Hyperspectral imaging near-infrared spectroscopy as a high-throughput analysis tool for rapid identification of pearls with different cultured origins and imitation pearls

Abstract

In this study, we sought to develop an analytical method enabling non-destructive and high-throughput evaluation of pearls, which have shown increased demand in current jewelry markets. For this purpose, hyperspectral imaging near-infrared (HSI-NIR) spectroscopy was exploited first to discriminate (i) beaded salt water-cultured (B-SW), (ii) beaded fresh water-cultured (B-FW), (iii) non-beaded fresh water-cultured (NB-FW), and (iv) imitation pearls. A tray hosting 50 pearls was fabricated for high-throughput analysis, and the pearl-loaded tray was moved under the slanted NIR illumination with 20° to the surface for acquisition of HSI-NIR data. The use of 20°-illumination was to avoid strong reflection around the top of each pearl, where the focal plane of the HSI camera was set. Next, NIR spectra corresponding to the individual pearls were extracted for subsequent discriminant analysis from the HSI-NIR data of 50 pearls. The imitation pearls made of plastic yielded apparent differences in spectral features compared to the cultured pearls. Also, the distinction of B-SW, B-FW, and NB-FW pearls was feasible since their NIR spectra were dissimilar depending on the characteristics of pearls such as chemical composition, irradiance (orient), and gloss. When linear discriminant analysis (LDA) was used for the four-group classification, the F1 scores identifying the B-SW, B-FW, NB-FW and imitation pearls were 95.6, 96, 100 and 100 %, respectively, and the overall prediction accuracy was superior, 97.9 %.