Compared to traditional CNN architectures, the Transformer architecture can effectively utilize the entire image information to restore details in any given region without being limited by the size of the convolution kernel.To address the limitation in blurry image restoration accuracy caused by the reliance on single-scale input in existing Transformer architectures, this study proposes a multi-scale input method based on the Transformer architecture. Additionally, to enhance the network's ability to extract richer and more detailed features, we designed a new module called DECT (DiNAT Enhanced Channel Transformer). This module effectively captures dependencies between channels and enhances the network's integration of different channel feature information. Our proposed MTNet incorporates both the multi-input module and the DECT module on top of DeblurDiNAT. The proposed MTNet was trained on the GoPro public dataset, and compared to the state-of-the-art DeblurDiNAT-L, MTNet improved the PSNR and SSIM values by 0.42 and 0.003, respectively. Finally, MTNet was tested on the HIDE dataset for deblurring, and the results demonstrated that MTNet exhibited good generalization performance.