Finding Ship Radars in SAR Images: Localizing Radio Frequency Interference Using Unsupervised Deep Learning

Abstract

Synthetic Aperture Radar (SAR) satellite images are used increasingly more to observe the maritime environment, but they sometimes experience image degradation caused by interfering signals from external radars. Few on-ground radars can cause Radio Frequency Interference (RFI) and the RFI information can therefore increase domain awareness. Localizing and characterizing RFI signals in the ocean might help classify otherwise overlooked ships as, e.g., potential navy ships. In this study, we detect and localize RFI signals automatically in Sentinel-1 quick-look images. The spatial structure of RFI signals vary greatly and unsupervised deep learning was therefore used to reconstruct RFI-free Sentinel-1 images. Anomaly heat-maps were then computed to localize RFI anomalies in the images under varying environmental and geographical conditions. We localized several RFI signals mid-sea believed to be caused by ship-borne air-surveillance radars. This study shows that more information can be extracted from certain detected objects, such as ships, from SAR images.