The effects of limiting spectral resolution and the number of bands on HSI target detection algorithm performance

A number of object recognition techniques including spectral matched filter algorithms have been successfully applied to the detection of targets and the characterization of backgrounds in hyperspectral images (HSI) over the past decade or so. Normally, advantage is taken of information in as many spectral bands as are available and the spectral regions most crucial to effective performance are never identified. The study reported in this paper demonstrates the effects on algorithm performance of degrading the spectral resolution and limiting the spectral extent of the HSI data. The results, although they display strongly target specific effects, provide indications of the extent to which HSI sensors may be simplified by reducing resolution. They also suggest which bands and how many may be required to accomplish various levels of confidence in target recognition.