Feature extraction based on kernel sparse representation for hyperspectral image classification

Feature extraction is a promising technique for hyperspectral image classification. Recent research has shown that the criterion of sparse representation classification (SRC) can help to design a feature extraction method. This method is called the SRC steered discriminative projection (SRCDP). Motivated by the fact that kernel trick can exploit the nonlinear case of features, this paper generalizes SRCDP to its kernel case named KSRCDP. Extensive experiments show that KSRCDP can obtain excellent classification performance on two classic hyperspectral images.