Discrimination of precious and semi-precious gemstones using laser-induced breakdown spectroscopy and machine learning approaches

Abstract

Laser-Induced Breakdown Spectroscopy (LIBS) is a very simple but capable spectroscopic analytical method that performs simultaneous multi-element analysis in a single laser shot. By using LIBS, we can identify artificial gemstones that look like real ones. In this study, the samples that are used include five different gemstones namely Sapphire, Emerald, Amethyst, Tourmaline, Topaz, and two artificial replicas of Amethyst and Emerald that closely resemble their original counterparts. The LIBS spectra were collected from each of the gemstones by ablating them with a Q-Switched Nd: YAG Laser of wavelength 1064 nm, pulse width (6–10 ns), and maximum energy of 1000 mJ. The radiation from the resulting plasma was recorded with a compact miniature USB2000+ spectrometer in the wavelength range of 600–900 nm. Each sample underwent at least five cleaning laser shots before 10 measurement shots. Two different datasets were prepared namely dataset 1, comprising full-length spectra, and dataset 2, comprising spectral lines only. We applied Principal Component Analysis (PCA) on LIBS spectral data to visualize its discriminability and Random Forest (RF) to perform classification. Scree plots describe the discriminability of samples generated in PCA and suggested PC1 and PC2. Loading plots showed the most significant spectral regions for PC1, PC2, and PC3 which covered the data variance of 67.95%, 20.16%, and 6.28% respectively. For the RF classification accuracy, the best splitting for training and testing sets was 80–20%. It demonstrated a perfect 100% accuracy in discriminating gemstones with both dataset 1 and dataset 2.