Case Study - AI-Driven Mineral Exploration
Award winning AI approach to mineral exploration
Summary
New Gradient's machine learning engineers developed an innovative 3D scene completion deep neural network method for predicting new areas of high mineral concentration. Our AI approach to mineral exploration won recognition for its innovation as part of an international minerals exploration challenge in collaboration with Dundee Precious Metals.
Novel AI approach to mineral exploration
Predicting new ore deposits for exploration and extraction of precious metals is an extremely challenging task with a success rate of less than 1%. Despite the vast amount of available data from drill holes, including geological assay data, lithological classification, and mineral concentration, along with other sources such as gravitational and magnetic data, traditional geological interpretations and attempts to extrapolate trends to new areas have proven ineffective.
The success of our AI-driven mineral exploration method demonstrates New Gradient's expertise in applying cutting-edge machine learning techniques to solve complex problems in traditional industries. The flexibility of our approach makes it valuable not only for mineral exploration but also potentially for other 3D exploration and extractive industries, such as oil field exploration. This project showcases our ability to develop innovative artificial intelligence solutions that can significantly improve the efficiency and success rate of resource discovery.
We developed a 3D scene completion deep neural network that leverages state-of-the-art sparse transformer networks. Our approach applied a novel 3D Masked Autoencoder (MAE) method for patch masked training and dense scene completion. This innovative technique proved highly successful at predicting the quality of ore deposits in new locations, achieving a 0.74 weighted mean Average Precision (mAP) at predicting ore class quality in unexplored areas across four classes: poor quality, mild quality, good quality, and high quality ore.
Our method was recognized for its innovation, earning a notable submission award in the 'Future Explorers' international mineral exploration challenge backed by Dundee Precious Metals. This achievement was particularly impressive given our solely AI-based approach, which is entirely geologically agnostic and flexible in terms of input and output, making it usable without prior geological human understanding.
