Abstract: Geological analysis can provide important information and basis for geological resource exploration. Laser-induced breakdown spectroscopy (LIBS) can provide a rapid, accurate, and in-situ discriminant method for rock analysis. The application of high repetition rate LIBS technology to lithology analysis of rock samples, combined with the convolutional neural network (CNN) model for classification, not only solves the problems of large volume and heavy weight of traditional high-energy single-pulse lasers, but also overcomes the shortcomings of poor universality of traditional machine learning algorithm models, and also comforms to the development trend of LIBS technology towards portability, miniaturization, precision and intelligence. A high repetition rate LIBS experimental platform is used to collect the spectra of rock compression samples, and the rock samples are divided into 5 and 9 categories according to the origin and lithology of the rocks, and then 1D-CNN and ResNet34 convolutional neural network models are used to classify them. The results show that the combination of high repetition frequency LIBS and CNN can achieve rock classification, with the optimal results of 99.43% and 97.14% respectively. Finally, based on MATLAB App Designer, a system software for rock lithology analysis is developed, which realizes the rapid real-time classification of rocks and greatly improves the efficiency of geological exploration.

Key words: spectroscopy, rock classification, high repetition rate laser-induced breakdown spectroscopy, convolutional neural network