Complexity expert-guided visual perception enhancement for boosting maritime navigational safety under low-visibility scenes

• Enhances perception in low-visibility maritime environments. • Proposes dynamic complexity-aware expert selection and boosts system efficiency. • Improves the reliability of intelligent maritime transport systems. In maritime intelligent transportation systems, visual sensors serve as critical components for maritime environmental perception and navigation assistance, yet their imaging quality is susceptible to degradation under inclement weather conditions, manifesting as restricted visibility range and chromatic distortion, etc. Such degradation directly impair the perceptual system’s output accuracy and induce decision-making latency, thereby substantially elevating navigational risks. To mitigate the adverse impacts of low-visibility imaging conditions on maritime safety, this paper proposes a low-visibility image enhancement framework guided by a complexity-aware expert mechanism, designed to improve visual perception performance across diverse weather conditions. In detail, a complexity-aware routing strategy is introduced, employing the complexity bias term to achieve autonomous adaptation between input and expert model, ensuring dynamic adjustment of processing capabilities according to varying weather-induced degradation scenarios. This mechanism integrates with a hierarchical network architecture that optimizes computational efficiency and enhancement quality through differentiated computational complexity configurations and receptive field adjustments, selectively bypassing non-essential expert models during inference. Extensive experiments on multiple degradation-type datasets demonstrate that the proposed model effectively enhance the low-visibility images while maintaining low computational overhead, surpassing existing comparative methods in both quantitative metrics and visual quality assessments. Experimental results confirm its capability to enhance accuracy and stability in target detection under marine low-visibility conditions, highlighting its potential for constructing robust navigational visual perception systems.