<p>Underwater image enhancement is crucial for underwater vision research. To address the problems of blurred details, color distortion, and low contrast in underwater images, this paper proposes a generative adversarial network-based underwater image enhancement algorithm (SER-GAN). The generator of SER-GAN is constructed by spatially enhanced residual modules (SERM), hierarchical attention-intensive aggregation module (HAAM), and feature enhanced bridging module (FEM). To solve the problem of color distortion, this paper designs a spatially enhanced residual block, which employs the overall spatial information to enhance the semantic feature learning in the key regions and uses the residual structure to promote the reuse of features to improve the accuracy of the color representation. In addressing low image contrast, this paper introduces a hierarchical attention dense aggregation module, which extracts image feature information at different scales and dynamically adjusts the weights of different regions through hierarchical attention, so that regions with low contrast receive more attention and enhancement. For the problem of detail blurring, this paper proposes the feature enhancement bridging block, which transforms the original feature space through an intermediate layer and can effectively extract and enhance the edge details and texture information while retaining the important information. Experiments have proved that the algorithm in this paper performs well in solving the problems of blurred details, color distortion, and low contrast in underwater images. The average optimization of image objective evaluation metrics such as PSNR, SSIM, and UIQM on public datasets are 2.49%, 1.34%, and 2.14%, respectively.</p>