Automated Pore Type Classification from Thin Section Images Using Image Analysis and Topological Metrics

Petrographic thin sections provide critical insights into rock microstructures, especially pore spaces that control fluid flow, storage, and reservoir quality in sandstones and carbonates. Manual analysis is labor-intensive, time-consuming, and prone to bias, limiting its scalability for large datasets. Advances in digital imaging and computational methods have enabled automated approaches that improve efficiency and reproducibility.

This study presents an automated system for pore segmentation and classification. It uses a hybrid approach: color-based thresholding for epoxy-dyed images and an Attention U-Net deep learning model for non-dyed or complex cases. The system extracts topological metrics, including pore size distribution, aspect ratios, and solidity, to classify pores into interparticle, vuggy, and fracture types, aiding understanding of pore connectivity and reservoir performance.

Building on previous work in image processing and deep learning, the system integrates hybrid detection, quality verification, and exportable metrics in Excel, making it accessible to geologists without programming expertise. Unlike methods that focus solely on segmentation, this tool adapts to dyed and non-dyed images while providing topological classification. Tested on real datasets, it achieves accuracy comparable to manual analysis while drastically reducing processing time, offering a scalable and user-friendly solution for modern petrographic studies.