Panoptic Segmentation for Microcrystalline Calcite in SEM Petrographic Images for Carbonate Reservoir Characterization

Summary The paper presents MicroCrystalNet, a novel deep learning framework for automated characterization of microcrystalline calcite in carbonate reservoir SEM petrographic images. Microporosity, typically manifesting as intercrystalline pore systems bounded by microcrystals (<10 μm), significantly influences reservoir properties. Traditional SEM analysis methods have been limited by sample charging effects, matrix occlusions, and operator bias. The framework integrates Meta’s Segment Anything Model for instance segmentation (97.6% accuracy) with a custom convolutional neural network featuring a Feature Selection and Reduction block. Trained on 1,852 labeled microcrystals from 48 SEM images, including both ancient reservoirs and modern analogs, MicroCrystalNet achieves 92.12% Top-1 accuracy in classifying crystals into four morphological categories: Euhedral/Subhedral-Rhombic, Euhedral/Subhedral-Polyhedral, Anhedral-Amorphous, and Anhedral-Spherical. The model demonstrates superior performance compared to benchmark CNNs like VGG16 and ResNet50, with precision, recall, and F1 scores exceeding 0.90 across all classes. This advancement enables high-throughput quantitative analysis of microporous carbonate reservoirs, facilitating improved understanding of pore system characteristics and their impact on reservoir properties. The framework’s applicability extends to diverse carbonate lithologies, offering broad applications in reservoir characterization.