx.230600050
      Abstract ( 26 )   PDF(4264KB) ( 139 )   
      References | Related Articles | Metrics
      The magnitude error and phase error(MEPE) in the system transfer function of the stepped-frequency synthetic aperture radar(SAR) introduces periodic MEPE in the synthesized wideband signal,resulting in periodic grating lobes in the high-re-solution range profile(HRRP).After subsequent SAR imaging processing,these periodic grating lobes appear as false targets in the SAR image,seriously affecting SAR image target detection and recognition.Therefore,the GLS algorithm based on SAR images has been proposed,which is based on the assumption of point targets and effectively suppresses image grating lobes by utilizing strong point targets in SAR images.However,for non-point target scenes,existing GLS algorithms based on SAR images can cause image defocusing while suppressing image grating lobes.Therefore,this paper proposes a GLS algorithm based on target information differences,named the target information difference method,which is not based on the assumption of point-shaped targets.By estimating the information difference between the ideal HRRP and the actual HRRP of the target after wideband synthesis,it can robustly estimate the periodic MEPE in the synthesized wideband signal.By compensating for this periodic MEPE,the algorithm can suppress the range grating lobes to the SAR image background level.By comparing the performance of different GLS algorithms,it is found that the proposed GLS algorithmis less affected by the image signal-to-noise ratio and is suitable for both non-point target and point target scenes,thus having obvious advantages comparing to existing GLS algorithms.Experimental data processing results demonstrate the effectiveness and superiority of the proposed method compared to existing GLS algorithms.
      Video Anomaly Detection Method Based on Dual Discriminators and Pseudo Video Generation
      GUO Fangyuan, JI Genlin
      Computer Science. 2024, 51 (8): 217-223.  doi:10.11896/jsjkx.230600148
      Abstract ( 30 )   PDF(2454KB) ( 149 )   
      References | Related Articles | Metrics
      In unsupervised video anomaly detection tasks,deep autoencoders are typically trained on datasets containing only normal events and use reconstruction(prediction) error to identify anomalous frames.However,this assumption does not always true in practice because sometimes autoencoders can reconstruct(predict) anomalous events well,leading to false alarms.To address this issue,this paper proposes a video anomaly detection method based on dual discriminators and pseudo video generation,which enhances the generation model’s prediction capability of normal frames and suppresses its prediction capability of pseudo video frames through adversarial training between the discriminator and the generator.Moreover,the introduction of coordinated attention in the generation model further improves its detection performance.Additionally,by predicting intermediate frames instead of future frames in previous methods,the model can learn forward and backward motion information,which further enhances its detection performance.Experimental results on the publicly available datasets UCSD Ped2 and CUHK Avenue demonstrate that the proposed method achieves AUC values of 98.6% and 85.9%,respectively,outperforming other video anomaly detection methods significantly.
      Few-shot Semi-supervised Semantic Image Translation Algorithm Based on Prototype Correction
      HE Zhilin, GU Tianhao, XU Guanhua
      Computer Science. 2024, 51 (8): 224-231.  doi:10.11896/jsjkx.230500038
      Abstract ( 28 )   PDF(4476KB) ( 154 )   
      References | Related Articles | Metrics
      Image translation plays a vital role in computer vision and has extensive applications in visual fields,such as image sty-lization and image super-resolution generation.Datasets are frequently challenging to label,and semantic labeling has substantial costs.This paper proposes a few-shot semantic image translation framework based on prototype correction,mainly encompassing the StyleGAN module,semantic similarity regressor module,and pSp encoder module.First,to decrease the dependence of the model on the labeled image,our framework utilizes the StyleGAN pre-trained model as a generator,which expands the number of training samples in few-shot and the diversity of image generation.Second,considering the variations within the sample semantic class,our framework designs a semantic similarity regressor to correct the prototype,improving the accuracy of the pseudo-label and enhancing the model optimization effect.Third,the cyclic synthesis of semantic information is realized by combining label feature maps,synthetic feature maps and prototype vectors.Meanwhile,a self-supervised loss function is constructed to avoid the label information requirements of semantic similarity regressor training.Then the pSp encoder is trained with pseudo-tag images,and the task of semantic image synthesis is achieved.Experimental results show that the proposed framework is superior to classical frameworks in terms of excellent generalization performance and diversity of synthesized images.
      Infrared Human Action Recognition Method Based on Multimodal Attention Network
      WANG Chao, TANG Chao, WANG Wenjian, ZHANG Jing
      Computer Science. 2024, 51 (8): 232-241.  doi:10.11896/jsjkx.230600143
      Abstract ( 44 )   PDF(7515KB) ( 200 )   
      References | Related Articles | Metrics
      Human behavior recognition has become one of the research hotspots in the field of machine vision and pattern recognition,and has important research value.Many intelligent services require rapid and accurate recognition of human behavior.Human behavior recognition has important research significance and wide application value in fields such as intelligent monitoring and smart home,and has been widely studied by scholars at home and abroad.Human behavior recognition usually uses visible light video data,but visible light videos are easily affected by light and cannot adapt to nighttime recognition.Due to the characteristics of infrared information such as being less affected by light and protecting privacy,human behavior recognition methods based on infrared video have great significance.Deep learning network has some limitations on the learning and representation ability of infrared single mode data.To solve the above problems,an infrared human behavior recognition method based on multimodal attention network is proposed.Because the deep learning network model cannot directly train and classify the video information,first,the preprocessing module preprocesses the video information obtained into infrared views,and then extracts the edge information and optical flow information of the infrared view through Sobel operator and L1 norm based total variation optical flow method to obtain the edge view and optical flow view respectively.Secondly,input the infrared view,edge view,and optical flow view into the three stream network fused with the attention mechanism module for feature learning.Then,fuse the multimodal features extracted from each network in the three stream network.Finally,the fusion feature vector is input to random forest for training and classification.Experimental results on the public dataset NTU RGB+D and the self-built dataset indicate that the proposed me-thod has good recognition performance.In the future,we will consider expanding our method to more datasets to verify its effectiveness.
