Lightweight three-dimensional superresolution reconstruction technique for rocks using a stochastic degradation...

Area of Science:

• Geoscience and Imaging Technologies
• Petroleum Engineering
• Materials Science

Background:

• Computed tomography (CT) is crucial for analyzing rock microstructures and properties.
• Real-world CT imaging suffers from degradation, limiting analysis accuracy.
• Current superresolution (SR) methods struggle with real rock CT data due to reliance on paired or inadequate simulated data.

Purpose of the Study:

• To develop a robust 3D superresolution (SR) reconstruction method for rock CT images under real-world degradation conditions.
• To address the limitations of existing SR techniques that fail to adequately represent actual degradation processes.
• To enable high-precision digital core analysis through improved image quality.

Main Methods:

• Proposed a novel approach integrating a physics-driven stochastic degradation model with a lightweight 3D network architecture.
• The degradation model uses randomized operations to emulate diverse real-system degradations, enhancing synthetic-to-real distribution alignment.
• A batch normalization-free lightweight 3D network was employed for computational efficiency and preservation of spatial fidelity.

Main Results:

• Achieved superior reconstruction quality and preservation of physical characteristics on real rock CT images.
• Demonstrated a robust synthetic-to-real framework trained solely on physics-compliant synthetic data.
• The proposed method significantly enhances image quality, overcoming low-resolution limitations.

Conclusions:

• This work presents an efficient and reliable solution for rock CT image enhancement.
• The developed method is valuable for high-precision digital core analysis in petroleum exploration.
• The approach offers insights into geoscientific image reconstruction challenges.