Extraction of Floating Raft Aquaculture Areas from Sentinel-1 SAR Images by a Dense Residual U-Net Model with Pre-Trained ResNet34 as the Encoder

Marine floating raft aquaculture (FRA) monitoring is significant for marine ecological environment and food security assessment. Synthetic aperture radar (SAR) based monitoring is considered to be the effective way for FRA identification because of its capability for the all-weather application. Considering the poor generalization ability and extraction accuracy of traditional monitoring methods, we proposed a semantic segmentation model called D-ResUnet to extract FRA areas from Sentinel-1 images. It has the U-Net-like structure but combines the pre-trained ResNet34 as the encoder and dense residual units in the decoder. The experiments showed the effectiveness and superiority of FRA extraction based on the proposed D-ResUnet compared with the other three state-of-the-art semantic segmentation models.