With the widespread application of wide-angle lenses and panoramic cameras, geometric distortions such as barrel distortion frequently occur in images,which severely degrade image quality and adversely affect subsequent image understanding and computer vision tasks.To address this problem,a deep collaborative network integrating an appearance flow estimation module and Swin Transformer was proposed for automatic barrel distortion correction.Taking distorted images as input,the proposed method predicts the corresponding dense optical flow fields within a unified deep learning framework and generates corrected undistorted images through a resampling operation.To enhance training effectiveness,a panoramic image dataset covering diverse scene categories was constructed, and paired distorted and undistorted images with corresponding distortion parameters were generated based on the unified spherical camera model.Experimental results demonstrated that the proposed method achieved a peak signal-to-noise ratio (PSNR) of 22.0140 dB,a structural similarity index (SSIM) of 0.7955,a mean pixel displacement error (MPDE) of 16.5980,and a line curvature error (LCE) of 1.8153,demonstrating good performance in both image distortion correction and geometric structure recovery.Moreover,the proposed approach shows clear superiority over existing methods in terms of subjective visual quality,further validating its effectiveness for geometric distortion correction tasks.The constructed dataset is publicly available at: https://github.com/ssssssch/yuyan-dataset