Core Grain Size Analysis Using a Handheld Device

Abstract This study presents a novel handheld device for automated grain size analysis in geological cores, integrating hardware, software, and machine learning to optimize traditional labor-intensive methods. The developed tool combines a high-resolution camera, touchscreen display, and single-board computer to capture and process core images in near real-time. A YOLOv9-based segmentation model, trained on a custom dataset augmented by the Segment Anything in High Quality (SamHQ) model, identifies individual grains, while contour extraction and ellipse fitting algorithms quantify grain morphology. Post-processing employs a gradient boosting model to correct biases in median grain size estimates, aligning results with expert interpretations. Field tests demonstrate the device’s ability to generate accurate grain size distributions and depth profiles, with applications in reservoir characterization and lithological analysis. The solution offers a scalable, efficient alternative to manual core description, enhancing objectivity and reducing analysis time in both lab and field environments.