Characterization of hyperspectral data using a genetic algorithm

Spectral mixture analysis plays an important role in analyzing hyperspectral data. Mixture models are typically built from spectral libraries with model coefficients determined by least squares techniques. As a result of noise, instrument artifacts, and other factors, the mixture model can be inadequate. This paper reports a genetic algorithm that generates multiple mixture models using subsets of reference spectra. Additionally, models of similar fit are also generated thereby providing the analyst with alternative explanations of the data. Details of the algorithm are provided along with some initial results of applying the algorithm to AVIRIS data.