Weak Energy Interference Suppression in InSAR Image Using Semantic Segmentation Network

Spaceborne interferometric synthetic aperture radar (InSAR) satellite, serves multiple industries, including land management, earthquake analysis, mapping, environmental monitoring, disaster reduction, and forestry. However, the presence of radio frequency interference (RFI) makes it difficult to obtain high-quality digital elevation model (DEM) and to perform deformation monitoring. In this paper, a RFI suppression model based on Residual Attention UNet (RA-UNet) is proposed for the Lutan-1 ground processing system, which is capable of processing massive images quickly and efficiently. The model combines attention mechanism and residual block to quickly focus on the changes in the local interference region, and uses UNet to extract the interference features in the spectrum to distinguish the interference. Under different RFI conditions, a wide range of experimental results are provided in different scenarios of the real measured data of Lutan-1, including single-pass and repeat-pass modes. The experimental results demonstrate the effectiveness, efficiency and robustness of the proposed method in interference suppression.