Abstract

Spectral pattern recognition in IR remote sensing is the process of identifying an "event," usually the presence of a particular substance, from a spectrum produced or in some way influenced by the event. Ideally, the process will correctly classify all events of interest and reject all background, interference and noise events associated with the sensing process. It is a statistical problem, since the classes and/or their associated spectra are not unique. For example, a gas of interest may have a well-defined absorption coefficient signature, but can be present in varying amounts; its influence in an observed spectrum may also depend on concentration/temperature distributions along the line-of-sight.

© 1983 Optical Society of America