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....
 
 
2019 Vol.32 Issue.1, Published 2019-01-25
   
0
2019 Vol. 32 (1): 0-0 [Abstract] ( 366 ) [HTML 1KB] [ PDF 311KB] ( 331 )
1 Target-Directed Locomotion of a Snake-Like Robot Based on Path Integral Reinforcement Learning
FANG Yongchun, ZHU Wei, GUO Xian
Path integral is derived from stochastic optimal control. It is a numerical iteration method and solves the problem of the optimal control about continuous nonlinear systems at a high convergence speed without system model. A policy improvement algorithm based on path integral reinforcement learning is proposed for the target-directed locomotion of a snake-like robot in this paper. The path integral reinforcement learning approach is employed to learn the parameters of the snake-like robot serpentine equation, and the robot is controlled to arrive at the target position fast without contacting obstacles in simulation environment. Moreover, the robot with the priori knowledge from the simulation in real environment can complete the task well. Experimental result verifies the validity of the propose algorithm.
2019 Vol. 32 (1): 1-9 [Abstract] ( 895 ) [HTML 1KB] [ PDF 1653KB] ( 488 )
10 Multi-task Learning Based Face Analysis for Machine Bionic Eyes
FAN Di, Hyunwoo Kim, CHEN Xiaopeng, LIU Yunhui, HUANG Qiang
The performance of human-machine interaction is crucial for intelligence robot, and face analysis makes human-machine interaction more friendly. In this paper, a multi-task learning convolutional neural network is proposed. The tasks of smile recognition and gender classification are solved simultaneously. Inherent correlated tasks are learned, and the performance of individual task is improved. On CelebA test dataset, the proposed network achieves high accuracy on a smile recognition task and a gender classification task. The proposed model is tested on the designed machine bionic vision eyes, achieving satisfactory result on smile recognition and gender classification. The research on face analysis in this paper improves the human-machine interaction ability with the machine bionic eyes.
2019 Vol. 32 (1): 10-16 [Abstract] ( 497 ) [HTML 1KB] [ PDF 1187KB] ( 422 )
17 Eye Center Localization Based on Improved SVR
ZHANG Wanqi, WANG Zhiyong, LIU Honghai
The accuracy of eye center localization is reduced due to the low resolution of input image, poor lighting condition, side face and eyes status. To handle this issue, an improved support vector regression (SVR) method is proposed to detect the eye center based on the facial feature localization. Several image processing techniques are tried to improve the accuracy. Results show that the SVR combining a Gaussian filter achieves a better accuracy.
2019 Vol. 32 (1): 17-23 [Abstract] ( 458 ) [HTML 1KB] [ PDF 2000KB] ( 202 )
24 Robot Vision under Complex Weather Conditions
TIAN Jiandong, LIU Lianqing
This paper systematically introduces the research achievements of modeling and removal algorithms for rain, snow and fog in recent years from authors' team. It includes a depth estimation and scattering removal algorithm based on near-filed illumination, a fog removal algorithm based on far parallel illumination and region optimization, and a snowflake removal algorithm based on low-rank decomposition as well as a raindrop & snowflake removal algorithm based on matrix decomposition.
2019 Vol. 32 (1): 24-35 [Abstract] ( 465 ) [HTML 1KB] [ PDF 2684KB] ( 448 )
36 Guidewire Tracking Based on Regional Proposal Network and Residual Structure
LIU Shiqi, SUN Xiaobo, XIE Xiaoliang, HOU Zengguang
X-ray image navigation is a breakthrough of improving the accuracy and safety of robotic interventional surgery. A method based on region proposal network, residual structure and canny edge detection is proposed in this paper. It is specifically designed for guidewire segmentation framework. In the image calibration, multi-scale marking strategies are adopted to enable detection networks to learn accurate features. In the image augmentation, a multi-filter fusion strategy is employed to increase the recognizability of the guidewire and improve the tracking accuracy and the system robustness. The experiment is conducted on 22 sets of X-ray video sequences. Experimental results demonstrate the superiority of the proposed algorithm in terms of speed, accuracy and robustness.
2019 Vol. 32 (1): 36-42 [Abstract] ( 343 ) [HTML 1KB] [ PDF 924KB] ( 221 )
43 Real-Time Updatable Globally Consistent 3D Grid Mapping
YI Xiaodong, YANG Sining, YANG Shaowu

