Deep Learning based Techniques on Hyperspectral Imaging

Abstract

There are certain challenges when dealing with hyperspectral images including limited training samples, high dimensionality of the data, Hughes phenomena, handling the noisy information, problem of overfitting and information loss. The work presented here describes the various techniques to handle the challenges above while working with the hyperspectral images and their applications. The mostly used classification algorithms are the k nearest neighbor algorithm, the support vector machine, Convolutional neural networks, Decision Trees, Random Forests, and other deep learning algorithms. The work reviews in detail the above algorithms and their performance based on few publicly available hyperspectral images datasets.