Research On SAR ATR Based On Image Domain Feature Sparse Representation

Synthetic aperture radar(SAR)is an active high resolution microwave imaging radar,which can work all day and all weather.Automatic target recognition based on high resolution SAR image has great application in the field of military reconnaissance,mineral exploration,environmental monitoring,which receives much attention from the researchers all around the world,and it is also the focus of this paper.The main work done in this paper is as follows:(1)The characteristics of the target in SAR image are well studied.Based on the study of the distribution characteristics of the scatterers,it is found that the position distribution information of the strong and weak scatterers in SAR image is an identification feature,therefore,a new feature extraction method two dimensional slice Zernike moments(2DS-ZMs)is proposed for extracting that feature in this paper and it is analyzed theoretically,which proves that the 2DS-ZMs is a discriminative feature.(2)Sparse representation theory are studied in this paper.The main focus is centered on the approximation algorithm of sparse representation and sparse representation based classification method.The advantages and disadvantages of the standard sparse representation system are well analyzed.The standard sparse representation based classification recognition algorithm has the advantages of accurately recognizing target and robust to noise corruption.On the other hand,it is pointed out that the standard sparse representation classification recognition algorithm has the disadvantages of solving high dimensional data and runs slowly,which means that it is not suitable for constructing real-time target recognition system.(3)The SAR target recognition algorithm based on 2DS ZMs and non-negative least square sparse representation are proposed in this paper and well researched.The non-negative least squares sparse representation classification model overcome the shortcomings of the standard sparse representation model.Several experiments based on MSTAR dataset are conducted in this paper,and the results proved that the proposed SAR target recognition method based on 2DS-ZMs and non-negative least squares sparse representation classification model performs well in extracting discriminative feature and recognizing target.Furthermore,it runs efficiently and robust to noise corruption.