Federated Deep Learning-Based Framework to Avoid Collisions Between Inland Ships

Abstract

With the rapid growth of inland shipping and the increased number of inland ships required to convey the freight, the ultimate goal is to design an intelligent shipping system to make inland shipping safer and more efficient. Cooperative ships safety systems are an emerging approach to supporting reliable collision detection. Two critical requirements of cooperative safety applications are position accuracy and ultra-low communication latency. Therefore, this paper proposes a new collision detection system for inland ships based on Federated Deep Learning, which is expected to provide a robust positioning prediction model. In addition, it guarantees collaborative learning among all ships while preserving ships' privacy. Furthermore, our safety system is deployed at Multi-access Edge Computing (MEC) nodes to ensure low latency communication and guarantee real-time reaction to avoid collisions between ships. Extensive simulation results show the system's accuracy and, hence, the efficiency of the collision detection system to ensure timely and trusted communications and avoid collisions between ships.