Spectral mixing of camouflaged targets

Abstract

A fundamental problem in target detection is the separation of a target and its background, particularly when the target is camouflaged. It is possible to discern camouflaged objects in vegetative backgrounds using reflected light in the visible and infrared range. Reflectance data was taken of five camouflage nets draped over various vehicles with a predominately green background. The aim of this analysis was to reconstruct the spectrum of the observed scene using a linear combination of individual basis spectra called "pure" endmembers. Linear spectral mixing assumes that the observed spectral radiance may be modeled as a linear combination of members of a "pure" endmember spectral mixing library. The computer algorithm written for this analysis demonstrated the ability to use linear spectral mixing to reconstruct an observed spectrum. The analysis of the abundance mixtures showed that consistent exploitable patterns exist with this type of data. The task of reconstructing the observed spectra was performed with a crude, "non-pure" endmember library. Even greater success could be achieved with a more sophisticated and complete library.