Abstract:
The traditional methods of assessing gold ore grades, although reliable often involve intensive fieldwork, extensive laboratory analyses and substantial costs. This research aimed to develop a cost-effective, reliable and environmentally friendly alternative by assessing gold ore grades using eucalyptus leaf imagery and concentration analysis. The specific objectives included determining gold concentrations in eucalyptus leaves and underlying ore deposits, modelling relationships between leaf gold concentrations and ore grades and predicting gold grades through a regression model based on eucalyptus leaf imagery features. Leaf and soil samples were systematically collected from Tira sub county, analysed for gold concentrations and associated with distinct visual characteristics (colour, texture, vein patterns). A convolutional neural network (CNN) regression model was developed to predict continuous leaf gold concentrations from leaf images. A calibrated cubic polynomial equation was then applied to translate predicted leaf gold concentrations into soil gold grades. Results demonstrated significant correlations between predicted and measured gold concentrations, highlighting distinct leaf visual features associated with varying gold concentrations. The model exhibited strong predictive capability, indicating potential applications in biogeochemical mineral exploration. This research not only provides an effective tool for miners to rapidly and economically estimate gold ore grades but also promotes sustainable mining practices through reduced environmental impact
