Underwater image processing is gaining increasing attention for improving underwater exploration tasks, including terrain scanning and autonomous underwater vehicle (AUV) navigation. Due to the attenuation of light in water and wavelength-dependent scattering effects, images captured underwater often suffer from severe color distortion, reduced contrast, and detail loss, which adversely impact subsequent tasks such as target recognition and environmental reconstruction. To address the issues of uniform scattering and noise-induced degradation in underwater images, as well as the high computational cost of dense attention mechanisms, we propose an underwater image enhancement method based on a Top-k channel sparse self-attention mechanism. This approach retains only the top-k high-response channels for cross-channel information fusion, thereby reducing unnecessary computations and suppressing noise interference. In the feed-forward network, a GELU-activated gating mechanism is employed to dynamically select and filter channel features, effectively suppressing redundant information. Additionally, residual connections are incorporated into the underwater imaging model to facilitate image restoration. Experimental results indicate that the proposed method has a positive effect on several key metrics, both qualitative and quantitative.