Enhancing spectral classification using Adaboost

Abstract

Spectral classification for hyperspectral image is a challenging job because of the number of spectral in a hyperspectral image and high dimensional spectral. In this paper, we proposed a method to enhance the spectral classification using the Adaboost for hyperspectral image analysis. By applying the Adaboost algorithm to the classifier, the classification can be executed iteratively by giving weight to the spectral data, thus will reduce the classification error rate. The Adaboost is implemented to spectral angle mapper (SAM), Euclidean distance (ED), and city block distance (CD). From the experimental results, the Adaboost increases the average classification accuracy of 2000 spectral up to 99.63% using the CD. Overall, Adaboost increases the average classification accuracy of ED, CD, and SAM by 2.54%, 1.95%, and 1.67%.