Flooded Cropland Identification Using SAR Data and Transfer Deep Learning

Synthetic aperture radar (SAR) technology has significant advantages in flood detection. However, deep learning models trained on one type of SAR satellite data often perform poorly on other SAR data sources. To address the challenge of transferring deep learning models between different SAR datasets, we propose a flood detection method that combines self-supervision with ResUNet. Experiments conducted in the vicinity of Wuchang City demonstrate that our proposed method effectively transfers the Sentinel-I data training model to LT -1 data and improves the accuracy of flood detection.