Nondestructive Detection For Panax Notoginseng Otoginseng Saponins Based On Saponins Based On Hyperspectral Imaging Technology

Panax notoginseng is a rare Chinese medicine unique to China,mainly produced in Yunnan Province.Saponin is the most important active component of panax notoginseng,which is an important index to evaluate the quality of panax notoginseng.Liquid chromatography is usually used in the quality inspection process of panax notoginseng procurement,acceptance,warehousing and feeding.But liquid chromatography is not easy to promote because of its high cost,long cycle,large reagent consumption and sample destruction.With the development of detection technology,spectroscopy has been applied to the detection of saponins.However,due to the uneven distribution of chemical information in panax notoginseng,the information obtained by point source sampling is not comprehensive,and the detection accuracy is not high.Hyperspectral imaging technology can effectively make up for the defects of spectral technology by region sampling.In this study,ginsenoside Rg₁,ginsenoside Rb₁,notoginsenoside R₁ and total saponins in panax notoginseng were predicted by hyperspectral imaging combined with stoichiometry.The main contents and conclusions are as follows:1.Hyperspectral images of 160 panax notoginseng samples were collected by visible-near infrared hyperspectral imaging system and spectral information was extracted.The true values of ginsenoside Rg₁,ginsenoside Rb₁ and notoginsenoside R₁ in panax notoginseng were determined by HPLC.The calibration set and prediction set were divided according to the true value of panax notoginseng saponins.Ensure uniform distribution of calibration set and improve the stability of model.In order to explore the nondestructive detection mechanism of panax notoginseng saponins by using hyperspectral imaging technology,this study observed the spectral manifestations of panax notoginseng with different saponins content in the visible spectra（480～780 nm）and near infrared spectra（780～1000 nm）,and analyzed the correlation between the difference of organic matter content/change of cell structure and spectral characteristics of panax notoginseng.It can provide a feasible basis for nondestructive detection of panax notoginseng saponins by using hyperspectral imaging.2.The original spectra were preprocessed by using savitzky-golay mixed multiplication scatter correction（SG-MSC）to improve the signal to noise ratio of spectra.Variable iterative space shrinkage approach（VISSA）,competitive adaptive reweighed sampling（CARS）and bootstrapping soft shrinkage（BOSS）were used to extract the feature wavelengths that could characterize the information of saponins.The combination of feature wavelengths with the highest correlation with saponin information was selected by comparing the root mean square error of cross validation（RMSECV）of feature wavelengths and the modeling accuracy.Least squares support vector regression（LSSVR）model was established based on CARS,VISSA and BOSS.It was found that the VISSA-LSSVR model had the highest prediction accuracy for ginsenoside Rg₁ and ginsenoside Rb₁,and CARS-LSSVR model had the highest prediction accuracy for notoginsenoside R₁.The feature wavelength numbers of ginsenoside Rg₁,ginsenoside Rb₁ and notoginsenoside R₁ were 180,211 and 70,respectively.In order to further improve the accuracy of the model,snake optimizer（SO）,equilibrium optimizer（EO）and arithmetic optimization algorithm（AOA）were introduced to optimize the parameters（γ,σ²）of LSSVR model.The results showed that the optimal prediction models of ginsenoside Rg₁,ginsenoside Rb₁ and notoginsenoside R₁ were SO-VISSA-LSSVR、EO-VISSA-LSSVR、AOA-CARS-LSSVR.The R_p²,RPDP,MAEP and RMSEP of ginsenoside Rg₁ were 0.9645,5.3076,0.1133 and 0.0860.The R_p²,RPDP,MAEP and RMSEP of ginsenoside Rb₁ were0.9582,4.8911,0.1160 and 0.0908.The R_p²,RPDP,MAEP and RMSEP of notoginsenoside R₁ were 0.9861,8.4916,0.0371 and 0.0287.Therefore,it is feasible to detect the content of panax notoginseng saponin by using hyperspectral imaging technology.3.CARS,VISSA and BOSS were used to extract the feature wavelengths that could characterize the information of panax notoginseng total saponins.By comparing the results of LSSVR model and RMSECV with three feature wavelengths,it was found that VISSA had the best effect,and the feature wavelength numbers of panax notoginseng total saponins was 192.Finally,SO,AOA and EO were introduced to optimize the LSSVR model.The results showed that the optimal prediction model for panax notoginseng total saponins was AOA-VISSA-LSSVR.The R_p²,RPDP,MAEP and RMSEP of panax notoginseng total saponins were 0.9808,7.2119,0.2074 and0.1716.The predicted values of ginsenoside Rg₁,ginsenoside Rb₁ and notoginsenoside R₁ were added and fitted to the true values of panax notoginseng total saponins.The R_p²,RPDP,MAEP and RMSEP of panax notoginseng total saponins were 0.9714,5.9131,0.2474 and 0.1849.It can be found that the accuracy of direct prediction for panax notoginseng total saponins was higher.Therefore,it is necessary to predict the panax notoginseng total saponins.In this study,hyperspectral imaging technology combined with stoichiometric modeling method was used to realize the nondestructive and accurate detection of panax notoginseng saponin content,which provided the theoretical basis and technical support for the development of a rapid detection device for panax notoginseng saponin content,and also provided a reference for the nondestructive and accurate detection of other araliaceae materials.