Software Tools for Full Spectrum Analysis of Hyperspectral Data

Current hyperspectral analysis, used to extract information related to nuclear fuel cycle signatures, relies on principal component and matched filter processing, which may not fully exploit all of the information in the data. This project will develop an innovative method for: (1) using Independent Component Analysis (ICA) as a means to find end-members, which will serve as an alternative to principal components; (2) adding rule-based spectroscopic methods as an alternative to matched filter processing, and (3) assembling the tools in a user-friendly software package. The approach will be based on a previously-developed N-FINDR end-member determination and unmixing package, along with an associated microcorder spectral-identification tool. In Phase I, the requirements for the new tools, to supplement principal components and matched filter, will be analyzed. The N-FINDR Visualization Package will be modified to include ICA. Finally, LWIR spectra will be incorporated, and the requirements for gas identification using these spectroscopic techniques will be determined.Commercial Applications and other Benefits as described by the awardee: The efficient and accurate determination of spectral end-members should lead to a number of commercial remote sensing application for hyperspectral sensors, including mineral exploitation, agriculture, and product inspection. For example, anomalous spectral end-members can be important indicators of mineral deposits (for geology) or diseased crops (for agriculture). In addition, the military has multiple hyperspectral applications underway that can benefit from improved tools, including terrain categorization and the detection of camouflaged vehicles.