Hyperspectral target detection in a whitened space utilizing forward modeling concepts

Abstract

This paper addresses the issue of radiance domain target detection in hyperspectral imagery based on forward modeling of a target reflectance spectrum. The work focuses on taking advantage of generated target spaces and how to incorporate them into a detection scheme. Analysis was performed in a whitened space where lack-of-fit issues can be magnified. From this, two types of detectors were generated, one based on utilizing all vectors in a target space and another, similar detector, utilizing all target space vectors in a lower dimensional space. Receiver operating characteristic (ROC) curve results show that the new detectors perform better than previously implemented methodologies.