Deep Learning Enhanced Pore Segmentation and Classification and Pore Type-Based Porosity and Pore Size Analysis in Shale with High-Resolution Large-Area Sem

Abstract

This study applies a deep learning-based workflow to enhance pore segmentation and classification in high-resolution, large-area scanning electron microscopy (SEM) images of shale samples from the Delaware Basin and Eagle Ford Formation. To overcome limitations of traditional SEM workflows—such as small field-of-view and qualitative analysis—we utilize image areas exceeding the representative elementary area (250–1000 μm in length) to enable statistically robust quantitative analysis. An iterative U-Net segmentation model, refined through local correction, achieves high accuracy across five shale samples, effectively addressing challenges including overlapping grayscale intensities, milling artifacts, and charging effects. A stepwise classification algorithm further categorizes pores into organic matter-lined (OML), clean mineral (CM), and intraparticle (Intra) types. Quantitative analyses of pore size distribution and porosity by pore type provide critical insights on the contributions from different types of pores that are relevant to oil and water saturation and flow, critical insights not accessible through conventional techniques such as mercury intrusion capillary pressure (MICP) or nuclear magnetic resonance (NMR). The pore size distribution derived from image analysis shows good agreement with MICP-derived pore throat sizes, while porosity estimates are compared against helium pycnometry measurements. The causes and implications of observed similarities and discrepancies are discussed, including the influence of sub-resolution pores and sample heterogeneity. This innovative workflow establishes a scalable, image-based methodology for pore-scale reservoir characterization, offering a powerful tool for advancing quantitative shale reservoir evaluation and improving predictions of oil and water production.

Keywords: deep learning, large-area SEM, pore segmentation, pore type, Pore size distribution, porosity