Successive feature extraction from hyperspectral data

A feature extraction method for hyperspectral data is proposed. Extracting significant features is essential for processing and transmission of hyperspectral data. Conventional ways of extracting features are not always good for hyperspectral data in terms of computation time and optimality. The authors present a feature extraction method designed for significance-weighted supervised classification. After all the data are reduced by principal component analysis, a set of adequate features for the prescribed purpose is extracted successively as linear combinations of reduced components. The method is applied to 500 dimensional hyperspectral data which are required to be classified into five categories. Three features are extracted, which are found to yield high accuracy for classification