Shale sample permeability estimation using fractal parameters computed from TransUnet-based SEM image segmentation

Abstract

Microscopic pore structure forms the foundation for studying shale gas adsorption and transport mechanisms and for establishing geological models. However, most current methods for analyzing microporous structure through physical experiments are time-consuming and labor-intensive. Hence, there is a need to automate pore segmentation and extract pore microstructural information from shale SEM images quickly and accurately. This will significantly enhance the efficiency of digital rock analysis and related computational simulations. This study used scanning electron microscopy (SEM) images of shale from a certain region in China to investigate the relationship between the microscopic structure of shale pores and the macroscopic permeability. Firstly, a semantic image segmentation model called TransUnet, based on deep learning, was used to segment the pore images and extract the micro-pore structure parameters. Then, the relationship between the macroscopic permeability parameters and the micro-pore structure was analyzed using a fractal apparent permeability calculation model. Finally, the permeability of the shale was calculated to improve the efficiency of geological exploration and reduce experimental costs. The experimental results show that this study provides an effective image processing method for the SEM quantification of shale microstructure and extraction of permeability parameters.