Considering the problems of severe scattering, unclear target imaging, and low contrast in high-turbidity water environments, a residual Unet (Mu-UNet)-based de-scattering algorithm for underwater Mueller matrix images is proposed on the basis of the traditional UNet structure and polarization imaging theory. According to the intensity and polarization information of targets provided by Mueller matrix images, this algorithm establishes the image data sets of multiple targets under different turbidities. The residual module is introduced on the basis of UNet, and the Mu-UNet is used to extract the underlying information of the targets, which learns the characteristics of the labeled images and finally reconstructs the underwater target images with high contrast and detailed information. The comparative experimental results reveal that compared with the original image, the image restored by the proposed algorithm is improved by 89.40% in the peak signal-to-noise ratio, and the structural similarity is improved by 82.37%. Compared with traditional algorithms and UNet, the proposed algorithm can obtain restored images with a more significant de-scattering effect and finer details, which provides a new idea for the detection and high-quality imaging of underwater polarization.