模式识别与人工智能
Home      About Journal      Editorial Board      Instructions      Ethics Statement      Contact Us                   中文
Pattern Recognition and Artificial Intelligence
22 Judgement and Disposal of Academic Misconduct Article
22 Copyright Transfer Agreement
22 Proof of Confidentiality
22 Requirements for Electronic Version
More....
22 Chinese Association of Automation
22 National ResearchCenter for Intelligent Computing System
22 Institute of Intelligent Machines,Chinese Academy of Sciences
More....
 
 
2018 Vol.31 Issue.5, Published 2018-05-25
Papers and Reports    Researches and Applications    Surveys and Reviews   
   
Papers and Reports
389 Loop Closure Detection Based on Maximizing Rényi Divergence of Multiple Features
WANG Xiaolong, PENG Guohua
Multiple image features provide more discriminative information of the scenes compared with individual feature, and thus the performance of loop closure detection(LCD) is improved. However, a suitable combination criterion is vital. A weighting method of multiple feature combination is proposed. The accuracy of LCD of the feature combination is expressed as the Rényi divergence of the distance distributions of true matches and false matches in the feature space. The optimal feature combination maximizes the Rényi divergence. The relationship between the parameter of Rényi divergence and the performance of LCD of the optimal feature combination is analyzed and experimentally verified. The experiments show that the proposed method improves the performance of LCD significantly and the best performance is achieved with the parameter of Rényi divergence being from 0.75 to 1.
2018 Vol. 31 (5): 389-397 [Abstract] ( 650 ) [HTML 1KB] [ PDF 3821KB] ( 422 )
398 Self-adapting PSO Algorithm with Efficient Hybrid Transformation Strategy for X-Architecture Steiner Minimal Tree Construction Algorithm
LIU Genggeng, CHEN Zhisheng, GUO Wenzhong, CHEN Guolong

