Radio Frequency Interference Artifacts Mitigation in SAR Images Exploiting Semantic Segmentation

Spaceborne synthetic aperture radar (SAR) satellites suffer from the challenge of radio frequency interference (RFI). Existing operational filtering processes fail to detect and mitigate RFI of different intensities, which significantly affects image interpretation applications such as target detection, land monitoring, etc. To deal with this deficiency, this paper proposes an interference mitigation approach that utilizes the enhanced semantic segmentation network to detect and mitigate mixed strong and weak RFI. The proposed network combines residual blocks with encoder-decoder architecture, enabling rapid detection of localized interference variations. Additionally, it incorporates an augmented attention mechanism to extract interference characteristics from the spectrum, enabling efficient interference discrimination. It demonstrates the capability to process massive real-measured SAR images without RFI pre-detection, as well as showcase the generalization performance on unlabeled data using SLC data from multiple scenes and multiple spaceborne SAR systems.