AI-Assisted Thin Section Image Processing for Pore-Throat Characterization in Tight Clastic Rocks

Abstract

The characterization of pore-throat structures in tight sandstones is crucial for understanding fluid flow in hydrocarbon reservoirs and groundwater systems. Both thin-section and Mercury Intrusion Capillary Pressure (MICP) offer insights rock petrophysical parameters. However, thin-section analysis is limited by its 2D nature and subjective interpretation, while MICP provides 3D pore-throat distributions, it lacks direct visualization of pore morphology. This study evaluates AI-assisted thin-section image analysis for pore-throat characterization by comparing its results to MICP-derived measurements. A machine learning-based workflow was developed using color thresholding, K-Means clustering, and medial axis transformation to segment pore structures in thin-section images. Throat width, porosity, and permeability were quantitatively assessed against MICP to determine the accuracy and reliability of the technique. The analysis of 26 sandstone samples outlined differences between the two methods. Thin-section analysis showed porosity values from 1.37% to 53.37%, with average pore-throat sizes between 5.63 micron and 30.09 micron, while permeability estimates ranged from 0.01 mD to 344.35 mD. Correlation analysis showed moderate agreement for throat size (r=0.62) and permeability (r=0.61), but weaker for porosity (r=0.32), highlighting the differences in how each method captures pore connectivity. Results demonstrate that the AI-assisted segmentation provides a scalable and reproducible approach but is constrained by thin-section imaging resolution. While MICP remains reliable for permeability evaluation, its comparison with AI-driven image analysis helps assess the reliability of the method. Future research should refine segmentation algorithms, incorporate pretrained data to validate AI-derived pore-throat attributes for improved reservoir quality assessment and predictive modeling.