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....

2016 Vol.29 Issue.11, Published 2016-11-30 Papers and Reports    Researches and Applications        Papers and Reports 961 Hybrid Heuristic Value Iteration POMDP Algorithm LIU Feng Point-based value iteration methods are a kind of algorithms for effectively solving partially observable Markov decision process (POMDP) model. However, the algorithm efficiency is limited by the belief point set explored in most of the algorithms by single heuristic criterion. A hybrid heuristic value iteration algorithm (HHVI) for exploring belief point set is presented in this paper. The upper and lower bounds on the value function are maintained and only the belief points with its value function bounds difference greater than the threshold are selected to expand. Furthermore, the furthest belief point away from the explored point set among the subsequent belief points with the above difference also greater than the threshold is explored. The convergence effect of HHVI is guaranteed by making the explored point set fully distributed in the reachable belief space. Experimental results of four benchmarks show that HHVI can guarantee the convergence efficiency and obtain better global optimal solution. 2016 Vol. 29 (11): 961-968 [Abstract] ( 529 ) [HTML 1KB] [ PDF 423KB] ( 635 ) 969 Double-Level Absolute Reduction for Multi-granulation Rough Sets DENG Dayong, HUANG Houkuan Multi-granulation rough set is a rough set model for heterogenous data in essence. However, it is still not employed to deal with heterogenous data. From the viewpoints of absolute attribute reduction, double-level absolute reduction for multi-granulation rough sets is proposed, including multi-granulation absolute recducts and multi-granulation absolute granulation reducts, and properties of double-level absolute reduction are analyzed from the perspective of heterogenous data. The algorithms for double-level absolute reduction are presented. Theoretical analysis and example show the validation of multi-granulation absolute reducts, multi-granulation absolute granulation reducts and double-level absolute reducts. 2016 Vol. 29 (11): 969-975 [Abstract] ( 483 ) [HTML 1KB] [ PDF 367KB] ( 479 ) 976 Rules Acquisition and Attribute Reduction of Ordered Formal Decision Contexts ZHANG Jing, WEI Ling Based on the formal decision context containing multi-valued decision information, the concept of the ordered formal decision context and the correlated theory of ordered decision concept lattice are proposed. In addition, the decision rule of the ordered formal decision concept lattice as well as the confidence level and the support degree of the rule is given, and the significance of decision rules in the practical application is discussed. Furthermore, the rule-preserved attribute reduction is defined. Meanwhile, the attribute reduction method of ordered decision concept lattice structure preservation is obtained. Finally, the relationship between the rule preserved reduction and the lattice structure preserved reduction of the ordered formal decision context is discussed. 2016 Vol. 29 (11): 976-984 [Abstract] ( 371 ) [HTML 1KB] [ PDF 408KB] ( 335 ) 985 An Improved Compressive Tracking Algorithm Adapting to Variable Target Scales ZHANG Yuting, YE Dongyi, KE Xiao, CHEN Zhaojiong Compressive tracking algorithms based on compressive sensing theory for reducing the dimension of Haar-like feature of the target utilize a fixed tracking scale, and therefore they are prone to tracking drift or even target missing when the size of the target changes. To overcome the drawback, the variation of Haar-like feature according to the target scales is analyzed. It is found that the values of Haar-like feature of target in the tracking rectangular frame change with the area of the tracking frame in an approximately linear way within certain range of scales. Grounded on this relationship, an improved compressive tracking algorithm adapting to variable target scales (CTVS) is proposed. Experimental results show that CTVS can adapt to the change of target size and perform well in reducing the influence of interferences like occlusion, light illumination variation, background clutter and deformation. Moreover, CTVS is capable of real-time tracking with higher robustness, accuracy and computation efficiency. 2016 Vol. 29 (11): 985-996 [Abstract] ( 478 ) [HTML 1KB] [ PDF 1907KB] ( 761 ) 997 Biomedical Named Entity Recognition Based on Deep Conditional Random Fields SUN Xiao, SUN Chongyuan, REN Fuji Biomedical named entity recognition is the fundamental and key step in bioinformatics. In this paper, a biomedical named entity recognition method based on deep conditional random fields is proposed. The deep conditional random fields of multi-layer structure are constructed by stacking the linear-chain conditional random fields and the optimal feature set is built by incremental learning strategy. Finally, error correction algorithm based on full name-abbreviation and error correction algorithm based on domain knowledge are adopted for further modifying the recognition results. Experiments are conducted on the biomedical named entity recognition corpus JNLPBA, and the results demonstrate the effectiveness of the proposed method. 