In the pursuit of advancements in underwater image recognition technology, this study innovatively introduces an advanced image enhancement strategy based on MuLA-GAN (Multi-Level Attention Generative Adversarial Network) to address the unique challenges of contrast attenuation, color distortion, and image blurring in underwater environments. This strategy deeply integrates the robust data generation capabilities of Generative Adversarial Networks (GANs) with the refined feature extraction advantages of a multi-level attention mechanism, markedly enhancing the model’s ability to discern and learn critical information from underwater images, thereby demonstrating exceptional performance in restoring image details. Through a comprehensive comparative analysis of standard test datasets, the results indicate that RtMuLA-GAN achieves optimal levels in the restoration of intricate image details, surpassing current mainstream methods. Additionally, to validate the practical application potential of this approach, tailored datasets specific to scenarios such as biological pollution monitoring and aquaculture management were constructed, and a series of rigorous experimental evaluations confirmed the model’s high robustness in complex underwater environments. This research not only enriches the theoretical framework of underwater image enhancement but also profoundly elucidates the pivotal role of multi-level attention mechanisms in enhancing GAN models, offering an innovative and comprehensive solution for improving underwater image quality.