<p>Visual surveillance technologies play an indispensable role in advanced maritime traffic systems. However, the quality of visual images is often compromised by adverse weather conditions, such as haze. The degradation of visual information leads to inaccurate environmental perception by vessels, increasing navigational risks and presenting significant challenges to maritime vessel monitoring. To mitigate the impact of adverse weather, a novel maritime image dehazing network is required to enhance real-time visibility. Existing deep learning-based dehazing networks are primarily designed for dehazing in land background and perform poorly in maritime water-sky conditions. In this paper, we propose a Frequency-Guided and Enhanced Fusion Dehazing Network (FGEF-Net), which employs the frequency feature selection mechanism combined with the frequency information fusion strategy based on spatial and channel attention to effectively extract global features. The designed feature refinement block utilizes a dual-branch structure to integrate both non-local and local features, avoiding the loss of details in different feature levels. Extensive experiments are conducted on standard datasets and maritime intelligent vessel-related datasets. The results show that FGEF-Net outperforms other dehazing networks in terms of image restoration quality, achieving optimal dehazing performance with fewer parameters, making it more efficient for deployment in maritime intelligent traffic systems. Moreover, it delivers superior results in visibility enhancement and ship detection tasks under maritime hazy conditions, significantly improving the safety of vessel navigation in complex weather.</p>