Wang, Zhuolin (Dalian Maritime University), Zhang, Yingjun (Dalian Maritime University), Lu, Hongrui (China, Dalian Maritime University), Zhang, Haoze (China, Dalian Maritime University, Navigation College)

MADRL：A Framework for MASS Sea Surface Image Defogging

Scheduled for presentation during the Regular Session "S13c-Localization, Mapping, and Sensing for Robust Navigation in ITS" (WE-LA-T13), Wednesday, November 19, 2025, 16:00−16:20, Stradbroke

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 19, 2025

Abstract

Marine environment exhibits high complexity and dynamic variability, where sea fog interference significantly degrades image quality and poses substantial risks to the safety of vision-based MASS navigation systems. Conventional defogging methods suffer from inadequate adaptability and suboptimal performance in complex maritime scenarios. This paper proposes a collaborative MADRL framework for adaptive marine image defogging. The image defogging process is modeled as an interactive game between agent groups and the image environment, achieving multi-dimensional feature optimization through constructing scalable agent sets. The framework employs a D2QN integrated with prioritized experience replay mechanism, complemented by a multi-objective reward function that evaluates visual quality improvement, guiding agents group to explore optimal defogging strategies in the state-action space. Experiments demonstrates that in comparison with conventional algorithms, the MADRL paradigm exhibits superior performance in terms of PSNR, SSIM and other image quality evaluation metrics, effectively enhancing target clarity. This method provides high-quality image support for maritime target detection and visual navigation.