Computer vision and image-based sensing in mine-to-plant operations and extractive metallurgy: A bibliometric and systematic literature review
Cotrina-Teatino M.A.; Marquina-Araujo J.J. · Minerals Engineering (2026) · DOI: 10.1016/j.mineng.2026.110256
Resumen
The accelerating digitalization of the mining and metallurgical industries has driven the adoption of computer vision and image-based sensing technologies. However, research in this field remains fragmented across isolated applications in exploration, processing, and structural monitoring. This fragmentation hinders a comprehensive understanding of technological evolution, key scientific contributors, and existing methodological gaps. Therefore, the main objective of this study was to comprehensively characterize the intellectual structure, trends, thematic clusters, and future research directions in computer vision and image-based sensing applied to mine-to-plant operations and extractive metallurgy. A combined bibliometric and systematic analysis was conducted, drawing from publications between 1955 and 2026 sourced from Scopus, Taylor & Francis, SpringerLink, ScienceDirect, and IEEE Xplore, following the PRISMA protocol. A total of 933 articles were analyzed using performance metrics, co-occurrence maps, collaboration networks, and thematic evolution analysis, complemented by a qualitative review of the most influential studies. The results revealed three application-grounded clusters: (1) remote sensing and geological intelligence for resource characterization and prospectivity; (2) process vision and metallurgical control in mineral processing, including soft sensors and control-relevant visual analytics; and (3) real-time operational perception and safety monitoring in mining systems, emphasizing edge-ready detection/segmentation pipelines, geotechnical mapping/digital twins, and safety-critical monitoring. Building on these findings, the main future research directions include multimodal and interpretable prospectivity models under mixed/vegetated cover, operational hyperspectral pipelines, label-efficient learning (self-/semi-supervised and active learning) to reduce annotation burden, robust edge deployment under harsh conditions and domain drift with standardized reporting of latency/hardware, and tighter integration of perception outputs with operational decision-making and early-warning frameworks. Collectively, this study provides a quantitative and systematic overview of the field and a deployment-oriented foundation for advancing toward more autonomous, predictive, and resilient mine-to-metallurgy systems.
Instituciones: Universidad Nacional de Trujillo