2016 Vol. 29 (11): 997-1008 [Abstract] ( 469 ) [HTML 1KB] [ PDF 545KB] ( 652 ) Researches and Applications 1009 Fractional Total Variation Denoising Model Based on Adaptive Projection Algorithm ZHANG Guimei, SUN Xiaoxu, LIU Jianxin To preserve weak edges and texture details of an image during image denoising, a fractional order total variation denoising model is presented based on adaptive projection algorithm. Firstly, Grünwald-Letnikov fractional order differential is used as a substitute for the first order derivative in the regularization term of total variation model. Secondly, the image is projected to a total variation ball to handle the optimization problem. The image is divided into the texture area and the non-texture area according to the local information of the image, and thus soft threshold values can be calculated adaptively. Both theoretical analysis and experimental results show that the proposed method eliminates the block effect as well as preserves the texture details effectively for removing noise. 2016 Vol. 29 (11): 1009-1018 [Abstract] ( 465 ) [HTML 1KB] [ PDF 1510KB] ( 701 ) 1019 Adaptive Weighted Object Tracking Algorithm Based on Multi-appearance Models ZHU Zhenfeng, YANG Haobo, YE Yangdong Partial least squares (PLS) tracking algorithm ignores the differences among features and those among appearance models. The corresponding tracking is easily affected by the factors, such as illumination and occlusion, and thereby the tracking accuracy decreases. To address these problems in application, an adaptive weight object tracking algorithm based on multi-appearance model (AWMA) is proposed. Firstly, the PLS method is used to gradually establish multiple appearance models for the target region. Then, according to the importance of features and significant degree of object in each appearance model, a comprehensive model with adaptive weights is built. Furthermore, the error analysis between object and sample is accomplished by integrating multiple appearance models. Finally, particle filter is used to achieve object tracking. The experimental results show that the proposed algorithm can effectively filter the noise data and improve tracking robustness and efficiency. 2016 Vol. 29 (11): 1019-1027 [Abstract] ( 444 ) [HTML 1KB] [ PDF 1128KB] ( 532 ) 1028 Task Scheduling Algorithm Based on Q-Learning and Programming for Sensor Nodes WEI Zhenchun, XU Xiangwei , FENG Lin, DING Bei To improve the learning policy and obtain better application performance of sensor nodes, a task scheduling algorithm based on Q-learning and programming (QP) for sensor nodes is proposed with the task model of data collection applications. Specifically, some basic learning elements, such as state space, delayed reward and the exploration-exploitation policy, are defined in QP as well. Moreover, according to the characteristics of wireless sensor network(WSN), the programming process based on the expired mechanism and the priority mechanism is established to improve the learning policy by making full use of empirical knowledge. Experimental results show that QP has the ability to perform task scheduling dynamically according to current WSN environments. Compared with other task scheduling algorithms, QP achieves better application performance with reasonable energy consumption. 2016 Vol. 29 (11): 1028-1036 [Abstract] ( 447 ) [HTML 1KB] [ PDF 477KB] ( 1022 ) 1037 Multi-manifold Learning Based on Boundary Detection ZOU Peng, LI Fanzhang, YIN Hongwei, ZHANG Li, ZHANG Zhao In manifold learning algorithms, the data are assumed to be aligned on a single manifold. The application of algorithms is limited due to the general distribution of practical datasets on multiple manifolds. In this paper, multi-manifold learning based on boundary detection(MBD) is proposed. By the proposed method, data of distribution on several manifolds are efficiently learned through boundary detection and intra and inter manifolds geodesic distances can be kept faithfully. Firstly the boundary of data manifolds is detected and then the dimensionality of the manifolds is reduced separately. Finally, low dimensional coordinates are relocated into a global coordinate system. The effectiveness of the proposed multi-manifold learning algorithm is demonstrated through experiments on both synthetic and real datasets. 2016 Vol. 29 (11): 1037-1047 [Abstract] ( 461 ) [HTML 1KB] [ PDF 2310KB] ( 658 ) 1048 Continuous K-Nearest Neighbor Queries for Uncertain Moving Objects YU Yanwei, QI Jianpeng, SONG Peng, ZHANG Yonggang An urgent problem in location-based services is continuous K-nearest neighbor (KNN) queries for uncertain moving objects. An efficient algorithm for continuous K-nearest neighbor queries for uncertain moving objects is proposed. Firstly, two solutions, MaxMin and Rate, are proposed to predict the possible location range of the moving object in the time interval by utilizing the sampling points with velocities in the recent time window. A closed interval of minimum and maximum distances is employed to represent the distance between the query object and the moving object. Secondly, an optimized ranking method based on vague possibility decision is proposed to quickly find KNNs of the query object. Finally, experimental results on real and synthetic large-scale datasets demonstrate the effectiveness of the proposed algorithm. 2016 Vol. 29 (11): 1048-1056 [Abstract] ( 414 ) [HTML 1KB] [ PDF 549KB] ( 455 )

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