Fishing vessel tracking and monitoring is paramount to maintaining safe and proper activity in the high seas. Suspicious fishing activities or transshipment events can result in serious biosecurity risks, exchange of illegal goods, and even human rights violations. However, with over hundreds of thousands of fishing vessels, manually monitoring and predicting anomalous activity for each vessel becomes a highly intractable task.
Luckily, this task can be structured with the help of historical classification data and vessel interaction activity. We can form the problem space into a dynamic network of fishing vessels, where vessels are more likely to be anomalous if it has history of such activity or is associated with other anomalous vessels. Our approach can leverage this temporal graph structure through a machine learning model called Dynamic Graph Neural Networks. This model aims to learn the spatio-temporal representations of complex fishing networks, and aid in automatically classifying and forecasting potential anomalous vessels in a more sophisticated and scalable manner. Through this, we aim to ultimately enhance the efficiency of maritime monitoring and better enforce fishing regulations on a global scale.
Keywords: Graph Neural Networks, Maritime Monitoring, Tracking