A Binary Multi-objective CLONAL Algorithm for Band Selection in Hyper-Spectral Images

Abstract

A hyperspectral image collects the spectral information across the electromagnetic spectrum as a continuous set of hundreds of bands with a very narrow bandwidths ranging from 5-10nm. The bands are high correlated with each other and some of the bands carry negligible information. So, there is a need to reduce the dimensionality of the hyperspectral image. Thus the aim is to extract only those bands which are informative and eliminate the noisy and redundant bands. If the redundant bands are eliminated, the dimensionality of the hyperspectral image reduces. This reduction in dimensionality reduces the cost of computation and increases the accuracy of the analysis. In this paper, a dimension reduction technique based on binary multi-objective CLONAL algorithm has been proposed. The method uses two objective functions as entropy and Pearson correlation. The extracted spectral bands are taken and extracted hyperspectral image obtained is segmented using K-modes algorithm. Comparative results reveal superior performance of band selection by the proposed method over NSGA-II, BSSO and PCA based reduction.