Mineral Prospectivity Mapping via Gated Recurrent Unit Model

The application of deep learning algorithms in mineral prospectivity mapping (MPM) is a hot topic in mineral exploration. However, few studies have focused on recurrent neural networks (RNNs) in terms of integrating different evidential layers to map mineral potential. In this study, a gated recurrent unit (GRU) model was employed for MPM using a case study on the Baguio District of Philippines. To generate sufficient training samples for GRU, data augmentation with geological constraints was employed. To explore the influence of different orders of evidence layers, as inputs of RNN, a total permutation of the evidence layers was built. Meanwhile, a nonlinear controlling function was used to capture the spatial relationships between known mineral deposits and geological controlling factors and assign the value for each pixel of evidence layers. The obtained results demonstrated the excellent performance of GRU in MPM. The delineated high-anomaly areas show close spatial relationships with known mineral deposits and therefore can provide significant clues for the next round of mineral exploration in the study area.

Automated summary

This study utilized a GRU model for mineral prospectivity mapping in the Baguio District of Philippines, incorporating data augmentation and building total permutations of evidence layers for training samples. The results demonstrated the excellent performance of GRU in MPM, with delineated high-anomaly areas closely related to known mineral deposits.