      Artificial Intelligence
      Evaluation of Hyperparameter Optimization Techniques for Traditional Machine Learning Models
      LI Haixia, SONG Danlei, KONG Jianing, SONG Yafei, CHANG Haiyan
      Computer Science. 2024, 51 (8): 242-255.  doi:10.11896/jsjkx.230600164
      Abstract ( 47 )   PDF(1676KB) ( 166 )   
      References | Related Articles | Metrics
      Reasonable hyperparameters ensure that machine learning models can adapt to different backgrounds and tasks.In order to avoid the inefficiency caused by manual adjustment of a large number of model hyperparameters and a vast search space,various hyperparameter optimization techniques have been developed and applied in machine learning model training.At first,Paper reviews eight common hyperparameter optimization techniques:grid search,random search,Bayesian optimization,Hyperband,Bayesian optimization and Hyperband(BOHB),genetic algorithms,particle swarm optimization algorithm,and covariance matrix adaptation evolutionary strategy(CMA-ES).The advantages and disadvantages of these methods are analyzed from five aspects:time performance,final results,parallel capability,scalability,robustness and flexibility.Subsequently,these eight methods are applied to four traditional machine learning models:LightGBM,XGBoost,Random Forest,and K-Nearest Neighbors(KNN).Regression,binary classification and multi-classification experiments are performed on four standard datasets:Boston house price dataset,kin8nm power arm dataset,credit card default customer dataset and handwritten digit dataset.Different methods are compared by evaluating their performance using output evaluation metrics.Finally,pros and cons of each method and are summarized,and the application scenarios of different methods are given.The results highlight the importance of selecting appropriate hyperparameter optimization methods to enhance the efficiency and effectiveness of machine learning model training.
      Contrastive Learning-based Prompt Generation Method for Large-scale Language Model ReverseDictionary Task
      TIAN Sicheng, HUANG Shaobin, WANG Rui, LI Rongsheng, DU Zhijuan
      Computer Science. 2024, 51 (8): 256-262.  doi:10.11896/jsjkx.230600204
      Abstract ( 46 )   PDF(2523KB) ( 160 )   
      References | Related Articles | Metrics
      Reverse dictionary task is an emerging task that aims to find the corresponding word based on a given definition.Large-scale language models offer new possibilities for this task,but the quality of the prompt sentences affects the performance of the large models.To this end,this paper proposes a contrastive learning-based prompt generation method.This method extracts definition semantics from multiple semantic levels.It also enhances the model’s generalization ability by incorporating negative examples through contrastive learning.With this method,we can narrow down the target word to a small range,and use a large model to select the most semantically consistent word from this range.Experimental results show that the proposed method can effectively improve the performance of large-scale language models on the reverse dictionary task.The prompt generation model has a 94.7% probability of generating a range that contains the target word.The large-scale language model has a 58.03% pro-bability of directly selecting the target word,and a 74.55% probability of including the target word when five candidate words are given.
      Semi-supervised Text Style Transfer Method Based on Multi-reward Reinforcement Learning
      LI Jingwen, YE Qi, RUAN Tong, LIN Yupian, XUE Wandong
      Computer Science. 2024, 51 (8): 263-271.  doi:10.11896/jsjkx.230600184
      Abstract ( 32 )   PDF(2260KB) ( 161 )   
      References | Related Articles | Metrics
      Text style transfer is an important task in natural language processing that aims to change the stylistic attributes of text while preserving necessary semantic information.However,in many tasks where large-scale parallel corpora are lacking,existing unsupervised methods suffer from issues such as insufficient text diversity and poor semantic consistency.To address these problems,this paper proposes a semi-supervised multi-stage training framework.It first constructs a pseudo-parallel corpus using a style labeling model and a masked language model to guide the model to learn diverse transfer styles in a supervised manner.Then,adversarial similarity reward,Mis reward,and style reward are designed to conduct reinforcement learning on unlabeled data to enhance the model’s semantic consistency,logical consistency,and accuracy of style transfer.In the sentiment polarity conversion task based on the YELP dataset,the proposed method’s BLEURT score increases by 3.1%,the Mis score increases by 2.5%,and the BLEU score increases by 9.5%.In the formal style conversion experiment based on the GYAFC dataset,its BLEURT score increases by 6.2%,and the BLEU score increases by 3%.
      Word-Character Model with Low Lexical Information Loss for Chinese NER
      GUO Zhiqiang, GUAN Donghai, YUAN Weiwei
      Computer Science. 2024, 51 (8): 272-280.  doi:10.11896/jsjkx.230500047
      Abstract ( 35 )   PDF(2676KB) ( 165 )   
      References | Related Articles | Metrics
      Chinese named entity recognition(CNER) task is a natural language processing technique that aims to recognize entities with specific categories in text,such as names of people,places,organizations.It is a fundamental underlying task of natural language applications such as question and answer systems,machine translation,and information extraction.Since Chinese does not have a natural word separation structure like English,the effectiveness of word-based NER models for Chinese named entity recognition is significantly reduced by word separation errors,and character-based NER models ignore the role of lexical information.In recent years,many studies have attempted to incorporate lexical information into character-based models,and WC-LSTM has achieved significant improvements in model performance by injecting lexical information into the start and end characters of a word.However,this model still does not fully utilize lexical information,so based on it,LLL-WCM(word-character model with low lexical information loss) is proposed to incorporate lexical information for all intermediate characters of the lexicon to avoid lexical information loss.Meanwhile,two encoding strategies average and self-attention mechanism are introduced to extract all lexical information.Experiments are conducted on four Chinese datasets,and the results show that the F1 values of this method are improved by 1.89%,0.29%,1.10% and 1.54%,respectively,compared with WC-LSTM.
