Hybrid CNN-GCN Network for Hyperspectral Image Classification
In recent years, convolutional neural networks (CNNs) have been impressive due to their excellent feature representation abilities, but it is difficult to learn long-distance spatial structures information. Unlike CNN, graph convolutional networks (GCNs) can well handle the intrinsic manifold struct...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10915579/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1850177755012923392 |
|---|---|
| author | Cuiping Shi Diling Liao Liguo Wang |
| author_facet | Cuiping Shi Diling Liao Liguo Wang |
| author_sort | Cuiping Shi |
| collection | DOAJ |
| description | In recent years, convolutional neural networks (CNNs) have been impressive due to their excellent feature representation abilities, but it is difficult to learn long-distance spatial structures information. Unlike CNN, graph convolutional networks (GCNs) can well handle the intrinsic manifold structures of hyperspectral images (HSIs). However, the existing GCN-based classification methods do not fully utilize the edge relationship, which makes their performance is limited. In addition, a small number of training samples is also a reason for hindering high-performance HSI classification. Therefore, this article proposes a hybrid CNN-GCN network (HCGN) for HSI classification. First, a graph edge enhanced module (GEEM) is designed to enhance the superpixel-level features of graph edge nodes and improve the spatial discrimination ability of ground objects. In particular, considering multiscale information is complementary, a multiscale GEEM based on GEEM is proposed to fully utilize texture structures of different sizes. Then, in order to enhance the pixel-level multi hierarchical fine feature representation of images, a multiscale cross fusion module based on the CNN framework is proposed. Finally, the extracted pixel-level features and superpixel-level features are cascaded. Through a series of experiments, it has been proved that compared with some state-of-the-art methods, HCGN combines the advantages of CNN and GCN frameworks, can provide superior classification performance under limited training samples, and demonstrates the advantages and great potential of HCGN. |
| format | Article |
| id | doaj-art-4a3b2fcdea8e4db9bc401a2d74b3741e |
| institution | OA Journals |
| issn | 1939-1404 2151-1535 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing |
| spelling | doaj-art-4a3b2fcdea8e4db9bc401a2d74b3741e2025-08-20T02:18:55ZengIEEEIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing1939-14042151-15352025-01-0118105301054610.1109/JSTARS.2025.354857110915579Hybrid CNN-GCN Network for Hyperspectral Image ClassificationCuiping Shi0Diling Liao1https://orcid.org/0000-0002-8979-5246Liguo Wang2https://orcid.org/0000-0001-9373-6233School of Information Engineering, Huzhou University, Huzhou, ChinaSchool of Electronics and Communication Engineering, Sun Yat-sen University, Shenzhen, ChinaSchool of Information and Communication Engineering, Dalian Nationalities University, Dalian, ChinaIn recent years, convolutional neural networks (CNNs) have been impressive due to their excellent feature representation abilities, but it is difficult to learn long-distance spatial structures information. Unlike CNN, graph convolutional networks (GCNs) can well handle the intrinsic manifold structures of hyperspectral images (HSIs). However, the existing GCN-based classification methods do not fully utilize the edge relationship, which makes their performance is limited. In addition, a small number of training samples is also a reason for hindering high-performance HSI classification. Therefore, this article proposes a hybrid CNN-GCN network (HCGN) for HSI classification. First, a graph edge enhanced module (GEEM) is designed to enhance the superpixel-level features of graph edge nodes and improve the spatial discrimination ability of ground objects. In particular, considering multiscale information is complementary, a multiscale GEEM based on GEEM is proposed to fully utilize texture structures of different sizes. Then, in order to enhance the pixel-level multi hierarchical fine feature representation of images, a multiscale cross fusion module based on the CNN framework is proposed. Finally, the extracted pixel-level features and superpixel-level features are cascaded. Through a series of experiments, it has been proved that compared with some state-of-the-art methods, HCGN combines the advantages of CNN and GCN frameworks, can provide superior classification performance under limited training samples, and demonstrates the advantages and great potential of HCGN.https://ieeexplore.ieee.org/document/10915579/Cross fusionedge enhancedgraph convolutional network (GCN)hyperspectral image (HSI) |
| spellingShingle | Cuiping Shi Diling Liao Liguo Wang Hybrid CNN-GCN Network for Hyperspectral Image Classification IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing Cross fusion edge enhanced graph convolutional network (GCN) hyperspectral image (HSI) |
| title | Hybrid CNN-GCN Network for Hyperspectral Image Classification |
| title_full | Hybrid CNN-GCN Network for Hyperspectral Image Classification |
| title_fullStr | Hybrid CNN-GCN Network for Hyperspectral Image Classification |
| title_full_unstemmed | Hybrid CNN-GCN Network for Hyperspectral Image Classification |
| title_short | Hybrid CNN-GCN Network for Hyperspectral Image Classification |
| title_sort | hybrid cnn gcn network for hyperspectral image classification |
| topic | Cross fusion edge enhanced graph convolutional network (GCN) hyperspectral image (HSI) |
| url | https://ieeexplore.ieee.org/document/10915579/ |
| work_keys_str_mv | AT cuipingshi hybridcnngcnnetworkforhyperspectralimageclassification AT dilingliao hybridcnngcnnetworkforhyperspectralimageclassification AT liguowang hybridcnngcnnetworkforhyperspectralimageclassification |