Hyperspectral Image Classification Using a New Dictionary Learning Approach with Structured Sparse Representation

This paper introduces a new dictionary learning approach for hyperspectral images classification with structured sparse representation based on Compressed Sensing (CS), An important contribution of our paper is partition the pixels of a hyperspectral image into a number of spatial neighborhoods called pixel groups and the pixel group can be modeled of different size. The idea is to use of hyperspectral remote sensing image spatial correlation between pixels and the aim is to obtain a dictionary of each pixel. The dictionary is a linear combination of a few dictionary elements learned from the hyperspectral data and can accurately represent hyperspectral remote sensing images with less coefficients. The pixels are induced a common sparsity pattern and have a implicitly spectral correlation between pixels which are in a identical pixel group. The sparse coefficients are then used for classification hyperspectral images by a linear Support Vector Machine. The experiments show that the proposed method can get a better representation of hyperspectral images and has a higher overall accuracy and Kappa coefficients.