Abstract
Identifying the type of phosphor is a critical step in the light emitting diode (LED) e-waste recycling process. However, the similarities in color and chemical composition among different phosphors pose significant challenges for their accurate identification. In recent years, laser-induced breakdown spectroscopy (LIBS) and machine learning have shown great potential in the field of rapid material identification. In this study, the spectra of eight common LED phosphors (rare-earth phosphor, heavy-metal-containing phosphor, and common phosphor) were collected separately using LIBS, and based on these spectra, random forest (RF), support vector machine (SVM), and two-dimensional convolutional neural network (CNN) classification models were established for phosphor identification. The results show that both RF and SVM models have accuracy above 95%, while CNN showed the best prediction with the best accuracy of up to 99.47% for the test set for different types of phosphors. This study demonstrates the great potential of combining LIBS technology with CNN algorithms in the field of LED e-waste recycling.