![[2602.19190] FUSAR-GPT : A Spatiotemporal Feature-Embedded and Two-Stage Decoupled Visual Language Model for SAR Imagery](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.19190] FUSAR-GPT : A Spatiotemporal Feature-Embedded and Two-Stage Decoupled Visual Language Model for SAR Imagery
Summary
FUSAR-GPT is a novel visual language model designed for interpreting SAR imagery, enhancing performance through spatiotemporal feature embedding and a two-stage decoupling approach.
Why It Matters
This research addresses the limitations of existing visual language models in the SAR domain, which is critical for remote sensing applications. By introducing a specialized dataset and innovative model architecture, FUSAR-GPT significantly improves the interpretation of SAR images, which can lead to advancements in various fields such as environmental monitoring and disaster response.
Key Takeaways
- FUSAR-GPT introduces a new dataset for SAR imagery interpretation.
- The model employs spatiotemporal feature embedding to enhance performance.
- A two-stage decoupling strategy improves knowledge injection and task execution.
- FUSAR-GPT outperforms existing models by over 12% on benchmark tests.
- The advancements can significantly impact remote sensing applications.
Computer Science > Computer Vision and Pattern Recognition arXiv:2602.19190 (cs) [Submitted on 22 Feb 2026] Title:FUSAR-GPT : A Spatiotemporal Feature-Embedded and Two-Stage Decoupled Visual Language Model for SAR Imagery Authors:Xiaokun Zhang, Yi Yang, Ziqi Ye, Baiyun, Xiaorong Guo, Qingchen Fang, Ruyi Zhang, Xinpeng Zhou, Haipeng Wang View a PDF of the paper titled FUSAR-GPT : A Spatiotemporal Feature-Embedded and Two-Stage Decoupled Visual Language Model for SAR Imagery, by Xiaokun Zhang and 8 other authors View PDF HTML (experimental) Abstract:Research on the intelligent interpretation of all-weather, all-time Synthetic Aperture Radar (SAR) is crucial for advancing remote sensing applications. In recent years, although Visual Language Models (VLMs) have demonstrated strong open-world understanding capabilities on RGB images, their performance is severely limited when directly applied to the SAR field due to the complexity of the imaging mechanism, sensitivity to scattering features, and the scarcity of high-quality text corpora. To systematically address this issue, we constructed the inaugural SAR Image-Text-AlphaEarth feature triplet dataset and developed FUSAR-GPT, a VLM specifically for SAR. FUSAR-GPT innovatively introduces a geospatial baseline model as a 'world knowledge' prior and embeds multi-source remote-sensing temporal features into the model's visual backbone via 'spatiotemporal anchors', enabling dynamic compensation for the sparse representation of targ...
