<p>Underwater suspended particles can cause attenuation and scattering of light, resulting in color distortion, low contrast, and blurred details in underwater images. The degradation of the images has had an impact on the application of advanced visual tasks. We propose a method for underwater image processing based on dual Transformer aggregation and multi-scale feature fusion (DTAM-UIE) to address the above issues. First, features of underwater images are extracted by combining the channel self-attention Transformer with the bi-level routing dynamic sparse self-attention Transformer. Then, the extracted features are fused through the designed parallel channel spatial attention block. Next, a multi-scale fusion block is designed to aggregate features of multiple scales in the decoder part of the network model, improving the model’s adaptability to different underwater environments. Finally, we design the post-processing block for the U-shaped network’s output layer to fuse each part’s output features, improving the performance of feature fusion and thus enhancing the overall output effect of the model. Experimental results show that the DTAM-UIE can effectively improve image color distortion, enhance contrast, and enhance image details. In both qualitative and quantitative comparisons, the proposed method outperforms the comparison methods, better restoring image color and texture details.</p>