Classification of Hyperspectral Images with Support Vector Machines

Abstract

In this study we investigate the performance of the Support Vector Machines SVM) classifier when applied to the classification of high dimensional remotely sensed image data. As SVM deals with a pair of classes at a time, we propose its implementation in a binary tree approach where two classes only are dealt with at each node. The accuracy of the thematic  image produced by this classification
scheme was evaluated for two different  kernel functions and different data dimensionality. Tests were performed using hiperspectral image data collected by the sensor system AVIRIS. Results are presented and discussed.