Abstract:

The Multiple Instance Adaptive Cosine Estimator and the Multiple Instance Subspace Match Filter are algorithms used in target detection, where a target class of interest is attempted to be detected amongst a non-target, background class. These algorithms learn a single feature vector representation to estimate a target class in a transformed feature space that normalizes the data to the background class. In this thesis, a number of algorithms are proposed to learn multiple target representations. These are evaluated using their respective performance and computation time using experiments containing a simulated hyperspectral dataset and the MUUFL Gulfport hyperspectral dataset captured over the campus of Southern Mississippi – Gulfport. The results are analyzed to conclude which variation of the multiple target algorithms is best in terms of performance and computation time.

Links:

Citation:

J. Bocinsky, “Learning Multiple Target Concepts from Uncertain, Ambiguous Data Using the Adaptive Cosine Estimator and Spectral Match Filter,” M.Sc. Thesis, Gainesville, FL, 2019. 

@MastersThesis{Bocinsky2019Thesis,
author = {James Bocinsky},
title = {Learning Multiple Target Concepts from Uncertain, Ambiguous Data Using the Adaptive Cosine Estimator and Spectral Match Filter},
school = {Univ. of Florida},
year = {2019},
address = {Gainesville, FL},
month = {May},
}