Determination of uranium in uranium ore by laser-induced breakdown spectroscopy (LIBS) and a two-dimensional convolutional neural network (CNN)
Authors: Da Huang, Kaiqiang Shu, Yue‐Qin Tang, Qingyu Lin, Yixiang Duan
Journal: Instrumentation Science & Technology
Publication year: 2025
DOI: 10.1080/10739149.2025.2529308
Abstract
Uranium ore is a vital energy mineral resource, and its accurate exploration is of great significance in ensuring energy security and promoting the development of the industry. Laser-induced breakdown spectroscopy (LIBS) enables real-time quantitative analysis without the need for complex sample pretreatment, offering a unique advantage in elemental analysis. However, accurate quantification is still challenging for LIBS due to complex matrix effects. Herein, a method based on LIBS combined with a convolutional neural network (CNN) is reported for the accurate determination of uranium in uranium ore. The root mean square error of prediction (RMSEP), coefficient of determination (R2), relative error prediction (REP), and relative standard deviation (RSD) values of prediction on the test groups were used to evaluate the quantitative performance of CNN. The results showed that the RMSEP and R2 of the prediction of a two-dimensional convolutional neural network (2D-CNN) model were 0.0347 and 0.9874, respectively. In addition, the performance of 2D-CNN was compared with that of one-dimensional convolutional neural networks (1D-CNN), partial least squares regression (PLSR), and artificial neural networks (ANN). The results demonstrate that LIBS coupled with 2D-CNN achieved accurate quantification of the uranium and showed potential in the characterization of uranium ore.