Abstract: To address issues such as severe color cast, haziness, and loss of detail in underwater images, this paper proposes an underwater image enhancement algorithm based on a multi-scale weighted CycleGAN. First, a multi-scale weighted module based on a self-attention mechanism is proposed and embedded into the residual structure of CycleGAN, allowing the network to dynamically adjust its focus on different regions of the underwater image to correct color cast and color distortion. Then, multi-scale feature fusion is employed to integrate the weighted features, enriching the local details and global structural information in the enhanced image. Finally, the Structural Similarity Index Measure (SSIM) loss function is incorporated to supervise the training process of the generator, further improving the quality of the generated images. Experimental results show that, on the test set, the proposed method achieves average values of 0.6022, 4.426, and 7.562 for the Underwater Color Image Quality Evaluation (UCIQE), Underwater Image Quality Measures (UIQM), and information entropy, respectively. These represent improvements of 17.2%, 32.7%, and 3.2% over the original images. The proposed algorithm not only enhances the quality of underwater images but also preserves more detailed information.