Stereo Matching of High-Resolution Satellite Images via Hierarchical ViT and Self-Supervised DINO
Dense matching plays an important role in 3D modeling from satellite images. Its purpose is to establish pixel-by-pixel correspondences between two stereo images. This study presents a learning-based dense matching approach that integrates selfsupervised learning with a multi-head attention mechanis...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Copernicus Publications
2025-07-01
|
| Series: | ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences |
| Online Access: | https://isprs-annals.copernicus.org/articles/X-G-2025/357/2025/isprs-annals-X-G-2025-357-2025.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849321117675683840 |
|---|---|
| author | X. He M. Yang S. Jiang S. Jiang W. Jiang Q. Li |
| author_facet | X. He M. Yang S. Jiang S. Jiang W. Jiang Q. Li |
| author_sort | X. He |
| collection | DOAJ |
| description | Dense matching plays an important role in 3D modeling from satellite images. Its purpose is to establish pixel-by-pixel correspondences between two stereo images. This study presents a learning-based dense matching approach that integrates selfsupervised learning with a multi-head attention mechanism to achieve feature fusion. Since stereo matching in satellite datasets is restricted by the disparity range, the pixel-by-pixel method can reduce the limitation. In the feature extraction module, we have performed attention-based in-depth learning on the smallest-scale feature using the self-supervised DINO. In addition, a CEP (Context-Enhanced Path) module is added outside the main matching path, and continuously enhanced position embedding is used to improve relative position encoding. The effectiveness of this method has been demonstrated through experiments on the US3D and WHU-Stereo datasets. |
| format | Article |
| id | doaj-art-de74f4b2e7e14387ba3ba2e65eea1c69 |
| institution | Kabale University |
| issn | 2194-9042 2194-9050 |
| language | English |
| publishDate | 2025-07-01 |
| publisher | Copernicus Publications |
| record_format | Article |
| series | ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences |
| spelling | doaj-art-de74f4b2e7e14387ba3ba2e65eea1c692025-08-20T03:49:50ZengCopernicus PublicationsISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences2194-90422194-90502025-07-01X-G-202535736410.5194/isprs-annals-X-G-2025-357-2025Stereo Matching of High-Resolution Satellite Images via Hierarchical ViT and Self-Supervised DINOX. He0M. Yang1S. Jiang2S. Jiang3W. Jiang4Q. Li5School of Computer Science, China University of Geosciences, Wuhan 430074, ChinaSchool of Computer Science, China University of Geosciences, Wuhan 430074, ChinaGuangdong Key Laboratory of Urban Informatics, Shenzhen Univeristy, Shenzhen 518060, ChinaEngineering Research Center of Natural Resource Information Management and Digital Twin Engineering Software, Ministry of Education, Wuhan 430074, ChinaState Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430072, ChinaGuangdong Key Laboratory of Urban Informatics, Shenzhen Univeristy, Shenzhen 518060, ChinaDense matching plays an important role in 3D modeling from satellite images. Its purpose is to establish pixel-by-pixel correspondences between two stereo images. This study presents a learning-based dense matching approach that integrates selfsupervised learning with a multi-head attention mechanism to achieve feature fusion. Since stereo matching in satellite datasets is restricted by the disparity range, the pixel-by-pixel method can reduce the limitation. In the feature extraction module, we have performed attention-based in-depth learning on the smallest-scale feature using the self-supervised DINO. In addition, a CEP (Context-Enhanced Path) module is added outside the main matching path, and continuously enhanced position embedding is used to improve relative position encoding. The effectiveness of this method has been demonstrated through experiments on the US3D and WHU-Stereo datasets.https://isprs-annals.copernicus.org/articles/X-G-2025/357/2025/isprs-annals-X-G-2025-357-2025.pdf |
| spellingShingle | X. He M. Yang S. Jiang S. Jiang W. Jiang Q. Li Stereo Matching of High-Resolution Satellite Images via Hierarchical ViT and Self-Supervised DINO ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences |
| title | Stereo Matching of High-Resolution Satellite Images via Hierarchical ViT and Self-Supervised DINO |
| title_full | Stereo Matching of High-Resolution Satellite Images via Hierarchical ViT and Self-Supervised DINO |
| title_fullStr | Stereo Matching of High-Resolution Satellite Images via Hierarchical ViT and Self-Supervised DINO |
| title_full_unstemmed | Stereo Matching of High-Resolution Satellite Images via Hierarchical ViT and Self-Supervised DINO |
| title_short | Stereo Matching of High-Resolution Satellite Images via Hierarchical ViT and Self-Supervised DINO |
| title_sort | stereo matching of high resolution satellite images via hierarchical vit and self supervised dino |
| url | https://isprs-annals.copernicus.org/articles/X-G-2025/357/2025/isprs-annals-X-G-2025-357-2025.pdf |
| work_keys_str_mv | AT xhe stereomatchingofhighresolutionsatelliteimagesviahierarchicalvitandselfsuperviseddino AT myang stereomatchingofhighresolutionsatelliteimagesviahierarchicalvitandselfsuperviseddino AT sjiang stereomatchingofhighresolutionsatelliteimagesviahierarchicalvitandselfsuperviseddino AT sjiang stereomatchingofhighresolutionsatelliteimagesviahierarchicalvitandselfsuperviseddino AT wjiang stereomatchingofhighresolutionsatelliteimagesviahierarchicalvitandselfsuperviseddino AT qli stereomatchingofhighresolutionsatelliteimagesviahierarchicalvitandselfsuperviseddino |