      Semi-supervised Emotional Music Generation Method Based on Improved Gaussian Mixture Variational Autoencoders
      XU Bei, LIU Tong
      Computer Science. 2024, 51 (8): 281-296.  doi:10.11896/jsjkx.230500124
      Abstract ( 32 )   PDF(6507KB) ( 148 )   
      References | Related Articles | Metrics
      Music can transmit audio content and emotions through serialized audio features.Emotion is an important component in the semantic expression of music.Therefore,music generation technology should not only consider the structural information of music but also incorporate emotions.Most existing emotional music generation technologies use the complete supervised methods based on emotion labeling.However,the music field lacks a large number of standard emotional labeling datasets,and emotional labels are insufficient to express the emotional features of music.To solve these problems,this paper proposes a semi-supervised emotional music generation method(Semg-GMVAE) based on improved Gaussian mixture variational autoencoders(GMVAE),which connects the rhythm features and mode features of music with emotions,incorporates a feature disentanglement mechanism into GMVAE to learn the potential variable representations of these two features,and performs semi-supervised clustering infe-rence on them.Finally,by manipulating the feature representation of music,our model can achieve music generation and emotion switching on happy,tense,sad,and calm emotions.Meanwhile,this paper conducts a series of experiments on the problem that GMVAE is difficult to distinguish different emotional categories of data.The key reason for the problem is that the variance regularization term and mutual information suppression term in the evidence lower bound of GMVAE make the Gaussian components of each category less dispersed,thus affecting the performance of learned representation and the quality of generation.Therefore,Semg-GMVAE penalizes and augments these two factors respectively,and uses Transformer-XL as the encoder and decoder to enhance the modeling capabilities on long sequence music.Experimental results based on real data show that,compared to existing methods,Semg-GMVAE achieves better separation of music with different emotions in potential space,enhances the correlation between music and emotions,effectively disentangles different music features,and finally achieves better emotional music generation and emotion switching by changing the feature representation.
      Chinese Geological Entity Relation Extraction Based on RoBERTa and Weighted Graph Convolutional Networks
      ZHANG Lu, DUAN Youxiang, LIU Juan, LU Yuxi
      Computer Science. 2024, 51 (8): 297-303.  doi:10.11896/jsjkx.230600231
      Abstract ( 38 )   PDF(2448KB) ( 174 )   
      References | Related Articles | Metrics
      Knowledge is the cornerstone of big data and artificial intelligence.Knowledge graphs offer interpretability and sca-lability advantages,making them crucial in intelligent systems.Intelligent decision has urgent application demand in various fields,providing decision support and application scenarios for knowledge graphs based on data analysis and reasoning.However,constructing and applying knowledge graphs face challenges due to complex domain scenarios,multi-source data,and extensive knowledge dimensions.To address the problem of incomplete domain knowledge patterns during geological domain knowledge graph construction and the limitations of existing entity relationship extraction methods in dealing with non-Euclidean data,a graph structure-based entity relationship extraction model RoGCN-ATT is proposed.This model utilizes RoBERTa-wwm-ext-large,a Chinese pre-trained model,as the sequence encoder combined with BiLSTM to capture richer semantic information.It also employs weighted graph convolutional networks along with attention mechanisms to capture structural dependency information and enhance the extraction performance of relation triplets.Experimental results show that the F1 value reaches 78.56% on the geological dataset.Compared with other models,RoGCN-ATT effectively improves the entity-relationship extraction performance and provides strong support for the construction and application of geological knowledge maps.
      Multi-channel Graph Convolutional Networks Enhanced by Label Propagation Algorithm
      YUAN Lining, FENG Wengang, LIU Zhao
      Computer Science. 2024, 51 (8): 304-312.  doi:10.11896/jsjkx.240100139
      Abstract ( 34 )   PDF(3151KB) ( 136 )   
      References | Related Articles | Metrics
      Most graph convolutional networks(GCN) improve the experimental performance of node classification tasks by designing efficient methods for information propagation and preservation,while ignoring the propagation of node label information in the topological and attribute spaces.Aiming at the above problems,the paper proposes a multi-channel graph convolution model MGCN-LPA enhanced by the label propagation algorithm(LPA).The model enhances the propagation of node features and label information by increasing the weights of relationship between nodes of the same class in the attribute space and topology space.Firstly,it calculates the similarity values of different node attributes and generates an attribute relation graph using the k-nearest neighbor algorithm.Then,it combines the GCN and LPA in the graph convolution layer GCN-LPA to extract potential features from the attribute graph and attribute relation graph,generating topological node representations and attribute node representations.Finally,the method combines the topological and attribute representations and utilizes the final representation for node classification tasks.On three benchmark graph datasets,the experimental performance of MGCN-LPA can match the current state-of-the-art baseline models.The classification results on the Cora and Citeseer datasets show improvements of 9.3% and 12% respectively compared to the best-performing baseline.The experimental results demonstrate that MGCN-LPA can increase the weights of paths between nodes of the same class and enhance the propagation of information among nodes of the same class,thereby enhancing the performance of node classification tasks.In addition,the ablation experiments demonstrate that the fusion of both topological space and attribute space information in MGCN-LPA enhances the model’s representational capacity and ge-neralization compared to variants using only one type of information.This fusion allows for the full extraction and preservation of latent features present in the original graph.
      Study on Recommendation Algorithms Based on Knowledge Graph and Neighbor PerceptionAttention Mechanism
      CHEN Shanshan, YAO Subin
      Computer Science. 2024, 51 (8): 313-323.  doi:10.11896/jsjkx.230500143
      Abstract ( 34 )   PDF(2709KB) ( 141 )   
      References | Related Articles | Metrics
      In order to solve the cold start problem caused by traditional recommendation algorithms when they face the recommendation task with sparse data,this paper introduces the knowledge graph into the recommendation algorithm,combing a new neighbor perception attention mechanism to replace the traditional graph attention mechanism to mine the higher-order connected information between entities,and proposes a recommendation model KGNPAN based on the knowledge graph and neighbor perce-ption attention mechanism.Thanks to the knowledge graph,recommendations can be accurate,diverse and interpretable.This model can effectively alleviate issues of data sparsity and cold start.Firstly,this model utilizes the graph embedding method RotatE based on self adversarial negative sampling to expand the semantic information of the original item and user representations,mapping entity and relationship vectors into low dimensional embedding vectors.Secondly,based on the different types of collaborative neighbors,neighbor perception attention mechanisms are applied to aggregate neighbor node information,enrich the semantics of target nodes,and recursively mine high-order connected information in convolutional form.Finally,the inner product operation is applied to calculate the interaction probability between the user and the project vector,and the recommendation result is obtained.Experiments are conducted on two common benchmark datasets,Amazon-book and Last-FM,and compared with six benchmark models,namely CKE,BPRMF,RippleNet,KGAT,KGCN,and CAKN,KGNPAN.The results show that KGNPAN improves the recall rate by 1.30% and 1.37%,and normalized discounted cumulative gain(NDCG) increases by 1.26% and 1.14%,respectively,compared with CAKN model,which has the best performance in the benchmark modes,verifying the effectiveness and interpretability of the model.
