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....

2018 Vol.31 Issue.11, Published 2018-11-25 Papers and Reports    Researches and Applications        Papers and Reports 965 Incremental Dynamic Community Detection Algorithm Based on Density Clustering GUO Kun, PENG Shengbo, CHEN Yuzhong, GUO Wenzhong The community structures of social networks in the real world are always varying with nodes and edges of social networks increasing or disappearing dynamically as time goes by. In this paper, an incremental dynamic community detection algorithm based on density clustering is proposed. Firstly, the initial communities are generated according to the improved DBSCAN algorithm. Then, an index of edge variation rate is proposed and it is combined with the cosine similarity index to determine the community belonging adjustment process of the nodes whose neighbors vary in adjacent moment. In addition, both direct and indirect neighbor nodes are taken into account during the calculation of community belongingness.Finally, the communities are merged by iteratively updating the modularity gain to reduce the interference of noise communities. Experimental results on artificial datasets and real networks show that the proposed algorithm effectively copes with the variation of the network structures and incremental calculation cumulative errors with a low time complexity. 2018 Vol. 31 (11): 965-978 [Abstract] ( 748 ) [HTML 1KB] [ PDF 1075KB] ( 455 ) 979 Component Symmetric Positive Definite Descriptor Based on Gaussian Mixture Model CHU Li, WU Xiaojun The Gaussian mixture model(GMM) can use multiple Gaussian components to capture the variation information of image sets, and therefore it is a fine description method for image sets. Combining image set component symmetric positive definite descriptor]a component symmetric positive definite(SPD]model based on Gaussian mixture model(G-CSPD) is proposed. The image set is divided into sub-image sets with the same size, and the Gaussian mixture model of each sub-image set is calculated. A G-CSPD matrix in the form of the kernel matrix for all the sub-image sets is obtained, and the element in the matrix is used to denote the similarity between sub-image sets. The experimental results of 4 classification algorithms on 3 image sets show that G-CSPD is a more discriminative representation method for image sets. 2018 Vol. 31 (11): 979-985 [Abstract] ( 454 ) [HTML 1KB] [ PDF 637KB] ( 421 ) 986 Hesitant Fuzzy Rough Set Approach Based on Optimistic and Pessimistic Strategies LI Jianzhuo The existing hesitant fuzzy rough set model does not take the multi-source information into account. To solve this problem, an optimistic strategy based multigranulation hesitant fuzzy rough set model and a pessimistic strategy based multigranulation hesitant fuzzy rough set model are proposed, respectively. The properties of the two models are shown. Finally, the approximate sets under optimistic version and pessimistic version are analyzed by an example of a multi-source information system. 2018 Vol. 31 (11): 986-996 [Abstract] ( 626 ) [HTML 1KB] [ PDF 570KB] ( 309 ) 997 Enhanced Image Style Transferring Method with Primary Structure Maintained LIN Xing, CHEN Zhaojiong, YE Dongyi In the existing image style transferring methods, the primary structure of the target image is often deformed while a distinctive style is transferred. Therefore, a loss function for style transferring based on DCNN is designed. In addition to the original items]two more regularization terms are introduced in the function. To maintain the primary structure of the target image, the edge information extracted by the LoG operator is used as the feature for the primary structure. The first regularization term is constituted by feature difference between the resultant image and the target image. The second regularization term is composed of features obtained by Gabor filter to enhance the description of directional style features since artistic style is closely related to directional characteristics such as strokes, texture orientation and color flow while depth features are more focused on depicting global information. This item avoids the weakening effect of transferred style due to the maintenance of the primary structure. The experimental results show that the proposed method maintains better primary structure of the target image while successfully transferring a distinctive style. 2018 Vol. 31 (11): 997-1007 [Abstract] ( 384 ) [HTML 1KB] [ PDF 7106KB] ( 444 ) 1008 Anomaly Detection of Cloud Computing Platform Based on Multi-features Fusion ZHANG Jing, REN Yonggong A multi-feature fusion model based on distance constraint and solution space optimization is proposed to utilize the information of different sub-systems in the cloud computing platform and enhance the performance of anomaly detection. The minimum of the errors-sum from all of the single sub-system is obtained to achieve the optimal solution and the fusion of multi-features by iterating, and the high power coefficient is introduced to avoid the degenerating. Moreover, the proposed method is developed as an incremental learning method to ensure the real-time performance. The proposed method reduces the redundant information between high-dimension features and meanwhile mines the latent knowledge of different sub-systems in cloud platform. Thus, the performance in anomaly detection is improved. The private cloud platform based on OpenStack is constructed, and the real-time collection of data is implemented to verify the effectiveness of the proposed method. Compared with the state-of-the-art methods of anomaly detection in cloud platform, the proposed method achieves better accuracy. 