The existence of blue-green bias, low clarity and contrast of underwater images seriously affects the accuracy and reliability of underwater research. To address the above problems, this paper proposes an underwater image enhancement network based on multi-scale residual fusion. Firstly, a multi-scale channel feature extraction module MSCFE is proposed. The MSCFE module models each channel independently to avoid information interference between channels, and at the same time, channel attention is introduced to enhance the key features to effectively enhance the colour and details. Then, a global-local colour correction module GLCC is proposed, and the GLCC module adopts two branches, local and global, to model the local colour details and long-range dependencies respectively to correct the image colour. The experiments show that on the UIEB dataset, the structural similarity of the enhanced image reaches 0.937 8, the peak signal-to-noise ratio reaches 23.768 7, the underwater colour image quality evaluation index reaches 0.568 9, and the image information entropy reaches 7.572 3; on the EUVP dataset, the structural similarity of the enhanced image reaches 0.910 5, the peak signal-to-noise ratio reaches 25.169 9, underwater colour image quality evaluation index reached 0.525 3, and image information entropy reached 7.347 9, which are better than other mainstream methods.