Discriminative Feature Learning for Real-Time SAR Automatic Target Recognition With the Nonlinear Analysis Cosparse Model

This letter presents an efficient application of the nonlinear analysis cosparse model (NACM) to the task of real-time synthetic aperture radar automatic target recognition (ATR). In contrast to the conventional synthesis sparse representation model, NACM enables efficient sparse feature extraction and selection using a feed-forward mechanism. Furthermore, NACM does not require a sparsity-inducing regularizer. This model uses a task-driven learning framework, in which a naive Bayes or a discriminative classifier is adaptively learned along with the regularized features. Experimental results with the moving and stationary target acquisition and recognition benchmark demonstrate the effectiveness and efficiency of our proposed approach. Compared with traditional classification algorithms using sparse representation, our approach not only achieves higher or comparable recognition accuracy but also dramatically reduces the execution time for real-time ATR.