      Fixed-time Recurrent Neural Networks for Time-variant Matrix Computing and Its Application in Repeatable Motion Planning
      LI Xing, ZHONG Guomin
      Computer Science. 2024, 51 (8): 324-332.  doi:10.11896/jsjkx.230500052
      Abstract ( 31 )   PDF(4022KB) ( 151 )   
      References | Related Articles | Metrics
      Fixed-time recurrent neural network(RNN) models with logarithmic settling time are proposed for solving time-variant neural computing problems.Two novel RNN models are designed and analyzed in detail,deriving the explicit expressions of settling time functions and providing the upper bounds of the settling times under any initial condition.Compared with the existing RNN models with fixed-time convergence,the two novel models with logarithmic settling time have a smaller upper bound on the settling time and faster convergence speeds.Taking into account initial conditions located within a region with a definite finite radius,the settling time functions of the RNN models with logarithmic settling time are given,and the upper bounds on the settling time functions in the semi-global sense are derived.Modified RNN models adopt the inverse of the bound to ensure that the semi-global predefined time converges to the exact solution,and its prescribed time is an adjustable parameter.Simulation results of the proposed RNN model for solving time-variant Lyapunov and Sylvester equations are given.The proposed RNNs are applied to the repetitive motion planning of a redundant manipulator with initial errors,and numerical results are presented to verify the effectiveness of the proposed RNN models.
      Dynamic Multi-objective Optimization Algorithm Based on RNN Information Accumulation
      CHENG Xuefeng, DONG Minggang
      Computer Science. 2024, 51 (8): 333-344.  doi:10.11896/jsjkx.230500046
      Abstract ( 46 )   PDF(6826KB) ( 174 )   
      References | Related Articles | Metrics
      Dynamic multi-objective optimization problems exist widely in real life.After the environment changes,it is necessary for the evolutionary algorithm to have the abilities of fast convergence,fast tracking Pareto optimal frontier and maintaining diversity.For severe and frequent environmental changes,the traditional forecasting method can not effectively obtain Pareto optimal frontier solution.For this problem,a dynamic multi-objective optimization algorithm based on recurrent neural networks information accumulation(IA-RNN) is proposed.Firstly,a nonlinear prediction method based on RNN information accumulation is proposed,which uses RNN recursion for information accumulation,improves the utilization rate of historical information and enhances the ability of prediction.Secondly,a linear prediction method based on individual is designed,which uses parameter matrix to predict the linear changes of individual.Linear prediction and RNN nonlinear prediction co-evolve,which can quickly track the Pareto optimal frontier.Finally,a parameter correction strategy based on the least square method is designed to guide the parameter correction by the approximate Pareto optimal frontier solution in the current environment,which reduces the influence of error accumulation.IA-RNN is compared with five representative dynamic multi-objective optimization algorithms on 14 DF benchmark problems.Experiments show that the IA-RNN algorithm has better convergence and diversity.
      Computer Network
      Safe Placement of Multi-antenna Wireless Chargers
      REN Meixuan, DENG Peng, ZHAO Yue, WANG Xiaoyu, WANG Chao, DAI Haipeng, WU Libing
      Computer Science. 2024, 51 (8): 345-353.  doi:10.11896/jsjkx.240400156
      Abstract ( 57 )   PDF(3250KB) ( 141 )   
      References | Related Articles | Metrics
      This paper studies the problem of safe placement of multi-antenna wireless chargers(SPINNER),that is,given a set of wireless rechargeable devices and a set of wireless chargers,each equipped with multiple directional antennas,scheduling both the power level and the strategy(the position and the orientations of antennas) of each charger,so that the overall charging utility can be maximized and any position in the field satisfies electromagnetic radiation(EMR) safety constraints.In this paper,we consider two different scenarios,that is,safe placement of multi-antenna wireless chargers with a given position set(SPINNER-G) and safe placement of multi-antenna wireless chargers with arbitrary points(SPINNER-A).First,we adopt a piecewise constant function to approximate the nonlinear charging power function and partition the 2D field into a limited number of subareas.Thus,the number of EMR constraints is limited.Then,for SPINNER-G,we propose a maximal coverage set extraction method to further limit the number of orientations of chargers.For SPINNER-A,we construct maximal intersection condition set to limit the searching space for the positions and orientations of chargers.Then,for SPINNER-G and SPINNER-A,we propose two linear programming based greedy schemes,both of which achieve an approximation ratio of 1/2－ε.Simulations show that the charging utility of our algorithm improves by at least 54.2% comparte to the three comparison algorithms.
      Variable-length Shaping Queue Adjustment Algorithm in Time-sensitive Networks
      CAI Changjuan, ZHUANG Lei, YANG Sijin, WANG Jiaxing, YANG Xinyu
      Computer Science. 2024, 51 (8): 354-363.  doi:10.11896/jsjkx.230500214
      Abstract ( 35 )   PDF(2361KB) ( 137 )   
      References | Related Articles | Metrics
      A variable length shaping queue adjustment algorithm based on an improved krill herd algorithm and traffic prediction is proposed to address the issues of low buffer resource utilization and high average delay of schedulable streams using fixed length shaping queues for traffic shaping in asynchronous traffic shaper(ATS).Considering the queue allocation rules of flows,bounded delay requirements,and limited buffer resources,transmission constraints for schedulable flows are defined in time-sensitive networks.The improved krill herd algorithm is used to find the maximum adjustable upper limit of the shaping queue,using a combination of chaos mapping,opposition-based learning,elite policy,and adaptive location update strategy to enhance the algorithm’s solving ability.The traffic is predicted based on convolutional neural network and long short-term memory model(CNN-LSTM),and the queue length is calculated according to the predicted value to adjust the step.Simulation results show that compared with the method of using fixed-length shaping queues,the proposed algorithm can effectively increase the number of sche-dulable flows,reduce the average delay of scheduled traffic(ST),and improve the utilization rate of network buffer resources.
