On Appraisal of Spectral Features Based Supervised Classifications for Hyperspectral Images

Abstract

The classification of hyperspectral images is a challenging task since it contains unbalanced ratio between the training and testing samples, and number of spectral bands. The detailed spectral data of hyperspectral images increases the ability to individualize the different classes and achieving accurate classification maps. Hence, in this paper, we use spectral data for classification and we address the performance of different supervised classification techniques like logic-based, ensemble-based, statistical-based, non-probabilistic-based and instance-based techniques on spectral features. Experiments are carried out using hyperspectral imagery captured by AVIRIS sensor such as Indian Pines, Salinas and Salinas-A. The appraisal of these supervised classification methods are held with each other in terms of performance metrics such as overall accuracy, precision, recall, F1-score and execution time.