Mineral Texture Classification Using Deep Convolutional Neural Networks: An Application to Zircons From Porphyry Copper Deposits

The texture and morphology of igneous zircon indicates magmatic conditions during crystallization can be used to constrain provenance. Zircons from porphyry copper deposits are typically prismatic, euhedral, strongly oscillatory zoned which may differentiate them zircons associated with unmineralized systems. Here, cathodoluminescence images the Quellaveco district, Southern Peru, were collected compare textures between premineralization Yarabamba Batholith deposit. zoned, whereas batholith subhedral-anhedral weaker zoning. We adopt a deep convolutional neural network (CNN) approach demonstrate that CNN classify high success. trial several architectures images: LeNet-5, AlexNet VGG, including transfer learning where we weights VGG model pretrained on ImageNet data set. is most effective approach, accuracy receiver operating characteristic-area under curve (ROC-AUC) scores 0.86 0.93, indicating CL image ranked higher than 93% probability. Visualizing layer outputs demonstrates models recognize crystal edges, zoning, mineral inclusions. implementing trained as unsupervised feature extractors, empirically quantify morphology. Therefore, provides tool for extraction information large, imaged-based petrographic sets facilitate petrologic provenance studies.