Super-resolution x-ray imaging technology based on U-shaped Residual Transformer Generative Adversarial Network

Abstract Current x-ray imaging technologies have difficulties in achieving higher resolution, because the fabrication of optical components with superior performance remains a challenge. Here, a super-resolution technology based on the U-shaped-residual-transformer-generative adversarial network (URT-GAN) network is proposed, which synergizes a transformer-based denoising module and a GAN-based resolution enhancement module. Through resolution test patterns and real sample datasets, the effectiveness and wide applicability of this method has been demonstrated and validated from both quantitative and qualitative perspectives. Without altering the existing hardware conditions, it can still achieve higher resolution x-ray imaging through adoption of the URT-GAN network. The source code will be released along with this paper and we welcome our colleagues in the field to adopt and improve upon it.