Joint sparse representation of hyperspectral image classification based on optimized dictionary

Inspired by hyperspectral classification algorithm with kernel function, a joint sparse representation classification method based on dictionary optimization (DO-JSRC) is presented to lower the cost of information collection for hyperspectral, as spectral information of hyperspectral images (HSIs) and size of sparse representation dictionary dramatically increases. In this proposed method, we initially select a small number of atoms, calculate the spectral similarity between each atom and the cluster center of the sample through the Gaussian kernel function, and then take the average value. For an atomic dictionary with low spectral similarity, we increase the number of atoms to make it sufficiently representative of this class. The principal component analysis is adopted to extract the principal components of the dictionary after reselecting the atoms, which help to reduce redundant components of the dictionary and facilitates the sparse representation classification. Experiments on the Indian pines datasets show that the presented method can better classify hyperspectral datasets with fewer atoms.