Focus Area 8 - (SPADE) Spatial Anomaly Detection

Coastal and marine navigation, as well as doing so safely, requires the correct identification of vessel type and prediction of behaviors. While the development of Automatic Identification System (AIS) has eased the burden of vessel localization and identification, errors still occur due to improperly configured messages, whether done innocently or maliciously.In order to increase the safety of our waterways, we propose the Spatial Anomaly Detection (SPADE) system. Utilizing recent advances in deep learning and reinforcement learning techniques, SPADE will provide a system for identifying vessel types, detecting anomalous behavior based on self-reported vessel type, and identify shipping lanes as well as likely destination. By treating the problem as a reinforcement problem, SPADE will learn decoupled behavior dynamics of different vessel types as well as their intentions/destinations. This allows for global behavior models to be adapted to local regions by replacing the intention/destination functions with those more appropriate for the region, allowing for a powerful-yet-flexible model that adapts to regionally specific problems. The SPADE system will allow for the identification of pirate vessels sending incorrect identifiers over AIS, predict where other vessels will be going to aid in safer navigation, and identify ships in distress based on unlikely movement patterns