Automatic Target Recognition (ATR) in Complex Underwater Environments

Black River Systems Company, Inc. proposes a novel online machine learning (OML) solution for underwater ATR using acoustic, magnetic, and EO sensors. The proposed solution, called TRACE (Target Recognition Aided by Continuous Evolution), will utilize tailored feature extractors and a robust online Bayesian classifier to maximize ATR performance in challenging underwater environments. TRACE is designed to minimize data overhead and to remain robust in situations when sensor streams are degraded or unavailable. Variational Bayesian ML techniques, the key driver of TRACE, will provide a flexible, and explainable framework for OML. Online updates will be achieved using compact representations of historic samples stored on disk to minimize data overhead and eliminate the need for persistent acoustic comms (ACOMMS). TRACE is easily extended to learn from a fleet of AUVs by populating the database with samples from various assets in real-time or post-mission.