Abstract: The diffusion model has good performance in image deblurring, especially in detail reconstruction. However, the denoising U-Net network in the diffusion model is difficult to process high-resolution images, and its non-linear representation ability is weak, making it difficult to accurately focus on key information, resulting in significant artifacts in the restored images. A multi-scale gated extended unit image deblurring algorithm is proposed based on the stochastic differential equation diffusion model to address the above issues. Firstly, in combination with the attention mechanism, a multi-head deep residual block with residual connection is designed. While increasing the receptive field, the model's ability to process high-resolution images has been improved, and the original information has been retained. Secondly, in order to enhance the nonlinear representation ability of the model and obtain an accurate noise distribution, a multi-scale gated expansion unit module is designed. Finally, the model is verified on multiple datasets. The experimental results show that the images restored by the proposed method are visually closer to the real situation, with richer texture details. Its PSNR value and SSIM value are superior to those of other common models.