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....

2020 Vol.33 Issue.6, Published 2020-06-25 Papers and Reports    Researches and Applications    Surveys and Reviews        Papers and Reports 479 Aspect Level Sentiment Analysis Based on Knowledge Graph and Recurrent Attention Network DENG Liming, WEI Jingjing, WU Yunbing, YU Xiaoyan, LIAO Xiangwen The existing aspect level sentiment analysis methods cannot solve the problem of polysemous word in different contexts. Therefore, a method for aspect level sentiment analysis based on knowledge graph and recurrent attention network is proposed. The text representation of the bidirectional long short-term memory network is integrated with synonym information in knowledge graph using dynamic attention mechanism to obtain the state vector of knowledge perception. To classify aspect level sentiment, the memory content is constructed by combining the location information and inputting the multi-level gated recurrent unit for calculating the sentiment characteristics of aspect terms. The experimental results show that the proposed method achieves better classification results on three open datasets. 2020 Vol. 33 (6): 479-487 [Abstract] ( 840 ) [HTML 1KB] [ PDF 852KB] ( 613 ) 488 Multi-graph Kernel Based Transfer Learning Method JIANG You, ZHANG Daoqiang, ZHANG Junyi Labeling medical data is costly and there are differences in the distributions of the neuroimaging data provided by different research centers. Therefore, it is nearly impossible to improve the diagnosis results by integrating the data. A multi graph-kernel based transfer learning method is proposed to tackle with this problem. Several different graph kernels are employed to mine structure information from brain network data and measure the similarity between brain networks. Then, the performance of transfer learning model is improved by a proposed multi-kernel learning framework. Experiments on the multi-center dataset of autistic spectrum disorder(ASD) indicate the classification performance on brain network data is improved and the advantages of different graph kernels are efficiently utilized by multi-kernel learning framework. 2020 Vol. 33 (6): 488-495 [Abstract] ( 409 ) [HTML 1KB] [ PDF 713KB] ( 383 ) 496 Salient Object Detection Based on Deep Center-Surround Pyramid CHEN Qin, ZHU Lei, HOU Yunlong, DENG Huiping, WU Jin Center-surround based contrast calculation is rarely applied in deep learning-based algorithms. Therefore, a salient object detection method based on deep center-surround pyramid is proposed. Center-surround based contrast and convolutional neural network are combined for salient object detection. Firstly, deep semantic features are introduced into each stage of the network. Then, the dilated convolution is employed to build the center-surround pyramids to capture the contrast information of different scales and generate the corresponding multi-scale conspicuous maps. Finally, all conspicuous maps are further fused to produce final salient object detection result. Comparative experiments on four public datasets verify that the proposed algorithm achieves lower mean average error and higher F measure. 2020 Vol. 33 (6): 496-506 [Abstract] ( 627 ) [HTML 1KB] [ PDF 1281KB] ( 454 ) 507 Feature Pyramid Object Detection Network Based on Function Maintenance XU Chengqi, HONG Xuehai To solve the problem of feature pyramid network in multi-scale and small object detection, a feature pyramid object detection network based on function maintenance is proposed. Firstly, feature maps are selected in the backbone convolutional architecture to build feature pyramid. For these feature maps of different scales, feature fusion with low loss is carried out from top to bottom using function maintenance fusion module. The strong high-level semantic information is maintained more effectively, and the representation ability for small object of low-level feature maps is greatly enhanced. The detection precision is improved by two-stage features of the proposed network to describe the objects. Finally, context information is fully utilized to further enhance the ability to distinguish multi-scale object. Experiments on PASCAL VOC public dataset show that the detection result of the proposed network is satisfactory. Moreover, the proposed network achieves better results in the case of object occlusion and blur as well. 2020 Vol. 33 (6): 507-517 [Abstract] ( 414 ) [HTML 1KB] [ PDF 2523KB] ( 579 ) 518 Label Noise Filtering via Perception of Nearest Neighbors JIANG Gaoxia, FAN Ruixuan, WANG Wenjian Label noise filtering algorithms based on k nearest neighbor are sensitive to the neighbor parameter k. Aiming at this problem, a label noise filtering algorithm based on perception of nearest neighbors(PNN) is proposed to solve the problem of intra-class label noise in binary classification datasets effectively. Positive and negative samples are considered separately in PNN, and thus the label noise detection problem in classification is transformed into two outlier detection problems with single-class data. Firstly, the personalized neighbor parameter is determined automatically by the neighbor perception strategy to avoid the sensitivity of neighbor parameter. Secondly, all samples are divided into core samples and non-core samples by noise factor. The non-core samples are taken as the candidates of label noise. Finally, the noise is identified and filtered by combining the label information of the nearest neighbors of the candidate samples. Experiments indicate that the proposed algorithm performs well in noise filtering and classification prediction. 