TOWARDS AUTOMATED VESSEL DETECTION USING SPATIAL SCAN STATISTICS AND DISTRIBUTED ACOUSTIC SENSING

Kurzfassung

Situational awareness is a central task to managing potential risks to maritime infrastructure; however, conventional methods for capturing vessel movements are often susceptible to spoofing or sensitive to weather conditions. This issue motivates the methodology presented in this paper, which investigates distributed acoustic sensing (DAS) as a method to measure acoustics relating to vessels across wide regions, and proposes an anomaly detection method to process data. This paper aims to address two challenges for anomaly detection using DAS data: 1) increasing detection power via leveraging multiple DAS channels and 2) reducing the number of false alarms when monitoring large regions. For this purpose, a spatial scan statistic is implemented, as it presents a method for testing regions with various sizes, and often reduces false alarms related to testing multiple regions independently. Furthermore, it is proposed that model residuals can be used as features for the spatial scan statistic, since predictive models can be designed such that their residuals are insensitive to changes in the data relating to benign events. The proposed system is validated using an open dataset collected off the coast of Valencia. The results suggest that the method is able to consistently detect ships, with each ship crossing event in the test set being detected prior to crossing, and on average alarms are raised when ships are 2km from the cable. Furthermore, the spatial scan statistic is shown to both increase the detection range, while reducing false alarms when compared to testing each region independently