VLGeo: Bridging Viewpoints in UAV-View Geo-Localization Using a Large Vision–Language Model
IEEE Transactions on Geoscience and Remote Sensing · 2026 · 64 · 1-13
DOI: 10.1109/tgrs.2026.3680280 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Geoscience and Remote Sensing, 2026. Verfasst von Zhengxin Cao und 7 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 52 Literaturangaben. Gefördert von: Special Projects of the National Natural Science Foundation of China, Key Research and Development Program of Shaanxi Province, Shanghai Xiaoyuan Innovation Center Youth Fund.
Abstract (Originalsprache)
Unmanned aerial vehicle (UAV) geo-localization has garnered significant attention due to its potential to enhance UAV autonomy in scenarios where traditional GPS signals are unavailable or unreliable. Several innovative visual-based approaches have been developed, including deep image- and stereo-matching localization techniques. Nevertheless, extracting common features remains difficult, primarily because of significant appearance discrepancies caused by changes in viewpoint, altitude, and orientation. In this article, we introduce VLGeo, the first text-assistant visual geo-localization benchmark, to enhance UAV-view localization by facilitating cross-view and cross-modal interaction. We crafted detailed textual descriptions to enhance retrieval capabilities by enabling text–image-guided extraction of common features. Our approach entails generating attribute-based prompts through triplet images, thus effectively highlighting the intrinsic differences between images. Furthermore, a bidirectional-guided semantic alignment mechanism and a contrastive learning network are introduced to extract discriminative common features across views and modalities. Finally, a cross-modality dataset, UAV-VLL, is also introduced for evaluating and advancing the integration of visual and textual data for UAV geo-localization. Experimental results indicate that our method significantly enhances localization accuracy, achieving improvements of 4.88% in top-1 recall and 1.99% in average precision (AP) compared to state-of-the-art methods. The code is available at https://github.com/ballball9/VLGeo
Förderung
- Special Projects of the National Natural Science Foundation of China
- Key Research and Development Program of Shaanxi Province
- Shanghai Xiaoyuan Innovation Center Youth Fund
Zitieren
APA
Cao, Z., Tang, L., Zhou, X., Liang, Y., Chen, L., Xu, R., Wang, K., & Gao, T. (2026). VLGeo: Bridging Viewpoints in UAV-View Geo-Localization Using a Large Vision–Language Model. IEEE Transactions on Geoscience and Remote Sensing, 64, 1-13. https://doi.org/10.1109/tgrs.2026.3680280
Vancouver
Cao Z, Tang L, Zhou X, Liang Y, Chen L, Xu R, et al. VLGeo: Bridging Viewpoints in UAV-View Geo-Localization Using a Large Vision–Language Model. IEEE Transactions on Geoscience and Remote Sensing. 2026;64:1-13. doi:10.1109/tgrs.2026.3680280
BibTeX
@article{Cao2026,
author = {Cao, Zhengxin and Tang, Lei and Zhou, Xin and Liang, Yunji and Chen, Longbin and Xu, Ruixiao and Wang, Kaiping and Gao, Tao},
title = {VLGeo: Bridging Viewpoints in UAV-View Geo-Localization Using a Large Vision–Language Model},
journal = {IEEE Transactions on Geoscience and Remote Sensing},
year = {2026},
volume = {64},
pages = {1--13},
doi = {10.1109/tgrs.2026.3680280}
}
EUVIMED