In February 2020 Danish police halted and boarded a containership finding 100 kilograms of cocaine in its cargo, but how did police know to stop this one ship? How do we identify pirates endangering international shipping, and fishermen endangering species of marine animals with illegal overfishing? In order to recognize the small amount of maritime traffic engaged in illegal or dangerous activities we can identify small anomalies in ship movements which deviates from the normal picture.
Current methods for detecting abnormal maritime behavior are largely based on the Automatic Identification System (AIS). AIS messages contain several static and kinematic ship data including position and heading. This information is used to derive vessel trajectories, which are compared to historical data for the purpose of detecting abnormal behavior. However, the use of AIS data alone is problematic as it is only required for larger vessels to carry an AIS transmitter, and the signal can be lost due to bad weather, turned-off, jammed, spoofed or otherwise manipulated. Furthermore, methods often require the full trajectories from start to end to perform anomaly detection, but in a real world setting that will be too late. The illegal goods will have been delivered, the fish will have been caught and the pirates will have boarded the ship.
The purpose of the project is to evaluate and demonstrate the potential of deep learning based predictive analytics methods centered on time series of radar tracks captured at several locations in Denmark. This include detection of anomalies and performing classification of marine vessels and aircraft behavior indicating dangerous, suspicious or malicious behavior in real-time.
The project will incorporate techniques to produce uncertainty estimate of predictions. This is important to meet the relevant skepticism of blackbox methods and would help increase user trust of any system. We study the effects of conditioning normality on external data sources and available AIS messages to investigate how external factors may influence the normal behavior. Ultimately a sailing ship might change its route to achieve better winds, rather than engaging in illegal activities.
By making real time anomaly detection of maritime radar tracks we will be able to recognize vessels engaged in piracy or smuggling and identify vessels in need of assistance.