Extracting consistent geotechnical data from drill core imagery using Computer Vision at the Carrapateena deposit

As the global mining industry seeks to meet the ever-increasing demand for minerals, there is a corresponding increase in drilled metres. However, many operations are struggling to find skilled geologists and geotechnical engineers available to produce high quality logging data. Subjective data commonly logged by geologists and geotechnical engineers are increasingly not being included in critical mining models due to a lack of auditability and consistency. Recent advances in computer vision - specifically in the field of deep learning - have provided algorithms that can efficiently augment and automate the manual drill core logging process. Using these computer vision-based workflows, we analyse traditional RGB drill core photography to generate fracture frequency, a commonly logged geotechnical dataset. These outputs are evaluated against experienced loggers to create a detailed, quantitative comparison to highlight the strengths and weaknesses of both data collection methods. The geotechnical data generated from the computer vision-based methods produced data that was highly comparable to that of the traditionally logged datasets in terms of accuracy but offered significant benefits in terms of speed and auditability. In addition, the computer vision generated data was able to measure more features in greater detail than would be practical to do with a human logger. This work demonstrates that detailed, consistent, and auditable geotechnical data can be generated using computer vision and core imagery, and in turn greatly improve data collection workflows for the mining industry.