Matched Filter Stochastic Background Characterization for Hyperspectral Target Detection

摘要

Algorithms exploiting hyperspectral imagery for target detection have continually evolved to provide improved detection results. Adaptive matched filters, which may be derived in many different scientific fields, can be used to locate spectral targets by modeling scene background as either structured (geometric) with a set of end members (basis vectors) or as unstructured (stochastic) with a covariance matrix. In unstructured background research, various methods of calculating the background covariance matrix have been developed, each involving either the removal of target signatures from the background model or the segmenting of image data into spatial or spectral subsets. The objective of these methods is to derive a background which matches the source of mixture interference for the detection of sub pixel targets, or matches the source of false alarms in the scene for the detection of fully resolved targets. In addition, these techniques increase the multivariate normality of the data from which the background is characterized, thus increasing adherence to the normality assumption inherent in the matched filter and ultimately improving target detection results. Such techniques for improved background characterization are widely practiced but not well documented or compared. This thesis will establish a strong theoretical foundation, describing the necessary preprocessing of hyperspectral imagery, deriving the spectral matched filter, and capturing current methods of unstructured background characterization. The extensive experimentation will allow for a comparative evaluation of several current unstructured background characterization methods as well as some new methods which improve stochastic modeling of the background. The results will show that consistent improvements over the scene-wide statistics can be achieved through spatial or spectral sub setting, and analysis of the results provides insight.