Hyperspectral Imagery Classification Based on Gentle AdaBoost and Decision Stumps
Guopeng Yang, Xin Zhou, Xuchu Yu · 2009
Hyperspectral imagery organically includes the spectral information and space information of the ground objects, so it can bring opportunity to ground objects recognition more precisely. Because the performance of many kinds of classifiers can often be dramatically improved by AdaBoost algorithm, in this paper, we introduce the basic procedure of the Discrete AdaBoost algorithm for two-class classification problem, describe the decision stump classifier used as weak learner, and then we bring forward the multiclass Gentle AdaBoost algorithm using hamming loss for hyperspectral imagery classification. Through the experiments of the AVIRIS imagery classification, we can conclude that this method has better generalization capability, faster performance and lower implementation complexity, compared with other common imagery classification methods.