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.2, Published 2019-02-25
Papers and Reports    Researches and Applications   
   
Papers and Reports
97 Deep Human Pose Estimation Method Based on Mixture Articulated Limb Model
LIU Binghan, LI Zhenda, KE Xiao
A flexible mixture model is proposed to solve the problems of human pose estimation. The model is composed of joint appearance and inner-joint relationship models, and it is trained through a deep convolutional neural network (DCNN). Firstly, a graphical model is constructed to represent joints and limbs of human body. Secondly, images are decomposed into several image blocks centered on the joints and used as training input data. Finally, a multiple classification DCNN network is obtained to perform human pose estimation.The proposed method is more flexible for human body representation, and the detection rate of joint points and the correct detection rate are effectively improved.
2019 Vol. 32 (2): 97-107 [Abstract] ( 689 ) [HTML 1KB] [ PDF 1407KB] ( 431 )
108 Self-adaptive Ejector Particle Swarm Optimization Algorithm
ZHU Jingwei, FANG Husheng, SHAO Faming, JIANG Chengming
Particle swarm optimization(PSO) is easily trapped in local optimum and stagnation, and therefore a self-adaptive ejector particle swarm optimization algorithm(SAEPSO) is proposed. To keep the vitality of the particle swarm, the ejector operation is introduced into the algorithm. While the satisfying the condition, the particle is given a high speed at the current position to fly to a faraway area. Full-dimensional ejection and probabilistic ejection can be selected for the ejection mode. To cope with the ejector operation, a new quality judgment for particles is proposed, so particles can be ejected out of the feasible region. A self-adaptive discrimination function is introduced in the proposed algorithm to judge whether the particle should be ejected. While satisfying the function, the particles are ejected. Numerical experiments show that the proposed algorithm possesses relatively strong global search ability and fast search speed.
2019 Vol. 32 (2): 108-116 [Abstract] ( 460 ) [HTML 1KB] [ PDF 1048KB] ( 437 )
117 Spatial-Temporal Texture Cascaded Feature Method for Face Liveness Detection
GAN Junying, ZHAI Yikui, XIANG Li, CAO He, HE Guohui, ZENG Junying, TAN Haiying, DENG Wenbo
To solve the security problem in identity authentication, the face liveness detection method is always employed. Therefore, a spatial-temporal texture cascaded feature method is proposed to improve the robustness of living face detection. Firstly, local binary pattern(LBP) is utilized to calculate the differential excitation of Weber local descriptor(WLD), and Prewitt operator is exploited to calculate the directional angle of WLD to extract texture features in time domain and space domain. Secondly, the histogram of texture features obtained from three orthogonal space-time planes, XY, XT and YT, is cascaded. Finally, the dynamic texture features, namely spatial-temporal texture cascade features, can be used to determine whether the real face or the disguised face. Experimental results on CASIA face anti-spoofing database and replay-attack database show that the proposed method obtains higher recognition rate than the existing mainstream local texture feature methods and it can be widely used in identity authentication and security monitoring systems.
2019 Vol. 32 (2): 117-123 [Abstract] ( 534 ) [HTML 1KB] [ PDF 1155KB] ( 445 )
124 Decision-Making Evaluation for Collaborative Alliance Based on Cloud Model Theory Orienting to Big Data
YIN Lei, JIANG Jianguo, ZHANG Guofu
Aiming at the strong uncertainty of decision-making evaluation for alliance, a multi-level decision-making evaluation method of multi-task collaborative alliance based on cloud model theory orienting to big data is proposed. Firstly, a decision-making evaluation framework for collaborative alliance on the basis of big data is established, and the evaluation data of basic evaluation indexes of alliance members are obtained from the processing and analysis platform of big data. The reverse cloud generator algorithm is applied to create the corresponding evaluation cloud. Meanwhile, the cloud characteristic parameters of alliance evaluation indexes are generated using integrated cloud computing. Then, combining the evaluation index weight and task weight of alliance, the decision-making evaluation cloud of single-task alliance and multi-task collaborative alliance are gained by applying cloud weighted arithmetic averaging operator on the basis of cloud clustering algorithm, respectively. Next, the alternative schemes of multi-task collaborative alliance are evaluated and selected to determine the optimal one. Finally, by comparing with the traditional alliance evaluation method based on the D-S evidence theory, the effectiveness of the proposed method is verified.
2019 Vol. 32 (2): 124-132 [Abstract] ( 349 ) [HTML 1KB] [ PDF 827KB] ( 318 )
133 Entity Relations Extraction in Chinese Domain Based on Distant Supervision with Multi-feature Fusion
WANG Bin, GUO Jianyi, XIAN Yantuan, WANG Hongbin, YU Zhengtao
Aiming at the extraction of Chinese domain entity relationship from unlabeled text, a hybrid method of domain entity attribute extraction based on distant supervision is proposed. The structured relational three tuples in the knowledge base are applied to obtain the training corpus automatically from the natural language text. Due to the large amount of noise in the annotation data of distant supervision method, the latent Dirichlet allocation(LDA) topic model for topic keyword extraction is adopted, and then the similarity calculation with relationship type and keyword pattern matching for denoising are performed. Finally, the part-of-speech feature, the dependency feature and the phrase syntax tree feature are extracted, and the relationship extraction model is trained. Experiments show that the method fusing three features produces higher F value and better extraction performance.
