Joint low rank and sparse representation-based hyperspectral image classification

Representation-based classification has gained great interest recently. In this paper, we present a novel joint low rank and sparse representation-based classification (JLRSRC) method for hyperspectral imagery. For a testing set, JLRSRC seeks weight coefficients to represent a testing pixel as linear combination of atoms in an over-complete dictionary. Since the low rank model is capable of preserving global data structures of data while sparsity can select the discriminative neighbors in the feature space, the resulting representation is both representative and discriminative. Experimental results demonstrate the effectiveness of the proposed JLRSRC when compared with the traditional counterparts.