Development of an AI-based system for automated mineral deposit mapping from geological data

Thesis Abstract

Accurate identification and mapping of mineral deposits are critical components of mineral exploration, yet existing methods rely heavily on manual interpretation of geological data, which is often time-consuming, subjective, and limited in scalability. This study addresses the need for a more efficient, consistent, and scalable approach by developing an AI-based system capable of automated mineral deposit mapping from geological datasets. The primary aim is to design, implement, and validate a machine learning framework that leverages integrated geological, geophysical, and geochemical data to accurately identify mineralized zones. Specific objectives include (1) reviewing existing GIS and remote sensing techniques employed in mineral deposit mapping; (2) developing an integrated data processing pipeline that consolidates multidimensional geological data; (3) designing and training a convolutional neural network (CNN) model for pattern recognition and deposit classification; (4) evaluating the model’s performance through comparative analysis with manual mapping; and (5) establishing operational guidelines for deploying the AI system in diverse geological settings. The research adopts a mixed-methods approach, combining quantitative machine learning techniques with qualitative assessments of model accuracy. The study’s population comprises geological datasets collected from mineral-rich regions, specifically sampled from a database of 3,000 borehole and outcrop records across a mineralized belt of a North American region. A stratified random sampling method selects 200 datasets, balanced across known mineralized and non-mineralized zones, to train and test the AI model. Data collection instruments include remote sensing imagery, geological maps, geophysical survey reports, and geochemical analyses. Data preprocessing involves normalization, feature extraction, and augmentation to enhance model robustness. The model’s training employs deep learning architectures, notably CNNs, optimized using cross-validation, loss function tuning, and hyperparameter adjustment, with performance metrics such as accuracy, precision, recall, and F1-score guiding iterative improvements. Validation of the model’s efficacy involves a comparative analysis with traditional manual mapping performed by expert geologists, employing statistical tests such as paired t-tests and ROC curve analysis. Expected findings suggest that the AI system will significantly outperform manual mapping in terms of accuracy, processing speed, and reproducibility. Preliminary tests forecast an overall classification accuracy exceeding 90%, with particular improvements in delineating subtle mineralized zones that are often overlooked or misinterpreted by manual methods. The system’s ability to integrate diverse geological datasets and learn complex spatial patterns demonstrates the viability of AI-driven approaches in mineral exploration. The research's contribution offers a substantial advancement to geospatial mineral mapping by providing an automated, scalable, and transparent decision-support tool, thus bridging gaps between traditional geological interpretations and modern data-driven technology. The study concludes that AI-based mineral deposit mapping systems can revolutionize mineral exploration workflows by reducing reliance on subjective interpretation, decreasing operational costs, and accelerating resource discovery processes. Recommendations include expanding the dataset diversity to improve model generalizability, integrating real-time sensor data for dynamic mapping, and developing user-friendly interfaces to facilitate adoption by geoscientists. Future research directions propose exploring transfer learning techniques for application in other mineral provinces and integrating 3D geological modeling to enhance spatial understanding. Overall, this thesis substantiates that intelligent systems, grounded in advanced machine learning algorithms, can significantly enhance the accuracy, efficiency, and reliability of mineral deposit identification, promising a new paradigm in geoscientific exploration and resource management.