To address the problem that the long-range dependence established by the shift window of the attention mechanism does not allow sufficient interaction of global information and how to effectively fuse multi-layer features, this paper proposes a global dynamic joint enhanced attention algorithm (GDUEA). First, the superimposed convolutional aggregation module is used to aggregate local low-frequency features in the shallow layer of the network to help the model better understand the input features; second, the proposed deep global attention algorithm uses the attention mechanism only for global information acquisition; finally, the dynamic joint enhancement module is used to perform dynamic joint equalization of features at different levels and deep feature enhancement. By comparing SwinIR network on Urban100, B100, Set5, and Set14 test sets, GDUEA obtains faster and more stable training and significantly improves the performance of the network in peak signal-to-noise ratio (PSNR) by 0.075~0.32db.