Fast Automatic Targeting Algorithms for Hyperspectral Sensors

“Smart” and precision-guided Army weapons systems are being developed that use infrared (IR) hyperspectral imaging (HSI) sensors for target detection and classification. Both the long-wavelength IR and mid-wavelength IR regions provide day/night, all-weather capabilities and discrimination based on thermal and emissivity contrast. New efficient and fast algorithms are needed to process the image signals rapidly enough for weapons targeting. A major challenge is that the IR spectral signature of a given object varies with surface temperature, orientation, sensor range sensor, and atmospheric conditions, forming a multidimensional subspace. Rapid targeting requires the ability to distinguish spectra in this subspace from other, non-target spectra selectively and rapidly (much faster than by using look-up tables). Spectral Sciences, Inc. (SSI) hereby proposes to develop a novel IR HSI data processing system suitable for real-time weapons targeting. A combination of subspace compression and constrained fitting algorithms will yield both high speed and high target selectivity. A hybrid approach that merges measurements with first-principles simulations will be used to generate test imagery containing realistic spectral-spatial properties. The results from Phase I will be used to define algorithms to be implemented in Phase II real-time processing hardware.