Towards a more automated mineralogy

Automated mineralogy has been around for more than thirty years as the reference technique to support process mineralogy and geometallurgical studies. Pioneering techniques such as the QEM-SEM have been further improved to benefit from the latest hardware developments in terms of stable, sensitive and efficient imaging. FEG sources, solid state x-ray detectors and almost real-time signal processing have made the most significant contributions in the last decade. A quick overview of existing technologies reveals that further improvement could be gained by merging different imaging modes and using more advanced classification algorithms such as those well known in machine learning and remote sensing. If such algorithms were complemented with hierarchical databases wherein only important economic minerals are listed with their associated probabilities of occurrences, significant improvements could be gained in automated mineral identification. This paper contributes to demonstrate that a real “mineral intelligence” of ores and materials is now within reach