Date of Award
Doctor of Philosophy (PhD)
Department of Electrical and Computer Engineering
Michael J. Mendenhall, PhD.
Hyperspectral imagery is collected as radiance data. This data is a function of multiple variables: the radiation profile of the light source, the reflectance of the target, and the absorption and scattering profile of the medium through which the radiation travels as it reflects off the target and reaches the imager. Accurate target detection requires that the collected image matches as closely as possible the known "true" target in the classification database. Therefore, the effect of the radiation source and the atmosphere must be removed before detection is attempted. While the spectrum of solar light is relatively stable, the effect of the atmosphere on this profile varies significantly depending on multiple atmospheric parameters. There are several data processing methods available to researchers for removing the influence of these parameters; however, little research has been done to describe, in a general way, how the uncertainty and error associated with these methods affects target detection. Our objective is to characterize the uncertainty in the detection method due to the uncertainty in the estimation of atmospherics. We apply a range of atmospheric profiles, correlated with relative humidity, to a radiative transfer model-based prediction of the atmospheric extinction effect using simulated hyperspectral imagery. These profiles are taken from known distribution percentiles as obtained from historic meteorological measurements at the simulated sites. We quantify the expected detection error, given the range of atmospheric conditions in the historic profile. We show that temporal variation in atmospheric parameters across their distribution impacts the accuracy of target detection. We show that this impact is more acute at high humidity than at low humidity. We show that, given the uncertainty associated with atmospheric profile estimation, the optimum assumption for purposes of target detection may be other than their median values, and that this effect is target dependent.
Yarbrough, Allan W., "Hyperspectral-Based Adaptive Matched Filter Detector Error as a Function of Atmospheric Profile Estimation" (2011). Theses and Dissertations. 1306.