Deep Learning For Lithological Mapping: A New Paradigm In Geological Interpretation

Abstract

The lithological mapping is the key factor to explore the subsurface structure of the Earth which has traditionally been based on the manual interpretation of geophysical data, field surveys, and experience of experts. Nevertheless, the existing methods are rather slow and subject to human error. As deep learning technologies and especially convolutional neural networks (CNNs) and recurrent neural networks (RNNs) are being developed, lithological mapping is taking a new and revolutionary turn. In this paper, the use of deep learning models to automatize and increase the accuracy of the lithological classification based on remote sensing data, hyperspectral imagery, and borehole logs is discussed. We introduce an entire process of data preprocessing, model training, and model verification of predictions on a test geological area. The inferences indicate substantial increase in the accuracy of classification and spatial adherence than the conventional machine learning applications. The suggested strategy suggests a paradigm change in the geological interpretation, which can enable scaleable, efficient, and more objective lithological mapping.