Abstract—Flooding, a common natural disaster, causes widespread damage globally. Detecting flood extents rapidly and accurately using Synthetic Aperture Radar (SAR) images is crucial for effective disaster response and mitigation. This paper proposes a novel machine learning model specifically designed for SAR image analysis to detect floodwaters. The model leverages change detection techniques and operates on pairs of satellite images captured at different time points. The feature extraction module employs a parallel Siamese architecture with a Swin-Transformer backbone to extract features at various levels. Prior to entering the decoding module, the features undergo enhancement by computing the difference between feature maps at the same level. The decoding process predicts changing regions at each level and integrates them into the final result. Experimental results demonstrate that our proposed model outperforms other methods, achieving a recall of 94.6%, a precision of 96.9%, and an F1-score of 95.7%, with a computational cost of 32.3 G FLOPs.
 
Keywords—flood detection, deep learning model, Synthetic Aperture Radar (SAR) image, Swin-Transformer, vision transformer


Copyright © 2025 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).