Application of deep neural network to discriminating chalcopyrite deposits, ore types, and textures

The trace elements of chalcopyrite have indicative significance in the formation process and genetic types of its deposits. They can also function as indicator minerals for mineral exploration. This study investigates the chalcopyrite in the global representative Ni-Cu sulfide deposits and Reef-type PGE deposits, and establishes deep neural network (DNN) models to discriminate the chalcopyrite deposits, ore types, and textures. The feature importance of DNN models is analyzed by calculating the mean SHapley Additive ExPlanations (SHAP) value of 26 elements in chalcopyrite, revealing the important discriminant elements in DNN models. Dimensionality reduction algorithms such as principal component analysis (PCA) are employed to the dataset and determine the decision boundary for the subsequent visualization analysis. A series of comprehensive experiments are conducted and the results reveal the DNN models to discriminate chalcopyrite deposits and ore types at accuracies of 99 % and 98 %, respectively. Based on the sulfide texture classification scheme proposed by the previous studies, we classify the chalcopyrite textures using the DNN models, with an accuracy of 96 %. The high-precision classification of sulfide textures helps to systematically explore the formation mechanism of sulfide-rich magmatic ores. Moreover, we compare our DNN models with traditional machine learning algorithms using the chalcopyrite dataset. This further illustrates the successful application of DNN models in the discrimination of chalcopyrite deposits, ore types, and textures. •High-precision discrimination of chalcopyrite deposits, ore types, and textures•Analyzing the contribution of trace elements in chalcopyrite to model predictions•Visualizing high-dimensional sample distribution and model prediction performance•Comparing proposed models with various machine learning models to show superiority