In order to address the limitations in learning global feature information and the restricted receptive field of the existing image deblurring models，an improved multi-scale image deblurring network that integrates Transformer is proposed. Firstly，a multi-feature multi-scale fusion module is designed to enhance the model’s ability to learn global features and capture distant pixels. This module effectively combines local and global feature information by using a dual bypass structure while simplifying the Transformer to improve computational efficiency. Secondly，in order to alleviate the drawback of convolution operations lacking in input content adaptability，channel attention is introduced into the feature fusion module to dynamically learn useful information. On the benchmark dataset GoPro，the peak signal-to-noise ratio is 31.87 dB and the structural similarity is 0.952. The experimental results demonstrate that compared with mainstream methods，the proposed approach effectively restores image detail features and enhances the robustness of subsequent computer vision tasks.