Hyper-Spectral Image Classification with Support Vector Machine

Abstract

A broad range of works have been proposed and assessed related to airborne and satellite imagery. In this paper, supervised classification learning methods such as K-nearest neighbor (KNN) and support vector machine (SVM) are compared for hyper-spectral data classification. This paper analyzes the potentials of the SVM classifier in hyper-dimensional feature spaces. Remote sensing images are acquired from the hyper-spectral sensors. These sensors are characterized by the higher spectral dimensionality. The results are obtained from the Spyder version 3.7. The accuracy from both the classifiers is assessed and compared. The accuracy obtained from the KNN classifier before applying principal component analysis (PCA) is 60.09% and that obtained after applying PCA is 72.74%. The SVM classifier acquired an accuracy of 88.18%.