%A %T Research on Variety Identification of Fritillaria Based on Terahertz Spectroscopy %0 Journal Article %D 2021 %J SPECTROSCOPY AND SPECTRAL ANALYSIS %R 10.3964/j.issn.1000-0593(2021)11-3357-06 %P 3357-3362 %V 41 %N 11 %U {https://www.gpxygpfx.com/CN/abstract/article_12303.shtml} %8 2021-11-01 %X Fritillary is widely used in clinical practice of Chinese medicinal materials, especially Fritillaria cirrhosa Don. There are adulteration and fake phenomenon, fake fritillary will have a negative impact on the health of the drug users. Terahertz Time-Domain spectroscopy has many advantages of transient, broadband, safety, penetration, etc. In recent years, Terahertz Time-Domain spectroscopy is very active in drug and food non-destructive detection. In this experiment, four common fritillaria species (Fritillaria cirrhosa Don, Fritillaria ussuriensis Maxim, Fritillaria pallidiflora Schrenk, and Fritillaria thunbergii) were taken as the research objects to explore the feasibility of using terahertz time-domain spectroscopy to identify fritillaria species. In this experiment, the TAS7500TS Terahertz spectrum system was used to collect the spectra of fritillate samples in the range of 0.6~3.0 THz, and the stoichiometric method was combined for pretreatment and classification model establishment. When the number of categories is 2, it is called Binary classification; when the number of categories exceeds 2, it is called Multiple classifications. Four kinds of fritillary were established by Partial Least Squares Discriminant Analysis (PLS-DA). Initial spectra are treated with Savitzky-Golay (S-G) smoothing, Multiplicative Scatter (MSC) Correction, Standard Normal Variable Transformations, moving averages, or Baseline. Principal Component Analysis is performed.PCA can reduce the dimensionality of the preprocessed data to reduce the amount of data computation and simplify the operation. Finally, a multi-classification model of Random Forest (RF), Support Vector Machine (SVM) and Back Propagation Neural Network (BPNN) can be established. The discriminant accuracy rate of the model was 93.333% for Fritillaria cirrhosa Don-Fritillaria pallidiflora Schrenk, 98.333% for Fritillaria cirrhosa Don-Fritillaria thunbergii, and 100% for all the other four biocalcification models. The accuracy of the other four dichotomies was 100%. By comparing and analyzing the established multi-classification models, it was found that the SVM combining SNV modeling effect is best, the Fritillaria cirrhosa Don accuracy is 95.349%, the Fritillaria pallidiflora Schrenk accuracy is 96.552%, the accuracy rate of Fritillaria ussuriensis Maxim and Fritillaria thunbergii was 100%. The overall accuracy rate was up to 97.490%. This research shows that it is feasible to use Terahertz Time-Domain spectroscopy to identify different fritillaria varieties, and a SNV-SVM multi-classification model with good classification effect is established, which provides a new means to control the quality of traditional Chinese medicine and is of great significance to maintain the normal operation of the traditional Chinese medicine market.