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.9, Published 2020-09-25 Papers and Reports    Researches and Applications    Multi-granulation Attribute Representation and Its Application        Multi-granulation Attribute Representation and Its Application 767 Attribute Logical Formulas Description of Object Granules Based on Property-Oriented(Object-Oriented) Concepts WU Xia, ZHANG Jialu The attribute logical formula descriptions of object granules based on the property-oriented concept and the object-oriented concepts are studied. The relationships between common attribute concept lattice and property-oriented concept lattice as well as common attribute concept lattice and object-oriented concept lattice are discussed. On the basis of the object granule description method of concept lattice of common attribute analysis, the object granule descriptions based on property-oriented concept lattice of possible attribute analysis and object-oriented concept lattice of necessary attribute analysis are given, respectively. Structural characterizations of attribute logic formulas are analyzed respectively, the semantic of these attribute logic formulas are exactly the extent of property-oriented concept or object-oriented concept. The attribute logical formula descriptions of object granules are helpful to construct the property-oriented and the object-oriented concept lattices. 2020 Vol. 33 (9): 767-775 [Abstract] ( 524 ) [HTML 1KB] [ PDF 543KB] ( 287 ) 776 Boolean Matrix Approach for Multi-scale Covering Decision Information System CHEN Yingsheng, LI Jinjin, LIN Rongde, CHEN Dongxiao In multi-scale decision information system, one condition attribute corresponding to a certain scale forms a partition of the universe. A multi-scale covering decision information system (MSCDS) is proposed and the partition is generalized to a covering. A boolean matrix method is applied to simplify the complexity of information expression in this system. Firstly, boolean matrix is employed to describe the covering decision information system, including upper and lower approximations, consistency and generalized decision function. Secondly, the definitions of MSCDS, consistency and generalized decision invariant of the system are expressed in boolean matrix method. Finally, the boolean matrix method is utilized to define the significance of a combination scale with both consistency and inconsistency, and the relevant algorithms and examples of optimal granularity selection of MSCDS are presented. 2020 Vol. 33 (9): 776-785 [Abstract] ( 394 ) [HTML 1KB] [ PDF 677KB] ( 220 ) 786 Attribute Reductions of Formal Context Based on Information Entropy CHEN Dongxiao, LI Jinjin, LIN Rongde, CHEN Yingsheng Attribute significances and attribute reduction are crucial in formal concept analysis. Some approaches to attribute reduction of formal context are proposed based on information entropy. Firstly, information entropy, conditional entropy and mutual information of formal context are defined, and attribute reduction by means of conditional entropy is conducted in consistent decision formal context. The equivalence between the granular consistency and the entropy consistency in decision formal context is produced. Secondly, limitary information entropy, limitary conditional entropy and limitary mutual information are proposed, and attribute reductions are conducted by means of limitary conditional entropy in inconsistent formal decision context. Finally, the attribute reduction algorithms of consistent and inconsistent formal decision contexts are proposed by the significance of attributes, and numerical experiments show the efficiency of the proposed algorithms. 2020 Vol. 33 (9): 786-798 [Abstract] ( 449 ) [HTML 1KB] [ PDF 748KB] ( 307 ) 799 Incomplete Multi-granulation Reduction Based on Discernibility Matrix LIU Kai, TAN Anhui, GU Shenming For the incomplete data with missing attribute values, the multi-granularity reduction structures of incomplete information systems and incomplete decision systems are constructed from the perspective of discernibility matrix. Firstly, the reduction attributes of incomplete information systems based on pessimistic and optimistic multi-granularity approximations are discussed, and three types of multi-granularity discernibility matrices of incomplete information systems and incomplete decision systems are constructed. Then, it is theoretically proved that all the multi-granularity reductions of incomplete information systems and incomplete decision systems can be computed accurately by the disjunctive and conjunctive logical operations of the constructed discernibility matrices. Finally, examples are given to demonstrate the effectiveness and practicability of the proposed method. 2020 Vol. 33 (9): 799-810 [Abstract] ( 380 ) [HTML 1KB] [ PDF 615KB] ( 184 ) Papers and Reports 811 Adaptive Undersampling Based on Density Peak Clustering CUI Caixia, CAO Fuyuan , LIANG Jiye Undersampling based on K-means clustering is only suitable for hypersphere shape data, the impact of overlapping regions on classification is not taken into account, and the density of samples in the clusters is neglected. Therefore, an adaptive undersampling method based on density peak clustering is proposed. Firstly, the samples of the majority class in the overlapping region are identified by the nearest neighbor search algorithm and deleted. Secondly, a number of clusters of different shapes, sizes and densities are automatically obtained by improved density peaks clustering. Then, undersampling is performed according to the sampling weights calculated by the density of the samples in the subclusters, and bagging ensemble classification is conducted on the obtained balanced dataset. Experiments indicate that the performance of the proposed method is better on most datasets. 