Portable Intelligent Lithology Identification via Deep Learning of Mesoscopic Rock Images

A field intelligent lithology identification method was proposed based on the deep learning of mesoscopic rock image to address the problems of traditional method relying on manual experience,strong subjectivity,and time-consuming and labor-intensive.Convolutional neural networks combined with transfer learning methods was used to constructed the lithology identification model.The identity sampler was used to select samples in a diverse way to construct training batches,ensuring that the model does not bias towards samples with large amounts of data and improving the accuracy of inference.Then,the trained Pytorch model was transformed into an ONNX model and deployed on the portable device,and a software specifically designed for lithology identification was designed and developed.In this software,electronic magnifying glasses can be used to capture mesoscopic rock images,achieving accurate lithology identification.The results show that the identification accuracy of this method can reach up to 99%,which can meet the need of geological investigators to quickly obtain rock lithology in field operations.