Deep Learning and Sentinel-2 data for artisanal mine detection in a semi-desert area

In sub-Saharan Africa, artisanal and small-scale mining (ASM) represents a source of subsistence for a significant number of individuals. While 40 million people officially work in ASM across 80 countries, more than 150 million rely indirectly upon ASM. However, because ASM is often illegal, and uncontrolled, the materials employed in the excavation process are highly dangerous for the environment, as well as for the people involved in the mining activities. One of the most important aspects regarding ASM is their localization, which currently is missing in most of the African regions. ASM inventories are crucial for the planning of safety and environmental remediation interventions. Furthermore, the past location of ASM could be used to predict the spatial probability of the creation of newborn mines. To this end, we propose a Deep Learning (DL) based approach able to exploit Sentinel-2 open-source data and a non-complete small-size mine inventory to accomplish this task. The area chosen for this study lies in northern Burkina Faso, Africa. The area is chosen for its peculiar semi-desert environment which, in addition to being a per se challenging mapping environment, presents a wide spatial variability. Moreover, given the high level of danger involved in field mapping, it is fundamental to develop reliable remote sensing-based methods able to detect ASM. Performance comparison of two convolutional neural networks (CNNs) architectures is provided, along with an in-depth analysis of the predictions when dealing both with dry and rainy seasons. Models’ predictions are compared against an inventory obtained by manual mapping of Sentinel-2 tiles, with the help of multitemporal interpretation of Google Earth imagery. The findings show that this approach can detect ASM in semi-desertic areas starting with a few samples at a low cost in terms of both human and financial resources.