Self-Supervised Graph Contrastive Learning for Mineral Prospectivity Mapping

The application of machine learning algorithms (MLAs) for mineral prospectivity mapping (MPM) is a significant frontier in mineral exploration. Supervised MLAs require a substantial number of labeled samples for training models; however, the rarity of mineralization leads to a scarcity of labeled training samples. Self-supervised learning can leverage large amounts of unlabeled data, providing a suitable solution for MPM in areas with few known mineral deposits. This paper introduces a graph-based self-supervised learning framework called implicit graph contrastive learning (IGCL) for MPM. This method uses augmentation in the latent space learned from a variational graph autoencoder by reconstructing the topological structure of a graph, thereby improving graph-contrastive learning efficiency without manual data augmentation. The model was employed to map potential iron polymetallic mineralization in southwestern Fujian Province, China. The results showed that the high-probability zones identified by IGCL were closely associated with known iron deposits. In comparison experiments with a supervised graph convolutional network model, the IGCL achieved a higher success rate and greater area under the receiver operating characteristic curve. The mineral prospective map obtained in this study provides guidance for further exploration of the study area.