Target detection for SAR images based on beamlet transform

Target detection for SAR images has many important applications; however there is a challenge that inherent speckle noise in SAR images may cause serious interference. Beamlet transform is a multi-scale image analysis method to extract line features in an image with strong anti-noise capacity. In this paper a method based on Beamlet transform is proposed for target detection for SAR images. It takes the advantage of Beamlet transform in feature extraction. Firstly Beamlet transform is applied on a SAR image to obtain Beamlet coefficients,which are then processed by a coefficient filtering algorithm to remove unreal Beamlet features caused by noise. The remained Beamlet features are fed to the BD-RDP (Beamlet-decorated recursive dyadic partition) algorithm for optimization and then clustered by NEC (Nearest Endpoint Clustering) algorithm to detect targets. The experimental results show that this method is able to detect target directly in a SAR image without pre-filtering. Further more, it still works well under the background of strong speckle noise.