Spectral Band Selection Using Binary Gray Wolf Optimizer and Signal to Noise Ration Measure

Abstract

In remote sensing, spectral band selection has been a primordial step to improve the classification of hyperspectral images. It aims at finding the most important information from a set of bands by eliminating the irrelevant, noisy, and highly correlated bands. In this paper, the band selection problem is regarded as a combinatorial optimization problem. We propose a new band selection approach for hyperspectral image classification based on the Gray Wolf Optimizer (GWO) which is a new meta-heuristic that simulate the hunting process of gray wolf in nature. A new binary version of GWO based on transfer function is proposed. In addition, a new fitness function is designed using two terms: the first term is the SVM classifier and the second term of the fitness function is SNR measure (Signal to Noise Ration) which measures the capacity of discrimination. The proposed approach is benchmarked on three hyperspectral images widely used in band selection and hyperspectral images classification. The experimental results show that this approach is suitable to the challenging problem of spectral band selection and provides a higher classification accuracy rate compared to the other band selection methods.