Development of a Dual-Attention U-Net Model for Sea Ice and Open Water Classification on SAR Images

This study develops a deep learning (DL) model to classify the sea ice and open water from synthetic aperture radar (SAR) images. We use U-Net, well-known fully convolutional network (FCN) for pixel-level segmentation, as backbone. employ DL-based feature extracting model, ResNet-34, encoder of U-Net. To achieve high accuracy classifications, we integrate dual-attention mechanism into original U-Net improve representations, forming (DAU-Net). The SAR images are obtained Sentinel-1A. dual-polarized information incident angle inputs. used 15 acquired near Bering Sea train other three test model. Experiments show that DAU-Net could classification; can classification accuracy. Compared with improves intersection over union (IoU) by 7.48.% points, 0.96.% 0.83.% points on recently published DenseNet FCN , improvement IoU values 3.04.% 2.53.% 2.26.% respectively.