Johnson, S, Dagasan, Y & Banff, C 2024, 'Automated extraction of discing regions from core photography using computer vision at Sunrise Dam ', in P Andrieux & D Cumming-Potvin (eds), Deep Mining 2024: Proceedings of the 10th International Conference on Deep and High Stress Mining, Australian Centre for Geomechanics, Perth, pp. 137-152, https://doi.org/10.36487/ACG_repo/2465_03
(https://papers.acg.uwa.edu.au/p/2465_03_Johnson/)
Abstract: As the global mining industry seeks to meet the ever-increasing demand for metals, mineral deposits are trending deeper, larger, and lower grade. Engineers play a crucial role in ensuring the extraction process is efficient. A key aspect of decision-making involves using subjectively logged geotechnical datasets, commonly collected by geologists, geotechnical engineers, and technicians. However, these datasets are often excluded from critical mining models due to a lack of quality, auditability, consistency and completeness.
Recent advances in computer vision, particularly deep learning, have provided algorithms capable of efficiently identifying features of interest in core photography. Coupled with domain expertise, these algorithms can provide a computer vision-assisted core logging process, significantly increasing the value of downhole datasets. We have developed a novel workflow using computer vision to automatically identify discing regions, highlighting areas of the rock mass exposed to high in situ stress. By detecting every fracture and measuring their angles, we can define discs, group adjacent discs to create a discing region, and then group nearby regions to extract a consistent dataset. This workflow allows geotechnical engineers to establish a standard discing definition, facilitating high confidence in the data.
The outputs are evaluated against traditionally logged geotechnical data to create a detailed comparison. This study demonstrates that detailed, consistent, and auditable geotechnical data can be extracted using computer vision and core imagery, significantly improving data collection workflows for the mining industry. This approach enables mine planners and geotechnical engineers to proactively manage potential hazards, integrate these risks into mine design and scheduling, and ultimately ensure safer and more efficient mining operations.
The developed workflow was implemented at the Sunrise Dam Gold Mine (SDGM), demonstrating the advantages of computer vision-assisted core logging over traditional methods. This implementation underscores the potential of our approach to enhance the efficiency and reliability of geotechnical data collection in the mining sector.

Keywords: geotechnical data, drillcore, discing, geotechnical hazards, artificial intelligence, computer vision, machine learning