Computer Assisted Petrography and Core Logging

Geological interpretation relies on petrographic descriptions for its interpretation. These descriptions are subjective, observer-dependent, and difficult to reproduce faithfully. This lack of reproducibility represents a fundamental weakness in a discipline aspiring to scientific rigour.

Beyond reproducibility, the textual and qualitative nature of traditional petrography complicates its integration into modern analytical pipelines and artificial intelligence. Numerical data, by contrast, are unambiguous, reproducible at scale, and directly exploitable by machine learning algorithms. Petrographic descriptions are difficult to integrate into metallogenic fertility prediction maps or geographic information system databases. The absence of robust quantitative petrographic frameworks limits geological advancement.

The solution proposed here consists of building a structured set of mathematically grounded petrographic descriptors, derived from descriptive geometry and analytical functions. These descriptors quantify grain shape (aspect ratio, convexity, circularity, roughness, etc.) and the spatial arrangement of minerals. They produce feature vectors exploitable by AI while preserving geological traceability: each descriptor retains physical and geometric meaning, maintaining geological logic throughout the analytical chain. This approach enables rigorous and reproducible documentation of petrographic observations.

A major challenge remains: segmentation—obtaining the outline of each mineral, which is complicated by twinning, inclusions, exsolutions, hydrothermal alteration, metamorphism (particularly greenschist facies), etc.

Quantitative petrography is the revolution that geology needs.

This research presents an advanced deep learning framework for the automated identification of mineral grains from high-resolution photomicrographs. The study is situated in a context in which mineral identification plays an essential role in mineral exploration, civil engineering and environmental applications. In practice, this task is often still carried out manually with a microscope or using analytical techniques, such as X-ray fluorescence (XRF) or X-ray diffraction (XRD). Although accurate, these methods remain slow, sequential and difficult to apply on a very large scale. And yet, in many contexts, between 100,000 and 1,000,000 grains need to be analyzed. Traditional approaches are inefficient in such situations. The main objective of this presentation is therefore to demonstrate that an artificial intelligence-based approach can make the identification process simple, quick and efficient.

Gold exploration in glaciated terrain traditionally relies on the analysis of gold grains in sediments. The morphology and abundance of these grains are used as indicators to locate mineralization. However, traditional visual methods remain qualitative and painstaking. This study proposes a novel quantitative approach based on principal component analysis of 73 diverse shape descriptors, divided into two groups, in order to capture both large- and small-scale morphological characteristics. The results are compared with those of ARTMorph, an artificial intelligence-based image processing routine trained and corrected by geologists. The results of this quantitative method combined with ARTMorph classification make it possible to characterize gold grain morphologies by classifying the grains according to their intensity of deformation, thus reflecting their morphological evolution as a function of sedimentary processes. Large-scale shape analysis organizes shapes into a dispersion cone, the extremes of which correspond to smooth, stubby grains with complex outlines or elongated, rough grains. For small-scale (roughness) analysis, a single component is sufficient, as it explains 63% of the variance. This factor allows grains to be classified from smooth to rough with a complex outline. The approach reveals evolutionary trends consistent with geological observations and constitutes an objective alternative to traditional visual classifications, improving the morphological classification of gold grains.

This presentation will provide an update on the core digitization initiative underway at Agnico Eagle since 2021 and highlight recent advances in leveraging geoscientific data acquired on drill core. Through the integration of high-resolution imaging, spectral and elemental analysis and artificial intelligence tools, core digitization is becoming a key lever for improving understanding of deposits and supporting decision-making across the mining value chain.

The presentation will also highlight the technical progress made over this period, including the extraction of vein and texture signatures from drill core imaging to characterize mineral associations and morphological characteristics. It will also demonstrate how these signatures, combined with high-resolution mineralogical and elemental data, are used to develop predictive models for metallurgical performance and environmental behaviour, including hardness and acid mine drainage potential. Examples from ongoing projects will demonstrate how these approaches help transform data generated by drilling programs into quantitative, reproducible and directly actionable information for exploration, engineering and mining operations.