Speaker
Description
Super-resolution deep-learning models are increasingly used in Digital Rock workflows to address the inherent trade-off between field of view and resolution in rock micro-CT imaging. This trade-off limits analyses requiring high-resolution characterization across broad spatial domains, particularly critical for heterogeneous rocks such as Brazilian pre-salt carbonates. These carbonates exhibit complex, multi-scale pore structures with significant micro-porosity, where representative elementary volumes (REVs) demand large sample sizes to capture geological variability, yet essential pore-scale features controlling fluid flow and storage require high-resolution imaging.
Super-resolution models address this challenge by computationally enhancing lower-resolution images acquired over larger fields of view to approximate high-resolution scan quality. Neural networks trained on paired high- and low-resolution datasets learn to reconstruct fine-scale pore structures and textural details otherwise requiring prohibitively expensive scanning protocols or exceeding hardware capabilities. This enables acquisition of lower-resolution micro-CT scans across representative volumes followed by super-resolution enhancement to recover pore-scale features critical for accurate property estimation. Consequently, super-resolution techniques can eliminate the traditional choice between spatial coverage and resolution, enabling comprehensive multi-scale characterization where micro-porosity networks and macroscopic heterogeneity are simultaneously represented.
This research establishes a benchmark for super-resolution in the publicly available dataset "16 Brazilian Pre-Salt Carbonates: Multi-Resolution Micro-CT Images" [1]. This dataset consists of micro-CT images of high and low resolutions, along with their corresponding segmentations, from 16 carbonate samples from the Brazilian pre-salt formations. We explore a 2D super-resolution task with 4× amplification using distinct neural-network architectures, data augmentation strategies, and different methods to couple the super-resolution and segmentation tasks. Results demonstrate that super-resolution models effectively enhance image detail while preserving pore network statistical properties. Comparative analysis of petrophysical properties, including porosity and pore size distributions, from super-resolved images shows strong agreement with ground truth high-resolution acquisitions. These findings indicate that super-resolution techniques effectively mitigate the field of view/resolution trade-off in micro-CT analysis of pre-salt carbonates, enabling multi-scale characterization workflows balancing computational efficiency with physical accuracy.
| References | [1] - Vidal, A.D., Neta, A.P.M., de Castro Vargas Fernandes, J. et al. Multi-resolution X-ray micro-computed tomography images of carbonate rocks from brazilian pre-salt. Sci Data 11, 1361 (2024). https://doi.org/10.1038/s41597-024-04198-9 |
|---|---|
| Country | Brazil |
| Student Awards | I would like to submit this presentation into both awards |
| Acceptance of the Terms & Conditions | Click here to agree |
