Pattern Recognition最新文献

筛选
英文 中文
Underwater image enhancement via color constraints and transmission-guided modeling 通过颜色约束和传输引导建模的水下图像增强
IF 7.5 1区 计算机科学
Pattern Recognition Pub Date : 2025-05-29 DOI: 10.1016/j.patcog.2025.111840
Kaichen Chi , Qiang Li
{"title":"Underwater image enhancement via color constraints and transmission-guided modeling","authors":"Kaichen Chi ,&nbsp;Qiang Li","doi":"10.1016/j.patcog.2025.111840","DOIUrl":"10.1016/j.patcog.2025.111840","url":null,"abstract":"<div><div>Underwater images suffer from color deviation and turbidity due to absorption and refraction caused by media. To alleviate severe degradation, we devise an underwater image enhancement method via color constraints and transmission-guided modeling, dubbed CTGAN. Specifically, the critical insight of CTGAN is to break down the overall enhancement process into more manageable steps, thereby enjoying the mutual benefits between color correction and turbidity removal. We develop an interactive constraint color recovery module, which integrates the mean value and mode priors of color channels to render the realistic color. Coupled with a transmission-guided strategy, turbidity traces are gracefully eliminated by integrating heterogeneous degradation cues. To bridge the gap between enhanced and reference images, a frequency-driven triple discriminator is implemented to guide the generation of visually pleasing appearances. We also contribute an Underwater Image Visual Perceptual Enhancement Benchmark (UVPE) to support qualitative and quantitative analysis. Extensive experiments demonstrate the superiority of CTGAN against state-of-the-art methods.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"168 ","pages":"Article 111840"},"PeriodicalIF":7.5,"publicationDate":"2025-05-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144166969","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
AGFormer: An anchor-guided transformer for class imbalance in remote sensing change detection AGFormer:遥感变化检测中类不平衡的锚导变压器
IF 7.5 1区 计算机科学
Pattern Recognition Pub Date : 2025-05-28 DOI: 10.1016/j.patcog.2025.111839
Jiaen Chen , Da Wu , Quanqing Ma, Shengjie Xu, Yuchen Zheng
{"title":"AGFormer: An anchor-guided transformer for class imbalance in remote sensing change detection","authors":"Jiaen Chen ,&nbsp;Da Wu ,&nbsp;Quanqing Ma,&nbsp;Shengjie Xu,&nbsp;Yuchen Zheng","doi":"10.1016/j.patcog.2025.111839","DOIUrl":"10.1016/j.patcog.2025.111839","url":null,"abstract":"<div><div>Remote Sensing Change Detection (RSCD) aims to assess changes by comparing two or more images recorded for the same area but taken at different time stamps. Mainstream research improves the representation of models through the optimization of model architecture design, ignoring the importance of correcting classifiers. However, the issue of class imbalance in the RSCD field inevitably introduces biases into the classifier, damaging the model performance. In this paper, we propose an Anchor-Guided transFormer-based model, named AGFormer, to address this problem. Specifically, the HAR (Hypersphere Anchor Regularization) calibrates the classification layer from an anchor view, which ensures both inter-class separability and intra-class balance between compactness and diversity by initializing class anchors on the hypersphere and applying similarity-based contrastive learning in different phases. In addition, a disentanglement anchor optimization strategy is designed to avoid the influence of class imbalance in the RSCD field. By supervising the main features and calibrating classifiers with mapped class anchors, more discriminative representations and robust classifiers are obtained. In addition, we design the CEM (Change Enhancement Module) based on flow to highlight the changed features. The proposed HAR and CEM are plug-and-play and can be integrated into existing architectures. Extensive experiments are conducted on four benchmark datasets, and state-of-the-art performance is achieved by the proposed AGFormer. All the codes are available at <span><span>https://github.com/jiaenchen2024/AGFormer</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"168 ","pages":"Article 111839"},"PeriodicalIF":7.5,"publicationDate":"2025-05-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144166938","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
Dual-space Co-training for Large-scale Multi-view Clustering 大规模多视图聚类的双空间协同训练
IF 7.5 1区 计算机科学
Pattern Recognition Pub Date : 2025-05-28 DOI: 10.1016/j.patcog.2025.111844
