Sparse representation based on stacked kernel for target detection in hyperspectral imagery

Conventional sparse representation gets poor performance in nonlinear information processing for target detection in hyperspectral images (HSI). In this paper, a novel sparse representation based on stacked kernel is proposed for target detection in HSI. This method uses several different kinds of stacked kernel function to project nonlinear information contained by the hypercube into a new feature space in which the data becomes linear separable to promote high level of detection accuracy. Then, the algorithm, simultaneous orthogonal matching pursuit (SOMP), is used to solve the convex relaxation techniques. Experiment results demonstrate that the sparse representation method with stacked kernel for target detection further increases the detection accuracy. (C) 2015 Elsevier GmbH. All rights reserved.