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.3, Published 2019-03-25 Papers and Reports    Researches and Applications        Papers and Reports 193 Real-Time Detection of Underwater Fish Based on Improved YOLO and Transfer Learning LI Qingzhong, LI Yibing, NIU Jiong To fast detect underwater fish in unrestricted underwater environment based on underwater video collected by underwater robots, a real-time detection algorithm for underwater fish based on improved you only look once(YOLO) and transfer learning is proposed. Firstly, an underwater-YOLO for the embedding computer system of underwater robots is designed to overcome the shortcomings of traditional YOLO. Then, transfer learning strategy is employed to train the underwater-YOLO network and alleviate the limitation of known underwater fish samples. A preprocessing algorithm based on contrast limited adaptive histogram equalization is proposed to overcome the problem of underwater image degradation. Finally, a video frame selection method for foreground computation of underwater-YOLO based on structure similarity between inter-frames is proposed to increase the detection frame rate. The experimental results show that the proposed algorithm achieves the goal of real-time detection of underwater fish in unconstrained underwater environment. Compared with the traditional YOLO,the proposed underwater-YOLO generates better detection performance in complex scenes with small fish and overlapped fishes. 2019 Vol. 32 (3): 193-203 [Abstract] ( 1923 ) [HTML 1KB] [ PDF 1432KB] ( 1098 ) 204 A Fuzzy Style Clustering Algorithm on Stylistic Data SHEN Hao , WANG Shitong In the stylistic data, different organizational styles obviously or implicitly exist in different clusters. The classical partitional clustering methods represented by K-means and fuzzy C-means are ineffective for the stylistic data. Therefore, a fuzzy style clustering(FSC) is proposed. A style normalization matrix is utilized to represent the style information of the samples within each cluster, and the distance matrix is calculated with samples transformed by style normalization matrices. Besides, the fuzzy membership is exploited to describe the representable degree of a sample for a certain cluster. The membership matrix and style normalization matrix are optimized simultaneously by the commonly-used alternating optimization technique. FSC can make use of the style information of samples and the information between samples and clusters effectively, and the experimental results on synthetic and real datasets indicate the effectiveness of the proposed algorithm. 2019 Vol. 32 (3): 204-213 [Abstract] ( 540 ) [HTML 1KB] [ PDF 1756KB] ( 413 ) 214 Zero-Shot Image Recognition Algorithm via Semantic Auto-Encoder Combining Relation Network LIN Kezheng, LI Haotian, BAI Jingxuan, LI Ao A semantic auto-encoder structure improved by relation network is proposed and used for zero sample identification algorithm to handle the projection domain shift problem and improve the robustness of distance similarity measure in the traditional model of zero-shot recognition. The feature map between image visual features and semantic vectors is constructed by the proposed algorithm based on the semantic auto-encoder, and then the reconstructed vector is sent to the neural network after concatenating the true value of the corresponding vector. Finally, the prediction category is determined by the output scalar. The experimental results show that compared with the traditional distance measurement method, the recognition rate of the proposed algorithm on the public datasets AWA, CUB and ImageNet-2 is improved and its semantic-visual projection has a better effect than back projection on some datasets. 2019 Vol. 32 (3): 214-224 [Abstract] ( 507 ) [HTML 1KB] [ PDF 3018KB] ( 445 ) 225 Image Segmentation Using Generalized Information Entropy for Eigenvector Selection ZHANG Daming, ZHANG Xueyong, LI Lu, LIU Huayong The clustering result is determined by the quality of the eigenvectors (spectral) of the related graph Laplacian in spectral clustering, and therefore eigenvector selection is crucial. To solve this problem, spectral discrimination(SD), spectral discrimination availability(SDA) and spectral discrimination degree(SDD) are defined based on generalized information entropy. SD is exploited to measure the clustering information of each eigenvector. SDA is utilized to remove the ineffective eigenvectors for clustering. SDD is employed to construct a selective clustering ensemble scheme based on contribution for clustering. Thus, a spectral clustering algorithm based on eigenvector selection is proposed. The experimental results on varied natural images show that the proposed algorithm is simple and effective. 