The Segment Anything Model (SAM) for Mining Engineering Applications: From Zero to One Shot

In the era of Mining 4.0, the global mining community is witnessing the convergence of technical nuances of mining with the exceptional capabilities of Artificial Intelligence (AI). One emerging sphere within the AI domain in the mining sector is computer vision. A notable instance is the Segment Anything Model (SAM), a robust pre-trained machine-learning model for image segmentation developed by Meta AI and released in April 2023. SAM gained instant recognition among researchers for its capability of predicting masks via zero-shot learning, i.e., enabling the segmentation of objects that have not been seen before. SAM has been applied and tested and has shown immense potential in various research fields like medical imaging, remote sensing, and satellite imagery. However, the potential implementation of SAM in the mining industry remains relatively underexplored. This paper aims to explore the potential applications of SAM within the mining industry by catering to academic researchers and mining practitioners. The broad areas of application of SAM include rock fragmentation analysis, drill-core analysis, geotechnical monitoring, land use analysis, and real-time equipment monitoring. Given SAM’s extensive and robust training on a vast dataset, the requirement for additional customized training, particularly for its mining applications, seems minimized. This inherent adaptability establishes SAM’s potential to integrate seamlessly into the mining domain.