An optimal nonparametric weighted system for hyperspectral data classification

Abstract

In real situation, gathering enough training samples is difficult and expensive. Assumption of enough training samples is usually not satisfied for high dimensional data. Small training sets usually cause Hughes phenomenon and singularity problems. Feature extraction and feature selection are usual ways to overcome these problems. In this study, an optimal classification system for classifying hyperspectral image data is proposed. It is made up of orthonormal coordinate axes of the feature space. Classification performance of the classification system is much better than the other well-known ones according to the experiment results below. It possesses the advantage of using fewer features and getting better performance.