Chest X-ray images of patients with pneumonia can quickly and visually show the extent and location of the lung infection, helping doctors to more accurately diagnose and assess the condition. The necessity of segmenting pneumonia X-ray images lies in that it can accurately locate and extract the infected lung area, helping doc-tors to more clearly identify the scope of the lesion and its severity. Therefore, we propose a combined structure U-GANs, which is a pyramid convolutional attention fusion network (PCAF-net) based on Convolutional Networks for Biomedical Image Segmentation (U-net) and a Super-Resolution Generative Adversarial Network (SRW-GAN) based on Generative Adversarial Networks (GAN). This combination not only improves the training quality of the model and enhances the robustness of the diagnostic system, but also provides physicians with more valuable tools for image analysis and assisted diagnosis.