      Construction of Internet of Vehicles Covert Channel Based on Websocket Protocol
      ZHAO Hui, PENG Jianyou, QIN Yulin, HAN Lili
      Computer Science. 2024, 51 (8): 364-370.  doi:10.11896/jsjkx.230500037
      Abstract ( 26 )   PDF(2273KB) ( 126 )   
      References | Related Articles | Metrics
      Aiming at the problem that the construction method of covert channel under traditional Internet of Vehicles protocol is single and cannot be applied to complex network environment,a construction method of covert channel in Internet of Vehicles based on WebSocket protocol is proposed by analyzing the data frame format and communication mechanism of Websocket－a lightweight application layer protocol of Internet of Things.This method uses information separation and aggregation algorithm to transmit the covert information in multiple dimensions to enhance the transmission rate and anti-exposure of the covert channel.Besides,considering the dynamic topological characteristics of the Internet of Vehicles network,the information separation and aggregation mode and coding mapping table are transformed adaptively based on frequency hopping technology.Finally,in order to improve the concealment of the channel,the least square algorithm is used to simulate the transmission characteristics of normal network traffic.The results of simulation experiments show that the constructed covert channel is less affected by network fluctuations and has better robustness when facing poor network environment.And compared with the covert channel with single-dimension transmission,it has certain improvement in terms of concealment and transmission rate.
      Information Security
      Survey of Detection Techniques for Domain Generation Algorithm
      WANG Xuxian, HUANG Jinhua, ZHAI You, LI Chu’nan, WANG Yu, ZHANG Yupeng, ZHANG Yipeng, YANG Liqun, LI Zhoujun
      Computer Science. 2024, 51 (8): 371-378.  doi:10.11896/jsjkx.230700189
      Abstract ( 39 )   PDF(1670KB) ( 154 )   
      References | Related Articles | Metrics
      The C&C server is an intermediate server used by cyber attackers to control bots,and plays a key role in botnet.In order to enhance the concealment of the C&C server,cyber attackers use domain generation algorithms to hide the IP address of C&C server.In recent years,domain generation algorithm detection technology,as an important means of detecting botnets,has become a research hotspot.This paper first introduces the current development trend of cyber security and the topological structure of botnets.Secondly,the domain generation algorithm and the related dataset are introduced.Then,the classification of domain generation algorithm detection techniques is introduced,and these detection techniques are summarized.Finally,the pro-blems existing in the domain generation algorithm detection technology at the present stage are discussed,and the future research directions are prospected.
      Abnormal Traffic Detection Method for Multi-stage Attacks of Internet of Things Botnets
      CHEN Liang, LI Zhihua
      Computer Science. 2024, 51 (8): 379-386.  doi:10.11896/jsjkx.230700197
      Abstract ( 33 )   PDF(2535KB) ( 126 )   
      References | Related Articles | Metrics
      To address the problem of how to efficiently detect multi-stage attack behavior of IoT botnet from massive network traffic data,an IoT botnet attack detection method based on multi-scale hybrid residual network(IBAD-MHRN)is proposed.Firstly,in order to reduce the calculation parameters of the detection model,a feature selection algorithm based on variance threshold(FS-VT)method is proposed in data preprocessing.Secondly,a data image processing strategy that converts data samples into image samples is adopted to fully tap the potential of the deep learning model.Then,in order to solve the deficiency of the traditional botnet detection model with limited representation ability,a multi-stage attack detection model of IoT botnet based on multi-scale hybrid residual network is proposed.The model integrates the feature information extracted at different scales and depths in a hybrid way,and then eliminates the effect of network degradation caused by network deepening through residual connection.Finally,an IBAD-MHRN method for IoT botnet attack detection is proposed by integrating the above models and algorithms.Experimental results show that the detection accuracy and F1 value of the proposed IBAD-MHRN method reaches 99.8%,and the accuracy and F1 value is improved by 0.14% and 0.36% respectively compared with the better convolutional neural network method,which can effectively and efficiently detect multi-stage attacks of Internet of Things botnets.
      Privacy-preserving Principal Component Analysis Based on Homomorphic Encryption
      ZHANG Jindou, CHEN Jingwei, WU Wenyuan, FENG Yong
      Computer Science. 2024, 51 (8): 387-395.  doi:10.11896/jsjkx.230800177
      Abstract ( 38 )   PDF(1635KB) ( 136 )   
      References | Related Articles | Metrics
      In real life,data is not interconnected between different industries,or even between different departments within the same industry.With the improvement of computer computing power,it is not computing power but data volume that restricts the effectiveness of model training.Therefore,in order to obtain a better algorithm model,relying solely on one party’s data is not enough.It needs the participation of two or more parties,which requires privacy protection for all parties.In addition,as data collection becomes more detailed,the data dimension also increases.For high dimension data,dimension reduction is an indispensable step.And in terms of dimension reduction,principal component analysis(PCA) is a commonly used method.Homomorphic encryption is a solution when two parties want to collaborate on privacy protection data dimension reduction.Homomorphic encryption can compute encrypted data while protecting data privacy,and can be used to compute the PCA on encrypted data.In this paper,a two party encrypted data PCA scheme is designed using the CKKS homomorphic encryption scheme and the power method for dominant eigenvectors,achieving the goal of dimension reduction while protecting the privacy of both parties’ data.By improving the traditional power method iteration steps,the expensive homomorphic ciphertext division is avoided,allowing for more iterations with small encryption parameters,thereby reducing the computing time and improving the accuracy of the computed results.Through testing on public datasets and comparing it with some existing schemes,the scheme reduces the computational time by about 80%,and reduces the mean squared error to within 1% compared to the plaintext computation results.
      Encrypted Traffic Classification of CNN and BiGRU Based on Self-attention
      CHEN Siyu, MA Hailong, ZHANG Jianhui
      Computer Science. 2024, 51 (8): 396-402.  doi:10.11896/jsjkx.230500032
      Abstract ( 40 )   PDF(3040KB) ( 140 )   
      References | Related Articles | Metrics
      To address the problems of low accuracy of traditional encrypted traffic classification methods,the use of traffic load will violate user privacy and weak generalization ability,an encrypted traffic classification method of CNN and BiGRU based on self-attention(CNN-AttBiGRU) is proposed,which can be applied to both regular encrypted and VPN and Tor encrypted traffic.The method converts traffic into intuitive pictures based on packet size,packet arrival time and packet arrival direction.To improve the accuracy of the model,CNN is used to extract the spatial features of traffic pictures,while BiGRU and self-attention models are designed to extract temporal features,making full use of the temporal and spatial features of traffic pictures.The traffic can be classified at different levels by traffic category,encryption technique and application type.The proposed method achieves an average accuracy of 95.2% for classification of encrypted traffic categories,which is 11.65% better than before;95.5% for classification of encryption technologies,which is 7.1% better than before;and 99.8% for classification of applications used by traffic,which is 11.03% better than before.Experimental results show that the CNN-AttBiGRU method has strong ge-neralization ability and only utilizes some statistical features of encrypted traffic,which effectively protects user privacy while achieving high accuracy rates.
