To maximize the utility of spectral imagery sensors it is essential that processing techniques for data from these sensors are highly automated, provide high probability of target detection with low false alarm rates, and offer a reliable confidence assessment capability. The detection and false alarm probabilities required for a hyperspectral sensor to be a useful intelligence asset varies based on specific applications, therefore this activity responds in general to the prevailing opinion that current algorithms for spectral processing are not meeting the needed performance goals of the present and potential users of spectral data. Current processing short-falls are especially notable when the spectral backgrounds are complex or highly structured both in the VIS/NIR/SWIR and MWIR/LWIR regions of the spectrum. This Phase I program will be directed toward the development of: 1) Innovative spectral signal processing algorithms and flows that improve current detection, identification, and quantification techniques while simultaneously minimizing false alarm rates. 2) Novel clutter suppression/contrast enhancement algorithms that demonstrate reliable and acceptable performance in the structured environments that inherently exhibit high dimensionality and significant spectral variability. 3) Software prototypes compatible with existing NGA software tools that provide for objective evaluation using data with wavelengths ranging from .4 - 13 microns.
Keywords: Hyperspectral, Detection, Orthogonal Subspace Projection, Identification, Subpixel Mixture, Imaging Spectrometry, Automatic, Exploitation