Target detection from SAR images based on wavelet transform de-noise and improved CFAR

Abstract

Target detection is an important part of an automatic target recognition (ATR) system. There would be many false alarms if using constant false alarm rate (CFAR) algorithm directly on complex synthetic aperture radar (SAR) images with tremendous speckle. Usually, the speckle should be reduced previously before CFAR. In this paper, a wavelet transform de-noise and an improved CFAR algorithm have been combined to detect military targets from SAR image. Different threshold methods were used in the wavelet domain when dealing with the detail information and non-detail information in the image to receive the edge information and reduce the speckle. Then a three-stage CFAR algorithm was used to detect the de-noised image. This algorithm contains global CFAR, local CFAR and count filters. Good results are obtained when the method is used to process high-resolution, HH polarization SAR images. Such algorithms could be arranged in the SAR image based automatic target recognition system.