Home      About Journal      Editorial Board      Instructions      Ethics Statement      Contact Us                   中文

Judgement and Disposal of Academic Misconduct Article Copyright Transfer Agreement Proof of Confidentiality Requirements for Electronic Version More.... Chinese Association of Automation National ResearchCenter for Intelligent Computing System Institute of Intelligent Machines,Chinese Academy of Sciences More....

2020 Vol.33 Issue.3, Published 2020-03-25 Papers and Reports    Researches and Applications    Surveys and Reviews        Papers and Reports 191 Improved NSGA-III Algorithm Based on Reference Point Selection Strategy GENG Huantong, DAI Zhongbin, WANG Tianlei, XU Ke The traditional multi-objective evolutionary algorithm ignores distribution information of the population in the decision space and Pareto front shape of the optimization problem is not taken into account. To solve the problems, an improved NSGA-III algorithm based on reference point selection strategy is proposed. Firstly, according to the entropy thought in information theory, the entropy difference between two adjacent generations is calculated in line with the distribution characteristics of the population in the decision-making space, and the evolutionary status of the population is determined. Then, in the light of the distribution characteristics of the population in the target space, the importance of reference points is evaluated via statistical information of the number of the individuals associated with reference points. Finally, redundant and invalid reference points are removed according to the importance characteristics of reference points in the middle and late period of population evolution. Reserved reference points can adapt to the population size and Pareto frontier, and the selected reference points are exploited to guide the evolution direction of the population and accelerate the convergence and optimization efficiency. Experiments on test function sets indicate the significant advantages of the proposed algorithm in convergence and distribution. 2020 Vol. 33 (3): 191-201 [Abstract] ( 1218 ) [HTML 1KB] [ PDF 1425KB] ( 557 ) 202 Concept Drift Detection of Multi-label Data Stream Based on Hierarchical Verification ZHANG Yong, LIU Haoke, CHEN Tianzhen Most of the existing concept drift detection methods focus on single label data stream, and therefore they cannot meet the requirements of multi-label data stream concept drift detection. A concept drift detection algorithm of multi-label data stream based on hierarchical verification is proposed. The proposed algorithm consists of a test layer and a verification layer. The variation of data distribution is detected by the test layer to judge whether the concept drift occurs or not. The variation degree of tag confusion matrix is judged by the verification layer to verify whether the concept drift occurs indeed or not. Experiments on real multi-label and synthetic multi-label datasets indicate that the proposed algorithm detects the concept drift effectively and improves the classification performance. 2020 Vol. 33 (3): 202-210 [Abstract] ( 440 ) [HTML 1KB] [ PDF 653KB] ( 378 ) 211 Object Part Segmentation Network Based on DeepLab ZHAO Xia, NI Yingting The low precision exists in the existing part segmentation, and the generalization and precision can not be balanced. Aiming at the problems, a part segmentation network(DeepLab-MAFE-DSC) based on DeepLab is proposed. A multi-scale adaptive-pattern feature extraction(MAFE) module is proposed in encoder part of the network. The deformable convolution is exploited to enhance the processing capability to irregular contour, and sampling mode of cascade and concatenate in parallel is adopted to balance global and local information simultaneously. A decoder module based on skip connection(DSC) is designed to connect high-level semantic information and low-level character information. Experiments on the dataset show the advantages of DeepLab-MAFE-DSC in simplicity, high part segmentation accuracy and strong generalization. 2020 Vol. 33 (3): 211-220 [Abstract] ( 639 ) [HTML 1KB] [ PDF 3416KB] ( 402 ) 221 Sequential Subspace Clustering via Joint lp/l2,p-Norms Minimization HU Wenyu, LI Shenghao, TU Zhihui, YI Yun To extract the spatio-temporal information in sequential data effectively, a sequential subspace clustering method via joint lp/l2,p-norms minimization is proposed. Firstly, a l2,p-norm temporal graph is constructed to describe local similarity along the temporal direction by defining the sample-distance dependent weights. Secondly, since non-convex lp-norm(0<p<1) minimization delivers better results than convex l1-norm minimization does, and it removes more links between semantically-unrelated samples, lp-norm is adopted to measure the sparsity of representation matrix. Finally, the linearized alternating direction method is employed to solve the optimization model. Experiments on video dataset, motion dataset, and face dataset confirm the effectiveness of the proposed method. 