An Algorithm for Azimuth Ambiguities Detection on High Resolution SAR Images Using SSD

Abstract

Azimuth ambiguities in spaceborne synthetic aperture radar (SAR) will introduce multiple similar targets which will affect image quality seriously. Also, it will directly affect the correct rate of SAR image target detection. Many suppression methods were put forward, which made good results. But most of suppression methods need manually mark the azimuth ambiguities, which is obviously not suitable when processing huge amount of SAR images. For targets on the sea, this paper puts forward a new method to detect and mark azimuth ambiguities automatically in high resolution SAR images using single shot multibox detector (SSD) network of deep learning. Based on the Gaofen-3 spaceborne SAR images, the high resolution maritime ambiguities dataset (HRMAD) is conducted, including the main targets and ambiguities of boats, container and windmills. The SSD network is trained and tested using HRMAD. Experimental results show that the azimuth ambiguities on the sea can be detected automatically.