Multilayer CFAR Detection of Ship Targets in Very High Resolution SAR Images

Abstract

This letter proposes a new ship target detection method for very high resolution (VHR) synthetic aperture radar (SAR) images based on multilayer constant false alarm rate (CFAR). First, combined with log-normal distribution, a multilayer CFAR method is designed to overcome the holes and the fracture in the traditional detected results. This method can retain more details of ships and takes much less time than the traditional CFAR method for VHR SAR images. Second, based on a priori knowledge of ships, we use the sliding window to remove the false alarm targets. Finally, In order to measure the size and shape of a ship, we extract the outline of a ship and fill it by a level set method. Experimental results, carried out on real SAR images, demonstrate that the proposed approach outperforms the previous one in terms of the detection ratio of pixels instead of the number of ships.