Exploitable Physics for Recognition and Classification

ABSTRACT: The objective of this effort is to develop innovative methods for deriving a sparse set of physical target features that can be used for exploitation of air to ground signature data collected from sensor systems including electro-optical, infrared, and laser radar. Current classification methods require near exact replication of the original imaging parameters, or extensive modeling in order to generalize the signature to novel operating conditions. By understanding the physical constraints on the target information, one can better construct and refine a classification system. We will show how recent bounds, developed for electromagnetic scattering, work equally well for optical wavelengths, and we propose to use these bounds to develop tools to explore feature salience. BENEFIT: The primary benefit of successful completion of this effort is improved exploitation systems wherein we can explicate why and where features are informative. This capability has numerous commercial applications in various business sectors such as defense, communication, and medical imaging.