An Interference Suppression Method For Spaceborne Sar Image Via Space-Channel Attention Network

Abstract

Spaceborne synthetic aperture radar (SAR) is becoming a widely used sensor in the universe because of its ability to acquire high-resolution radar images under harsh conditions such as clouds, fog, rain, and snow. However, the presence of electromagnetic interference in cosmic space can seriously affect the efficiency of information detection of spaceborne SAR. To address the problem of degradation of spaceborne SAR images, we present the Space-Channel Attention Network (SCANet). Our model structure consists of two branches, the channel and spatial transform branch (CSTB) and the detail restoration branch (DRB). Among them, CSTB, with attention mechanisms in both channel and spatial dimensions, greatly preserves the texture features of the original images while eliminating interference. In addition, DRB recovers precise details of CSTB output feature maps guided by ground truth. We have verified the feasibility of the method by adopting real Sentinel-I data. Compared with some denoising algorithms in field of computer vision, Our method attains the best interference suppression performance with high-resolution image output.