The generative adversarial network is widely used in image generation, and the generation of images with different styles is applied to underwater image enhancement. The existing underwater image generative adversarial network does not realize color correction when processing underwater images Therefore, we propose an improved generative adversarial network for image color restoration. Firstly, the loss function in the network is improved to train the dataset. Then the improved network is used to detect the underwater image. After network testing, the underwater image is more satisfactory than the traditional image. Numerical results show that this method has a good color restoration and sharpening effects.