Abstract Digital imaging and modeling are essential tools for characterizing rock structures and understanding fluid flow behavior. These efforts often rely on X‐ray micro‐computed tomography (micro‐CT), which faces an inherent trade‐off between resolution and field‐of‐view (FOV). Deep learning super‐resolution (SR) methods have been developed to overcome this limitation, but their application to carbonate rocks is challenged by complex micro‐nanometer features. Due to the resolution limits, micro‐CT fails to capture sub‐micrometer features such as micropores in carbonates, and using such data as high‐resolution (HR) training images limits the SR model's ability to accurately reconstruct the micropore structures. We introduce a cascading SR pipeline designed to address these challenges and reveal sub‐micrometer features in carbonate rocks. The approach integrates multi‐stage 2D SR networks to progressively enhance low‐resolution (LR) images toward the HR domain, followed by a third‐plane SR network for 3D reconstruction. We evaluate this method on a three‐stage SR task: starting from a 3 m resolution micro‐CT image, super‐resolving to an intermediate 1 m resolution, and ultimately reaching 0.1 m resolution based on scanning electron microscopy (SEM), achieving a 30 scale factor. Validation with unseen SEM demonstrates that the reconstructed domains retain essential structural and physical properties. This approach provides a practical solution to current imaging limitations and enables the integration of multi‐resolution modalities for improved rock characterization.
Tang et al. (Fri,) studied this question.