- Slides: 1
Detecting Anomalies in Vessel Behavior Based on AIS Data Student Team: Eamon Bontempo, Khalil Hardy, Samuel Yakovlev Mentors: Chance Petersen, Dr. Barry Bunin, Dr. Hong Man HOMELAND SECURITY CHALLENGE OUTCOMES / RESULTS The Automatic Identification System (AIS) onboard maritime vessels broadcasts a wealth of information, including position, velocity, heading, draft, and vessel identifiers, to nearby vessels and stations. Although AIS is originally intended as a safety precaution to avoid collisions in lowvisibility conditions, the sheer amount of information it carries could potentially be used for security applications. This research aims to detect and classify anomalous behavior in vessel tracks, using deep learning techniques to predict the path vessels follow, and comparing the actual path to expected results. APPROACH / METHODOLOGY In order to predict vessel behavior, the team used a Recurrent Neural Network, a model designed for use with data sequences of variable length. This model structure links together sequences of 30 AIS points (containing Latitude, Longitude, Course over Ground, Speed over Ground, and Heading) and establishes relations over a sequences of 30 datapoints to predict the next point. This methodology was used to train individual models for pleasure craft, cargo, tanker and tug vessels in the port of NY/NJ, and for tanker, tug, and cargo vessels in the port of New Orleans. Finally, a model trained on tanker vessels was applied to real AIS data collected by the AIS server in the Maritime Security Center in order to establish real-world usefulness of this approach. CONCLUSION The Automatic Identification System, although designed as a safety measure before all else, can be used for security with the right tools. From the performance of per-vessel models, we conclude that deep learning is a viable approach to anomaly detection. With enough training, a model can scan new AIS data and detect unusual vessel behavior. However, the team’s approach is only one part of a larger set - a combination of sub-models and different data sources is required for a comprehensive anomaly detection system. ACKNOWLEDGEMENTS  The National Academies of Science, Engineering, and Medicine, “Nap. edu, ” 2003.  C. Peterson, “Trajectory Reconstruction Models for Maritime Vessel Anomaly Detection”, 2019.  P. Choudhary, “Introduction to Anomaly Detection”, 2017.  C. Olah, “Understanding LSTM networks, ” Aug 2015.  The National Oceanic and Atmospheric Administration and the Bureau of Ocean Energy. Management, “Marinecadastre. gov, ” June, July 2019. This material is based upon work supported by the U. S. Department of Homeland Security under Cooperative Agreement No. 2014 -ST-061 -ML 0001. The views and conclusions contained in this document are those of the authors and should not be interpreted as necessarily representing the official policies, either expressed or implied, of the U. S. Department of Homeland Security.