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....

2023 Vol.36 Issue.8, Published 2023-08-25 Papers and Reports    Researches and Applications        Papers and Reports 671 Fast Transferable Model for Cross-Dataset Finger Vein Recognition HUANG Zhe, GUO Cheng'an Deep learning-based methods show excellent recognition performance and potential in finger vein recognition. However, due to the expensive training costs and differences in categories and distributions across different datasets, a model that performs well on one dataset may struggle to efficiently adapt to new data or perform poorly on new data. A fast transferable model for cross-dataset finger vein recognition, including a two-stage solution, is proposed to realize high efficient application of model on different datasets with good performance in practical scenarios where recognition systems are applied to various user groups and devices. Firstly, in the first stage, a domain adaptation algorithm based on feature alignment and clustering is introduced to guide the network in extracting discriminative and robust features, aiming to obtain a deep model that can generalize well on unseen target data. Secondly, a bias field correction network is developed to reduce dataset gaps caused by bias fields in images and further adjust the latent distribution to make the datasets more similar to each other. Then, in the second stage of fast transfer, a modified classifier based on extreme learning machine with a faster learning speed is designed to accelerate model transfer training and make full use of the template information of new data.Experimental results on four public finger vein databases show that the proposed method realizes efficient transfer and achieves recognition performance as good as the best end-to-end training method dose in the target task. For common application scenarios, the proposed method can meet the requirements of real-time deployment and provide a feasible solution for the application of deep learning techniques in cross-dataset finger vein recognition. 2023 Vol. 36 (8): 671-684 [Abstract] ( 469 ) [HTML 1KB] [ PDF 1081KB] ( 1047 ) 685 Hash Image Retrieval Based on Category Similarity Feature Expansion and Center Triplet Loss PAN Lili, MA Junyong, XIONG Siyu, DENG Zhimao, HU Qinghua Convolutional neural networks are commonly employed in the existing deep hashing image retrieval methods. The similarity representation of the deep features extracted by convolutional neural networks is insufficient. In addition, the local triplet samples are mainly constructed for triplet deep hashing from the small batch data, the size of the local triplet samples is small and the data distribution is lack of globality. Consequently, the network training is insufficient and the convergence is difficult. To address these issues, a model of hash image retrieval based on category similarity feature expansion and center triplet loss is proposed. A hash feature extraction module based on vision transformer is designed to extract global feature information with stronger representation ability. To expand the size of mini-batch training samples, a similar feature expansion module based on category constraint is put forward. New feature is generated by the similarity among samples of the same category to enrich the triplet training samples. To enhance the global ability of triplet loss, a center triplet loss function based on Hadamard(CTLH) is constructed. Hadamard is utilized to establish the global hash center constraint for each class. With CLTH, the learning and the convergence of the network are accelerated by adding the center triplet of local constraint and global center constraint, and the accuracy of image retrieval is improved. Experiments on CIFAR10 and NUS-WIDE datasets show that HRFT-Net gains better mean average precision for image retrieval with different bit lengths of hash code, and the effectiveness of HRFT-Net is demonstrated. 2023 Vol. 36 (8): 685-700 [Abstract] ( 339 ) [HTML 1KB] [ PDF 2101KB] ( 768 ) 701 Instance-Level Sketch-Based Image Retrieval Based on Two Stream Multi-granularity Local Alignment Network HAN Xuekun, MIAO Duoqian, ZHANG Hongyun, ZHANG Qixian The goal of instance-level sketch-based image retrieval is to retrieve images by sketches. There is a significant modality gap and feature misalignment issue between sketches and images. In the existing methods, the modality gap between sketches and images cannot be effectively reduced, and only information at a single granularity is captured. Thus, features cannot be aligned effectively. To address these issues, a two stream multi-granularity local alignment network(TSMLA) is proposed. A two-stream feature extractor is introduced to extract both modality-shared and modality-specific local features. These features are simultaneously utilized to calculate the distance between the sketch and the image and reduce the differences between different modalities. Moreover, a multi-granularity local alignment module is adopted to pool the distance matrix at various granularities. Local features are aligned at different scales to effectively address the problem of feature misalignment. TSMLA can fully utilize the information of sketches and real images, while effectively utilizing the connections between features of different granularities. Experiments on multiple datasets validate the effectiveness of TSMLA. 