A Target Discrimination Method Based on Iterative Manifold SVM

To improve the false reject rate of discriminator for automatic target recognition based on synthetic aperture radar, we propose a new target discrimination method based on a modified manifold support vector machine (SVM). Covariance matrix features which combines texture features and their correlation information are used, and the distinguishability of these features are proved to be good by our experiment. An iterative manifold SVM discriminator is designed to better match the covariance matrix features in the non-euclidean space. The center of the hypersphere in SVM instead of the Karcher mean is selected as the base point by a novel iterative algorithm. Experimental results on RADARST-2 database demonstrate the superiority of the proposed method.