X-architecture Steiner minimal tree (XSMT) problem is an NP-hard problem, and it is the best connection model for a multi-terminal net in non-Manhattan global routing problem. An XSMT construction algorithm based on hybrid transformation strategy and self-adapting particle swarm optimization(PSO) is proposed. Firstly, an effective hybrid transformation strategy is designed to enlarge the search space and enhance the convergence of the algorithm. Secondly, the crossover and mutation operators based on union-find sets and a self-adapting strategy to adjust the learning factors are proposed to satisfy the robustness of particle coding and further speed up the convergence of algorithm. The experimental results show that the proposed algorithm efficiently produces a better solution than others. Moreover, it obtains a series of XSMTs with different topology but same length. Thus, it provides a variety of options for global routing and opportunities to reduce congestion.
2018 Vol. 31 (5): 398-408 [Abstract] ( 622 ) [HTML 1KB] [ PDF 866KB] ( 344 )
409 Subspace Clustering via Joint Feature Selection and Smooth Representation
ZHENG Jianwei, LU Cheng, QIN Mengjie, CHEN Wanjun
The performance of self-representation based methods is affected by redundant high-dimensional features. Therefore, a subspace clustering method via joint feature selection and smooth representation(FSSR) is proposed in this paper. Firstly, the idea of feature selection is integrated into the self-representation based coefficient matrix learning framework. Meanwhile, a weight factor is adopted to measure different contributions of correlated features. Furthermore, a group effectiveness constraint is imposed on the coefficient matrix for the preservation of locality property. An alternating direction method of multipliers(ADMM) based algorithm is derived to optimize the proposed cost function. Experiments are conducted on synthetic data and standard databases and the results demonstrate that FSSR outperforms the state-of-the-art approaches in both accuracy and efficiency.
2018 Vol. 31 (5): 409-418 [Abstract] ( 648 ) [HTML 1KB] [ PDF 1021KB] ( 503 )
419 Multi-label Learning Algorithm of Regression Kernel Extreme Learning Machine
WANG Yibin, CHENG Yusheng, HE Yue, PEI Gensheng
In the multi-label learning algorithms based on extreme learning machine(ELM), the ELM classification model is often used, and the correlation between labels is ignored. Accordingly, a multi-label learning algorithm of regression kernel extreme learning machine with association rules(ML-ASRKELM) is proposed in this paper. Firstly, the rule vectors between labels are extracted by analyzing the association rules of label space. Then, the prediction results are obtained by the proposed multi-label regression kernel extreme learning machine(ML-RKELM). Eventually, if the rule vectors are not empty, the final results are calculated by the rule vectors and the prediction results of ML-KRELM. Otherwise, the final results are predicted by ML-RKELM. The experimental results show that ML-ASRKELM and ML-RKELM are superior to other algorithms, and the effectiveness of the proposed algorithms are illustrated by the statistical hypothesis test.
2018 Vol. 31 (5): 419-430 [Abstract] ( 638 ) [HTML 1KB] [ PDF 623KB] ( 362 )
431 A Swarm Intelligence Algorithm-Lion Swarm Optimization
LIU Shengjian, YANG Yan, ZHOU Yongquan
Based on the natural division of labor among lion king, lionesses and cubs in a lion group, a swarm intelligent algorithm, loin swarm optimization(LSO), is proposed. LSO is inspired by intelligent behaviors of three populations including lion guarding, lioness hunting, cubs following. In LSO, policies of location updating are different for three populations. LSO follows the biological competition law of “survival of the fittest” in nature world, i.e., the lion king guards territory and possesses the priority of food, lionesses cooperate in hunting, and lion cubs fall into eating, learning to hunt, and being expelled after entering adulthood. The diversity of lion location updating guarantees that LSO converges fast and is not easily trapped into a local optimal solution. LSO is compared with the particle swarm optimization and the bare bones particle swarm optimization on six optimization test functions. Results show that LSO produces fast convergence and high precision, and it obtains a better global optimal solution.
2018 Vol. 31 (5): 431-441 [Abstract] ( 1980 ) [HTML 1KB] [ PDF 882KB] ( 889 )
Surveys and Reviews
442 Deep Learning for Gait Recognition: A Survey
HE Yiwei, ZHANG Junping
Gait recognition methods have difficulty in achieving satisfactory performance, since the gait is vulnerable to covariates such as occlusion, clothing, view angles and carrying condition. Based on the framework of end-to-end learning and multi-layer feature extraction technology, fruitful achievements are made by applying deep learning to the field of gait recognition. The status quo, pros and cons of deep learning in gait recognition are reviewed, and the key technologies and several potential research directions are discussed.
2018 Vol. 31 (5): 442-452 [Abstract] ( 1618 ) [HTML 1KB] [ PDF 778KB] ( 1237 )
Researches and Applications
453 Metal Fracture Image Classification Based on Adaptive Fusion of Multiple Features
LI Ming, XING Dongdong, WANG Yuling, LU Yuming
To enhance the discrimination ability of the extracted features of metal fracture images and improve the recognition rate of fracture images, a multi-feature fusion algorithm is presented by combining global and local texture features. Firstly, the global texture features of the image are extracted by Trace transform, and the local texture features are extracted by the local binary pattern. Then the dynamic weighted discrimination power analysis is employed for feature selection and adaptive weighted fusion. Finally, the classification are conducted by support vector machine. Experimental results on the image database of the metal fracture show that the proposed method produces a high recognition rate. The proposed algorithm also produces a high recognition rate and a strong generalization ability on other texture databases.
2018 Vol. 31 (5): 453-461 [Abstract] ( 637 ) [HTML 1KB] [ PDF 1189KB] ( 458 )
462 Travel Routing Mining Based on Multiple Latent Semantic Representation Model
SUN Yanpeng, GU Tianlong, BIN Chenzhong , SUN Lei
Aiming at mining and recommending the personalized travel behavior of tourists, a multiple latent semantic travel route representation model(MLSTR-RM) is proposed. With the consideration of the influence of different contexts on the travel route, the efficient representation of different latent semantics in travel routes is studied in MLSTR-RM. Firstly, the latent semantic contained by the different contexts in model is determined. Then, the negative sampling is applied to train parameters in the model, and a personalized attraction recommendation method is designed based on MLSTR-RM model. Experiments on real data sets show the effectiveness of the proposed model.
2018 Vol. 31 (5): 462-469 [Abstract] ( 424 ) [HTML 1KB] [ PDF 665KB] ( 328 )
470 Abnormal Activity Detection Based on Dense Trajectory Alignment and Motion Influence Descriptor in Crowded Scenes
YANG Xingming, HU Jun
Aiming at the defects of existing anomaly activity detection algorithms in terms of target tracking and description in crowded scenes, an algorithm based on dense trajectory alignment and motion influence descriptor is proposed to capture the key information of motion of video objects. Firstly, dense trajectory guarantees a valid proposal of video motion object. Then, the dense trajectory-aligned motion influence descriptor is extracted along the trajectory direction. Finally, an overall framework is developed to detect both global and local abnormal activities accurately. Experiments on UCSD public dataset prove that the proposed method outperforms other methods.
2018 Vol. 31 (5): 470-476 [Abstract] ( 413 ) [HTML 1KB] [ PDF 1684KB] ( 288 )
477 Large Scene Dense 3D Reconstruction System Based on Semi-direct SLAM Method
XU Haonan, YU Lei, FEI Shumin
The 3D reconstruction system is mostly based on the simultaneous localization and mapping(SLAM) system of the feature point method and the direct method. The SLAM of feature point method cannot obtain good reconstruction results in the absence of feature points, while the SLAM of direct method has difficulty in estimating the pose with a fast-moving camera, and consequently, reconstruction results are unsatisfactory. To solve these problems, a dense 3D scene reconstruction system with a depth camera (RGB-D camera) based on semi-direct SLAM is proposed in this paper. The feature point method is exploited to estimate the camera pose in feature-rich areas. In the area of missing feature points, the direct method is utilized to estimate the pose of the camera. Then, the three-dimensional map is constructed by the optimized camera pose. The furfel model and the deformation map are utilized to estimate the pose of the point cloud and fuse point cloud. Finally, the ideal 3D reconstruction model is obtained. Experiments show that the system can be applied to all three-dimensional reconstruction of various occasions and acquire the ideal three-dimensional reconstruction model.
2018 Vol. 31 (5): 477-484 [Abstract] ( 646 ) [HTML 1KB] [ PDF 1052KB] ( 419 )
模式识别与人工智能
 

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