Self-Supervised Deep Learning for Maritime Surveillance: Detecting Intentional AIS Shutdowns in Open Seas

Authors- Mrs.A. Srujana Jyothi., V.Siva.Sai Ram., K.Chandra Mouli., U.T.S.K.Krishnam Naidu., P.Isaac Ben Joseph., N. Likhita Pravallika

Abstract-Maritime traffic surveillance plays a vital role in detecting illegal activities such as unauthorized fishing or the trans shipment of illicit goods, making it a critical responsibility for coastal administrations. In open waters, authorities rely on Automatic Identification System (AIS) messages transmitted by on board transponders, which are captured by surveillance satellites. However, vessels engaged in illegal activities often intentionally disable their AIS transponders to evade detection. Distinguishing between intentional AIS shutdowns and signal loss due to protocol limitations, adverse weather conditions, or satellite positioning constraints remains a significant challenge. This paper introduces a novel approach for identifying abnormal AIS missing receptions using self-supervised deep learning techniques and transformer models. By leveraging historical AIS data, the trained model predicts whether a message should be received in the next minute, flagging anomalies by comparing the predictions with actual outcomes. Our method is designed for real-time processing, capable of handling over 500 million AIS messages per month, covering the movements of more than 60,000 ships. The approach has been tested on a year’s worth of real-world data from four Norwegian surveillance satellites, demonstrating effectiveness by successfully rediscovering previously identified intentional AIS shutdowns based on existing research findings.

DOI: /10.61463/ijset.vol.13.issue2.270