2019 Vol. 32 (2): 133-143 [Abstract] ( 627 ) [HTML 1KB] [ PDF 974KB] ( 402 )
Researches and Applications
144 Adaptive Weighted Online Extreme Learning Machine for Imbalance Data Steam
MEI Ying, LU Chengbo
It is problematic to classify data stream with imblanced class distributions for general online learning algorithms, especially in case of concept drift. In this paper, an adaptive weighted online extreme learning machine(AWO-ELM) is developed for imbalance data stream. AWO-ELM is an online learning method and it alleviates the class imbalance problem in chunk-by-chunk learning. Instead of adopting fixed weights, an efficient weight selection strategy is proposed to obtain better classification performance, and thus it can be applied to the task of learning static data stream with different imbalance ratio and the task of online learning with concept drift. The theoretical analysis and experimental results of several real data stream show that AWO-ELM obtains comparable or better classification performance than competing methods.
2019 Vol. 32 (2): 144-150 [Abstract] ( 461 ) [HTML 1KB] [ PDF 639KB] ( 383 )
151 Automatic Selection Method of Cluster Center Based on Positive Sequence Iterative Selection Strategy
WANG Wanliang, LÜ Chuang, ZHAO Yanwei, GAO Nan, YANG Xiaohan, ZHANG Zhaojuan
The decision function of density peak clustering algorithm cannot determine the clustering center automatically and effectively. Therefore, a density peak clustering algorithm, automatically clustering by fast search and find of density peaks(AUTO-CFSFDP), is proposed. Firstly, the normalization process is carried out to make the uneven distribution of variables in the decision function become uniform. Secondly, the selection strategy based on positive-sequence iteration is presented to search elbow point according to the variation trend of the number of cluster core points in the process of determining the cluster center. A set of points before the elbow point is used as the cluster centers to complete clustering. Finally, the performance of AUTO-CFSFDP is evaluated on UCI datasets. AUTO-CFSFDP can cluster the datasets of arbitrary distributions without extra time consumption. The adaptability and clustering results are improved effectively.
2019 Vol. 32 (2): 151-160 [Abstract] ( 419 ) [HTML 1KB] [ PDF 3748KB] ( 371 )
161 Chinese Multi-paragraph Reading Comprehension Model
ZHAO Junyao, PANG Liang, SU Lixin, LAN Yanyan, GUO Jiafeng, CHENG Xueqi

In the Chinese multi-paragraph reading comprehension task, three properties should be taken into account: the sparsity of evidence paragraph, the diversity of Chinese semantic and the validity of answer snippet. To solve these problems, a Chinese multi-paragraph reading comprehension model, CMPReader, is proposed. In CMReader, data augmentation is exploited to learn the paragraphs with no answer. Word level encoding and Chinese word tag are added to enrich the Chinese semantic representation, and the features of answer snippet are employed by the answer verifier model to choose the right answer. CMPReader is applied to the CIPS-SOGOU factoid question answer dataset, and the results show that the average of exact match score and F1 score are increased.
2019 Vol. 32 (2): 161-168 [Abstract] ( 460 ) [HTML 1KB] [ PDF 981KB] ( 452 )
169 CapsNet-Based Chinese Character Font Representation Model
XIE Haiwen, YE Dongyi, CHEN Zhaojiong
A CapsNet-based Chinese character font representation model is proposed to represent Chinese character font by the representation of components. Firstly, representative vectors of all categories are generated by the model. Then, a group of component representative vectors are selected by the Euclidean-distance-based outlier detection according to component probabilities. Finally, these vectors are utilized to form the Chinese character font representations. The experimental results show that the proposed model, merely trained on component fonts, is capable of identifying components of Chinese characters and automatically generating effective representation of Chinese characters.
2019 Vol. 32 (2): 169-176 [Abstract] ( 694 ) [HTML 1KB] [ PDF 913KB] ( 448 )
177 Bacteria Biotope Relation Extraction Based on a Fusion Neural Network
LI Mengying, WANG Jian, WANG Yan, LIN Hongfei, YANG Zhihao
To build a complete bacteria biotope relation database, a relation extraction system based on a convolutional neural network(CNN)-long short-term memory(LSTM) model is proposed. Combining CNN and LSTM, the deep learning of hidden features are realized, and the distributed word vector feature and entity position feature are extracted as feature input of the model.Comparative experiments verify the advantages of CNN-LSTM model after the addition of features.The feature output of the CNN model is taken as the feature input of the LSTM model, and the best result is obtained on the BB-event corpus published by the Bio-NLP 2016 shared task.
2019 Vol. 32 (2): 177-183 [Abstract] ( 418 ) [HTML 1KB] [ PDF 805KB] ( 424 )
184 Gated Dynamic Attention Mechanism towards Aspect Extraction
CHENG Meng, HONG Yu, TANG Jian, ZHANG Jiashuo, ZOU Bowei, YAO Jianmin
In the current aspect extraction researches, the attention modeling and training are fixed, and the sentence is modeled in one time step. However, the semantics of the words vary in contexts, and a fixed attention distribution lacks dynamic adaptability. Therefore, a gated dynamic attention mechanism towards aspect extraction is proposed in this paper. A bidirectional long short term memory network is exploited to obtain hidden representations of words in a target sentence. Then, a specific attention distribution is computed according to the target word and its context while the attention model labelling words. Thus, the attention-weight distribution can be automatically adjusted according to the changes of contexts. Next, a gate is adopted to adjust the quantities of information flowing to the next units. Finally, conditional random field is utilized to label the aspect. The official datasets of 2014-2016 semantic evaluation are employed to verify the effectiveness of the proposed method, and F1 scores are increased.
2019 Vol. 32 (2): 184-192 [Abstract] ( 536 ) [HTML 1KB] [ PDF 799KB] ( 321 )
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