Performance analysis of different classifier for remote sensing application

Abstract

The classification of remotely sensed data on thematic map is a challenging task from very long time and it is also a goal of today’s remote sensing because of complexity level of earth surface and selection of suitable classification technique. Hence selection of best classification technique in remote sensing will give better result. Classification of remotely sensed data is an important task within the domain of remote sensing and it is outlined as processing technique that uses a systematic approach to group the pixels into different classes. In this study, we have classified the multispectral data of Udupi district, Karnataka, India using different classifier including Support Vector Machine (SVM), Maximum Likelihood, Minimum Distance and Mahalanobis Distance classifier. The data of dimension 3980x3201 pixels are collected from a Landsat-3 satellite. Performance of the each classifier is compared by conducting accuracy assessment test and Kappa analysis. The obtained results shows that SVM will give accuracy of 95.35% and kappa value of 0.9408 respectively when compared other classifier, hence effectiveness of SVM is a good choice for classifying remotely sensed data.