Hyperspectral Band Selection using Mutual Information

Hyperspectral band selection (BS) is an intrinsic problem in hyperspectral image processing. This paper proposes a novel hyperspectral BS method based upon mutual information (MI). The proposed method tries to find the least redundant and most informative band subset from the whole image data cube iteratively. The informative and redundancy qualities of a band are measured by its MI with the set of unselected bands and the set of selected bands. Experimental findings on two different datasets reveal that the presented method can attain comparable performance for the classification of the hyperspectral image dataset compared to the other ranking based band selection method.