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....

2019 Vol.32 Issue.12, Published 2019-12-25 Papers and Reports    Researches and Applications        Papers and Reports 1061 Parallel Gastrointestine: An ACP-Based Approach for Intelligent Operations ZHANG Mei, CHEN Ling, WANG Fei-Yue, WANG Xiao, GUO Yuanyuan, YANG Tian The gastrointestinal system is an important organ for human to pick up energy from the foreign world. Causes of gastrointestinal diseases are multifactorial and complex. To develop intelligent and precise gastrointestinal diagnosis and excellent medical skills, a parallel gastrointestinal diagnosis system based on ACP theory is imposed in this paper. As the core of the parallel intelligence framework, the ACP theory consists of an artificial societies(A), computational experiments(C) and parallel execution(P). The artificial gastrointestinal systems are used to model the real complex diagnosis and treatment. The computational experiments are utilized to run various operations and evaluate the performance of results. Finally, the parallel execution is performed to constantly optimize the diagnosis schemes and realize virtual-real interaction guided diagnosis. With technologies of knowledge graph, deep learning, reality/augment reality and knowledge automation, the parallel gastrointestinal system is aimed to improve the accuracy and the efficiency of diagnosis and treatment, and contribute to a high level of national health. 2019 Vol. 32 (12): 1061-1071 [Abstract] ( 529 ) [HTML 1KB] [ PDF 1554KB] ( 778 ) 1072 Argumentation Mining Based on Multi-task Joint Learning LIAO Xiangwen, NI Jichang, WEI Jingjing, WU Yunbing, CHEN Guolong Most of the existing research on argumentation mining is focused on modeling single dataset, and the possible changes in feature of different datasets are neglected. And thus the generalization performance of the model is decreased. Therefore, an argumentation mining method based on multi-task learning is proposed to combine the argumentation mining tasks of multiple datasets for joint learning. Firstly, the input layers of multiple tasks are fused, and the sharing parameters of word level and character level are obtained via deep convolutional neural network and highway network. The joint task-related feature input into stacking long-short term memory is utilized to train the correlation information between multiple argumentation mining tasks in parallel. Finally, the results of sequence labeling are obtained by the conditional random field. The experimental results on six datasets of various fields verify the effectiveness of the proposed method with increased Macro-F1. 2019 Vol. 32 (12): 1072-1079 [Abstract] ( 466 ) [HTML 1KB] [ PDF 884KB] ( 369 ) 1080 Three-Way Decisions Method Based on Evaluations of Pythagorean Fuzzy Sets LIU Jiubing, WANG Tianxing, ZHOU Xianzhong, HUANG Bing, LI Huaxiong An optimization-based approach to determine the thresholds with Pythagorean fuzzy sets(PFSs) is proposed for threshold determination in three-way decisions(3WDs). Firstly, a pair of dual models from optimization angles are investigated, and it is proved that the dual models are equivalent to decision-theoretic rough sets models with the aid of the Karush-Kuhn-Tucker(KKT) condition. Next, the dual models are further generalized to the threshold determination of 3WDs with loss functions evaluated as PFSs, and a pair of nonlinear programming models are constructed based on nonlinear approaches for ranking PFSs. Meanwhile, the existence and the uniqueness of their optimal solution are proved and analyzed. Then, an optimization technique is exploited to solve these models, and a novel three-way decision approach under Pythagorean fuzzy evaluations is presented. Finally, an example and related comparison analysis indicate that the proposed method overcomes difficulties of the existing methods in determining the thresholds of Pythagorean fuzzy three-way decisions. 2019 Vol. 32 (12): 1080-1092 [Abstract] ( 351 ) [HTML 1KB] [ PDF 839KB] ( 384 ) 1093 Aspect Embedding on Memory Network for Aspect Sentiment Classification LIU Yiyi, ZHANG Jin, YU Zhihua, LIU Yue, CHENG Xueqi A comment contains multiple aspects and sentiments, and therefore it is difficult to classify sentiment polarity of different aspects correctly. A model combining aspect embedding with memory neural network is proposed to identify the sentiment of aspects in a comment. The aspect word vector is introduced into different modules of the memory network. The semantic information of the word is reinforced. The attention mechanism is guided to capture the relevant context, and thus the sentiment classification effect in the aspect is improved. Experimental results on short text English comments of SemEval 2014 Task 4 dataset and the long-text Chinese news dataset indicate that the proposed method achieves good classification effect and fully verifies the validity of the word embedding information introduced into the framework of the memory network. 2019 Vol. 32 (12): 1093-1099 [Abstract] ( 425 ) [HTML 1KB] [ PDF 821KB] ( 395 ) 1100 Event Temporal Relation Identification Based on Dependency and Textual Rhetoric Relation DAI Qianwen, ZHANG Longyin, KONG Fang In the identification of temporal relation between the existing events, only the local context of two events is taken into account and the relationship between the events from the perspective of discourse is neglected. To address this issue, a method to identify the temporal relation of events is proposed by combining the discourse rhetoric relation and intra-sentential dependency relation. The inter-event correlation is represented from two aspects, the shortest dependency path between events and the rhetorical relationship between the elementary discourse units of the events location. Based on this representation system, a neural network model is built to capture more effective information and improve the performance of event temporal relation identification. A series of experiments on Timebank-Dense corpus show the superiority of the proposed method. 