2020 Vol. 33 (6): 518-529 [Abstract] ( 448 ) [HTML 1KB] [ PDF 763KB] ( 367 ) Surveys and Reviews 530 An Overview of Natural Language Processing for Indonesian and Malay JIANG Shengyi, LI Shanshan, FU Sihui, LIN Nankai As the penetration rate of Indonesian and Malay rises, it is significant to carry out information processing on massive texts of these two languages. Extensive research is conducted on Indonesian and Malay. However, as low-resource languages, Indonesian and Malay draw less attention than common languages. Thus, the deep learning methods cannot be fully utilized. In this paper, research on Indonesian and Malay morphological analysis, syntactic parsing, machine translation, spelling check etc., is analyzed and summarized. In the most research findings, algorithms cannot be compared objectively due to their different corpus scales and evaluation metrics. Finally, problems and future directions of natural language processing on Indonesian and Malay are discussed with the consideration of the existing open language resources in various fields. 2020 Vol. 33 (6): 530-541 [Abstract] ( 670 ) [HTML 1KB] [ PDF 806KB] ( 1626 ) Researches and Applications 542 Deep Hamming Embedding Based Hashing for Image Retrieval LIN Jiwen, LIU Huawen, ZHENG Zhonglong The image features learned by deep convolutional neural networks have an obvious hierarchical structure. As the number of layers deepens, the learned features become more and more abstract and the discrimination of classes is gradually enhanced. Based on the above, deep hamming embedding based hashing for image retrieval is proposed. A hidden layer is inserted at the end of the deep convolutional neural network and then hash codes are obtained by the activation of each unit of the layer. According to the characteristics of hash codes, hamming embedding loss is proposed to preserve the similarity between the original data better. Experiments on commonly used benchmark image datasets CIFAR-10 and NUS-WIDE indicate that the proposed model improves image retrieval performance and performs better with short encoding length. 2020 Vol. 33 (6): 542-550 [Abstract] ( 426 ) [HTML 1KB] [ PDF 630KB] ( 676 ) 551 Automatic Short Text Summarization Based on Part-of-Speech Soft Template Attention Mechanism ZHANG Yafei, ZUO Yixi, YU Zhengtao, GUO Junjun, GAO Shengxiang Semantic integrity of the summary with intuitive subject-predicate-object structure is strong in the short text summarization task. However, part-of-speech combinations impose constraints on the structure. Aiming at this problem, an automatic short text summarization method based on part-of-speech soft template attention mechanism is proposed. Firstly, text is tagged with part-of-speech tags, and the tagged part-of-speech sequence is regarded as part-of-speech soft template of the text to guide the method to construct the structural specifications of a summary. Part-of-speech soft template is characterized at the encoder. Then, the part-of-speech soft template attention mechanism is introduced to enhance the attention of core part-of-speech information in text, such as nouns and verbs. Finally, the part-of-speech soft template attention and traditional attention are combined to generate a summary at the decoder. Experimental results verify the effectiveness of the proposed method on short text summarization datasets. 2020 Vol. 33 (6): 551-558 [Abstract] ( 384 ) [HTML 1KB] [ PDF 754KB] ( 337 ) 559 Stochastic Gradient Descent Method of Convolutional Neural Network Using Fractional-Order Momentum KAN Tao, GAO Zhe, YANG Chuang The stochastic gradient descent method may converge to a local optimum. Aiming at this problem, a stochastic gradient descent method of convolutional neural network using fractional-order momentum is proposed to improve recognition accuracy and learning convergence rate of convolution neural networks. By combining the traditional momentum-based stochastic gradient descent method with fractional-order difference method, the parameter updating method is improved. The influence of fractional-order on the training result of network parameters is discussed, and an order adjustment method is produced. The validity of the proposed parameters training method is verified and analyzed on MNIST dataset and CIFAR-10 dataset. The experimental results show that the proposed method improves the recognition accuracy and learning convergence rate of convolutional neural networks. 2020 Vol. 33 (6): 559-567 [Abstract] ( 688 ) [HTML 1KB] [ PDF 697KB] ( 635 ) 568 Occluded Pedestrian Detection Algorithm Based on Improved Network Structure of YOLOv3 LIU Li, ZHENG Yang, FU Dongmei Aiming at high missed detection rates of YOLOv3 for occluded pedestrian in surveillance video, a detection method for occluded pedestrian based on improved network structure of YOLOv3 is proposed. Firstly, the spatial pyramid pooling network is introduced into the fully connected layer to enhance the multi-scale feature fusion capability of the network. Secondly, the network structure pruning is employed to eliminate the network structure redundancy to avoid network degeneration and overfitting problem caused by the deepening of network layers and reduce the amount of parameters. Multi-scale training is performed on the corridor pedestrian dataset to obtain the best weight model. Experimental results indicate the improvement of average accuracy and detection speed of the proposed algorithm. 2020 Vol. 33 (6): 568-574 [Abstract] ( 911 ) [HTML 1KB] [ PDF 1502KB] ( 619 )

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