conference paper · SPE Nigeria Annual International Conference and Exhibition
Sedimentary structures found in drill cores provide critical information about past depositional environments and assist in hydrocarbon exploration. This research applied the YOLO11 deep learning model to automate the detection and classification of sedimentary structures in drill core photographs. Using an open-source dataset of 222 annotated images, the model was trained across five sedimentary facies: cross-bedded sandstones, low-angle cross-bedded sandstones, massive sandstones, parallel-laminated sandstones, and mud drapes, representing fluvial to shallow-marine environments. Following training, the model attained a mean average precision of 0.426 at IoU 0.5, a recall of 0.76, a precision of 1.00, and an F1 score of 0.86. Despite the small size of the training dataset, the findings demonstrate that the YOLO11 algorithm can effectively identify and classify sedimentary rocks according to their physical structures.
Drill cores offer direct physical evidence of subsurface geology, which is essential for understanding ancient environments and locating hydrocarbon reserves. Automating the identification of geological structures using computer vision reduces the manual effort required to analyse core samples. This helps geoscientists interpret subsurface formations more systematically, even when working with restricted quantities of photographic training data.
This technology could assist exploration geologists and energy companies by automating core logging and facies identification during subsurface resource assessment. Because the model was trained and evaluated on a modest dataset of 222 open-source images, the research represents an applied, early-stage proof of concept. Further development and testing on larger, proprietary industry datasets would be needed before integrating the system into routine commercial exploration workflows.
AI-generated from the published abstract. Always read the original work before citing.
Abstract Sedimentary structures are vital in revealing paleodepositional environments and in hydrocarbon exploration. Drill cores are the most direct reflection of geology, especially where outcrop data are not available. In recent years, machine learning algorithms have been developed and used for easier identification and classification of these sedimentary structures. In this study, we examined drill core image samples obtained from open-source data for identifying and classifying sedimentary structures contained therein using the YOLO11 model. The YOLO11 model was used to detect and classify sedimentary structures within the core images. A dataset containing 222 annotated core images with five classes of sedimentary facies, namely cross-bedded sandstones, low-angle cross-bedded sandstones, massive sandstones, parallel-laminated sandstones, and mud drapes, was used. The cross-bedded sandstone facies have strata that are inclined to the main bedding plane. The facies are medium-grained. The low-angle cross-bedded sandstones are characterized by inclined sandy layers, with a relatively low angle of inclination. They commonly form on the lee side of ripples and dune bedforms in the upper flow regime. The massive sandstone has been identified as ungraded, with medium to coarse grain size. It lacks recognizable internal sedimentary structures. The parallel-laminated sandstones are identified by their parallel laminae of a few millimeters thick. The mud drape facies contain thin clay (mud) layers. These facies may have originated from grain falls. The five sedimentary facies are believed to have been deposited in fluvial to shallow-marine environments. After training the model, its performance was evaluated using standard metrics, and the results were as follows: mAP@0.5 of 0.426, recall of 0.76, precision of 1.00, and F1 score of 0.86. The results presented here indicate that the YOLO11 algorithm is optimal and effective for the classification of sedimentary rocks, especially on the basis of their sedimentary structures, despite the limited dataset.
This page summarises published work. The authoritative version sits with the publisher.
DOI: 10.2118/234819-ms
Is something wrong with this record? Report it or request removal.
Discussion
Have you built on this work, tried to replicate it, or seen it applied in practice? Share what you know. Verified researchers and MARATTO™ domain experts can open a discussion, and any member can reply. Contributions are reviewed before they appear.
No discussion yet. Open the first thread.
New to MARATTO™? Create a free account.
