Application progress of machine learning to digital core research

Digital core has become a widely used approach for quantitative characterization of oil and gas reservoirs, and it plays an important role in unconventional reservoir evaluation and modeling. However, there exist two inherent trade-offs at the core scale: one between the field of view and image resolution, and the other between computing cost and model size. These trade-offs have become increasingly prominent as digital core research progresses. In recent years, the closer integration of machine learning with digital core has partially addressed these cross-scale challenges and thereby expanded the application scenarios of digital core. This paper begins by outlining the workflow of digital core research, followed by the algorithm examples to illustrate the applications of deep learning across different stages, including image segmentation, image fusion, super-resolution, rock feature recognition, and rock physical property simulation. Compared with conventional methods, deep learning-assisted digital core modeling and numerical simulation offer higher accuracy and efficiency. This paper concludes by discussing the potential of machine learning in complex core feature recognition and numerical simulation.