Smart Automated Target Recognition using
Weighted spectral
and Geometric Information
Overview
University of Tennessee, in partnership with North Carolina State
University, proposes multi-spectral and geometrical based automated
target recognition - a research effort aimed at developing a ?smart?
target recognition system which takes advantage of both the
multi-spectral and geometrical information available at different
ranges from the target.
The innovation of this research effort is that it proposes an
integrated approach to conduct ATR based on a weighted combination of
the multispectral and geometric information, where the weight is
adapted to the change of range between the target and the smart
ordnance. Four tasks are proposed: 1) Develop a feature space using a
multispectral snake technique which is both spectral- and
geometric-based, 2) Develop classifier fusion algorithm that uses
weights of evidence as a measure of the belief in different
classifiers (either spectral- or geometric-based) in the context of
Dempster?s Rule of Combination, 3) Develop feature fusion algorithms
based on a modified Hausdorff distance such that both spectral and
spatial features are embedded in the process of template matching, and
4) Algorithm implementation and evaluation against test data in three
forms, including synthetic data, calibrated data, and field data.
