The atmospheric polarization mode is a stable natural attribute with a wide range of applications in navigation, detection, and other fields. However, due to the influence of natural environment and surrounding structures, the obtained atmospheric polarization information at the same time is local and discontinuous, which affects its practical application. Existing methods mainly focus on the restoration of atmospheric polarization patterns on a large scale, but they have limited accuracy in repairing high-frequency signals, resulting in edge blurring. To address this problem, this paper proposes a method combining soft segmentation and soft synthesis to avoid the loss of high-frequency signals. Based on this, a atmospheric polarization mode generation method incorporating high-frequency perception is proposed, which preserves the missing region’s high-frequency polarization information, thus generating complete atmospheric polarization information. Experimental results demonstrate that this method can effectively reconstruct the missing polarization information in the atmospheric polarization mode. In the reconstruction experiment with cloud interference exceeding 40%, the SSIM and PSNR scores of this method are improved by 26% and 12% respectively compared to other methods.