      Study on Time Rotation Notary Group Model Based on Threshold Signature
      ZANG Wenyang, LYU Jinlai
      Computer Science. 2024, 51 (8): 403-411.  doi:10.11896/jsjkx.230500060
      Abstract ( 23 )   PDF(2073KB) ( 122 )   
      References | Related Articles | Metrics
      With the emergence of various blockchain systems,the demand for cross-chain interaction is increasing,and the security of cross-chain bridge verification becomes more and more important.The notary schemes have simple principle and high efficiency,and are often used in cross-chain transaction verification,cross-chain transaction confirmation and other processes.How-ever,the notary schemes have some problems,such as low degree of decentralization,verifiable transactions with low signature ratio,and information disclosure of verification nodes.In order to improve the security of cross-chain bridge verification,a time rotation notary group model based on threshold signature is proposed.The notary group in this model is composed of high credibility nodes that have both source and target blockchain accounts.The verification nodes in the notary group have tenure requirements and need to pledge the security deposit.The notary group uses threshold signature technology to sign cross-chain transactions.The cross-chain transaction can only be implemented after more than half of the verification nodes in the notary group sign.The candidate notary group provides some new verification nodes for the time rotation notary group.The analysis results of the time rotation notary group model proves that the proposed model has high degree of decentralization,low malicious attack rate of the verification nodes,high security of the verification signature links,high privacy of the verification nodes,and high efficiency of cross-chain message verification.
      New Type of UDP Reflection Amplification Protocol Recognition Method Based on Active-Passive Combination
      CHEN Hongwei, YIN Xiaokang, GAI Xianzhe, JIA Fan, LIU Shengli, CAI Ruijie
      Computer Science. 2024, 51 (8): 412-419.  doi:10.11896/jsjkx.230500227
      Abstract ( 23 )   PDF(2561KB) ( 119 )   
      References | Related Articles | Metrics
      Reflection amplification attack has gradually become a mainstream DDoS attack method because of its high-quality traffic doubling ability and anti-traceability capability.In recent years,new UDP reflection amplification attack methods represented by Internet of Things protocols such as OpenVPN have emerged constantly,showing a trend of multi-protocol combination reflection amplification.However,current UDP reflection amplification detection methods have some problems,such as inaccurate detection results and insufficient detection efficiency.In order to improve the UDP reflection amplification detection capability,a new type of UDP reflection amplification protocol recognition method based on active-passive combination is proposed.Firstly,the known Internet of Things reflection amplification protocol traffic is obtained through active detection method and is used as the experimental dataset.Secondly,in the process of automatic traffic analysis,dual threshold determination and multivariate feature matching are used to capture the unknown reflection amplification protocol and trigger mode.Finally,verify the authenticity through replay.Experimental results show that this method can effectively detect the reflection amplification traffic targeting UDP protocol,with an precision of 99.88%.The potential reflection amplification ability of the QUIC protocol has been disco-vered,effectively improving the protection ability against reflection amplification attacks.
      Integrity Interference Attack and Defense Methods for Network Traffic Measurement
      ZHENG Haibin, LIU Xinran, CHEN Jinyin, WANG Pengcheng, WANG Xuanye
      Computer Science. 2024, 51 (8): 420-428.  doi:10.11896/jsjkx.230500101
      Abstract ( 31 )   PDF(3799KB) ( 152 )   
      References | Related Articles | Metrics
      In recent years,network measurement has achieved good performance in evaluating network status and improving network self-adaptability,and is widely used in network management.However,there is a problem of network traffic data pollution caused by abnormal behavior in the current large-scale network.For example,malicious nodes in autonomous systems intentionally manipulate network metrics by forging malicious traffic data,affecting network measurements and misleading downstream task decisions.Based on this,this paper first proposes an integrity jamming attack method.By modifying the minimum cost of the traffic matrix,a multi-strategy jamming generator is used to generate an attack strategy that maliciously disturbs traffic,so as to achieve the purpose of jamming traffic measurement.Then,by providing a hybrid adversarial training strategy,a defense method against such attacks in the network is designed to achieve security hardening of the traffic measurement model.In the experiment,the attack target is limited accordingly,and the effectiveness of the integrity interference attack in the restricted scenario is verified.And through the comparison of the mixed training method,the robustness of the reinforcement method of the conventional model is verified.
      Efficient Quantum-secure Byzantine Fault Tolerance Consensus Mechanism Based on HotStuff
      CHENG Andong, XIE Sijiang, LIU Ang, FENG Yimeng
      Computer Science. 2024, 51 (8): 429-439.  doi:10.11896/jsjkx.230600200
      Abstract ( 28 )   PDF(2369KB) ( 136 )   
      References | Related Articles | Metrics
      The public-key digital signature used by Byzantine fault tolerance consensus mechanism in the classic blockchain exposes vulnerability to quantum computers that have the exponential acceleration of computing power,and therefore have security risks.To address the problem that the Byzantine fault tolerance consensus mechanism does not have quantum security,this paper proposes an efficient quantum secure Byzantine fault tolerance consensus mechanism based on HotStuff,known as EQSH(efficient quantum secure HotStuff).Firstly,an efficient multi-party ring quantum digital signatures(EMRQDSs) scheme is proposed to solve the problem of high complexity of unconditionally secure signatures(USS) communication.The scheme is based on a ring quantum network that guarantees post-quantum security,non-enforceability,non-repudiation,and transferability while the communication complexity is O(n).Secondly,the gated signature used in HotStuff is improved,instead,we propose an alternative scheme for post-quantum security,i.e.,a signature collection scheme based on a key distribution center,which could achieve the same effect as gated signature while guaranteeing post-quantum security with a communication complexity of O(n).Subsequently,the above two schemes are adopted in HotStuff to provide post-quantum security;a heartbeat is designed to ensure the activity;the consensus message format is simplified and the consensus efficiency is improved by using a pipelined consensus process.Costly techniques such as quantum entanglement are not used in EQSH,our scheme can be implemented under existing technology conditions and thusis of high practical value.Compared to HotStuff,EQSH has post-quantum security.Compared with other non-entangled quantum-secured Byzantine fault tolerance consensus mechanisms,EQSH reduces the communication complexity to O(n) for the first time and has better performance which requires less quantum circuit resourcesfor the client,which is beneficial to the construction of quantum networks.
