Automated Mineralogy and Effective Diffusivity Prediction from Drill Core Images Using Computer Vision and Deep Learning

We present an AI-based workflow for predicting mineralogy and effective diffusivity directly from highresolution drill core images using computer vision and deep learning. Convolutional neural networks, trained via transfer learning, accurately predict rock composition and mineral content. Especially, the predicted clay content is then used to estimate effective diffusivity through an empirical model. This approach demonstrates how AI and ML can support mineralogical or diffusivity estimation based on core image input. The approach could be expanded to train geophysical well log input, combined with MultiMin analysis for predictions of other properties, such as conductivity, porosity, etc and may thus support future applications for efficient, high-resolution subsurface analysis.