An Advanced Deep Learning-Based High-Resolution \(\mu{\text{CT}}\)  Images Construction Method for Cement Hydration Microstructure

Analyzing the microstructure images of cement can help identify and quantify the internal material structure, understand the hydration mechanism, and design cement materials. Micro-CT ( \(\mu{\text{CT}}\) ), as an advanced physical imaging device, can image internal structural information of cement hydration at the microscopic scale. However, due to limitations in the equipment price, maintenance costs, and single imaging time of the equipment itself, the acquisition of high-definition \(\mu{\text{CT}}\)  images for cement hydration microstructure faces high scanning economic costs. Additionally, limited by the imaging mechanism of the physical devices, the size of the cement samples affects image resolution. This paper proposes a high-resolution \(\mu{\text{CT}}\)  images construction method for cement hydration microstructure based on deep learning. Real-ESRGAN network is introduced to improve the resolution of \(\mu{\text{CT}}\)  images obtained from physical devices after imaging, on a super-resolution reconstruction approach. In addition, a refined high-order degradation is designed and applied to Real-ESRGAN to suit the characteristics of \(\mu{\text{CT}}\)  image data. Experiment results validated the good performance of the proposed method in constructing high-resolution \(\mu{\text{CT}}\)  images and were extended to historical images imaged by devices that are currently technologically backward.