Model acquisition and invariant tracking of unknown materials in hyperspectral images

consider the problem of acquiring models for unknown materials in airborne 0.4-2.5 mum hyperspectral imagery and using these models to identify the un known materials in image data obtained under significantly different condit ions. The material models are generated with use of an airborne sensor spec trum measured under unknown conditions and a physical model for spectral va riability. For computational efficiency, the material models are represente d by using low-dimensional spectral subspaces. We demonstrate the effective ness of the material models by using a set of material tracking experiments in HYDICE images acquired in forest and desert environments over widely va rying conditions. We show that techniques based on the new representation s ignificantly outperform methods based on direct spectral matching. (C) 2001 Optical Society of America.