[S-CG55] Driving Solid Earth Science through Machine Learning

[SCG55-P17] Knowledge-guided machine learning for recognizing geochemical anomalies with mineralization

*Zhang Chunjie^1,2 (1.Earthquake Research Institute , the university of tokyo, 2.China University of Geosciences)

Keywords:Knowledge guided machine learning , Anomalies detection , Mineral prospectivity mapping

The identification of geochemical anomalies is important in mineral exploration and deep learning algorithms have become popular for recognizing these patterns. However, purely data-driven deep learning algorithms may not always align with geologic knowledge. In this study, a geologically constrained deep learning algorithm was proposed to extract geochemical anomalies associated with W polymetallic mineralization in China. This algorithm used fractal analysis to quantify the known mineral deposits and then used this knowledge to constrain an adversarial autoencoder network in identifying geochemical anomalies. The results showed that this approach produced more reasonable and interpretable results that were more consistent with regional metallogenic laws, compared to purely data-driven deep learning algorithms. This geological constraint improves the generalization ability of the algorithm and enhances the interpretation of results in geosciences.