One-dimensional convolutional neural network for mapping mineral prospectivity: A case study in Changba ore concentration area, Gansu Province

Abstract

Predictive modeling of mineral prospects is a critical but challenging procedure for delineating undiscovered prospective targets in mineral exploration. In the present study, the one-dimensional convolutional neural network (1D CNN) was utilized for mineral prospectivity mapping (MPM). Local singularity analysis and deposit buffer analysis were used to determine the unfavorable metallogenic regions, and random points were taken as negative samples. Synthetic minority over-sampling technique (SMOTE) was used to generate the training data to enhance the generalization ability of the model. The effects of different hyperparameters on classification performance in 1D CNN were studied and the optimal combination of hyperparameters was determined. The average classification accuracy of this model was 96.2%, and the standard deviation was 2.14%, indicating that the model constructed by this hyperparameter set was robust. On this basis, the geochemical data of 19 elements were used as the input characteristic variable, and the prospecting prospect of lead–zinc deposits in Changba ore concentration area was predicted based on 1D CNN. The results showed that the prospectivity map of lead–zinc deposits generated by the 1D CNN model can effectively link the multivariate geochemical data with the known positions of lead–zinc deposits and greatly increase the precision of the potential exploration areas for lead–zinc deposits.