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....

2021 Vol.34 Issue.8, Published 2021-08-25 Papers and Reports    Researches and Applications        Papers and Reports 677 Listwise Adversarial Domain Adaption Algorithm for Image Cropping WANG Haowen, SANG Nong Image cropping is short of training data for its high threshold for annotation. Current research on image cropping is confined on public datasets. Grounded on domain shift between training domain and practical application scene, a listwise adversarial domain adaption algorithm for image cropping is proposed in this paper. Firstly, the domain shift between two image cropping datasets, GAICD and CPC, is proved. Then, an image cropping model composed of an aesthetic evaluation module and an adversarial domain adaptation module is constructed. Aesthetic evaluation module is employed to predict the aesthetic score of current image and assist the model to extract the invariant features for cropping task. Adversarial domain adaptation module is exploited to realize adversarial based domain adaptation learning. Domain migration experiments between different cropping datasets and between different scene domains verify the effectiveness of proposed algorithm. 2021 Vol. 34 (8): 677-688 [Abstract] ( 509 ) [HTML 1KB] [ PDF 1459KB] ( 346 ) 689 Local Optimal Scale Combination Selections in Inconsistent Generalized Multi-scale Decision Systems WU Weizhi, SUN Yu, WANG Xia, ZHENG Jiawen To investigate knowledge acquisition for objects in inconsistent generalized multi-scale decision systems, the concept of local optimal scale combination is presented. Firstly, the notion of scale combination in a generalized multi-scale information system is introduced. Information granules with different scale combinations as well as their relationships from generalized multi-scale information systems are formulated. Lower and upper approximations of sets with different scale combinations in generalized multi-scale information systems are further constructed and their properties are examined. Finally, concepts of seven types of local optimal scale combinations for an object in an inconsistent generalized multi-scale decision system are defined and their relationships are clarified. It is proved that there are five different types of local optimal scale combinations in fact. 2021 Vol. 34 (8): 689-700 [Abstract] ( 365 ) [HTML 1KB] [ PDF 693KB] ( 274 ) 701 Pessimistic Concept Lattices and Optimistic Concept Lattices Based on Attribute Classification GAO Le, WANG Zhen, WEI Ling, QI Jianjun In the formal context with attribute classification, the pessimistic classified formal context and optimistic classified formal context are firstly proposed, the operators and concepts in these contexts are defined, and the relationships between operators and concepts in classified contexts and those in the original contexts are studied. Then, for the pessimistic classified formal context, the mapping between the original concept lattice and the pessimistic classified concept lattice is established, and the generation method of the pessimistic classified concept lattice from the original concept lattice is presented. For the optimistic classified formal context, the relationships between the original concept lattice and the optimistic classified concept lattice are studied by introducing the concept inclusion mapping, and the corresponding generation method of the concept lattice is presented as well. Finally, the practical application and the semantic interpretation of both pessimistic and optimistic classified concept lattices are discussed. 2021 Vol. 34 (8): 701-711 [Abstract] ( 341 ) [HTML 1KB] [ PDF 737KB] ( 270 ) 712 Collaborative Filtering with Heterogeneous Neighborhood Aggregation XIA Hongbin, LU Wei, LIU Yuan In traditional collaborative filtering models, the feature vector generated by one-hot encoding is sparsely informative. Heterogeneous behavior data is only employed to describe the relationship between different behaviors and the relationship between behaviors of different users is ignored.Aiming at these problems, an algorithm of collaborative filtering with heterogeneous neighborhood aggregation is proposed. Firstly, the heterogeneous interaction between users and items is modeled by the graph, and neighborhoods are built through the connectivity of graph. Then, the neighborhood information integrated by the lightweight graph convolution method is merged into the feature vectors of the target users and items. Finally, the feature vectors of users and items integrating with neighborhood information are input into a multi-task heterogeneous network for training. The problem of data sparseness is alleviated by enriching the hidden information of feature vectors. Experiments on the datasets prove that the performance of the proposed model is better. 2021 Vol. 34 (8): 712-722 [Abstract] ( 393 ) [HTML 1KB] [ PDF 757KB] ( 399 ) 723 Construction Algorithm of Concept Set Based on Simulated Annealing Algorithm LIU Zhonghui, CHEN Jianyu, SONG Guojie, MIN Fan In formal concept analysis, the construction of concept lattice produces high time and space complexity, but only partial lattices or concept sets are applied in recommendation. To solve this problem, a construction algorithm of concept set based on simulated annealing algorithm is proposed. The candidate concepts generation technique is presented based on the simulated annealing algorithm. The objective function takes the extension similarity of a concept into account. The Metropolis criterion is employed to update the solution. The concept filtering technique is designed based on all candidate concepts. Strong concepts of each user are selected with the extension similarity as the evaluation indicator, and the filtered strong concepts constitute a concept set. The recommendation technique is proposed based on the strong concept set. It provides personalized recommendations to the target user using the preferences of neighbor users in the same extension. Experimental results on 5 public datasets demonstrate that the recommendation performance and the efficiency of proposed algorithm are superior. 