The Challenge
Mineral exploration is one of the most challenging and capital-intensive activities in the mining industry. Predicting new ore deposits has a success rate of less than 1%, with exploration drilling representing a major cost centre for mining operations. Despite vast amounts of available data—geological assay results, lithological classifications, mineral concentrations, gravitational surveys, and magnetic data—traditional geological interpretation methods struggle to extrapolate trends reliably to unexplored areas.
Dundee Precious Metals, a major mining operation, sought innovative approaches to improve exploration success rates and reduce the substantial costs associated with speculative drilling. The challenge: could machine learning identify patterns in complex, multi-dimensional geological data that human experts might miss, and use these patterns to predict high-value targets in unexplored terrain?
Our Approach
New Gradient developed a novel 3D scene completion deep neural network specifically designed for geological prediction. Our approach treats mineral exploration as a spatial completion problem—given partial observations from existing drill holes, can we predict the mineral composition in the spaces between and beyond known data points?
We built our solution on state-of-the-art sparse transformer networks, implementing a custom 3D Masked Autoencoder (MAE) architecture for patch-masked training. This technique, adapted from computer vision advances, allows the model to learn rich representations of geological structures by predicting masked regions from surrounding context. The network processes heterogeneous data sources including drill hole assays, lithological boundaries, and geophysical measurements in a unified framework.
Crucially, our method is entirely geologically agnostic—it learns patterns directly from data without requiring pre-programmed geological rules or human interpretation of structural features. This flexibility means the system can adapt to different geological settings and deposit types without manual reconfiguration, making it broadly applicable across diverse mining operations.
The Outcome
Our 3D scene completion network achieved 0.74 weighted mean Average Precision (mAP) at predicting ore class quality across four categories: poor, mild, good, and high-quality ore. This level of accuracy enables mining operations to prioritise drilling targets with significantly higher confidence than traditional methods.
The approach earned recognition for innovation in Dundee Precious Metals' international 'Future Explorers' challenge, validating the potential of purely AI-driven geological prediction. By reducing speculative drilling through better target prioritisation, the technology offers potential cost savings of 40% in exploration programmes.
Beyond mineral exploration, our 3D scene completion methodology applies to any domain requiring spatial prediction from sparse observations. Oil and gas exploration, groundwater mapping, and subsurface infrastructure detection all present similar challenges where partial information must be extrapolated into unmeasured volumes. The success of this project demonstrates how advances in deep learning architectures can transform traditional resource industries.