Transfer learning and siamese neural network based identification of geochemical anomalies for mineral exploration: A case study from the Cu[sbnd]Au deposit in the NW Junggar area of northern Xinjiang Province, China

Abstract

The identification of geochemical anomaly plays an important role in mineral exploration. The division of geochemical background and anomaly can be treated as a binary classification problem. Deep learning has shown encouraging performance in various fields of classification and prediction, and research has shown that geochemical anomalies can be effectively identified on the base of spatial characteristics and internal relations in geochemical data. However, deep learning models are often limited by the number of training samples, so new improved algorithms are needed to improve the accuracy of the model in anomaly recognition. Considering that the spatial patterns of geochemical usually have nested or hierarchical multi-scale characteristics, it is possible to improve model performance by integrating local and regional geochemical information. In this study, a combination of transfer learning and siamese neural network was used to improve the ability to extract multi-element geochemical anomalies, and multi-scale geochemical data was tried to improve model performance. The model accuracy of using both transfer learning and siamese neural network reached 85%, which showed that the improved deep learning method can greatly improve the ability of anomaly recognition of the model. The integration of multi-scale geochemical data made the model accuracy to 88%, which showed that the deep learning model also had a certain ability to recognize the information of the scale. Therefore, deep learning can effectively depict complex geochemical spatial patterns and implicit anomalies, which can be better applied to the recognition of geochemical anomalies through continuous improvement.