      Blockchain Certificateless Encryption Mechanism Based on National Secret Algorithm
      XIANG Yanjie, HUANG Xiaofang, XIANG Kefeng, ZHENG Ji’nan
      Computer Science. 2024, 51 (8): 440-446.  doi:10.11896/jsjkx.230400203
      Abstract ( 35 )   PDF(1743KB) ( 149 )   
      References | Related Articles | Metrics
      The blockchain has attracted widespread attention because of its distributed,non-tamperable and inherent immutability features.However,the international cryptographic algorithm used in the blockchain has certain backdoor security risks.Based on the national secret algorithm SM2,this paper proposes a blockchain-based certificateless public key encryption(CL-PKE)scheme,which combines with the certificateless cryptographic mechanism.The scheme does not use bilinear pairing,reduces computa-tional cost,and eliminates certificate management and key escrow issues.At the same time,due to tamper proof and traceable of blockchain,the scheme realizes the user’s updating and revocation of the public key,so as to fight against Type-1 and Type-2 adversaries in the certificateless mechanism.Based on the difficulty of the computational Diffie-Hellman problem(CDHP),it is proved that the scheme is indistinguishable under the adaptive chosen ciphertext attack in the random prediction model.Finally,after analysis and testing,compared with the existing CL-PKE schemes,the computational efficiency of this scheme is increased by at least 11%.
  • Computer Science
    NCTCS2017 (0)
    Network & Communication (202)
    Information Security (689)
    Software & Database Technology (3)
    Artificial Intelligence (779)
    Surveys (52)
    Graphics, Image & Pattern Recognition (58)
    WISA2017 (1)
    WISA2018 (1)
    NASAC 2017 (13)
    CGCKD 2018 (12)
    Netword & Communication (4)
    Graphics, Image & Pattern Recognition (7)
    CCF Big Data 2017 (7)
    NCIS 2017 (5)
    Graphics, Image & Pattem Recognition (10)
    Interdiscipline & Frontier (95)
    Review (33)
    Intelligent Computing (158)
    Pattern Recognition & Image Processing (82)
    Big Date & Date Mining (21)
    Interdiscipline & Application (228)
    ChinaMM 2017 (12)
    CCDM2018 (14)
    Graphics ,Image & Pattern Recognition (45)
    Pattem Recognition & Image Processing (25)
    Big Data & Data Mining (42)
    Surverys (5)
    NDBC 2018 (8)
    Graphics,Image & Pattern Recognition (11)
    Database & Big Data & Data Science (396)
    Computer Graphics & Multimedia (452)
    Intelligent Software Engineering (17)
    Databωe & Big Data & Data Science (14)
    Computer Software (83)
    Mobile Crowd Sensing and Computing (9)
    Software Engineering (11)
    Intelligent Edge Computing (14)
    New Distributed Computing Technologies and Systems (9)
    Image Processing & Multimedia Technology (166)
    Artificial Intelligence Security (12)
    Human-Machine Interaction (5)
    Intelligent Data Governance Technologies and Systems (14)
    Blockchain Technology (18)
    Invited Article (2)
    Multilingual Computing Advanced Technology (11)
    Novel Distributed Computing Technology and System (11)
    Survys (0)
    Network & Communication (0)
    Graphics, Image & Patten Recognition (0)
    Interdiscipline & Frontier (4)
    HPC China 2018 (6)
    Data Science (18)
    Advances on Multimedia Technology (11)
    Special Issue of Social Computing Based Interdisciplinary Integration (13)
    Smart IoT Technologies and Applications Empowered by 6G (10)
    Smart Healthcare (10)
    Software & Database Technology (48)
    ChinaMM2018 (14)
    ChinaMM2019 (0)
    Intelligent Mobile Authentication (10)
    Big Data & Data Science (182)
    Network & Cornmunication (4)
    Theoretical Computer Science (10)
    Computer Vision: Theory and Application (16)
    Software Engineering & Database Technology (8)
    Computer Architecture (26)
    Discipline Construction (2)
    High Performance Computing (26)
    Computer Science Theory (19)
    Computer Network (257)
    Computer Graphics& Multimedia (10)
    High Perfonnance Computing (10)
    Federated Leaming (11)
    Edge Intelligent Collaboration Technology and Frontier Applications (10)
    Database & Big Data & Data Science (44)
    Computer Graphics & Multimedia (41)
    Explainable AI (11)
    Database & Big Data & Data Science (15)
    Software & Interdiscipline (16)
    Data Security (11)
    Granular Computing & Knowledge Discovery (10)
    Big Data & Data Science (37)
    Network & Communication (13)
    Special Issue on the 50th Anniversary of Computer Science (1)
    Special Issue on the 54th Anniversary of Computer Science (1)
    Special Issue on the 55th Anniversary of Computer Science (1)
    Special Issue on the 56th Anniversary of Computer Science (1)
    Special Issue on the 58th Anniversary of Computer Science (1)
    Discipline Frontier (7)
    Artificial Intelligenc (36)
    Computer Networ (14)
    Information Security Protection in New Computing Mode (6)
    Compact Data Structure (7)
    Special Issue of Knowledge Engineering Enabled By Knowledge Graph: Theory, Technology and System (11)
    Computer Graphics & Multimedia (10)
    Granular Computing & Knowledge Discovery (6)
