Application of Machine Learning on Remote Sensing Data for Mineral Exploration (A Case Study)

This is a case study on the application of machine learning on remote sensing data. Remote sensing is the spectral data, the sun’s electromagnetic radiation reflected from the Earth, collected by satellites. A literature review was undertaken on the different machine learning applied to mineral exploration.  Mineral exploration is a multi-step process that includes determining the rock types of the region of interest and knowing the rock types determining the minerals present.

Machine learning methods are classified as supervised learning, where a labeled dataset is used to train a model, or unsupervised learning, which clusters the data into classes without using a labeled dataset.

The objective of this study was to apply supervised machine learning to remote sensing data collected by three different sensors carried by different satellites: Landsat-8 and ASTER sensors carried by satellites launched by NASA, and Sentinel-2 sensor taken by European Space Agency satellite. The data was for Sistan, a region in Iran. The remote sensing data, along with labeled data, were publicly available.

Ensemble methods Random Forest and Gradient Boosting were trained to classify the rock units of the region. The hyperparameters of the method were optimized using randomized search with cross-validation. Finally, the trained method was applied to the area, and a classified map of the rock types was produced. The unsupervised machine learning method, K-means clustering, was also used, and the resulting classification method was compared to the one made by the supervised method.