Research On Identification Method Of Salvia Miltiorrhiza Quality Based On Machine Learning And Hyperspectral

Salvia miltiorrhiza,as a kind of Chinese herbal medicine,is often used to treat cardiovascular and cerebrovascular diseases.The quality of Salvia miltiorrhiza from different producing areas is quite different due to climate,geographical conditions and other factors.At the same time,in the production process of salvia miltiorrhiza related drugs,due to the differences in raw materials and production processes used by different manufacturers,the drug effects may be different.Therefore,it is of great significance to develop a rapid and efficient method to identify the source of salvia miltiorrhiza.This paper takes salvia miltiorrhiza as the main research object,and uses hyperspectral technology and machine learning method to conduct the following research:(1)Hyperspectral discrimination of salvia miltiorrhiza from different sources based on machine learning.Firstly,spectral data of 9 kinds of salvia miltiorrhiza samples from different sources were obtained by hyperspectral technology.After data normalization and data standardization preprocessing and dimensionality reduction by principal component analysis and linear discriminant analysis,four recognition model methods,namely random forest,extra trees,support vector machine and K nearest neighbor algorithm,were established.After analyzing and comparing the experimental results,it is concluded that the data standardization-linear discriminant analysis-K nearest neighbor algorithm model is the best,and the prediction accuracy of its test set is 99.69%.(2)Identification of compound salvia miltiorrhiza tablets manufacturer based on boosting integrated learning method.The spectral data of compound salvia miltiorrhiza tablets from 7 different manufacturers were collected by hyperspectral technique,and four integrated learning and recognition models were established for these data.The accuracy of integrated learning model and other indicators were compared under three preprocessing conditions:no preprocessing and mean centering,savitzky-golay smoothing filtering,multiplicative scatter correction,principal component analysis,kernel principal component analysis,and locally linear embedding.The optimal model was finally obtained as multiple scattering correction-locally linear embeddinggradient boosting decision tree,and the evaluation indexes of Accuracy,Precision,Recall and F1 Score of the test set were 99.60%,99.63%,99.57%and 99.60%,respectively.(3)Multilayer perceptron based hyperspectral manufacturer identification of compound salvia miltiorrhiza tablets.Firstly,two important parameters of the multilayer perceptron are obtained.The experimental results show that the hidden layer parameters of the multi-layer perceptron model method are the second layer,and the neuron parameters are 40 for each layer.The multilayer perceptron model with mean centering preprocessing has the best effect,and the prediction accuracy of test set is 100%,and the ten-fold cross validation accuracy is 99.76%.It shows that the prediction effect of multilayer perceptron model is more accurate than that of the above four integrated learning model methods.