Autonomous Threat Assessment and Maneuver Control for Unarmed Ships Using ML-Based Pirate Detection

This study proposes an end-to-end, machine-learning-driven maritime security system that combines a YOLOv8-based object-detection pipeline trained on an augmented dataset encompassing low-light and adverse-weather conditions with a probabilistic threat-level assessment module. By incorporating vessel characteristics such as size, speed, and heading, our framework distinguishes between potential pirate skiffs and benign craft (e.g., refugee boats), thereby minimizing false positives. Upon detecting a hostile vessel, the system estimates its range and velocity relative to the merchant ship and computes an optimal evasive response. In particular, the results demonstrates that zigzag maneuvering consistently yields the highest probability of repelling an attack when the vessel is unarmed. Our approach is designed for real-time operation on commodity computing hardware and provides a proof-of-concept pathway toward early threat mitigation at sea.