Qingchen Fang

Xiaokun Zhang, Yi Yang, Ziqi Ye et al.

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 targets in SAR images. Furthermore, we designed a two-stage SFT strategy to decouple the knowledge injection and task execution of large models. The spatiotemporal feature embedding and the two-stage decoupling paradigm enable FUSAR-GPT to achieve state-of-the-art performance across several typical remote sensing visual-language benchmark tests, significantly outperforming mainstream baseline models by over 12%.

Yi Yang, Xiaokun Zhang, Qingchen Fang et al.

Cross-modal artificial intelligence has garnered widespread attention in recent years, achieving significant progress in the study of natural images. However, existing methods are mostly designed for RGB imagery, leaving a significant gap in modeling synthetic aperture radar (SAR) imagery. SAR, with its all-day, all-weather imaging capabilities, plays an irreplaceable role in remote sensing scene understanding. To address this gap, this paper proposes FUSAR-KLIP, the first universal SAR multimodal foundational model, along with reusable data and evaluation baselines. Specifically: (1) This work introduces the critical yet long-overlooked attribute of geographic information into remote sensing research, constructing FUSAR-GEOVL-1M (the first large-scale SAR dataset with complete geographic projection properties), covering multiple satellite platforms, 120,000 images, and 135 cities. (2) Aligned structured text is generated through a hierarchical cognitive chain-of-thought (HCoT), providing more than one million multi-dimensional semantic annotations of landforms, regional functions, target attributes, and spatial relationships. (3) We design a Self-Consistent Iterative Optimization mechanism that continuously enhances cross-modal alignment through a self-supervised closed loop of contrastive, matching, and reconstruction learning on a transferable multimodal encoder. (4) A unified evaluation benchmark is established across 11 representative downstream vision and vision-language tasks, with comparisons against 14 leading foundation models, where FUSAR-KLIP demonstrates leading performance, particularly in object counting and land-cover classification. We expect that FUSAR-KLIP's large-scale multimodal data, transferable model architecture, and comprehensive experimental benchmark will significantly advance the development of SAR multimodal baseline models.