3D Convolutional Neural Network-based 3D mineral prospectivity modeling for targeting concealed mineralization within Chating area, middle-lower Yangtze River metallogenic Belt, China
Xiaohui Li,Chen Xue,Yuheng Chen,Feng Yuan,Yue Li,Chaojie Zheng,Mingming Zhang,Can Ge,Dong Guo,Xueyi Lan,Minhui Tang,Sanming Lu
DOI: https://doi.org/10.1016/j.oregeorev.2023.105444
IF: 3.3
2023-04-24
Ore Geology Reviews
Abstract:The Chating area is situated within the Middle-Lower Yangtze River Metallogenic Belt, China. Several concealed skarn and porphyry-type deposits have been discovered in this area, indicating high potential for hosting hydrothermal deposits. However, due to the complex geological structure, exploration risks significantly increase with increasing depth. To overcome this challenge, three-dimensional mineral prospectivity modeling (3DMPM) has begun to be widely applied for mapping the prospectivity of deep-seated and concealed mineralization. However, most previous studies on 3DMPM were based on shallow supervised machine learning models and dimensionality-reduced 3D predictive maps. Although these models have shown good results, they may lose spatial correlation within the 3D predictive maps and fail to explore nonlinear correlations between the 3D predictive maps and mineralization. Meanwhile, 3D geological models are the most important basis of the 3DMPM, however, in the past, few studies have incorporated the optimization of the 3D geological models into the process of 3DMPM. Therefore, this paper initially builds and optimizes 3D geological models through implicit 3D geological modeling and "total litho-inversion" approach. Subsequently, the 3D predictive maps are generated by employing various 3D methods, which are further integrated using a 3D convolutional neural network (3D CNN) model to identify highly prospective areas for mineralization. The results show that the highly prospective areas identified by the 3DMPM include not only the training data but also other mineral deposits that have previously been discovered within the study area. In addition, compared with the Logistic Regression model (LR), Support Vector Machines (SVM), and Radom Forest (RF), the 3D CNN performs better prediction capabilities due to its enhanced ability to capture the correlations between 3D predictive maps and multiple types of mineral deposits. It suggests that the 3DMPM based on the 3D CNN model has commendable predictive capabilities in identifying prospective mineralization areas, and some new highly prospective areas can be considered as priority areas for future exploration of concealed mineralization within the Chating Area.
mineralogy,geology,mining & mineral processing