COMBINING DRILL-CORE HYPERSPECTRAL AND GEOCHEMICAL DATA BY DEEP LEARNING TO ENHANCE DRILL-CORE MINERAL MAPPING

Drill-core mineral mapping is critical in orebody modelling, but traditional analysis based on manual geological logging is time-consuming, subjective, and prone to bias. Hyperspectral (HS) imaging provides a quick, high-resolution, and non-invasive alternative; however, traditional approaches to analyse HS data also rely on human interpretation, which limits efficiency and reliability. Recent advances in deep learning (DL) offer ways to handle these challenges by automatically extracting complex spatial and spectral features from massive datasets (e.g., HS data), eliminating the need for human feature engineering. This study integrated drill-core HS data (in a form of reflectance spectra) and geochemical data (geochemical assay) with DL methods to enhance drill-core mineral mapping. The research was guided by three objectives: (i) quantifying uncertainty in drill-core mineral mapping using DL approaches, (ii) evaluating and comparing the performance of convolutional neural networks (CNNs) and recurrent neural networks (RNNs), specifically long-short term memory (LSTM) models, and (iii) developing a hybrid CNN-LSTM model to enhance mineral mapping. The results demonstrated that all three models had high predictive performances with low MAEs and uncertainty. The hybrid CNN-LSTM model outperformed the stand-alone models, with lowest MAE (0.0258), followed by the LSTM model (MAE = 0.0278), and CNN model (MAE = 0.0312). The improved performance of the hybrid model demonstrated the benefit of integrating spatial–spectral feature extraction with sequential dependency modelling, whereas the relative strength of LSTM over CNN emphasizes the importance of sequential features. This research demonstrated that integrating drill-core HS and geochemical data using DL methods can significantly enhance drill-core mineral mapping and mineral abundance estimation in drill-core datasets, while acquiring low MAEs.