An Algorithm for Azimuth Ambiguities Detection in SAR Images Using Faster-RCNN

In spaceborne synthetic aperture radar (SAR), finite sampling of azimuth Doppler signals will lead to azimuth ambiguities. Similar to the main signal, azimuth ambiguities will reduce the accuracy of target detection. Many suppression methods were put forward with good results. But few people paid attention to azimuth ambiguities' auto recognition and usually mark the ambiguities manually. Based on the research of azimuth ambiguities, for ships on the sea, this paper puts forward a new method to detect and mark azimuth ambiguities automatically in SAR images using Faster-RCNN network of deep learning and mutual information (MI). Based on the Gaofen-3 spaceborne SAR images, the maritime boat dataset (MBD) is conducted, including ships of different resolutions. The Faster-RCNN Network is trained and tested using MBD to detect boats on the sea. According to the azimuth ambiguity distance formula and the mutual information between two areas, the ambiguity will be detected. Experiment results shows this method is feasible.