Mineral prospectivity mapping by deep learning method in Yawan-Daqiao area, Gansu

Mineral prospectivity mapping is similar to probability prediction using multi-source geological data. However, the complexity of geological phenomena creates difficulties for research. In this study, a deep regression neural network was built to map the mineral prospectivity in the Daqiao Gold Mine in Gansu Province, China. The neural network was trained using multi-source data including geological, geophysical, and geochemical data for the study area. The proposed deep regression neural network reveals the complex relationships between the mineral prospectivity map and geological, geophysical, and geochemical features, improving the prediction results. Moreover, the training dataset does not require classified samples. Training samples with continuous values can help improve the fault tolerance of the training dataset and reduce the uncertainty of positive samples. The experimental results showed that the proposed neural network learned previous expert knowledge related to mineral prospectivity mapping and can be applied to deep regression neural networks to predict and evaluate mineral resources using multiple data sources. The prospectivity map obtained in this study benefits the search for gold mineralization in the study area.

Automated summary

Mineral prospectivity mapping, utilizing multi-source geological data for probability prediction, was enhanced in this study with the use of a deep regression neural network. The neural network learned expert knowledge in mineral prospectivity mapping, did not require classified samples, and could be applied for predicting and evaluating mineral resources.