Abstract

Imaging spectrometers can collect data in a large number of narrow bands providing a low spectral resolution spectrum for each pixel. Pixel to pixel spectral variations can be attributed to varying concentrations of constituents and to environmental variations. Identification of the material constituents provides for both interpretation and compression of the data. Identifying constituents by directly comparing data to laboratory spectra is hindered by pixel foot print size, typically more than one constituent contributes to each pixel, and by environment effects. Environmental effects, such as variations in illumination, atmospheric scattering and transport effects, are generally not included in the reference spectra. Analysis approaches which remove or reduce the effects of the environment and produce apparent reflectances are desired.¹ It is also desirable to determine constituents directly from the data and to rely on reference libraries for later confirmation or analysis. This effort focuses on the constituent analysis problem. The estimation of spectral components from the data has been based on finding convex hulls of data base spectra.^2,3 The approach reported on here finds bounds on the possible component spectra and abundances. These bounds can be narrowed by using additional information or constraints on spatial characteristics or knowledge of one or more of the constituents. In this summary, the method is outlined and preliminary results are discussed.

© 1995 Optical Society of America