<p>In underwater environments, light rapidly attenuates and scatters, causing captured images to often exhibit color distortion, low contrast, and blurriness. Although many existing underwater image enhancement methods have shown promising results, their high computational complexity and large model sizes make them challenging to deploy on resource-limited embedded devices or in real-time applications. This limitation not only hinders real-time image enhancement in specific underwater environments but also impedes its integration with high-performance computing (HPC) infrastructure. To address this issue, this paper proposes a lightweight network called CSCA-Unet for real-time underwater image enhancement. Specifically, we design a lightweight feature extraction module called CSC, which serves as the fundamental building block of the network. By incorporating an appropriate number of CSC modules during the encoding phase, the network achieves efficient multi-scale feature extraction while reducing computational complexity. The bottleneck layer integrates an appropriate attention mechanism and residual connections. Moreover, a joint feature loss function, combining pixel-level loss, structural similarity loss, and perceptual loss, is designed to generate enhanced images with finer textures. Experiments demonstrate that CSCA-Unet achieves superior color enhancement and excellent detail restoration in terms of visual quality. The proposed method achieves outstanding performance and remarkable processing speed compared to state-of-the-art techniques. In addition, the lightweight and modular design of CSCA-Unet facilitates efficient deployment in HPC systems, making it well-suited for large-scale underwater vision tasks.</p>