Automated Material Classification Toolset

A fast, inexpensive and accurate means of producing material mixture maps from remotely sensed imagery to support sensor simulation is the goal of this effort. Doing so requires some level of sub-pixel classification to be performed. In Phase I, SOCexamined four areas of potential improvements in sub-pixel classifiers; 1) hardware acceleration, 2) using spatial/spectral algorithms to improve speed and accuracy, and 3) using geo-specific material spectra to drive the classification.Our research showed that 1) hardware acceleration cost effectively improves performance by a factor of 100 or more, 2) spatial algorithms have potential, but require more research before incorporation in the the classifier, and 3) commercial satellitedata is spectrally poor and contaminated by the atmosphere; consequently, the use of geo-specific spectra could only offer marginal improvements.To reach our goals of fast, inexpensive and accurate, SOC proposes the following in Phase II: Construction of a Wintel-based automated classification system using SOC hardware. Continuing research into hybrid spatial/spectral algorithms to improveaccuracy. Design and delivery of a low cost hyperspectral imaging system to explore hyperspectral enhancements of multispectral data and as an end-to-end prototype database acquisition and classification system.