Multi-Scale Dense Graph Attention Network for Hyperspectral Classification

被引:0
|
作者
Wang, Chen [1 ]
Li, Lu [1 ]
Wang, Zhongqi [1 ]
Ma, Jingyao [1 ]
Kong, Yunlong [2 ]
Wang, Yanfeng [2 ]
Chang, Jianrui [1 ]
Zhang, Zimeng [1 ]
Lin, Xinyu [1 ]
机构
[1] Beijing Informat Sci & Technol Univ, Sch Automat, Dept Artificial Intelligence, Beijing, Peoples R China
[2] Chinese Acad Sci, Aerosp Informat Res Inst, Beijing, Peoples R China
基金
中国国家自然科学基金; 北京市自然科学基金;
关键词
IMAGE CLASSIFICATION; FUSION;
D O I
10.1080/07038992.2024.2333424
中图分类号
TP7 [遥感技术];
学科分类号
081102 ; 0816 ; 081602 ; 083002 ; 1404 ;
摘要
In recent years, numerous deep learning-based methods have gained increasing attention in hyperspectral classification, particularly the Graph Neural Network, which exhibits superior capabilities in structural description. However, a single graph structure is not suitable for hyperspectral feature representation. Therefore, we propose a novel Multiple-Scale graph network structure, known as the Multi-Scale Dense Graph Attention network for hyperspectral classification. Firstly, semi-supervised local Fisher discriminant analysis and superpixel segmentation were employed for dimensionality reduction and multi-scale graph construction, respectively. Secondly, Spectral-Spatial convolution is applied to extract shallow features from the image. Subsequently, an improved graph self-attention network is sequentially applied to each scale graph, and the different scale graphs are densely connected through spatial feature alignment modules, designed using twice matrix multiplication. Finally, the combined pixel-level feature map from multiple graph spaces is derived, and Spectral-Spatial convolution is employed to fuse the abundant feature maps for hyperspectral classification. Experimental results on various hyperspectral datasets demonstrate the superiority of our MSDesGATnet over many state-of-the-art methods. The code is available at https://github.com/l7170/MSDesGAT.git. Ces derni & egrave;res ann & eacute;es, de nombreuses m & eacute;thodes bas & eacute;es sur l'apprentissage profond ont suscit & eacute; une attention croissante dans la classification hyperspectrale, en particulier le r & eacute;seau neuronal graphique, qui pr & eacute;sente des capacit & eacute;s sup & eacute;rieures en termes de description structurelle. Cependant, une seule structure de graphe n'est pas adapt & eacute;e & agrave; la repr & eacute;sentation des caract & eacute;ristiques hyperspectrales. C'est pourquoi nous proposons une nouvelle structure de r & eacute;seau graphique & agrave; plusieurs & eacute;chelles, connue sous le nom de r & eacute;seau d'attention graphique dense & agrave; plusieurs & eacute;chelles pour la classification hyperspectrale. Tout d'abord, une analyze discriminante de Fisher locale semi-supervis & eacute;e et une segmentation de superpixel ont & eacute;t & eacute; utilis & eacute;es respectivement pour la r & eacute;duction de la dimensionnalit & eacute; et la construction de graphes & agrave; plusieurs & eacute;chelles. Ensuite, une convolution spectrale-spatiale est appliqu & eacute;e pour extraire des caract & eacute;ristiques de niveau sup & eacute;rieur & agrave; partir de l'image. Par la suite, un r & eacute;seau d'attention graphique am & eacute;lior & eacute; est appliqu & eacute; s & eacute;quentiellement & agrave; chaque graphe & agrave; diff & eacute;rentes & eacute;chelles, et les graphes & agrave; diff & eacute;rentes & eacute;chelles sont connect & eacute;s de mani & egrave;re dense gr & acirc;ce & agrave; des modules d'alignement des caract & eacute;ristiques spatiales, con & ccedil;us & agrave; l'aide de deux multiplications matricielles. Enfin, la carte des caract & eacute;ristiques combin & eacute;es au niveau des pixels & agrave; partir de plusieurs espaces de graphe est obtenue, et une convolution spectrale-spatiale est utilis & eacute;e pour fusionner les nombreuses cartes de caract & eacute;ristiques pour la classification hyperspectrale. Les r & eacute;sultats exp & eacute;rimentaux sur divers ensembles de donn & eacute;es hyperspectrales d & eacute;montrent la sup & eacute;riorit & eacute; de notre r & eacute;seau MSDesGAT par rapport & agrave; de nombreuses m & eacute;thodes de pointe. Le code est disponible sur https://github.com/l7170/MSDesGAT.git.
