Study On Target Recognition Based On Tensor Learning

Automatic target recognition(ATR) based on synthetic aperture radar(SAR) images is of great importance in the modern battlefield and has become a very hot research topic.According to exoterica of tensor learning which was proposed by Tao,Deng and so on,we extend conventional target recodnition,which included Principal Component Analysis(PCA),Linear Discriminant Analysis(LDA)and Support Vector Machine (SVM),to Tensor Principal Component Analysis(TPCA), Tensor Linear Discriminant Analysis(TLDA)and Support Tensor Machine(STM)。In this paper,we do lots of experiments to verify forenamed methods.In order to overcome the problem of experiments,we present two improved methods.To solve the problem of TLDA,whose recognition performances is low at small feature dimensions ,we present one improved method which can improve it.Existing STM algorithms is solved basing on iterative method.For an two-dimensional vector,one projective vector was in the optimisation-classified plane of another projective vector. It's not meaning the data of sample can be classified farthest by this projective vector.In this paper,according to optimal project method,we presented an mend STM method.Basing on Fisher discriminant analysis,we search for an optimal project along random direction of the image,and it effectively improves on the convengent rate of STM,saves lots of experimental time and improves the recognition,and the projective data has farthest divisibility.In the end,we extend multiclass classification method with SVM arithmetic to multiclass classification method with STM arithmetic and multiclass classification method with OPSTM arithmetic,give the algorithmic process.it is validated to real data by experiment.