Statistical Perspective on Hyper Spectral Classification Systems for Accuracy Improvement

Abstract

Classification on a hyperspectral imagery data is a multi-domain problem, it involves segmentation, followed by feature extraction (FE) & selection and finally classification. The vast majority of work in processing of hyperspectral imagery data is done in the field of image classification itself, due to the fact that most of the hyperspectral images are captured in order to evaluate the areas where a particular type of event is occurring, these events range from crop growth, forest covers and military applications. These systems use an algorithm for each of the given steps individually in order to evaluate the accuracy of the system under test. Thus, various algorithms have been proposed in order to evaluate the classification performance of hyperspectral systems. Due to so many algorithms in the field of research, there is a lot of confusion as to which approach should be selected for an effective system. Thus, we need to find approaches which have good accuracy. In order to find the best approaches for classification, researchers have to generally study a plethora of papers, so in this paper, we compare a set of algorithms used for hyperspectral image classification and compare their performance so that the researchers reading this text can analyses these algorithms and select the ones which are best suited for their particular application. Moreover, recommendations are also made in order to further improve the performance of these systems.