Research On Danshen (Salvia Miltiorrhiza) Traceability And Multivariate Calibration Method Based On Laser Induced Breakdown Spectroscopy Technology And Machine Learning Strategies

The quality supervision and control of traditional Chinese medicine(TCM)is the key to ensure the clinical efficacy,human's healthy and the steady development of the whole pharmaceutical industry.Laser induced breakdown spectroscopy(LIBS)can be used as an effective tool for rapid quality evaluation of TCM due to its unique advantages such as no complex sample pretreatment and rapid analysis.However,the chemical composition and structure of TCM show the characteristics of complexity and diversity,so it is still one of the difficulties to effectively extract,distinguish and accurately analyze the LIBS characteristic information of micro multi-component in complex TCM system.According to the actual needs of accurate analysis and intelligent evaluation of TCM quality,taking Salvia miltiorrhiza as the research object with three aspects of planting environment soil,salvia miltiorrhiza and preparation,traceability and multivariate calibration methods of Salvia miltiorrhiza is studied and proposed based on LIBS combined with machine learning strategy,of which will provide theoretical basis and technical support for the accurate analysis and intelligent evaluation of the quality of Salvia miltiorrhiza.The main research contents are as follows:Firstly,taking the soil of planting environment as the research object,LIBS technology combine with random forest(RF)was proposed for monitoring the quality of Salvia miltiorrhiza planting soil.(1)The quantitative analysis method of heavy metals in soil was carried out based on LIBS and RF.The influence of different spectral pretreatment methods on the prediction performance of RF calibration model was investigated,and the variation of prediction capability of RF calibration model with different input variables(full spectrum,characteristic band,backward interval and variable importance measurement)was explored.With the optimal spectral pretreatment,input variables and model parameters,RF calibration model based on Cu,Cr,Pb and Ni in soil was constructed.The RF calibration model based on backward interval shows a well prediction performance,and receives an accurate quantitative analysis of Cu,Cr,Pb and Ni in Salvia miltiorrhiza planting soil(Cu:RMSE = 8.0221 ?g/g;Cr: RMSE = 6.0120 ?g/g;Pb: RMSE = 1.7382 ?g/g;Ni:RMSE = 1.2851 ?g/g).(2)The quantitative analysis method of nutrient elements(N,K,Ca and Mg)in soil was studied based on LIBS and RF algorithm.Based on the 5fold cross validation(CV),the optimization of pretreatment method,input variables and model parameters were carried out,and the four nutrient elements in the prediction set soil samples were quantitatively analyzed based on the optimized RF calibration model.The results shows that the RF calibration model could be used for the accurate quantitative analysis of N,K,Ca and Mg in Salvia miltiorrhiza planting soil(N: MRE = 0.2677;K: MRE = 0.1117;Ca: MRE = 1.1744;Mg: MRE = 0.2955).Therefore,two quantitative analysis method of heavy metals and nutrients elements in planting soil based on LIBS combined with RF was established,which can provide new ideas and methods for accurate monitoring and intelligent evaluation of Salvia miltiorrhiza planting soil.Secondly,taking Salvia miltiorrhiza as the research object,the quality evaluation method of Salvia miltiorrhiza based on LIBS and chemometrics was established,includes: the perspective of origin identification and quantitative analysis of heavy metal elements.(1)The producing area discrimination of Salvia miltiorrhiza was studied by LIBS combined with Particle swarm optimization(PSO)-Kernel-extreme learning machine(KELM).Firstly,Principal component analysis(PCA)was used to judge the origin of Salvia miltiorrhiza,then KELM model based on full spectrum was established,and the influence of different variable selection methods(RF and PSO)on the predictive performance of KELM discriminant model was studied.With the optimized input variables and model parameters,Least squares support vector machine(LSSVM),RF and KELM discriminant models were established respectively.Compared with other two discriminant models,the PSO-KELM discriminant model shows a better predictive performance with the accuracy rate of 94.87%;(2)The quantitative analysis method of three heavy metal elements(Cu,Cd and Pb)in Salvia miltiorrhiza was studied and established based on LIBS combined with LSSVM.Firstly,the full spectrum as the input variable was used to construct the initial LSSVM model,and the 10 fold CV was applied to investigate the influence of different pretreatment methods on the predictive ability of LSSVM calibration model,and the selection and optimization of the input variables of LSSVM was carried out by the variable importance measure(VIM).With the optimized preprocessing method,input variables and model parameters,three calibration models of PLS,RF and LSSVM were constructed and applied for quantitative analysis respectively.Compared with RF and PLS calibration models,VIM-LSSVM calibration model shows a better predictive performance(Cu: RMSE = 0.2908 ?g/g;Cd: RMSE =0.0543 ?g/g;Pb: RMSE = 0.2274 ?g/g).Therefore,a novel method of origin identification and heavy metal quantitative analysis of Salvia miltiorrhiza based on LIBS technology combined with chemometrics was established,which can provide theoretical basis and technical support for intelligent origin traceability and comprehensive quality evaluation of Salvia miltiorrhiza.Finally,taking compound Salvia miltiorrhiza tablets as the research object,on the basis of data fusion strategy,the quality evaluation method of compound Salvia miltiorrhiza tablets was established based on LIBS-Infrared spectroscopy(IR)multispectral fusion(feature level)combined with RF.At first,the normalized method was used to eliminate the difference between the two spectral intensities,and the RF initialization discrimination model was constructed;Furthermore,the effects of different variable selection methods(Variable importance projection(VIP)and VIM)on the predictive performance of the RF discrimination models based on LIBS and IR spectra were investigated.The feature level fusion of LIBS-IR spectra was carried out under the optimized input variables,and the RF discriminant model was constructed to identify the manufacturers of compound Salvia miltiorrhiza tablets.Compared with the RF discrimination model based on LIBS and IR,the RF discrimination model based on VIM-LIBS-IR has shows the better predictive performance(sensitivity =0.9333,specificity = 0.9667,accuracy = 0.9619).Therefore,manufacturer identification of compound Salvia miltiorrhiza tablets was successfully completed by LIBS-IR spectral fusion technology coupled with RF,which can provide new technology and strategy for the production process and market supervision of compound Salvia miltiorrhiza tablets and its related preparations.