Classification of small specimen uranium ores using LIBS combined with machine learning and deep learning algorithms

Abstract

In this study, classification models based on laser-induced breakdown spectroscopy (LIBS) technology, combined with machine learning (ML) and deep learning (DL) algorithms, were proposed to enable the rapid and high-precision classification of small specimen uranium ores. LIBS spectral data from 12 types of uranium ore samples were collected and subsequently subjected to standard normal variate (SNV) preprocessing before model construction. The classification models were constructed using the random forest (RF) algorithm and three DL algorithms—feedforward neural network (FNN), convolutional neural network (CNN), and long short-term memory (LSTM)—incorporating two feature extraction methods: least absolute shrinkage and selection operator (LASSO) and principal component analysis (PCA). The classification performance and generalization ability of the different models and feature extraction strategies were systematically evaluated. It was found that the RF model exhibited significant overfitting when the training set size was small, and its performance improvement required an increase in training set proportion. When LASSO feature selection was incorporated, the DL models outperformed the RF model, although some misclassifications were still observed. In contrast, PCA, which utilized only the first five principal components, effectively retained the global discriminative information of the spectra. All DL models based on PCA features achieved 100% classification accuracy for both the training and testing sets. This study demonstrates that PCA can effectively extract global spectral information, overcoming the limitations posed by small specimens in LIBS classification tasks for uranium ores. When combined with DL algorithms, PCA significantly improves classification performance and generalization ability, offering a reliable technical pathway for the rapid and accurate identification of small specimen uranium ores.