Zisen Kong , Zhiqiang Fu , Dongxia Chang , Yiming Wang , Yao Zhao
{"title":"Dual-space Co-training for Large-scale Multi-view Clustering","authors":"Zisen Kong ,&nbsp;Zhiqiang Fu ,&nbsp;Dongxia Chang ,&nbsp;Yiming Wang ,&nbsp;Yao Zhao","doi":"10.1016/j.patcog.2025.111844","DOIUrl":"10.1016/j.patcog.2025.111844","url":null,"abstract":"<div><div>Anchor-based methods are popular for their low computational complexity and high efficiency. Existing solutions either construct anchors on each view and fuse them, or directly obtain a consistent view structure. However, these strategies do not fully utilize the information between views. To address this, we propose a Dual-space Co-training (DSCMC) model for Large-scale Multi-view Clustering, which learns the consistent anchor graph using a dual-space co-training strategy. Specifically, we introduce an orthogonal projection matrix in the original space enabling the learned consistent anchor graph to capture the inherent relationships in each view. Meanwhile, the feature transformation matrix maps samples to a shared latent space, facilitating information alignment and comprehensive data distribution understanding. The proposed joint optimization strategy allows us to construct a discriminative anchor graph that effectively captures the essential features of multi-view data. Extensive experiments demonstrate that our method reduces computational complexity while outperforming existing approaches in clustering performance.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"168 ","pages":"Article 111844"},"PeriodicalIF":7.5,"publicationDate":"2025-05-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144166930","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
Challenge-aware U-net for breast lesion segmentation in ultrasound images 基于挑战感知的U-net超声图像乳腺病灶分割
IF 7.5 1区 计算机科学
Pattern Recognition Pub Date : 2025-05-27 DOI: 10.1016/j.patcog.2025.111851
Dengdi Sun , Changxu Dong , Yuchen Yan , Bo Jiang , Yayang Duan , Zhengzheng Tu , Chaoxue Zhang
{"title":"Challenge-aware U-net for breast lesion segmentation in ultrasound images","authors":"Dengdi Sun ,&nbsp;Changxu Dong ,&nbsp;Yuchen Yan ,&nbsp;Bo Jiang ,&nbsp;Yayang Duan ,&nbsp;Zhengzheng Tu ,&nbsp;Chaoxue Zhang","doi":"10.1016/j.patcog.2025.111851","DOIUrl":"10.1016/j.patcog.2025.111851","url":null,"abstract":"<div><div>Deep learning methods can enhance the efficiency of tumor segmentation in breast ultrasound (BUS) images. However, noise interference, small tumors, and blurred boundaries can reduce segmentation accuracy. We design a three-branch challenge-aware U-net (CAU-net) to address these main challenges in BUS images. Our CAU-net extracts the features from three challenge-aware encoders in parallel first. Secondly, we propose an adaptive aggregation layer (AAL) to merge the multi-scale features of three challenging branches, enabling the network to adaptively handle different breast lesion samples with these main challenges. To further enhance the accuracy of segmentation, we introduce the graph reasoning module (GRM) to the network to model the correlation between the channels of the features and acquire the global information in the features. The result of our experiment on two datasets demonstrates the superiority of CAU-net over the advanced medical image segmentation methods. Our code can be downloaded from <span><span>https://github.com/tzz-ahu</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"168 ","pages":"Article 111851"},"PeriodicalIF":7.5,"publicationDate":"2025-05-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144166934","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
Hypergraph representation learning for identifying circRNA-disease associations 识别circrna与疾病关联的超图表示学习
IF 7.5 1区 计算机科学
Pattern Recognition Pub Date : 2025-05-24 DOI: 10.1016/j.patcog.2025.111835
Yang Li , Xuegang Hu , Peipei Li , Lei Wang , Zhuhong You
