LAUSR

ABSTRACT Floods are among the most devastating natural disasters. Most current methods for detecting flood-affected areas use single-polarized SAR images and traditional networks. However, multi-polarized SAR data and tailored feature extraction modules can enhance edge localization accuracy in complex environments, offering more detailed and reliable flood detection. Therefore, we generate a fused image based on dual polarized SAR data and propose a novel change detection network (PWU-Net) to improve detection accuracy. We extract a texture feature map from VV-polarized and VH-polarized data using the Gray-Level Co-occurrence Matrix(GLCM) and combine it with the VV and VH bands to create a three-channel fused image. Moreover, the proposed PWU-Net is an encoder-decoder network that extracts fusion change features through a Convolution-CoordAtt Module (CCAM) and enhances sensitivity to subtle changes within the fused image using the Wave-MLP module. Experimental results based on Sentinel-1 SAR images from the 2023 Beijing flood event demonstrate that the fused image not only amplifies detailed polarization information but also highlights distinct flood texture characteristics. Additionally, the PWU-Net outperforms existing advanced methods in terms of Precision, Recall, and F1 score for flood area detection. Notably, the combination of the fused image and PWU-Net results in a 9.1% improvement in F1 score.