Developing Algorithms for Band Selection in Airborne Hyperspectral Image using Optimization Technique

Abstract

A hyperspectral image (HSI) is made up of hundreds or thousands of continuous spectral bands that could be used to categorise any type of substance on the earth s surface using spectral reflectance. HSI contains similar spatial and spectral information attributes and can be used to improve classification accuracy. It poses considerable hurdles to the remotely sensed scientific community; it adds a higher dimension and also has a high correlation. newlineBand choosing has been one of the potential solutions for dealing with hyperspectral image newlinedata that has 100 to 300 bands. One solution to the high-dimensional issue would be to newlinebuild a model with an objective function that quantifies separability and use it to find a newlinesub-set of bands. Even though HSI contains a massive number of extremely linked spectral newlinebands, it is hard to analyse for three main reasons: 1) fast computation, 2) computing class newlineconditional likelihoods with such a tiny amount of training data, and 3) redundant bands that could behave like noise. The Hughes Phenomenon, also defined as the wrath of dimensional space, is composed of a large number of training datasets for data with much more worthwhile characteristics of bands. To resolve this issue, an effective feature selection or feature extraction method has been required. As a result, the primary aim of the research would be to create an effective and optimized method for selecting features in hyperspectral data. This study proposes an innovative metaheuristic optimization algorithm termed Moth Flame Optimization (MFO) for hyperspectral band selection. This metaheuristic optimization algorithm outperforms numerous different metaheuristic optimization algorithms such as Particle Swarm Optimization (PSO), Cuckoo Search Optimization (CSO), and Grey Wolf Optimization (GWO). In addition to the band selection method, feature extraction has been essential for features extracted from optimal subsets of HSI bands.