Abstract:

In this study， we established a model based on near-infrared data to predict the origin of Potentilla anserina tubers and rapidly detect their nutrient status. Tubers of P. anserina from 32 sampling sites were analyzed to quantify five key quality attributes， namely amylum farina， protein， polysaccharide， ellagitannin， and total saponin contents. These analyses were conducted according to the national standards and industry standards， and attenuated total reflection （ATR） and near infrared spectroscopy （NIR） infrared spectroscopic data were also collected for each material. The ATR and NIR spectral data combined with a modeling method， optical range type， and map type was used to conduct a three-factor， three-level orthogonal test. Selected P. anserina samples were divided into 430 correction sets and 215 prediction sets to construct and validate the origin discrimination model， and then the strengths and weaknesses of the model were evaluated. Comparative analyses were conducted to establish the optimal conditions for the ATR model， which were as follows： modeling method， diffusion model， light range type， standard normal variate， spectrogram type， original spectrogram. After optimization， the recognition rate of the ATR model was 99.07% and its prediction rate was 97.21%， indicating that it had a better discriminatory effect. On this basis， we established models for the quantitative detection of five compounds in P. anserina tubers. These models were established by optimizing the pre-processing method， modeling band， and other modeling conditions. The optimal model， which was for tannins， had the following conditions： PCR+MSC+D1+Norris smoothing （5， 5）； modeling band， 6148-5379 cm^-1. The predictive correlation coefficient （R_p） of the tannins model was 0.9393， and the ratio of standard deviation of the validation set to standard error of prediction （RPD） was 2.86 （>2.00）. The prediction model for polysaccharides was the second most effective. The optimal conditions for the polysaccharide model were as follows： PCR+MSC+spectrum+Norris smoothing （5， 5）； modeling band， 7000-4173 cm^-1. The R_p was 0.8470 and the RPD was 1.68 （>1.40）. Our study documents the development and optimization of models incorporating NIR spectroscopy data and chemometric data for the detection of the origin and quality of P. anserina materials. These results laid a foundation for the establishment of a rapid and reliable method for evaluating P. anserina based on NIR spectral data.

Key words: Potentilla anserina, infrared spectroscopy, origin discrimination model, quantitative detection model