2021 Vol. 34 (8): 723-732 [Abstract] ( 335 ) [HTML 1KB] [ PDF 791KB] ( 277 ) Researches and Applications 733 Chinese Medical Question Answering Matching Method Based on Knowledge Graph and Keyword Attention Mechanism QIAO Kai, CHEN Kejia, Chen Jingqiang Due to the lack of high-quality question and answer data in Chinese medical field, a Chinese medical question answering matching method combining knowledge graph and keyword attention mechanism is proposed. Firstly, the medical knowledge graph is introduced into the bidirectional encoder representation from transformers(BERT) model to obtain knowledge-enhanced sentence features, and a keyword attention mechanism is employed to emphasize the interaction between question and answer sentences. The experimental results on two open Chinese medical question-answer datasets, cMedQA and webMedQA , show that the proposed model is obviously better , especially for the small amount of samples. The ablation experiment also verifies that each of the new modules improve the performance of BERT to a certain extent. 2021 Vol. 34 (8): 733-741 [Abstract] ( 592 ) [HTML 1KB] [ PDF 705KB] ( 542 ) 742 Specular Reflection Separation with Hue Constraint ZHANG Zhen, REN Weihong, TIAN Jiandong, TANG Yandong The hue information in color images is not easily susceptible to inference from specular reflection. Grounded on this fact, an effective algorithm of specular reflection separation with hue constraint is proposed. Images are clustered by its hue information. The fusion coefficient of diffuse reflection and specular reflection is obtained by calculating the pixel chromaticity and the illumination chromaticity. The bilateral filter is conducted on fusion coefficients to eliminate the noise impact. A diffuse reflection image is acquired according to the calculated specular reflection coefficients. The experimental results show that the proposed method removes specular reflection effectively with the details and edge information of the image retained and it achieves satisfactory visual effects in processing natural highlight pictures without ground truth. 2021 Vol. 34 (8): 742-750 [Abstract] ( 379 ) [HTML 1KB] [ PDF 4870KB] ( 263 ) 751 Network Ensemble Model for Trend Analysis of Limit Order Books LÜ Xuerui, ZHANG Li To analyze the trend of limit order books(LOBs) better, a network ensemble model for trend analysis of LOBs(NEM-LOB) is proposed. Two long short-term memory(LSTM) sub-models and one convolutional neural network sub-model are integrated in NEM-LOB. One LSTM sub-model captures the global temporal dependence through the distribution information of LOBs. The other LSTM sub-model captures the global dynamics through the dynamic information of LOBs and order streams. The local features are extracted through the factual information of LOBs. Finally, three sub-models are combined to extract features to obtain prediction results. Experiments on FI-2010 dataset show that NEM-LOB makes a better trend analysis for LOBs by combining order streams. 2021 Vol. 34 (8): 751-759 [Abstract] ( 354 ) [HTML 1KB] [ PDF 893KB] ( 280 ) 760 Remote Sensing Image Super-Resolution Reconstruction Based on Dual-Parallel Residual Network LIU Cong, WANG Yaxin The image super-resolution reconstruction algorithm generates a poor effect for the remote sensing images due to different sizes of ground objects and high complexity in the images. Aiming at this problem, a dual-parallel lightweight residual attention network is proposed to increase the reconstruction result. Firstly, a multi-scale shallow feature extraction block(MFEB) is put forward to gain the feature information of different receptive field sizes. The problem of the ground objects with different sizes can be solved by MFEB. Secondly, a lightweight residual attention block(LRAB) is designed with asymmetric convolution and attention mechanism. And thus, the model parameters are reduced and more high-frequency information is captured. Then, the parallel network with different convolution kernels is designed to fuse different receptive fields. Besides, lots of skip connections are employed in residual blocks to increase the reusability of information. Finally, experiments show that the proposed model produces superior performance. 2021 Vol. 34 (8): 760-767 [Abstract] ( 551 ) [HTML 1KB] [ PDF 750KB] ( 419 ) 768 Two-Stage Image Classification Method Based on Three-Way Decisions CHEN Chaofan, ZHANG Hongyun, CAI Kecan, MIAO Duoqian A single model cannot handle the uncertainty in prediction results effectively, and therefore, the shadowed sets theory is introduced into image classification from the perspective of three-way decisions and a two-stage image classification method is designed. Firstly, samples are classified by convolutional neural networks to obtain the membership matrix. Then, a sample partitioning algorithm based on shadowed sets is employed to process the membership matrix and consequently the uncertain part of the classification results, the uncertain domain, for delayed decision making is obtained. Finally, feature fusion technique is utilized and SVM is regarded as a classifier for secondary classification to reduce the uncertainty of the classification results and improve the classification accuracy. Experiments on CIFAR-10 and Caltech 101 datasets validate the effectiveness of the proposed method. 2021 Vol. 34 (8): 768-776 [Abstract] ( 559 ) [HTML 1KB] [ PDF 1497KB] ( 392 )

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