From Pixels to Deposits: Porphyry Mineralization With Multispectral Convolutional Neural Networks

Mineral exploration is essential to ensure a sustainable supply of raw materials for modern living and the transition to green. It implies a series of expensive operations that aim to identify areas with natural mineral concentration in the crust of the Earth. The rapid advances in artificial intelligence and remote sensing techniques can help in significantly reducing the cost of these operations. Here, we produce a robust intelligent mineral exploration model that can fingerprint potential locations of porphyry deposits, which are the world's most important source of copper and molybdenum and major source of gold, silver, and tin. We present a deep learning pipeline for assessing multispectral imagery from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) with the objective of identifying hydrothermal alterations. Our approach leverages a convolutional neural network (CNN) to analyze the high-resolution images, overcoming computational challenges through a patch-based strategy that involves an overlapping window for partitioning the images into fixed-size patches. Through the utilization of manually labeled patches for image classification and identification of hydrothermal alteration areas, our results demonstrate the remarkable ability of CNN to accurately detect hydrothermal alterations. The technique is adaptable for other ore deposit models and satellite imagery types, providing a revolution in satellite image interpretation and mineral exploration.

Automated summary

Mineral exploration is crucial for sustainable supply of raw materials, and advancements in artificial intelligence and remote sensing technologies can significantly reduce the cost of these operations. A research team has developed an intelligent mineral exploration model using deep learning and satellite imagery, which accurately detects hydrothermal alterations, revolutionizing mineral exploration.