{"title":"Hypergraph representation learning for identifying circRNA-disease associations","authors":"Yang Li ,&nbsp;Xuegang Hu ,&nbsp;Peipei Li ,&nbsp;Lei Wang ,&nbsp;Zhuhong You","doi":"10.1016/j.patcog.2025.111835","DOIUrl":"10.1016/j.patcog.2025.111835","url":null,"abstract":"<div><div>CircRNA-disease associations (CDA) are crucial for identifying circRNA biomarkers, significantly aiding the prevention, diagnosis, and treatment of complex human diseases. Traditional wet-lab methods for CDA prediction, while useful, are time-consuming, labor-intensive, and not always successful. Recently, computational methods have emerged as promising alternatives, offering more efficient CDA detection. Nevertheless, existing computational methods often overlook the multifaceted nature of CDAs, where each circRNA can associate with multiple diseases simultaneously, and vice versa. These methods typically fail to capture the beyond pairwise relationships and higher-order complex associations between circRNA-disease pairs. To this end, we propose a novel and effective biomarker computational method named HyperGRL-CDA, which is based on biological attribute information and hypergraph representation learning strategies. Its cornerstone is a hypergraph representation learning module that employs circRNA and disease similarity attributes to construct biological hypergraphs. This module leverages a symmetric hypergraph convolutional network to learn and reveal hidden, high-quality embedding representations, capturing the complex associations within these hypergraphs. Enhancing computational efficiency, HyperGRL-CDA incorporates the Extra Trees algorithm to determine CDA matching scores. Tested through five-fold cross-validation on the circR2Disease dataset, HyperGRL-CDA achieved an impressive accuracy of 92.22% and an AUC score of 96.08%. Furthermore, it demonstrated superior predictive performance on various related CDA datasets. These extensive experiments confirm HyperGRL-CDA as an efficient, accurate, and robust method for CDA prediction based on hypergraph representation learning.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"168 ","pages":"Article 111835"},"PeriodicalIF":7.5,"publicationDate":"2025-05-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144134756","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
CAM: Causality-driven Adaptive Sparsity and Hierarchical Memory for robust out-of-distribution learning in GNNs 基于因果驱动的自适应稀疏性和层次记忆的gnn鲁棒分布外学习
IF 7.5 1区 计算机科学
Pattern Recognition Pub Date : 2025-05-24 DOI: 10.1016/j.patcog.2025.111812
Ran Chen , Huaguang Zhu , Bofei Xiao , TieFeng Ma
{"title":"CAM: Causality-driven Adaptive Sparsity and Hierarchical Memory for robust out-of-distribution learning in GNNs","authors":"Ran Chen ,&nbsp;Huaguang Zhu ,&nbsp;Bofei Xiao ,&nbsp;TieFeng Ma","doi":"10.1016/j.patcog.2025.111812","DOIUrl":"10.1016/j.patcog.2025.111812","url":null,"abstract":"<div><div>Graph Neural Networks (GNNs) face persistent challenges in out-of-distribution (OOD) generalization, a critical issue in advancing machine learning research. Existing methods primarily depend on statistical correlations, rendering them vulnerable to environmental shifts and leading to a sharp decline in generalization performance. This limitation stems from models inadvertently capturing environment-induced spurious correlations rather than leveraging robust causal mechanisms. To address this fundamental bottleneck, we reformulate graph learning through the lens of causal inference and propose a <strong>C</strong>ausality-driven <strong>A</strong>daptive Sparsity and Hierarchical <strong>M</strong>emory framework (<strong>CAM</strong>) to enhance OOD generalization in GNNs. Specifically, we introduce an Adaptive sparse expert selection module, which employs a data-driven dynamic activation strategy to identify causally relevant expert groups while mitigating environment-sensitive spurious correlations, thereby improving computational efficiency and robustness. Additionally, a Hierarchical memory mechanism module constructs a cross-layer causal inference framework, explicitly modeling inter-layer environmental dependencies to preserve essential causal features in deeper layers. A key advantage of CAM is its independence from prior environmental labels, relying solely on data distributions for causal deconfounding. This enables comprehensive modeling and elimination of environment-induced biases. Extensive experiments on multiple cross-domain OOD benchmarks demonstrate that CAM surpasses SOTA methods, significantly enhancing generalization performance. Beyond its empirical success, CAM introduces a novel causal inference perspective for GNNs research, laying a theoretical foundation for causal modeling in complex networks. The source code for CAM is available at: <span><span>https://github.com/12chen20/CAM</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"168 ","pages":"Article 111812"},"PeriodicalIF":7.5,"publicationDate":"2025-05-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144139762","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
