Abstract: An underwater image enhancement algorithm was proposed based on an improved CycleGAN to address issues such as color distortion, severe fogging, and low contrast in underwater images. First, a multi-scale color correction module was designed to solve the color bias problem in underwater images by capturing multi-level color and texture information. This module was integrated into the discriminator network of CycleGAN, enabling the discriminator to better distinguish between real and generated images. Secondly, a CBAM (convolutional block attention module) attention module was inserted into the discriminator to enhance the model's focus on important features. Additionally, a multi-scale feature extraction module was added to the U-Net generator to handle features and details at different scales, and the SmoothL1 loss function was used to replace the MSE (mean square error) loss in the CycleGAN network, which could improve the model's generalization ability. Finally, experiments were conducted on the UIEBD and EUVP datasets, and the results show that, compared to the original algorithm, the proposed method could improve the average UCIQE(underwater color image quality evaluation) by 10% and the UIQM by 17%, demonstrating that the proposed algorithm meets the requirements for underwater image enhancement.