Improved classification and segmentation of hyperspectral images using spectral warping

Abstract

This letter describes a method to increase hyperspectral image classification accuracy (CA) and segmentation accuracy (SA) using spectral warping, which is a nonlinear transformation that warps the frequency content of a signal. In the proposed approach, the frequency content corresponding to spectral data for the hyperspectral image was nonlinearly transformed along the spectral axis using warping. Classification and segmentation algorithms were estimated for the transformed spectral values to show the impact of warping. Experimental results are provided for different values of the warping parameter and it is shown that applying spectral warping increases CA and SA for appropriate warping parameters.