A Water Segmentation Algorithm for SAR Image Based on Dense Depthwise Separable Convolution

Abstract

Water segmentation of real SAR images is of great significance in military and civilian applications such as ship target detection and disaster monitoring. To solve the issues of poor robustness and inaccurate segmentation of traditional water segmentation algorithms, this paper first establishes a SAR water segmentation dataset based on the GF3 satellite and then presents a segmentation network architecture based on depthwise separable convolution. The network takes real SAR images as inputs, extracts high-dimensional features through depthwise separable and dilated convolutions, constructs an up-sampling and decoding module based on bilinear interpolation, and then outputs the corresponding segmentation results. The segmentation results of a water segmentation dataset show that the proposed segmentation method remarkably improves the segmentation accuracy, the segmentation robustness and running speed than traditional method. Therefore, the findings demonstrate the excellent practical engineering value of the proposed algorithm.

Keywords

• Synthetic Aperture Radar (SAR)
• Water segmentation
• Deep learning
• Dense separable convolution
• Feature extraction