SAR Target Pose Estimation Based On Ridgelet Transform

SAR(Synthetic Aperture Radar) ATR(Automatic Target Recognition) is crucial to the modern military battle and has become a very hot research topic.In recent years,radar target recognition has made steady progress in many fields,including target feature extraction, classification and recognition.Some ATR systems have been carried out and have been successfully used in some fields such as ground detecting and precision guidance in spaceborne/airborne SAR.In SAR ATR,the pose information of target is very important.Since SAR is highly sensitive to target pose,the appearance of a target in SAR image is different at different azimuth.It is very important to estimate target pose estimation before SAR ATR,which can improve the ATR system and improve the system's computational speed.In this dissertation, SAR image target pose estimation is deeply researched.The main contents of this paper are as follows.Firstly,the characteristics of wavelet transform and ridgelet transform are analyzed.The differences in expressing the features of signals among the wavelet transform,Radon transform and ridgelet transform are mainly discussed.The relationship between angular and ridgelet power spectrum is deduced for target pose estimation.Then,several methods of SAR target pose estimation are researched.According to the good direction features of ridgelet,an efficient method of target pose estimation based on ridgelet transform is proposed.In this method,with analyzing the difference ridgelet power at different direction,the maximum angular-ridgelet energy spectrum is used to estimate targets azimuth.Meanwhile,in order to estimate pose more accurately,it is necessary to segment the target from SAR image previously.SAR image statistic property and the distribution parameters estimation are also analyzed.A new approach of SAR target and shadow segmentation based on Weibull distribution is proposed.Experimental results with MSTAR (Moving and Stationary Target Acquisition and Recognition) datasets show a better estimation performance with our proposed approach.