Pedram Ghamisi, Kasra Rafiezadeh Shahi, Puhong Duan, Behnood Rasti, Sandra Lorenz, Rene Booysen, Samuel Thiele, Isabel Cecilia Contreras Acosta, Moritz Kirsch, and Richard Gloaguen
Geology map of an open-pit mine in southern Spain (Corta Atalaya, Rio Tinto) created by applying a random forest classifier to hypercloud data.
There is a growing demand for mineral resources worldwide, and more than half of global material extraction originates from mining. Responsible sourcing of raw materials requires technological innovations for an improved extraction and efficient monitoring. Imaging sensors are non-invasive technologies that assist in detection of mineral deposits, and improve exploration and monitoring of mining activities. In this review, we assess recent developments in machine learning for processing of imaging sensor data. We describe the imaging technologies used for mineral mapping and raw material extraction at different scales ranging from satellites to lab-based measurements. With the growing number of space-borne and airborne platforms, and remote location and often extreme environment of mining operations, remote sensing data plays an increasingly important role. Mining poses special requirements on data acquisition and correction due to demanding environmental conditions and subtle differences in material properties. Labeled training samples are scarce; their collection is often costly and time consuming. The review covers several key research areas such as image preprocessing, denoising, supervised and unsupervised classification, multi-sensor data fusion, and 3-D data interpretation. We focus on innovative processing workflows and illustrate most prominent approaches with examples. We also provide a list of available resources, codes, and libraries.
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021-09-01.