2020 Vol. 33 (9): 811-819 [Abstract] ( 459 ) [HTML 1KB] [ PDF 724KB] ( 399 ) 820 Online Streaming Feature Selection for High-Dimensional and Class-Imbalanced Data Based on Max-Decision Boundary LIN Yaojin, CHEN Xiangyan, BAI Shengxing, WANG Chenxi The feature space of data changes with time dynamically. The number of features on training data is high-dimensional and fixed, and the label space is imbalanced. Motivated by the above, an online streaming feature selection algorithm for high-dimensional and class-imbalanced data based on max-decision boundary is proposed. An adaptive neighborhood relation is defined with consideration of the effect of boundary samples based on neighborhood rough set, and then a rough dependency calculation formula with respect to max-decision boundary is designed. Meanwhile, three online feature subset evaluation metrics are proposed to select features with great discriminability in majority and minority classes. Experiments on eleven high-dimensional and class-imbalanced datasets indicate that the proposed method achieves better performance than some state-of-the-art online streaming feature selection algorithms. 2020 Vol. 33 (9): 820-829 [Abstract] ( 343 ) [HTML 1KB] [ PDF 732KB] ( 305 ) Researches and Applications 830 Generating Adversarial Example with GAN for White-Box Target Attacks ZHANG Gaozhi, LIU Xinping, SHAO Mingwen Deep neural networks(DNNs) are easily affected by adversarial examples and consequently generate wrong outputs. Adversarial examples are generated by the traditional methods from an optimization perspective. In this paper, a method for generating adversarial examples is proposed with generative adversarial network(GAN) and GAN is exploited for target attack in the white-box setting. Adversarial perturbations are generated by a trained generator to form adversarial examples. Four kinds of loss functions are utilized to constrain the quality of adversarial examples and improve attack success rates. The effectiveness of the proposed method is testified through extensive experiments on MNIST, CIFAR-10 and ImageNet datasets and the proposed method produces higher attack success rates. 2020 Vol. 33 (9): 830-838 [Abstract] ( 659 ) [HTML 1KB] [ PDF 1527KB] ( 440 ) 839 User Credit Ranking Based on Structured Non-linear Ordinal Regression REN Yonggong, GUO Jiaqi, ZHANG Jing The user fraud detection is realized by constructing the binary-classification model in the traditional methods, and therefore it is difficult to obtain the potential of applications. In this paper, an algorithm of user credit ranking based on structured non-linear ordinal regression and a robust structured non-linear ordinal regression model are proposed. Firstly, an adaptive global weight matrix is generated to solve overfitting and underfitting caused by the imbalanced distribution of samples. Then, the penalty constraint of ordered inter-categories is established to optimize the projection direction to avoid noises and enhance the robustness of the model. The user information from the actual internet applications is collected, and feature extraction and labelling of ordered categories are conducted. The experiment shows the proposed model achieves better performance. 2020 Vol. 33 (9): 839-851 [Abstract] ( 370 ) [HTML 1KB] [ PDF 1473KB] ( 198 ) 852 Rain Removal Method for Traffic Surveillance Video in Joint Spatial-Frequency Domain SONG Chuanming, HONG Xu, LIU Dingkun, WANG Xianghai The processing of traffic surveillance videos in rainy days is inefficient and unreliable. A rain removal method is proposed for traffic surveillance video in joint spatial-frequency domain with an ability to discriminate the amount of rainfall adaptively. After setting the low frequency coefficients of non-subsampled Shearlet transform to zero, the map of all edge information is computed by the Otsu method. Subsequently, the saliency mapping method is utilized to calculate a depth map. The map of main edge information is obtained by bilateral filtering and the high-frequency coefficients are retained. Through combining the map of all edge information, the map of main edge information and the frame difference, the areas of both raindrop and rain line are determined, and the amount of rainfall is analyzed. If the rain is moderate or heavy, a curvature-driven diffusion method is employed to restore the pixels in the areas of raindrop and rain line. Otherwise, the detection results are aggregated under two scales. Experimental results show that the proposed algorithm effectively removes the raindrops and rain lines in videos with shapes and texture details of the objects preserved. Moreover, the post-processing quality is improved, such as moving object tracking. 2020 Vol. 33 (9): 852-866 [Abstract] ( 345 ) [HTML 1KB] [ PDF 2971KB] ( 517 )

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