To address the problems of low image contrast, color distortion, and blurred details caused by turbid underwater environments, an underwater image restoration network based on multi-channel polarization and a dual-branch Transformer was proposed. A polarization feature adaptive fusion module based on the Stokes vector was used to extract the physical polarization features of water. A spatial-polarization dual-branch architecture was constructed, and a dynamic gating mechanism was utilized to adaptively fuse multi-source information, such as polarization intensity, degree of polarization, and polarization angle, with spatial features, overcoming the limitation of incomplete representation by single-modality information. An encoder-decoder structure integrating multi-scale feature interaction, channel attention, and multi-scale spatial Transformer modules was adopted for the network backbone to enhance the capabilities of degradation representation and global detail restoration. The experimental results show that this method has significant advantages in structure restoration, detail preservation, and color correction (SSIM: 0.854 0, PSNR: 32.93 dB, perceptual metric LPIPS: 0.363 3), and can significantly improve the clarity, contrast, and color fidelity of the restored images.