Cross-image modality flood change detection based on heterogeneous SAR data

Remote sensing flood change detection(FCD) plays a crucial role in disaster monitoring. This paper proposes a crossmodal FCD framework consisting of a translation module and a CD module to address the challenge of decreased detection accuracy caused by cross-modal disparities between two distinct SAR images. The translation module adopts a multiscale parallel encoding and multibranch decoding design, incorporating dilated convolutions and a spatial attention mechanism, which effectively correct the differences between Sentinel-1(S1) and UAVSAR images. The CD module performs image differencing, classifies changed pixels using a Gaussian mixture model, and refines results through morphological operations. Experimental results show that the proposed method surpasses existing advanced technologies in both detection accuracy and robustness.