A real time globally consistent three-dimensional(3D) grid mapping method is usually required for autonomous navigation of mobile robots in complex unknown 3D environments. Grid maps built by the existing simultaneous localization and mapping(SLAM) system are inconsistent with environments due to the lack of updating strategy. In this paper, information of environment provided by SLAM module are processed by the grid mapping module. A real-time updating strategy and an efficient data structure based on keyframe are proposed to produce globally consistent 3D maps and they are suitable for real time navigation of robots. Experimental results in dynamic indoor scenarios demonstrate that the 3D mapping method can update map in real time and build globally consistent 3D grid map to support the autonomous navigation.
2019 Vol. 32 (1): 43-50 [Abstract] ( 535 ) [HTML 1KB] [ PDF 3582KB] ( 321 )
51 Least p-Norm Based Broad Learning System
ZHENG Yunfei, CHEN Badong
Based on the broad learning system(BLS), a least p-norm based BLS(LP-BLS) is proposed, and it takes the p-norm of error vector as loss function and combines the fixed-point iteration strategy. With the proposed LP-BLS, the interferences from different noises can be well dealt with by flexibly setting the value of p(p≥1), so that the modeling task of unknown data can be better completed. Numerical experiments show that the good performance of the proposed method can always be maintained with Gaussian noise, uniform noise and impulse noise. Finally, the system is applied to electroencephalogram(EEG) classification task and achieves a higher classification accuracy on most subjects.
2019 Vol. 32 (1): 51-57 [Abstract] ( 437 ) [HTML 1KB] [ PDF 632KB] ( 447 )
58 An Optimal Projection Plane-Based Spatial Circle Measurement Method Using Stereo Vision System
LI Zhengyuan, MA Xin, LI Yibin
The accuracy of the existing measurement method based on binocular stereo vision depends on the accuracy of calibration, and the accuracy of measurement decreases when the spatial circle is occluded. Firstly, the reconstruction accuracy of point affected by the stereo matching error of points on projection curves is analyzed in the presence of external parameter error of the binocular stereo vision system. Then, based on the conclusion of error analysis, a new method for measuring the position and orientation of spatial circle is designed. Using the edge points selecting algorithm, the points of projection curves are selected. The circle is then reconstructed using the points with small stereo matching error. The projection of the reconstructive points on the optimal projection plane based on nonlinear optimization in the direction depth is utilized for fitting spatial circle to obtain the position and orientation. Experimental results prove the effectiveness of the proposed algorithm.
2019 Vol. 32 (1): 58-66 [Abstract] ( 452 ) [HTML 1KB] [ PDF 923KB] ( 229 )
67 A Review of Deep Reinforcement Learning Theory and Application
WAN Lipeng, LAN Xuguang, ZHANG Hanbo, ZHENG Nanning
Deep reinforcement learning(DRL) theory and applied research are deepening and it is now playing an important role in games, robot control, dialogue systems, automatic driving, etc. Meanwhile, due to shortcomings such as exploration-exploitation dilemma, sparse reward, sample collection hardness, poor model stability, DRL still has many problems for which researchers have proposed various solutions. New theories has further promoted the development of DRL, and opened up several new research fields of reinforcement learning, such as imitative learning, hierarchical reinforcement learning and meta-learning. This paper aims to explore and summarize future development of DRL, and a brief introduction of DRL theory, difficulties and applications is presented at the same time.
2019 Vol. 32 (1): 67-81 [Abstract] ( 1138 ) [HTML 1KB] [ PDF 1230KB] ( 1727 )
82 Salient Object Detection Based on Manifold Ranking and Co-connectivity
WANG Yanzhao, PENG Guohua, YAN Weidong
To improve the performance of saliency detection, a bottom-up saliency object detection model is proposed by integrating different features based manifold ranking and co-connectivity. Aiming at the calculation on edge and connection bewteen nodes of the graph in most manifold ranking based models, a manifold ranking based salient map is produced via different features to calculate the weight of edges and modified connection to construct the graph. Simultaneously, the co-connectivity based salient map is obtained by incorporating boundary connectivity and foreground connectivity. The final saliency map is achieved through fusing these two salient results with different scales. Compared with 16 state-of-the-art methods on four public benchmark datasets, the proposed algorithm performs better.
2019 Vol. 32 (1): 82-93 [Abstract] ( 429 ) [HTML 1KB] [ PDF 2351KB] ( 339 )
94
2019 Vol. 32 (1): 94-95 [Abstract] ( 264 ) [HTML 1KB] [ PDF 150KB] ( 292 )
96
2019 Vol. 32 (1): 96-96 [Abstract] ( 198 ) [HTML 1KB] [ PDF 145KB] ( 167 )
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