A super-resolution reconstruction network for lung CT images based on a locally enhanced Transformer and U-Net was proposed for the rich grey scale of the lung CT images leading to insufficient feature extraction and poor reconstruction details. The dilated convolution was used for deep feature extraction in multiple receptive fields, the global image information was obtained under the dilated convolution layers with different dilation rates, and the feature information under these different receptive fields was fused. The original features were obtained through the 3×3 convolutional layers, which were sent to the coding and decoding structure combining the proposed network, and the local enhancement window module reduced the computation and captured the local information. In the decoding stage, a skip connection was utilized, along with a segmentation attention block that fused spatial and channel attention to discard irrelevant information and utilize useful information, in order to obtain high-quality reconstructed images. Experimental results showed that, on the SARS-CoV-2 dataset, compared with the Transformer network, the proposed network improved the structural similarity index measure and the peak signal-to-noise ratio for 4-fold super-resolution by 0.029 and 0.186 dB, respectively.