2023 Vol. 36 (8): 701-711 [Abstract] ( 312 ) [HTML 1KB] [ PDF 1625KB] ( 845 ) Researches and Applications 712 Semi-Supervised Node Classification Algorithm Based on Hierarchical Contrastive Learning LI Yaqi, WANG Jie, WANG Feng, LIANG Jiye Most graph contrastive learning methods for semi-supervised node classification obtain two views by cumbersome data augmentation. Moreover, the above data augmentation inevitably changes the graph semantic information, limiting the efficiency and applicability of the existing graph contrastive learning methods. Therefore, a semi-supervised node classification algorithm based on hierarchical contrastive learning is proposed in this paper. In the proposed algorithm, graph data augmentation is unnecessary and the representations of different hierarchies of the graph neural network are learned as contrasted views to alleviate the tedious search and the semantic destruction. In addition, a semi-supervised contrastive loss is designed, and a small amount of labeled information and a large amount of unlabeled information are effectively utilized to provide rich supervised signals and improve the node representations. Finally, node classification experiments on four benchmark datasets validate the effectiveness of the proposed algorithm. 2023 Vol. 36 (8): 712-720 [Abstract] ( 407 ) [HTML 1KB] [ PDF 1049KB] ( 991 ) 721 Road Level Traffic Accident Risk Prediction by Incorporating Temporal Knowledge Graph TANG Weiwen, GUO Shengnan, CHEN Wei, LIN Youfang, WAN Huaiyu Exploring the law of accident occurrence from historical traffic accident data and realizing accurate road level traffic accident risk prediction can improve the travel safety and efficiency effectively. However, road level traffic accident risk prediction is faced with great challenges due to the influence of multiple factors, such as weather and traffic state, the complex temporal and spatial correlation between traffic accidents and the sparsity of accident data. To address these issues, a two-level and multi-view spatial-temporal graph neural network by incorporating temporal knowledge graph(STGN-TKG) is proposed. Firstly, a traffic accident temporal knowledge graph is constructed for the first time, and diachronic embedding for traffic accident temporal knowledge graph is designed to mine the high-order and dynamic correlation between multi-source influencing factor data. Then, a spatial graph convolution attention module and a temporal representation module are employed to fully model the complex spatial-temporal correlations between traffic accidents from two levels and multiple views. Finally, an accident risk propagation strategy is proposed to alleviate the zero-inflated issue. The experimental results on two real-world road level traffic accident risk datasets show that STGN-TKG achieves superior performance on the road level accident risk prediction task. 2023 Vol. 36 (8): 721-732 [Abstract] ( 559 ) [HTML 1KB] [ PDF 914KB] ( 999 ) 733 Differential Evolution Algorithm Based on Coupling and Coordinating Population State Assessment FENG Quanxi, JIN Peiyuan, CEN Jianmin, AI Wu, LIN Bin Differential evolution is a global stochastic search algorithm based on the differences between individuals within a population. The mutation operator is an important component of the differential evolution algorithm, and different mutation operators are suitable for different population distributions. To effectively identify the evolutionary state of the population, a differential evolution algorithm based on coupling and coordinating population state assessment(CCPDE) is proposed. The evolutionary state of the population in the iteration process is evaluated by calculating the coupling coordination degree between four different levels of fitness values and individual distances. The population is classified based on the evaluation results into three evolutionary states: search, balance and convergence, and corresponding mutation operator pools are constructed for different evolutionary states. In addition, the convergence speed of CCPDE is accelerated by adaptive adjustment of the Powell method. Numerical experiments on CEC2017 test functions show the effectiveness of CCPDE. 2023 Vol. 36 (8): 733-748 [Abstract] ( 251 ) [HTML 1KB] [ PDF 920KB] ( 947 ) 749 Weighted Attributes-Based Concept-Cognitive Learning Model LIANG Taoju, LIN Yidong, LIN Menglei, WANG Qijun In the existing concept-cognitive learning models, correlations among attributes and decisions are ignored, the concept space involved is redundant and the learning effectiveness is poor. Aiming at these problems, a weighted attributes-based concept-cognitive learning model(WACCL) is proposed in this paper. Firstly, the relevance of attribute-decision is discussed, and a weighting mechanism for attributes is proposed. With the consideration of the concept space redundancy, the positions of different concepts are explored and the compression of the concept space is achieved. Subsequently, conceptual clustering is implemented by combining similarities between concepts to provide a basis for learning clues. Finally, the effectiveness of WACCL is demonstrated by experiments on 13 datasets. 2023 Vol. 36 (8): 749-763 [Abstract] ( 310 ) [HTML 1KB] [ PDF 4516KB] ( 318 )

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