A Novel Super-Resolution-Based Multi-Scale Pore Network Modeling Method for Characterizing Complex Rock Inner Structures

Characterizing the complex inner structure of heterogeneous rocks with multi-scale pores presents significant challenges, as conventional imaging techniques cannot simultaneously achieve both high resolution and large field of view (FOV). While high resolution is essential for capturing fine microstructures, a large FOV is required to represent pore heterogeneity, yet these two objectives are fundamentally opposed in imaging systems. To overcome this limitation, we develop an innovative workflow that integrates deep learning-based super-resolution reconstruction with multi-scale pore network modeling (PNM), enabling comprehensive characterization of pore structures across different scales. Because there is no paired different-resolution images, we use down-sampled images (LR) as training data, and the original high-resolution images (HR) as reference. The workflow begins with high-resolution (2.68 μm/voxel) micro-CT imaging(HR) of a 6 mm core plug, which is downsampled to create low-resolution (10.72 μm/voxel) training images (SR) for a Super-Resolution Generative Adversarial Network (SR-GAN). The trained SR-GAN model then applied to all low-resolution images to generated the SR-GAN enhanced high-resolution image (SR-HR). These enhanced images (SR-HR) are segmented into macropores, solids, and microporous media, from which a macropore network model (macro-PNM) is extracted. For the sub-resolution microporous media, we employ high-resolution SEM image (0.1 μm/ pixel) to characterize micropore structures, which inform the generation of 3D microporous media and extraction of a micropore network model (micro-PNM). The final multi-scale pore network model is created by integrating both macro- and micro-PNMs, effectively capturing the heterogeneity of the rock sample and overcoming the resolution limitations of conventional imaging approaches. The proposed method is validated by comparing the permeability predicted from the multi-scale PNM with experimental measurements, showing excellent agreement and demonstrating its ability to accurately represent complex pore structures. The novelty of this study lies in its synergistic combination of super-resolution reconstruction, MPS-based 3D pore structure modeling, and multi-scale PNM generation. By bridging the gap between large-FOV imaging and fine-scale structural characterization, this approach provides a powerful framework for digital rock physics, significantly improving reservoir characterization and fluid flow predictions in heterogeneous formations. The integrated workflow not only addresses current imaging limitations but also opens new avenues for understanding multi-scale transport phenomena in porous media.