Exploiting Sparsity and Dictionary Learning to Efficiently Classify Materials in Hyperspectral Imagery