Enhancing Data Privacy in Maritime Operations with Federated Learning: A YOLOv7 Object Detection Approach

Open Access
Article
Conference Proceedings
Authors: Vo Ngoc Thy Thao VoAmin MajdMehdi AsadiJuha KalliovaaraTero JokelaJarkko Paavola

Abstract: The maritime industry is currently experiencing a period of rapid transformation, driven by the integration of artificial intelligence (AI) technologies. This integration is enabling advancements in autonomous navigation systems, remote monitoring capabilities, and operational efficiency. However, these innovations are accompanied by substantial privacy challenges, particularly in the management of sensitive data collected from vessels. In this work, we propose a Federated Learning (FL) framework tailored for the maritime environment. This framework aims to address privacy concerns while leveraging the capabilities of AI. Utilizing the TUAS dataset, which contains images, and employing the YOLOv7 object detection model, we demonstrate how FL enables vessels to collaboratively train robust machine learning models without sharing raw data.Our approach ensures that data collected on vessels, such as images for navigation and object detection, remains onboard, thereby safeguarding sensitive information. Each vessel trains a local YOLOv7 model on its image dataset and shares encrypted model updates with a central server for aggregation. This global model is then disseminated back to the vessels, ensuring enhanced performance across the fleet without compromising data privacy. A comparison of our FL-based approach to traditional centralized training methods is presented, highlighting the trade-offs in model accuracy, privacy preservation, and communication overhead. The findings demonstrate that Federated Learning with YOLOv7 attains object detection performance that is competitive with other methods, while addressing privacy concerns by keeping raw image data localized. Integrating FL into the maritime industry provides a scalable and secure solution for AI-powered applications, ensuring data privacy while promoting innovation. Experimental results are a substantial contribution to the development of privacy-preserving AI solutions for autonomous maritime operations and remote monitoring, demonstrating FL's potential to transform the maritime industry.

Keywords: Autonomous systems, AI, intelligent systems

DOI: 10.54941/ahfe1006554

Cite this paper:

Downloads
28
Visits
137
Download