Rock Classification through Knowledge-Enhanced Deep Learning: A Hybrid Mineral-Based Approach

TL;DR

This work tackles automated rock classification from Raman-identified mineral assemblages, addressing the core challenge that identical minerals can define different rocks depending on their proportions and formation context. It introduces a knowledge-enhanced hybrid system that fuses a data-driven 1D-CNN mineral classifier (and an uncertainty-aware variant) with a rule-based expert system guided by QAPF diagrams and sedimentary classifications, augmented by a confidence-based decision mechanism. The approach yields 1D-CNN mineral accuracy of $98.37\%$ and $97.75\%$ for the uncertainty-aware variant, while rock-type classification shows variable performance across granite, sandstone, and limestone, with limestone best and sandstone most challenging, highlighting compositional ambiguity as a key limitation. The study provides a methodological framework and open data/tools for automated mineral-to-rock characterization, enabling improved material sorting and resource management, and lays groundwork for broader, multi-modality rock classification future work.

Abstract

Automated rock classification from mineral composition presents a significant challenge in geological applications, with critical implications for material recycling, resource management, and industrial processing. While existing methods using One dimensional Convolutional Neural Network (1D-CNN) excel at mineral identification through Raman spectroscopy, the crucial step of determining rock types from mineral assemblages remains unsolved, particularly because the same minerals can form different rock types depending on their proportions and formation conditions. This study presents a novel knowledge-enhanced deep learning approach that integrates geological domain expertise with spectral analysis. The performance of five machine learning methods were evaluated out of which the 1D-CNN and its uncertainty-aware variant demonstrated excellent mineral classification performance (98.37+-0.006% and 97.75+-0.010% respectively). The integrated system's evaluation on rock samples revealed variable performance across lithologies, with optimal results for limestone classification but reduced accuracy for rocks sharing similar mineral assemblages. These findings not only show critical challenges in automated geological classification systems but also provide a methodological framework for advancing material characterization and sorting technologies.

Figures (6)

• Figure 1: Schematic overview of the mineral-based rock classification framework incorporating knowledge-enhanced deep learning.
• Figure 2: Hierarchical relationship between Granite and its essential minerals
• Figure 3: Hierarchical relationship between Sandstone and its essential minerals
• Figure 4: Hierarchical relationship between Limestone and its essential minerals
• Figure 5: Performance comparison of different machine learning models for minerals classification. The bar plot shows the mean accuracy with standard error bars for five different classifiers: Support Vector Machine (SVM), Random Forest, Multilayer Perceptron (MLP), One-dimensional Convolutional Neural Network (1D-CNN), and 1D-CNN with unknown class handling (1D-CNN-UNK). Results are averaged across five-fold cross-validation.