Application of the normal compositional model to the analysis of hyperspectral imagery

Abstract

Hyperspectral sensors have been deployed from airborne and spaceborne platforms to collect imaging spectrometry data for environmental, economic, and military applications including scene classification and material identification. A variety of models have been applied to hyperspectral imagery including the normal mixture (NMM), linear mixture (LMM), and subspace (SM) models, for purposes that include developing land cover classification maps, retrieving environmental parameters, detecting objects of interest, and predicting system performance. None of these models account for both subpixel mixing, i.e., multiple material types occupying the same pixel, and intra-class spectral variability. The stochastic mixture model and the normal compositional model (NCM) were defined to explicitly allow for these characteristics, and to bring second order statistical information to bear on compositional problems. In this paper, the normal compositional model is defined, methods of estimating the parameters are described, and applications that demonstrate its utility are presented.