The Application Of Infrared Spectroscopy And Chemometrics To Qualitation And Quantitative Study Models Of Codonopsis Pilosulae And Angelicae Sinensis

OBJECTIVE:(1) To establish classification model of Codonopsis Pilosulae from Wen County,Lin Tao and Long Xi by near infrared spectroscopy(NIR) combined with chemometrics.(2) To discriminate Angelicae Sinensis harvested from diverse places and different time by infrared spectroscopy combined with chemometrics.(3) To establish quantitative models of ethanol extract and ferulaic acid respectively by near infrared spectroscopy coupled with Partial least squares regression and Genetic algorithms-multiple linear regression,respectively.METHODS:(1) 36 samples of Codonopsis Pilosulae were collected.After the spectra region between 12000 cm^-1 and 4000 cm^-1 were preprocessed by the standard normal variate (SNV) 2st derivation smoothing,PCA was firstly performed,and then 3 predictive models of Wen County,Tan Chang,Long Xi and Lin Tao were built using RF, respectively.(2) 136 samples of Angelicae Sinensis collected from habitats of Wen County, Tan Chang,Long Xi and Lin Tao in the year 2007,2008 were sorted systematically and labeled.PCA was performed and 7 predictive models were built using RF respectively after the near infrared spectra were preprocessed by SNV 1st derivation smoothing;PCA was implemented and 7 classification models of Angelicae Sinensis were established by preprocessed mid-infrared spectra(4000-400 cm^-1) with RF, respectively.(3) The ethanol extract and ferulaic acid were determined by the official methods.PLSR,MLR models of ethanol extract and ferulaic acid for the four locations were built on original spectra,SNV 1st derivation smoothing spectra and SNV 2st derivation smoothing spectra.Correlation coefficients,RMSEC and cross validation correlation coefficient were employed to evaluate models.RESULTS:(1) The accuracy of classification model of Codonopsis pilosulae by RF was 88.89%.(2) Classification accuracy of model established by NIR coupled with RF was 94.85%while it was 93.75%by mid-infrared spectroscopy coupled with RF.(3) Quantitative models built by GA-MLR were selected by contrasting the values of Rcv calculated by the two algorithms.For Wen County,Tan Chang,Long Xi and Lin Tao,Correlative coefficients of cross validation of ethanol extract quantitative models were 0.8812,0.8976,0.8819 and 0.9423,respectively and RMSEC were 0.0107,0.0092,0.0110 and 0.0065;Correlative coefficients of cross validation of ferulaic acid quantitative models were 0.8922,0.8496,0.8870 and 0.9393, respectively and RMSEC were 0.0029,0.0069,0.0059 and 0.0035.CONCLUSION:As the research above shown,the NIR technique in tandem with RF can identify Codonopsis pilosulae samples from various places and discriminate angelicae sinensis from different locations and harvesting time without complex pretreatment comparatively rapidly and accurately.The NIR technique associated with GA-MLR could predict the contents of ethanol extract and ferulaic acid in Angelicae Sinensis.The models established needs further improvement and validation by sample addition and adjusting machine learning algorithm.