Clustering based band selection for hyperspectral images

An unsupervised band selection method for hyperspectral images is proposed in this article. Three steps are followed to carry out the algorithm. In the first step, characteristics (attributes) of the bands are generated. Next, redundancy among the bands is removed by using clustering. DBSCAN algorithm is used for clustering the bands. One representative band is selected from each cluster. Finally, the bands are ranked based on their discriminating capabilities for classification. To demonstrate the effectiveness of the proposed method, results are compared with a ranking based and two clustering based methods in terms of classification accuracy and Kappa coefficient. Results for the proposed methodology are found to be encouraging.