Improving identification ability of adulterants in powdered Panax notoginseng using particle swarm optimization and data fusion

Abstract Panax notoginseng (P. notoginseng) is frequently targeted for adulteration with lower-grade P. notoginseng because of its high price. This paper presents a novel method to improve the identification ability of adulterants using spectroscopic techniques and chemometrics. P. notoginseng powder of different grades were blended at different percentages (0–100%), in which the minimum blend ratio was 0.5%. NIR (12,497–4000 cm−1) and FT-MIR (4000–400 cm−1) spectra of samples were acquired. The wavenumbers of 7099–4200 cm−1, 3600–2750 cm−1, and 1750–400 cm−1 were selected manually as characteristic spectra, which included 3706 variables. Then principal component analysis (PCA) was used to further reduce the data dimension of characteristic spectra, and the first nine principal components (PCs) were applied to build classification models of 14 (the ratio of 0.5% excluded) and 15 kinds of blend ratios. Finally, Support vector machine (SVM) was built for classification of adulterants of different blend ratio. The accuracy of prediction set is 92.46% and 91.79%, respectively. On this basis, particle swarm optimization (PSO) was applied to optimize the parameters of SVM, and the accuracy of prediction set increased to 96.65% and 96.97%, respectively. The results demonstrate that the application of data fusion of NIR and FT-MIR spectra combined with SVM optimized by PSO can improve the identification ability of adulterants in powdered P. notoginseng.