Applications of machine vision and machine learning for deposit characterization at the Castelo de Sonhos paleoplacer gold deposit, Brazil

Paleoplacer gold deposits represent a significant source of global gold production. The Castelo de Sonhos deposit in Brazil is a promising candidate for development and contribution to the paleoplacer proportion of gold production. Mineralization in the deposit is associated with metamorphosed conglomeratic units. This study applies machine vision methods to RGB drill-core imagery to identify and quantify the characteristics of large clasts (pebble to cobble-sized) within these units. Using this workflow, we demonstrate that gold-bearing material is preferentially found in specific ranges of clast geometry criteria, corresponding to an optimal depositional environment, offering a scalable and internally consistent support for manual logging. This approach enables rapid assessment of mineralization potential to support drilling decisions, and in the production environment, would enable rapid and robust material sorting, lowering costs associated with siliceous cobble-rich ore. These findings highlight the value of machine vision methods integrated with geochemical data and interpretation to enhance both exploration and operational efficiencies.