Adaptable Multi-Layer Inference System for Distributed Sensor Networks

Adaptive Methods and Applied Research Laboratory at Penn State are developing a hierarchical inference approach for multi-modal unattended ground sensor (UGS) networks. that will enable significant performance gains via integrated machine learning techniques, to include In situ performance characterization and automated adaptation to site-specific environmental characteristics; unsupervised learning of activity patterns and establish a baseline for anomaly detection; flexible subject-matter expert knowledge capture and integration; and incorporation of historical and prototypical data sets. This is a multi-level fusion system distributed across sensor types and processing platforms. The deployment architecture takes advantage of specific physical characteristics and supports dynamic reconfiguration as nodes are lost or characteristics change. The architecture has four principle functional layers: (1) Level 1 Fusion layer which interfaces directly to single source sensors fusing them into a consistent view of entities are operating within the sensor field of regard; (2) Level 2 Fusion layer which evaluates the entity level picture characterizing the intent of those entities; (3) a prioritized data distribution system which insures that important information is shared in a timely fashion and (4) shared knowledge models used to organize, reason about and share information. A prototype demonstration, fusing recorded data, will be provided.