Fusing deep and hand-crafted features by deep canonically correlated contractive autoencoder for offline signature verification 采用深度正则相关压缩自编码器融合深度特征和手工特征,实现离线签名验证
IF 7.5 1区 计算机科学
Pattern Recognition Pub Date : 2025-05-24 DOI: 10.1016/j.patcog.2025.111834
Xingbiao Zhao , Lidong Zheng , Panli Yuan , Yuchen Zheng
{"title":"Fusing deep and hand-crafted features by deep canonically correlated contractive autoencoder for offline signature verification","authors":"Xingbiao Zhao ,&nbsp;Lidong Zheng ,&nbsp;Panli Yuan ,&nbsp;Yuchen Zheng","doi":"10.1016/j.patcog.2025.111834","DOIUrl":"10.1016/j.patcog.2025.111834","url":null,"abstract":"<div><div>Handwritten signatures are currently the most widely used and recognized form of identity authorization, which is a significant way for individuals to express their identity to information. Since the forgers learn information about the genuine signatures from the target signer in advance, there are usually only minor discrepancies between skilled forged and genuine signatures. Therefore, building an automatic handwritten signature verification system to recognize skilled forgeries is a worthy challenging task. In this paper, to learn a good representation for distinguishing skilled forged and genuine signatures, we propose an offline handwritten signature verification system that fuses deep learning-based and hand-crafted features, which combines the merits of different views of features. Specifically, a novel multi-view representation learning method is proposed, named Deep Canonically Correlated Contractive Autoencoder (DCCCAE) for learning combined representations between deep and hand-crafted features. After the feature learning process, we train Support Vector Machines (SVMs) as writer-dependent classifiers for each signer to build the completed verification system. Extensive experiments and analyses on four different language datasets, such as English (CEDAR), Persian (UTSig), Bengali and Hindi (BHSig), and Chinese (SigComp2011) demonstrate that the proposed system improves the learning ability compared with the single view features and achieve the competitive performance compared with the state-of-the-art verification systems.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"168 ","pages":"Article 111834"},"PeriodicalIF":7.5,"publicationDate":"2025-05-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144139760","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
Spatio-temporal Feature-level Augmentation Vision Transformer for video-based person re-identification 基于视频的人再识别的时空特征级增强视觉转换器
IF 7.5 1区 计算机科学
Pattern Recognition Pub Date : 2025-05-24 DOI: 10.1016/j.patcog.2025.111813
Minjung Kim , MyeongAh Cho , Heansung Lee , Sangyoun Lee
{"title":"Spatio-temporal Feature-level Augmentation Vision Transformer for video-based person re-identification","authors":"Minjung Kim ,&nbsp;MyeongAh Cho ,&nbsp;Heansung Lee ,&nbsp;Sangyoun Lee","doi":"10.1016/j.patcog.2025.111813","DOIUrl":"10.1016/j.patcog.2025.111813","url":null,"abstract":"<div><div>Video-based person re-identification (ReID) aims to match an individual across multiple videos, thus addressing critical aspects of security applications of computer vision. While previous transformer-based approaches have used various means to enhance performance, the growing complexities in network design have posed challenges in meeting the practical requirements of intelligent surveillance systems. To improve network efficiency, we introduce a Feature-level Augmentation Vision Transformer (FAViT), which reinterprets the attributes of video ReID. We leverage the property of maintaining identity even when backgrounds change or multiple persons appear in video frames. First, we introduce Token Representation Learning to distinguish foreground from background. We also employ spatio-temporal feature-level augmentation, along with conducting Altered Background ID classification and Anomaly Frame Detection, to strengthen the representation capacity of the transformer. Extensive experiments validate the effectiveness of FAViT with the least computational overhead among transformer-based models across five benchmarks. We substantiate our model’s generalization ability through analyses.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"168 ","pages":"Article 111813"},"PeriodicalIF":7.5,"publicationDate":"2025-05-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144148025","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