2018 Vol. 31 (11): 1008-1017 [Abstract] ( 445 ) [HTML 1KB] [ PDF 1379KB] ( 398 ) Researches and Applications 1018 Gait Parameters Optimization Algorithm of Humanoid Robot Based on Spiral Model LI Xiaoyu, WANG Hao, FANG Baofu Walking control is the key issue to motion control of humanoid robots. To achieve fast and stable gait, a humanoid robot spiral model algorithm based on the covariance matrix adaptation evolution strategy(CMA-ES) is proposed in this paper. In the walking optimization process, the optimization task is divided into three sub tasks, the parameters are selected according to the optimization goal to join the corresponding optimization group, and the CMA-ES optimizer is constructed. Each CMA-ES algorithm optimizer is designed according to different learning objectives. Based on the optimization results of the previous optimization group, the spiral iterative optimization is combined with new requirements, and finally the established learning objectives are achieved to obtain the optimal parameter values. The proposed algorithm is applied in the HfutEngine simulation 3D team. The relevant gait test data of the robot show the validity of the proposed algorithm. 2018 Vol. 31 (11): 1018-1027 [Abstract] ( 390 ) [HTML 1KB] [ PDF 1711KB] ( 402 ) 1028 Microcalcification Clusters Detection Method Based on Deep Learning and Multi-scale Feature Fusion ZHANG Xinsheng, WANG Zhe To accurately identify the microcalcification in X-ray images for diagnosis and early prevention of breast cancer, a microcalcification object detection method combining fine-grained cascading enhancement-network(FCE-Net) and multi-scale feature fusion(MFF]is proposed. Firstly, the residual weights of FCE-Net are accumulated and the multi-branch structure is enhanced within the convolution module to obtain hierarchical and fine-grained convolution plots. Then, a candidate detection network based on MFF is constructed and multi-scale features are merged by double-upsampling. Thus, the confidence and regional coordinates of the object of the microcalcification clusters are acquired. Finally, the object region is classified and the bounding box is adjusted in the pooling layer of the region of interest. The experimental results on MIAS breast cancer dataset show that the method combining FCE-Net and MFF has a better ability to extract deep feature with enhanced classification and positioning accuracies. 2018 Vol. 31 (11): 1028-1039 [Abstract] ( 503 ) [HTML 1KB] [ PDF 1066KB] ( 602 ) 1040 Image Segmentation Using Fast Generalized Fuzzy C-means Clustering Based on Adaptive Filtering WANG Xiaopeng, ZHANG Yongfang, WANG Wei, WEN Haotian When the fast generalized fuzzy C-means clustering(FGFCM]is directly used to segment serious noise images, the clustering center offset, inaccurate results and error of the image segmentation are easily caused due to the noise. Therefore, a fast generalized fuzzy C-means clustering algorithm based on adaptive filtering is proposed. Firstly, the parameter balance factor is adaptively determined according to the noise probability of nonlocal pixels to reflect the spatial structure information in the image more accurately. Then]the balance factor is used to effectively combine the linear weighted sum filtered image in the FGFCM algorithm with the median filtered image of the original image to create the adaptive filtered image. Since the filtering degree of the filtered image depends on the probability that the pixel is noise in the image, the dynamic noise suppression performance of the proposed method can be greatly improved. The experimental results show that compared with FCM and FGFCM, the proposed method obtains more accurate results in clustering segmentation of images with serious noise. 2018 Vol. 31 (11): 1040-1046 [Abstract] ( 701 ) [HTML 1KB] [ PDF 829KB] ( 476 ) 1047 Monte Carlo Noise Removal Algorithm Based on Adversarial Generative Network XIE Chuan, WANG Yongchao, LIN Zhijie, ZHENG Qiulan, QIAN Fei, ZHAO Lei To solve the problem of high frequency details loss in the existing Monte Carlo noise removal method, a Monte Carlo noise removal method based on adversarial generative network is proposed. An adversarial network structure, including the generative network of full convolution network and the discriminator network of deep convolution network, is employed to remove the Monte Carlo noise. The multi-dimensional auxiliary features, including the pixel color of the image, are added as the network input. Besides, the new loss function and local importance sampling technology based on the similarity deviation between normal vector variance and gradient size are applied to network training. Experimental results show that the proposed method achieves good quantization index in removing Monte Carlo noise and meanwhile preserves high-frequency detail features of the image. 2018 Vol. 31 (11): 1047-1060 [Abstract] ( 737 ) [HTML 1KB] [ PDF 3585KB] ( 638 )

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