<p>Due to the unique characteristics of the underwater environment, underwater images are inevitably affected by the selective absorption and scattering of different wavelengths of light by the water medium, resulting in color distortion, blurring, low contrast, and noise. To address these degradation problems simultaneously, this paper proposes a dual-coded underwater image enhancement network (MCMS-Net) that combines multi-color space and multi-scale approaches. Specifically, the proposed method employs residual networks to encode various color spaces, integrates the discriminative features derived from these multiple color spaces, and assigns weights accordingly. This strategy effectively enhances the diversity of feature representations. Additionally, a multi-scale coding structure is implemented to capture both detail and texture features of the image across various scales. The channel information and attention mechanism of the Transformer module are utilized to capture long-range dependencies in the image, enabling the understanding and processing of global features and addressing the weakened correlation between visual features in different regions of the image. The Inception Feature Prompting Module (IncFPM), integrated into the coding structure, allows the network to adaptively select optimal parameters, thereby improving the model’s generalization capabilities. Extensive experimental results demonstrate that the proposed method outperforms existing approaches in terms of quantitative metrics, significantly enhancing the visual quality of underwater images.</p>