One-pass online learning under evolving feature data streams: A non-parametric model 演化特征数据流下的一次在线学习:一个非参数模型
IF 7.5 1区 计算机科学
Pattern Recognition Pub Date : 2025-05-24 DOI: 10.1016/j.patcog.2025.111719
Han Zhou , Hongpeng Yin , Bin Wang , Chenglin Liao
{"title":"One-pass online learning under evolving feature data streams: A non-parametric model","authors":"Han Zhou ,&nbsp;Hongpeng Yin ,&nbsp;Bin Wang ,&nbsp;Chenglin Liao","doi":"10.1016/j.patcog.2025.111719","DOIUrl":"10.1016/j.patcog.2025.111719","url":null,"abstract":"<div><div>In real-world applications, data streams naturally evolves and thus may exhibit a dynamic feature space, wherein new features appear and old ones disappear. Online learning under such circumstances necessitates simultaneous learning from increasing data volume and adaptation to the dynamic feature space in real time. While several methodologies have been proposed to tackle this challenge, many of them rely on strong assumptions regarding evolving interaction manners. Instead, this study adopts a broader perspective aimed at facilitating learning from arbitrarily evolving features without any strict assumptions in the previous work. Then, we present an online learning method based on a non-parametric kernel model. This model accommodates data streams with both continuous instances and evolving features through simple deduction and addition operations. Theoretical analysis shows the sublinear regret <span><math><mrow><mi>O</mi><mrow><mo>(</mo><msqrt><mrow><mi>T</mi></mrow></msqrt><mo>)</mo></mrow></mrow></math></span> of the proposed method. Empirical studies show the capability to adapt not only to the previously constrained evolving features but also to the more arbitrarily evolving features.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"168 ","pages":"Article 111719"},"PeriodicalIF":7.5,"publicationDate":"2025-05-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144166936","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
Self-distillation salient object detection via generalized diversity loss 基于广义分集损失的自蒸馏显著目标检测
IF 7.5 1区 计算机科学
Pattern Recognition Pub Date : 2025-05-23 DOI: 10.1016/j.patcog.2025.111804
Yunfei Zheng , Jibin Yang , Haijun Tao , Yong Wang , Lei Chen , Yang Wang , Tieyong Cao
{"title":"Self-distillation salient object detection via generalized diversity loss","authors":"Yunfei Zheng ,&nbsp;Jibin Yang ,&nbsp;Haijun Tao ,&nbsp;Yong Wang ,&nbsp;Lei Chen ,&nbsp;Yang Wang ,&nbsp;Tieyong Cao","doi":"10.1016/j.patcog.2025.111804","DOIUrl":"10.1016/j.patcog.2025.111804","url":null,"abstract":"<div><div>Classic knowledge distillation (KD) via the Kullback–Leibler loss can improve the performance of small deep classification models effectively, but they are hard to be applied into salient object detection (SOD) models due to the lack of necessary multi-dimension knowledge representations in the logit layer. In this paper, a generalized diversity (GD) loss, inspired by ensemble learning, is proposed to constrain the student and teacher models to hold low diversity. This process drives the student to mimic the teacher’s salient knowledge representations while enhancing the student’s generalization ability. Secondly, a salient self-distillation (SD) framework based on the shared backbone and the salient SD loss is proposed. In a shared backbone network, a lightweight student sub-network and a large parameter teacher sub-network are constructed, respectively, to synchronously achieve coarse but rapid feature extraction, and refined but slow feature extraction. The SD loss is utilized to transfer refined salient knowledge from the teacher sub-network to the student sub-network, so that the performance of the student sub-network is improved. Extensive experimental results on five benchmark datasets demonstrate that the proposed GD loss can achieve salient knowledge transfer and outperforms recent six KD methods, and the proposed student network outperforms recent eleven SOD networks in performance and efficiency.</div></div>","PeriodicalId":49713,"journal":{"name":"Pattern Recognition","volume":"168 ","pages":"Article 111804"},"PeriodicalIF":7.5,"publicationDate":"2025-05-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144166937","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
引用次数: 0
0
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
相关产品
×
本文献相关产品
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信