Innovative Signature Exploitation for Long Range Object Discrimination

One of the principle challenges in developing a robust missile defense architecture is the characterization and classification of objects in a threat complex. The Ballistic Missile Defense System (BMDS) performance depends on data from diverse EO/IR and RF sensors. The objective of this proposal is to investigate innovative concepts for long range real-time discrimination of ballistic missile threats that exploit observations and sensor measurements collected by multiple sensors. We propose novel algorithms that identify and utilize discriminating characteristics of the objects using EO/IR and RF data collected in the form of short (several seconds) snippets and incorporating feature sets which are derived from fused Airborne Infrared (ABIR) sources. The goal is to classify (providing object"s class probability) numerous objects (raid environment) in the post-burnout phases of the flight. Proposed techniques are based on Sparse Bayesian Learning Theory [1-4] (Relevance Vector Machine classifiers) supported by results derived from Information Theory [5] (information-theoretic clustering using the minimum description length principle). To associate the exploitable information content with the underlying physics based measurements we construct context aware classification systems focusing on robust physical differences between classes including shape differences, non-uniformity in thermal distribution, and reflection effects.