Nondestructive detection of Panax notoginseng saponins by using hyperspectral imaging. (2nd May 2022)

• Record Type: Journal Article
• Authors: Shi, Lei; Li, Lixia; Zhang, Fujie; Lin, Yuhao
• Is Part Of: International journal of food science & technology. Volume 57:Number 7(2022)
• Page Start: 4537
• Page End: 4546
• Publication Date: 2022-05-02
• Subjects: Bootstrapping soft shrinkage -- equilibrium optimizer -- hyperspectral imaging -- Panax notoginseng saponins -- support vector regression; Food industry and trade -- Periodicals; 664
• DOI: 10.1111/ijfs.15790
• Languages: English
• ISSNs: 0950-5423

Abstract

Summary: Panax notoginseng saponin (PNS) is the most important physical and chemical index of panax notoginseng. In order to detect PNS rapidly and non‐destructively, 160 hyperspectral images of panax notoginseng rhizome and main root were acquired by using a visible‐near infrared hyperspectral image acquisition system (400–1000 nm), and the original spectrum were extracted from hyperspectral images. The signal‐to‐noise ratio of the spectrum was improved by savitzky‐golay mixed multiplication scatter correction (SG‐MSC) pretreatment. Feature wavelengths were extracted by using competitive adaptive reweighted sampling (CARS), variable combination population analysis (VCPA) and bootstrapping soft shrinkage (BOSS), and support vector regression (SVR) model was established based on the feature spectrum and the original spectrum. By comparing, it was found that BOSS had the best effect of feature selection. In order to improve the accuracy of the model, equilibrium optimizer (EO) was used to optimise the parameters (c, g) of the BOSS‐SVR model. The results showed that BOSS‐EO‐SVR of the optimal prediction model of PNS, achieving R P 2 and RMSEP of 0.95 and 0.32%, respectively. Therefore, hyperspectral imaging combined with BOSS‐EO‐SVR model is a feasible method to detect PNS. Abstract : Firstly, the spectrum of panax notoginseng was extracted by hyperspectral imaging system. Second, SG‐MSC was used to improve the signal‐to‐noise ratio of the spectrum. Thirdly, BOSS was the best feature wavelength selection method by comparison. Finally, BOSS‐EO‐SVR was the optimal prediction model of PNS by comparison.