Automation within current Navy sonar system implementations continues to require trained operational input in key detection, association, and contact management activities. Automation improvements that enable strides toward fully autonomous systems are required. Venator proposes to develop a Multi-Layer Association and Inference Graph (ML-AIG), a new architectural approach for signal association and spectral classification that leverages state of the art automation and builds on existing research activities. Our proposed solution builds upon graph-based track stitching capabilities, supports association based on spectral and kinematic features, establishes a multi-layer graph representation integrated with probabilistic reasoning via layered Probabilistic Graphical Models (PGMs), supports probabilistic inferencing on derived state variables and attributes using PGMs, and enables sequence-neutral evaluation of information without enumerating or pruning any association hypotheses. The approach is therefore robust to operator inputs and edits/changes at any time. Finally, the approach supports generation of a Common Operational Picture (COP) designed to adapt to addition or removal of information.
Benefit: The technology developed and demonstrated will support a path to fully automated DCL, a keystone of automated systems and reduced operator workload. The technology will be applicable to general autonomous DCL problems across all Navy surveillance, submarine, surface, and air sonar systems.
Keywords: Feature association, Feature association, PGM sonar classification, autonomous classification, integrated association and inference graph, graph-based tracking and association, Acoustic association, sonar automation
