Band Selection using Reinforcement Learning for Hyperspectral Images followed by CNN-based Classification

Hyperspectral imaging (HSI) is a hot topic in the analysis of remote sensing data due to the large quantity of information included in these kinds of images, which enables a better characterization and utilization of the Earth’s surface by integrating rich spectral and spatial information. The basic goal of hyperspectral imaging involves band selection, and feature extraction, followed by classification. The choice of bands is a crucial step in the effective processing of hyperspectral pictures. It describes the method of picking the bands in a hyperspectral image that are most important. By selecting a limited number of optimal bands, we aim at speeding up model training, improving accuracy, or both. In this research, we compare all the discussed methodologies in-depth by first providing quantitative findings utilizing well-known and often used HSI situations. In order to determine which paradigms and methodologies are most appropriate, we will evaluate and review the ones that are now in use. Second, we will use reinforcement learning to train an intelligent agent that, given a hyperspectral image, can automatically learn a policy to choose the best band subset