Sparse representation based hyperspectral imagery classification via expanded dictionary

Lin He, Weitong Ruan, Yuanqing Li

Abstract: Recently, pattern classification and recognition based on sparse representation have seen a surge of interest in many applications. In this article, we present a method of sparse representation based hyperspectral imagery classification via expanded dictionary. The original spectral signatures in hyperspectral imagery are transformed with 1-D dyadic wavelet transform. Then these wavelet features are combined with the original spectral signatures to form an expanded dictionary. Finally, linear programming is employed to calculate the sparse solution on such a dictionary which was further substituted into related decision rule. Results of experiment on real hyperspectral imagery validate the effectiveness of our method.