引用
收藏
页数:23
相关论文
共 50 条
  • [31] Deep Multi-scale Convolutional Neural Network for Hyperspectral Image Classification
    Zhang Feng-zhe
    Yang Xia
    [J]. NINTH INTERNATIONAL CONFERENCE ON GRAPHIC AND IMAGE PROCESSING (ICGIP 2017), 2018, 10615
  • [32] Hyperspectral Image Classification Based on A Multi-Scale Weighted Kernel Network
    Sun Le
    Xu Bin
    Lu Zhenyu
    [J]. CHINESE JOURNAL OF ELECTRONICS, 2022, 31 (05) : 832 - 843
  • [33] Hyperspectral Image Classification Based on A Multi-Scale Weighted Kernel Network
    SUN Le
    XU Bin
    LU Zhenyu
    [J]. Chinese Journal of Electronics, 2022, (05) : 832 - 843
  • [34] Multi-scale attention graph convolutional recurrent network for traffic forecasting
    Xiong, Liyan
    Hu, Zhuyi
    Yuan, Xinhua
    Ding, Weihua
    Huang, Xiaohui
    Lan, Yuanchun
    [J]. CLUSTER COMPUTING-THE JOURNAL OF NETWORKS SOFTWARE TOOLS AND APPLICATIONS, 2024, 27 (03): : 3277 - 3291
  • [35] UMGAN: multi-scale graph attention network for grid parameter identification
    Zou, Gang
    Xia, Min
    Zhang, Liudong
    Lei, Zhen
    Peng, Zhiqiang
    Liu, Jun
    [J]. ELECTRICAL ENGINEERING, 2024,
  • [36] HAMNet: hyperspectral image classification based on hybrid neural network with attention mechanism and multi-scale feature fusion
    Shen, Jinyue
    Zheng, Zhouzhou
    Sun, Yingwei
    Zhao, Mengmeng
    Chang, Yankang
    Shao, Yuyi
    Zhang, Yan
    [J]. INTERNATIONAL JOURNAL OF REMOTE SENSING, 2022, 43 (11) : 4233 - 4258
  • [37] MS3A-Net: multi-scale and spectral-spatial attention network for hyperspectral image classification
    Dai, Mengyun
    Sun, Qi
    Dai, Luanyan
    Lin, Yaohai
    Wei, Lifang
    Yang, Changcai
    Chen, Riqing
    [J]. INTERNATIONAL JOURNAL OF REMOTE SENSING, 2022, 43 (19-24) : 7139 - 7160
  • [38] DMAF-NET: Deep Multi-Scale Attention Fusion Network for Hyperspectral Image Classification with Limited Samples
    Guo, Hufeng
    Liu, Wenyi
    [J]. SENSORS, 2024, 24 (10)
  • [39] Hyperspectral Image Classification Based on Multi-Scale Convolutional Features and Multi-Attention Mechanisms
    Sun, Qian
    Zhao, Guangrui
    Xia, Xinyuan
    Xie, Yu
    Fang, Chenrong
    Sun, Le
    Wu, Zebin
    Pan, Chengsheng
    [J]. REMOTE SENSING, 2024, 16 (12)
  • [40] Dense Multi-Scale Graph Convolutional Network for Knee Joint Cartilage Segmentation
    Chadoulos, Christos
    Tsaopoulos, Dimitrios
    Symeonidis, Andreas
    Moustakidis, Serafeim
    Theocharis, John
    [J]. BIOENGINEERING-BASEL, 2024, 11 (03):