On Dimension Reduction of Hyperspectral Images

Abstract

Many dimension reduction (DR) methods have been proposed to deal with the large dimensions of hyperspectral images, but it is unclear which algorithm works best in a given situation. We compare the performance of several DR methods for remote sensing applications. The methods are evaluated by applying them to common datasets and then performing the common image processing tasks of classification, anomaly detection, and spectral unmixing. We evaluate the results against ground truth data and make recommendations for choosing a DR method for certain situations.