A NOVEL CLASSIFICATION PROCESSING BASED ON THE SPATIAL INFORMATION AND THE CONCEPT OF ADABOOST FOR HYPERSPECTRAL IMAGE CLASSIFICATION Bor-Chen Kuo 1 Shih-Syun Lin 1 Huey-Min Wu 1 Chun-Hsiang Chuang 2 kbc@mail.ntcu.edu.tw catchylss@hotmail.com lhswu@seed.net.tw cch.chuang@gmail.com 1 Graduate School of Educational Measurement and Statistics, National Taichung University, Taiwan, R.O.C. 2 Department of Electrical and Control Engineering, National Chiao Tung University, Taiwan, R.O.C. 1. INTRODUCTION In this paper, a novel classification processing based on the spatial information and the concept of Adaboost [1] for hyperspectral image classification is proposed. This classification process is named adaptive feature extraction with spatial information (AdaFESI). The main idea is adaptive in the sense that subsequent feature spaces are tweaked in favor of those instances misclassified by spectral or spatial classifiers in the previous feature space. This processing includes two concepts for classifying hyperspectral image: (1) For avoiding the Hughes phenomenon, the feature extraction is the important for hyperspectral image classification [2]-[3]. Hence, the feature space at the next round is varied at every round such that it suits for the misclassified samples at this round. The weights of the terms of the scatter matrices corresponding to the samples which are classified correctly at this round will be decreased in the next round. Otherwise, the weights will be increased for the samples which are misclassified. (2) Many studies [4]-[5] show that the performance of the classifier with spatial information outperforms than of the original one. Hence, which one of the spatial classifier and spectral classifier used at every round is determined by their classification performances at this round. The traditional hyperspectral image classification procedure is a special case of our proposed processing because it is the same to perform our proposed method one round without using spatial classifier. Fig. 1 shows the flowchart of AdaFESI. Note that any type of classifier and feature extraction method can be used in our proposed procedure. Fig. 1 The flowchart of AdaFESI Feature Extraction Spectral Classifier or Spatial Classifier Weighted Vote Misclassified Rate Weights for Classifier Weights for terms in Scatter Matrix Hyperspectral Data (Training Samples) Misclassified Samples