Authors:
- Remis Balaniuk: Universidade Catolica de Brasılia and Tribunal de Contas da Uniao, Brasılia, Brazil;[email protected],
- Olga Isupova: University of Bath, Bath, UK; [email protected]
- Steven Reece: Dept. Engineering Science, Oxford University, Oxford, UK; [email protected]
This work explores the combination of free cloud computing, free open-sourcesoftware, and deep learning methods to analyse a real, large-scale prob-lem: the automatic country-wide identification and classification of surfacemines and mining tailings dams in Brazil. Locations of officially registeredmines and dams were obtained from the Brazilian government open data re-source. Multispectral Sentinel-2 satellite imagery, obtained and processed atthe Google Earth Engine platform, was used to train and test deep neuralnetworks using the TensorFlow 2 API and Google Colab platform. FullyConvolutional Neural Networks were used in an innovative way, to searchfor unregistered ore mines and tailing dams in large areas of the Brazilianterritory. The efficacy of the approach is demonstrated by the discovery of263 mines that do not have an official mining concession. This exploratorywork highlights the potential of a set of new technologies, freely available,for the construction of low cost data science tools that have high social im-pact. At the same time, it discusses and seeks to suggest practical solutionsfor the complex and serious problem of illegal mining and the proliferationof tailings dams, which pose high risks to the population and the environ-ment, especially in developing countries.