2019 Vol. 32 (12): 1100-1106 [Abstract] ( 370 ) [HTML 1KB] [ PDF 634KB] ( 286 ) Researches and Applications 1107 Multi-partition Relaxed Alternating Direction Method of Multipliers for Regularized Extreme Learning Machine ZHANG Lijia, LAI Xiaoping, CAO Jiuwen To address the issue of overly heavy computational load of extreme learning machine(ELM) in the big data environment, parallel optimization for ELM is studied. A multi-partition relaxed alternating direction method of multipliers(ADMM) for regularized ELM along with two scalarwise implementations in the N- and N/2-equipartition cases is proposed. By the multi-partition, the proposed algorithm has a highly parallel structure and the combination with relaxation technique improves the convergence rate of the proposed algorithm. Through analysis, a necessary and sufficient convergence condition is established, and optimal convergence ratio and optimal parameters are obtained. Through simulations on bench-mark datasets, the relationship between the convergence ratio and the number of partitioned blocks is calculated, and convergence rates and GPU acceleration ratios of different algorithms are compared. Experimental results demonstrate that the proposed algorithm has low computational complexity and high parallelism. 2019 Vol. 32 (12): 1107-1115 [Abstract] ( 370 ) [HTML 1KB] [ PDF 771KB] ( 334 ) 1116 Group Activity Recognition Based on Regional Feature Fusion Network YANG Xingming, FAN Loumiao The existing group activity recognition methods cannot take full advantage of spatial information of the scene and the computational complexity of them is high. To solve these problems, a group activity recognition method based on regional feature fusion is proposed. Firstly, the convolution neural network is utilized to extract regional features of the scene, and then the regional features are split, arranged and combined into a series of regional feature sequences according to spatial position. Finally, long short term memory network is utilized to fuse regional feature sequences. Additionally, multilevel and multimodal strategies are adopted to improve the performance of the proposed method. Experiments on Collective and Volleyball datasets show that the proposed method achieves better performance. 2019 Vol. 32 (12): 1116-1121 [Abstract] ( 325 ) [HTML 1KB] [ PDF 519KB] ( 306 ) 1122 Land Cover Classification of Fully Polarimetric SAR with Encoder-Decoder Network and Conditional Random Field ZHAO Quanhua, XIE Kailang, WANG Guanghui, LI Yu Aiming at weak characterization of polarimetric synthesis aperture radar(PolSAR) image feature classification and low classification accuracy of the traditional fully convolutional network(FCN), a land cover classification algorithm of PolSAR with encoder-decoder network(E-D-Net) and conditional random field(CRF) is proposed. Firstly, Freeman decomposition and Pauli decomposition are employed to model PolSAR images and extract scattering features corresponding to each decomposition. Symmetric network model is built via semantic segmentation network model and multi-scale convolution unit. An E-D-Net network model is designed by embedding the multi-scale asymmetric convolution unit into the middle layer. PolSAR image and scattering features of Freeman decomposition are autonomously learned by E-D-Net to obtain the initial classification result. Finally, the CRF combined with Pauli coherent decompositionfalsecolorimageinformation isutilized todenoiseandsmooth the initialclassification results to obtain the final classification result. PolSAR image experiments on two areas verify the effectiveness and the feasibility of the proposed algorithm. 2019 Vol. 32 (12): 1122-1132 [Abstract] ( 305 ) [HTML 1KB] [ PDF 2770KB] ( 317 ) 1133 Image Matching Algorithm Combining Improved SURF Algorithm with Grid-Based Motion Statistics WANG Xiaohua, FANG Qi, WANG Wenjie To solve the problems of long operation time and low matching accuracy in the local invariant feature descriptor of speeded up robust features(SURF) algorithm, a image matching algorithm combining improved SURF algorithm with grid-based motion statistics(GMS) is proposed. Firstly, determinant of Hessian is utilized to determine the feature points of the image, and the main direction extraction method in SURF algorithm is improved by gradient direction to increase the accuracy of the main direction of the feature points. The binary feature descriptor rotation-aware binary robust independent elementary feature(rBRIEF) is employed to describe the feature points. Then, the feature points are roughly matched by Hamming distance. Finally, GMS is adopted to eliminate the mismatches. Experiment on Oxford VGG standard dataset indicates that the proposed algorithm achieves higher matching accuracy and efficiency with image changes in scale, illumination and rotation. 2019 Vol. 32 (12): 1133-1140 [Abstract] ( 392 ) [HTML 1KB] [ PDF 1960KB] ( 748 ) 1141 Rough Fuzzy K-means Clustering Algorithm Based on Mixed Metrics and Cluster Adaptive Adjustment ZHANG Xintao, MA Fumin, CAO Jie, ZHANG Tengfei Rough K-means clustering and its related derivative algorithms need the number of clusters in advance, and random selection of the initial cluster center results in low accuracy of data partition in the cross-region of clusters. To solve these problems, a rough fuzzy K-means clustering algorithm with adaptive adjustment of clusters is proposed. When the membership degree of the data objects belonging to different clusters in the intersection area of the cluster boundary is calculated, the mixed metrics of local density and distance are taken into account in the proposed algorithm. The optimal number of clusters is gained by adjusting the number of clusters adaptively. The midpoint of two samples with the smallest distance in the dense area of data objects is selected as the initial cluster center. The object with the local density higher than the average density is divided into the cluster, and then the re-maining initial cluster center can be selected. Thus, the selection of the initial cluster centers is more reasonable. The experiments on synthetic datasets and UCI datasets demonstrate the advantages of the proposed algorithm in adaptability and clustering accuracy for dealing with spherical clusters with blurred boundaries. 2019 Vol. 32 (12): 1141-1150 [Abstract] ( 371 ) [HTML 1KB] [ PDF 879KB] ( 317 )

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