2019 Vol. 32 (3): 225-236 [Abstract] ( 633 ) [HTML 1KB] [ PDF 2881KB] ( 484 ) 237 A Lightweight Convolutional Neural Network Architecture with Slice Feature Map ZHANG Yufeng, ZHENG Zhonglong, LIU Huawen, XIANG Daohong, HE Xiaowei, LI Zhifei, HE Yiran, KHODJA Abd Erraouf The capacities of mobile and embedded devices are quite inadequate for the requirement of the storage capacity and computational resources of convolutional neural network models. Therefore, a lightweight convolutional neural network architecture, network with slice feature map, named SFNet, is proposed. The concept of slice block is introduced. By performing the “slice” processing on the output feature map of the network, each feature map segment is respectively sent to a convolution kernel of different sizes for convolution operation, and then the obtained feature map is concatenated. A simple 1×1 convolution is utilized to fuse the channels of the feature map. The experiments show that compared with the state-of-the-art lightweight convolutional neural networks, SFNet has fewer parameters and floating-point operations, and higher classification accuracy with the same number of convolution kernels and input feature map channels. Compared with the standard convolution, in the case of a significant reduction in network complexity, the classification accuracy is same or higher. 2019 Vol. 32 (3): 237-246 [Abstract] ( 640 ) [HTML 1KB] [ PDF 1057KB] ( 480 ) Researches and Applications 247 Parallel Ensemble Learning Model Based on Hybrid Improved GSO and GRNN JIAN Shuqiang, NI Zhiwei, LI Jingming, ZHU Xuhui, NI Liping Glowworm swarm optimization(GSO) has problems of instability and slow convergence speed and accuracy. The error of general regression neural network(GRNN) is easily caused by the network structure. Aiming at these defects, a parallel ensemble learning model based on hybrid improved GSO and GRNN is proposed, and it is applied for haze prediction. Firstly, a hybrid improved glowworm swarm optimization(HIGSO) algorithm is constructed fusing multiple search strategies. The performance of the algorithm is verified via standard test functions. Next, the HIGSO algorithm is integrated with GRNN with disturbance parameter to construct a parallel ensemble learning model, and its validity and feasibility are verified on UCI standard dataset. Finally, the proposed model is applied for haze prediction in Beijing, Shanghai and Guangzhou areas to further verify its performance in haze prediction. 2019 Vol. 32 (3): 247-258 [Abstract] ( 533 ) [HTML 1KB] [ PDF 1067KB] ( 310 ) 259 Building Deep Neural Networks with Dilated Convolutions to Reconstruct High-Resolution Image ZHANG Zhuolin, ZHAO Jianwei, CAO Feilong To expand the perception field with the filter parameters unchanged, dilated convolution is introduced into very deep convolutional networks super-resolution model. Firstly, the perception field of the dilated convolution block with different expansion coefficients is analyzed and a better combination structure is selected as the dilated convolution block. Then, the deep convolution network is constructed by stacking convolution blocks and adding residual connection. Experiment shows that the reconstruction effect can be improved by the constructed network for the larger scaling factors of Set5 dataset. Besides, there are obvious visual advantages. 2019 Vol. 32 (3): 259-267 [Abstract] ( 708 ) [HTML 1KB] [ PDF 1416KB] ( 464 ) 268 Fabric Defect Detection Based on Template Correction and Low-Rank Decomposition JI Xuan, LIANG Jiuzhen, HOU Zhenjie, CHANG Xingzhi, LIU Wei To solve the problem of tensile deformation of periodic fabric, a fabric defect detection method based on template correction and low-rank decomposition is proposed. Firstly, the original image is corrected by the template to reduce the influence of stretching deformation on the detection results. Then, a low-rank correction decomposition model is proposed including a low-rank term, sparse term and correction term. The model can be solved by the alternating direction method to generate a low-rank matrix and a sparse matrix. Finally, the optimal threshold segmentation algorithm is utilized to segment the significant images generated by the sparse matrix. Experiments on standard databases show that the recall rate of the proposed method is improved. 2019 Vol. 32 (3): 268-277 [Abstract] ( 453 ) [HTML 1KB] [ PDF 1400KB] ( 373 ) 278 Unification of Multiple Class Data Based on Linguistic Ordered Pair 3-Tuple WANG Hongdong, HOU Xuehui, GAO Yunhui, ZOU Li To unify multiple class data, the linguistic ordered pair 3-tuple representation model based on linguistic 2-tuple is proposed, and some of its properties are studied. Then, the similarity between linguistic ordered pair 3-tuple is produced. Meanwhile, the linguistic ordered pair 3-tuple weighted average operator is discussed. The standardized transformation models of fuzzy sets, intuitionistic fuzzy sets, hesitant fuzzy sets, linguistic truth-valued intuitionistic fuzzy sets and linguistic ordered pair 3-tuple are constructed, respectively. Combined with the similarity between linguistic ordered pair 3-tuple, the standardized transformation models of fuzzy sets, intuitionistic fuzzy sets, hesitant fuzzy sets, linguistic truth-valued intuitionistic fuzzy sets are applied to pattern recognition. Finally, the rationality and the effectiveness of the proposed method are illustrated via an example of hospital intelligent triage. 2019 Vol. 32 (3): 278-286 [Abstract] ( 350 ) [HTML 1KB] [ PDF 793KB] ( 235 )

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