Accurate identification of geological samples using artificial neural network processing of laser-induced breakdown spectroscopy data

Mineral and rock identification is a fundamental analysis in geological study. It allows the retrieval of both physical and chemical information of an identified sample from an available database. In this study, laser-induced breakdown spectroscopy integrated with artificial neural network algorithm is proposed for geological sample identification. The training algorithm of the artificial neural network is modified from the conventional method which is used in our previous studies. The trained network is tested by a set of natural rock samples which include new rocks which are not in the certified training set. Despite the difference in surface texture and minor variation in chemical composition of the tested rocks as compared to the samples of the training set, the validation reports a higher correct identification rate. This demonstrates the robustness of the modified algorithm to assess the variation of samples and the readiness to recognize new samples for a detailed study.