    Interdiscipline & Frontier (0)
    Computer Software & Architecture (8)
    Interdiscipline & Application (14)
    Image Processing & Multimedia Technology (36)
    Special Issue on the 51th Anniversary of Computer Science (1)
    Special Issue on the 52th Anniversary of Computer Science (1)
    Special Issue on the 53th Anniversary of Computer Science (1)
    Special Issue on the 57th Anniversary of Computer Science (1)
    Special Issue on the 59th Anniversary of Computer Science (1)
    Image Processing & Multimedia Technolog (50)
    Computer Software & Architecture (10)
    Contents (12)

  • 2024 Vol. 51 No. 7

  • 2024 Vol. 51 No. 6A

  • 2024 Vol. 51 No. 6

  • 2024 Vol. 51 No. 5
  • 2024,51 (7) 
  • 2024,51 (6A) 
  • 2024,51 (6) 
  • 2024,51 (5) 
  • 2024,51 (4) 
  • 2024,51 (3) 
  • 2024,51 (2) 
  • 2024,51 (1) 
  • 2023,50 (12) 
  • 2023,50 (11A) 
  • 2023,50 (11) 
  • 2023,50 (10) 
  • 2023,50 (9) 
  • 2023,50 (8) 
  • 2023,50 (7) 
  • 2023,50 (6A) 

    • More>>
  • Review of Time Series Prediction Methods (3995)
    YANG Hai-min, PAN Zhi-song, BAI Wei
    Computer Science. 2019, No.1:21-28
    Abstract (3995) PDF (1294KB) (12222)
    Web Application Page Element Recognition and Visual Script Generation Based on Machine Vision (622)
    LI Zi-dong, YAO Yi-fei, WANG Wei-wei, ZHAO Rui-lian
    Computer Science. 2022, No.11:65-75
    Abstract (622) PDF (2624KB) (12095)
    Polynomial Time Algorithm for Hamilton Circuit Problem (6007)
    JIANG Xin-wen
    Computer Science. 2020, No.7:8-20
    Abstract (6007) PDF (1760KB) (11983)
    Optimization Method of Streaming Storage Based on GCC Compiler (1198)
    GAO Xiu-wu, HUANG Liang-ming, JIANG Jun
    Computer Science. 2022, No.11:76-82
    Abstract (1198) PDF (2713KB) (11839)
    Patch Validation Approach Combining Doc2Vec and BERT Embedding Technologies (436)
    HUANG Ying, JIANG Shu-juan, JIANG Ting-ting
    Computer Science. 2022, No.11:83-89
    Abstract (436) PDF (2492KB) (11247)
    Semantic Restoration and Automatic Transplant for ROP Exploit Script (442)
    SHI Rui-heng, ZHU Yun-cong, ZHAO Yi-ru, ZHAO Lei
    Computer Science. 2022, No.11:49-54
    Abstract (442) PDF (2661KB) (10936)
    Study on Effectiveness of Quality Objectives and Non-quality Objectives for Automated Software Refactoring (418)
    GUO Ya-lin, LI Xiao-chen, REN Zhi-lei, JIANG He
    Computer Science. 2022, No.11:55-64
    Abstract (418) PDF (3409KB) (10857)
    Decision Tree Algorithm-based API Misuse Detection (841)
    LI Kang-le, REN Zhi-lei, ZHOU Zhi-de, JIANG He
    Computer Science. 2022, No.11:30-38
    Abstract (841) PDF (3144KB) (10829)
    AutoUnit:Automatic Test Generation Based on Active Learning and Prediction Guidance (848)
    ZHANG Da-lin, ZHANG Zhe-wei, WANG Nan, LIU Ji-qiang
    Computer Science. 2022, No.11:39-48
    Abstract (848) PDF (2609KB) (10732)
    Research and Progress on Bug Report-oriented Bug Localization Techniques (770)
    NI Zhen, LI Bin, SUN Xiao-bing, LI Bi-xin, ZHU Cheng
    Computer Science. 2022, No.11:8-23
    Abstract (770) PDF (2280KB) (10593)
    Study on Integration Test Order Generation Algorithm for SOA (697)
    ZHANG Bing-qing, FEI Qi, WANG Yi-chen, Yang Zhao
    Computer Science. 2022, No.11:24-29
    Abstract (697) PDF (1866KB) (10195)
    Studies on Community Question Answering-A Survey (328)
    ZHANG Zhong-feng,LI Qiu-dan
    Computer Science. 2010, No.11:19-23
    Abstract (328) PDF (551KB) (10116)
    Research Progress and Challenge of Programming by Examples (610)
    YAN Qian-yu, LI Yi, PENG Xin
    Computer Science. 2022, No.11:1-7
    Abstract (610) PDF (1921KB) (9546)
    Survey of Distributed Machine Learning Platforms and Algorithms (1862)
    SHU Na,LIU Bo,LIN Wei-wei,LI Peng-fei
    Computer Science. 2019, No.3:9-18
    Abstract (1862) PDF (1744KB) (8495)
    Survey of Cloud-edge Collaboration (1801)
    CHEN Yu-ping, LIU Bo, LIN Wei-wei, CHENG Hui-wen
    Computer Science. 2021, No.3:259-268
    Abstract (1801) PDF (1593KB) (8071)
    Survey of Fuzz Testing Technology (1317)
    ZHANG Xiong and LI Zhou-jun
    Computer Science. 2016, No.5:1-8
    Abstract (1317) PDF (833KB) (7375)
    Overview on Multi-agent Reinforcement Learning (3147)
    DU Wei, DING Shi-fei
    Computer Science. 2019, No.8:1-8
    Abstract (3147) PDF (1381KB) (7085)
    Methods in Adversarial Intelligent Game:A Holistic Comparative Analysis from Perspective of Game Theory and Reinforcement Learning (2875)
    YUAN Wei-lin, LUO Jun-ren, LU Li-na, CHEN Jia-xing, ZHANG Wan-peng, CHEN Jing
    Computer Science. 2022, No.8:191-204
    Abstract (2875) PDF (4699KB) (7004)
    Survey of Code Similarity Detection Methods and Tools (1015)
    ZHANG Dan,LUO Ping
    Computer Science. 2020, No.3:5-10
    Abstract (1015) PDF (1428KB) (6854)
    Survey of Symbolic Execution (767)
    YE Zhi-bin,YAN Bo
    Computer Science. 2018, No.6A:28-35
    Abstract (767) PDF (1797KB) (6459)
Announcement
More...
Subject
More...
Download
More...
Links
渝ICP备16013121号-4
Add:18# Honghu West Rd., Yubei ,Chongqing,China    Post Code: 401121
Tel: 023-63500828/023-67039612/023-67039623
E-mail: jsjkx12@163.com
Powered by Beijing Magtech Co., Ltd.