<p>Degradations such as color casts, low illumination, and detail blurring are critical challenges for vision-based underwater exploration. To address these degradation challenges, we propose a low-resolution assisted multi-path collaborative network (LRMC-Net), a novel framework that leverages multi-path collaboration with low-resolution guidance to address color deviation and detail blur. First, the hierarchical residual downsampling module (HRDM) is used for progressive downsampling to extract low-resolution spatial information, compensating for details across different layers of the backbone network. Next, the dynamic triplet attention (DTA) is introduced to dynamically adjust feature weights, emphasizing critical features at different stages of degradation scenarios, effectively addressing color distortion. Finally, the adaptive feature fusion (AFF) module facilitates feature fusion between global contextual information and local texture details, which significantly reduces detail blurring artifacts while preserving structural integrity. Experimental results demonstrate that LRMC-Net outperforms state-of-the-art methods in both qualitative and quantitative evaluations across multiple underwater datasets, while exhibiting remarkable robustness under various degradation scenarios. Moreover, its efficient architecture supports real-time operation, making it highly suitable for practical underwater vision tasks.</p>