Machine Learning and Data Fusion Methods for Enhanced Maritime Surveillance

The monitoring of activities at sea is a key enabler for an effective Maritime Situational Awareness (MSA) capability. MSA can be thought as the means by which the operational picture is formed, which is constantly and continuously updated with data and information originating from deployed units and surveillance networks, both terrestrial and space-based. The achievement of an advanced MSA capability on the global scale remains an only partially solved challenge. This thesis offers advancements of machine learning and data fusion techniques, aiming at efficiently processing vast amounts of information with the final goal of supporting decision processes. Specifically, analytics methods are proposed to predict future vessel positions even in complex maritime traffic environments, thanks to the use of signal processing and deep-learning techniques jointly with large volumes of historical traffic data. Based on the prediction model, maritime anomaly detection methodologies are then developed to determine if a target deviates from a standard route even in the case of no data available, i.e., when the vessel goes “dark.” Finally, an assessment of the COVID-19 impact on the maritime commercial traffic based on big data processing is also presented.