Geohazard Identification in Underground Mines: A Mobile App

Abstract

Mining operations are indispensable to the global economy, supplying crucial resources across various industries. However, the risks of underground mining, particularly geotechnical hazards like rockfalls and structural collapses, pose significant safety challenges. Traditional methods of hazard detection rely on periodic visual inspections, which can beinefficient, subjective, and dangerous. The need for more accurate, real-time hazard detection methods is crucial to prevent accidents and improve mine safety. Recent advancements in computer vision technology have drawn significant interest in the mining sector as a viable alternative for continuous and automated monitoring of environments that demand visual inspection. While computer vision has been widely adopted in surface mining and mineral processing, its application in the more challenging underground settings has been slower to develop due to obstacles such as limited visibility, connectivity issues, and dust. The goal of this thesis is to present a comprehensive methodology for developing and implementing a computer vision-based system for geotechnical hazard identification in underground mines. This methodology aims to be replicable and adaptable to others mining environments. This thesis provides a comprehensive literature review on the application of computer vision techniques for identifying geotechnical hazards in underground mines. It also introduces the Hazard recognition in underground mines application (HUMApp), a mobile application developed to enhance safety within underground mines by efficiently identifying geotechnical hazards, particularly focusing on roof falls, thereby enhancing traditional safety measures. HUMApp has been trained using real data captured from San Xavier Mining Laboratory, encompassing a total of 2,817 images from underground environments. A fully functional mobile application has been developed and implemented. The effectiveness of HUMApp was validated through a comparative analysis with assessments from two field experts, demonstrating a strong correlation between the app’s predictions and expert evaluations.