<p>Due to the light scattering and absorption of impurities, underwater images often suffer from severe degradation, which seriously affects underwater exploration and research. To address the problem, we propose a dual-branch cooperative learning underwater image enhancement (UIE) algorithm based on generative adversarial network (GAN), in which each branch adopts a U-Net like encoder-decoder structure. Specifically, the proposed algorithm first employs an image decolorization block to yield grayscale images, thereby removing the interference of color bias in the original images. Then, the generated grayscale images and the original inputs are passed through the detail enhancement module to intensify the detail and texture representation of the images respectively, and then the enhanced images are fed into the two main branches to learn different types of features respectively. We conduct experiments on three datasets, including challenging underwater scenarios, to demonstrate the effectiveness and robustness of the proposed algorithm.</p>