2020 Vol. 33 (3): 221-233 [Abstract] ( 439 ) [HTML 1KB] [ PDF 1106KB] ( 667 ) Surveys and Reviews 234 Vision-Based Abnormal Vehicle Behavior Detection: A Survey HUANG Chao, HU Zhijun, XU Yong, WANG Yaowei Vision-based abnormal vehicle behavior detection can detect abnormal vehicle behaviors in the traffic surveillance video promptly and give an alarm. It plays an important role in improving the efficiency of traffic enforcement and traffic conditions and reducing traffic accident rate. Despite the progresses in abnormal vehicle behavior detection, there are still many challenges in practical application, such as lack of labeled data, uncertain anomaly, occlusion and poor real time capability. To make a clear understanding of abnormal vehicle behavior detection, the algorithms proposed in recent years are summarized. Firstly, the typical features representing vehicle behaviors are introduced, and the advantages and disadvantages of model learning methods of the algorithms are discussed from the perspectives of supervised and unsupervised learning. Then, the existing algorithms are categorized into model-based, reconstruction-based and deep neural network-based methods. Each category is introduced and analyzed. Finally, problems and prediction of the future of abnormal vehicle behavior detection are discussed. 2020 Vol. 33 (3): 234-248 [Abstract] ( 1178 ) [HTML 1KB] [ PDF 1299KB] ( 1339 ) Researches and Applications 249 Cost Sensitive Random Forest Classification Algorithm for Highly Unbalanced Data PING Rui, ZHOU Shuisheng, LI Dong For highly unbalanced data, insufficient learning of minority class samples is caused by self-sampling method of the traditional cost sensitive random forest algorithm, and the cost sensitive mechanism of the algorithm is easily weakened by the large proportion of majority class samples. Therefore, a weak balance cost sensitive random forest algorithm based on clustering is proposed. After clustering the majority class samples, the weak balance criterion is used to reduce the samples of each cluster repeatedly. The selected majority class samples and the minority class samples of the original training set are fused to generate a number of new unbalanced datasets for the training of cost sensitive decision tree. The proposed algorithm not only enables the minority class samples to be fully learned, but also ensures that the cost sensitive mechanism is less affected by reducing the majority class samples. Experiment indicates the better performance of the proposed algorithm in processing highly unbalanced datasets. 2020 Vol. 33 (3): 249-257 [Abstract] ( 734 ) [HTML 1KB] [ PDF 886KB] ( 366 ) 258 Deep Unsupervised Hashing with Pseudo Pairwise Labels LIN Jiwen, LIU Huawen It is difficult to obtain high-quality hash codes for unsupervised deep hashing methods due to the lack of similarity supervised information. Therefore, an end-to-end deep unsupervised hashing model based on pseudo-pairwise labels is proposed. Statistical analysis is performed on the image features extracted by the pre-trained deep convolutional neural network to construct the semantic similarity labels for data. Supervised deep hashing based on pairwise labels is then conducted. Experiments on commonly used image datasets CIFAR-10 and NUS-WIDE indicate that hash codes obtained by the proposed method perform better on image retrieval. 2020 Vol. 33 (3): 258-267 [Abstract] ( 444 ) [HTML 1KB] [ PDF 633KB] ( 345 ) 268 Locality Feature Aggregation Loss and Multi-feature Fusion for Facial Expression Recognition WANG Hao, LI Yongze, FANG Baofu Each individual makes facial expressions in a unique way. In this paper, a locality-feature-aggregation(LFA) loss function is proposed. Differences between images of the same class are reduced and those between images of different classes are expanded during the training of deep neural network. Thus, the influence of expression polymorphism on feature extraction by deep learning is weakened. The local areas with rich expressions can express facial expression features better. A deep learning network framework incorporating LFA loss function is proposed. Local features of facial images are extracted for facial expression recognition. Compared with other methods, the proposed method is more effective on real world RAF datasets and CK+ datasets under laboratory conditions. 2020 Vol. 33 (3): 268-276 [Abstract] ( 557 ) [HTML 1KB] [ PDF 751KB] ( 859 ) 277 Relationship Prediction for Literature Network under Meta-Structure WANG Xiu, CHEN Lu, YU Chunyan To solve the problem of relationship prediction among literature network nodes, the similarity of nodes is regarded as the probability of relationship among nodes, and a network representation learning method is utilized to embed nodes into a low-dimensional space to calculate the similarity. Therefore, a meta-structure-based network representation learning model is proposed. According to the correlation between nodes based on different meta-structures, the network is mapped to a low-dimensional feature space by fusing their corresponding feature representations. The relationship prediction of literature network is realized by the distance measure in the low-dimensional feature space. Experiments indicate that the proposed algorithm obtains good relationship prediction results in literature network. 2020 Vol. 33 (3): 277-286 [Abstract] ( 359 ) [HTML 1KB] [ PDF 716KB] ( 286 )

Supervised by China Association for Science and Technology Sponsored by Chinese Association of Automation NationalResearchCenter for Intelligent Computing System Institute of Intelligent Machines, Chinese Academy of Sciences Published by Science Press

Copyright © 2010 Editorial Office of Pattern Recognition and Artificial Intelligence Address: No.350 Shushanhu Road, Hefei, Anhui Province, P.R. China Tel: 0551-65591176 Fax:0551-65591176 Email: bjb@iim.ac.cn Supported by Beijing Magtech 　Email:support@magtech.com.cn