In today’s society, hospitals are treated with images generated by medical examination equipment for disease diagnosis, and high-resolution images can greatly improve the accuracy of doctors’ disease diagnosis. The study constructs an ultrasound image dataset US-Dataset suitable for the task of super resolution reconstruction of ultrasound images. Based on this ultrasound image dataset, a degradation model is proposed, which in turn constructs ultrasound image matching pairs containing high – low resolution images for training the model proposed in this paper. To improve the perceptual quality of endometrial images, a super-resolution reconstruction model UN-SRGAN based on generative adversarial network is proposed in this paper. The network structure of this model consists of a generator and a discriminator. To validate the effectiveness of the model proposed in this paper, it is evaluated on Accuracy, Precision, Recall, Specificity, and F1-score metrics. The proposed model achieves the lead on PSNR and SSIM metrics and subjective quality evaluation, and the UN-SRGAN model has an accuracy of 0.9721, which is better than the other models, verifying the effectiveness of the model.