ARETE ASSOCIATES — Department of Defense SBIR Phase I: N151-028

Description

This proposal addresses issues concerning operation of the Coastal Battlefield Reconnaissance and Analysis (COBRA) system in unseen locales and against unseen target types. The COBRA target detection algorithms are based on statistical classifiers that are trained using samples from past data collects. A serious issue is the difficulty of estimating system performance in these situations. Arets approach to dealing with this issue is to develop a comprehensive model of the COBRA system with the objective of estimating detection performance for unseen locales and target types. Development of this model will lead to the creation of a software tool for COBRA performance prediction called the COBRA Modeling Tool (COBRA MT). COBRA MT will utilize previous COBRA imagery and associated performance to predict performance in unseen but similar locales. For dissimilar backgrounds in new locales, COBRA MT will automatically generate COBRA synthetic imagery. In those situations the synthetic imagery will be injected into the COBRA algorithms to estimate performance.