FWENet: a deep convolutional neural network for flood water body extraction based on SAR images

As one of the most severe natural disasters in world, floods caused substantial economic losses and casualties every year. Timely accurate acquisition flood inundation extent could provide technical support for relevant departments field emergency response disaster relief. Given accuracy existing research works extracting based on Synthetic Aperture Radar (SAR) images deep learning methods is relatively low, this study utilized Sentinel-1 SAR as data source proposed a novel model named water body extraction convolutional neural network (FWENet) information extraction. Then three classical semantic segmentation models (UNet, Deeplab v3 UNet++) two traditional (Otsu global thresholding method Object-Oriented method) were compared with FWENet model. Furthermore, paper analyzed area change situations Poyang Lake. The main results follows: Compared other five methods, achieved highest accuracy, its F1 score mean intersection over union (mIoU) 0.9871 0.9808, respectively. This guarantee subsequent images.