To address issues such as color distortion, blurriness, and low contrast in underwater images, a Multiple Attention Aggregation U-Net for Underwater Image Enhancement (MAAU-UIE) is proposed. The network is constructed using an encoder-decoder structure, with a multiple attention block designed to enhance the ability to extract features from low-quality underwater images. First, axial rectangular window attention and shifted axial rectangular window attention are alternately applied to learn local context and establish global dependencies, respectively. Additionally, channel convolution and spatial convolution are incorporated into the process of window attention calculation to further supplement local information. A channel enhancement module is then added to improve the modeling capability in the channel dimension. Finally, gradient loss and multi-scale structural similarity loss are used to enhance the network’s ability to extract edge detail information and multi-scale structural features. Ablation experiments demonstrate the significant role of each proposed module plays in improving network performance. Quantitative experiments show that this method surpasses existing methods on various objective metrics. The peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) on the benchmark dataset UIEB test set reach 24.467 and 0.920, respectively. The underwater image quality measurement (UIQM) and underwater color image quality evaluation (UCIQE) on the UIEB challenge set and in three color-bias environments of UCCS outperform cutting-edge methods. Qualitative experiments indicate a clear advantage in subjective visual effects, effectively restoring underwater images with natural colors and clear texture structures.