Foggy-UAVCD and DSRF-Net: Benchmarking robust satellite–UAV change detection under fog degradation

Synergistic observation using multimodal remote sensing data is crucial for rapid geographic monitoring. However, acquiring large-scale multimodal datasets remains a significant challenge. While Synthetic Aperture Radar (SAR) is often used for all-weather monitoring, its inherent speckle noise and limited semantic interpretability constrain its application in fine-grained change detection (CD). In contrast, Unmanned Aerial Vehicle (UAV) imagery offers superior spatial resolution and flexible deployment, yet its optical sensors are highly susceptible to adverse weather. In practical emergency response, fog degradation frequently introduces severe visibility reduction and pseudo-changes, creating a gap between ideal laboratory settings and real-world deployment. To address these challenges, we first construct Foggy-UAVCD, a novel heterogeneous benchmark specifically designed for SAT–UAV CD under foggy conditions. By simulating controllable fog-induced visibility degradation, this dataset serves as a specialized benchmark for evaluating cross-modal geometric alignment and robust feature extraction. Furthermore, we propose the Dual-Stage Rectification and Fusion Network (DSRF-Net), a lightweight yet robust architecture. Specifically, a Cross-modal Feature Rectification (CFR) module is designed to suppress fog-induced noise by leveraging satellite priors, while a Dual-Domain Fusion Upsampler (DDFU) module is developed to recover high-frequency boundary details. Experimental results on the Foggy-UAVCD and HTCD datasets demonstrate that DSRF-Net achieves a competitive trade-off between accuracy and efficiency, with only 3.43 M parameters and 1.84 G floating point operations (FLOPs), indicating potential for efficient adverse-weather remote sensing analysis, while further validation across broader geographic regions, climate zones, and seasonal conditions is still required. Code and dataset will be available at https://github.com/xiaobing-min/DSRFNet .