Study On Drug Component Analysis And Calibration Transfer Method Based On Near Infrared Spectroscopy

Near infrared spectroscopy（NIRS）has been widely applied in the de-tection of pharmaceutical materials,productive processes and drug ingre-dients for its characteristics.The quality of drugs has caused extensive con-cern.Traditional drug detection methods are time-consuming and difficult to operate,which is not conducive to the on-site monitoring of drugs.NIRS has strong application potential in the field of on-line and rapid detection of pharmaceutical ingredients,and it is of great significance to improve the level of drug quality supervision.Model transfer technology is the key to solve the problem that models can not be shared among different NIRS instruments.Studying the model transfer method between different NIRS instruments could be valuable and significant for the popularization of drug detection methods.In this paper,Cefradine granules were used as experimental objects,and the establishment and transmission of quantitative model of drug com-ponents were studied by NIRS and chemometrics.The main contents and conclusions are as follows:（1）Different pretreatment methods and abnormal values elimination methods were compared and analyzed.The quantitative models of cefradine,Cefalexin and water in samples were established.The results show that the Monte Carlo interactive validation method combined with SNV and SG smoothing has achieved good results.The determinant coef-ficients R² of the three components models established on the two spec-trometers were 0.988 and 0.982,0.623 and 0.611,0.874 and 0.767,respec-tively.（2）SLRDS,DS,PDS and Shenks’algorithm were used to study the model transfer between Frontier^TM and Antaris^TM.The results show that the prediction results of slave spectrum transmitted by SLRDS,PDS and Shenks’algorithm are basically the same as that of master spectrum.The spectral correction rates of cefradine model,Cefalexin model and water model are 89.0%,87.8%and 85.3%respectively,93.1%,93.6%and 93.0%respectively,and 92.3%,94.2%and 92.9%respectively.（3）A sample selection method based on Markov chain（MC）was pro-posed and applied to the selection of conversion sets,considering the the-ory and properties of Markov chain.In MC method,the sample selection process was randomized,and the spectral correction rate was selected to construct the Markov chain transfer matrix,and its stationary probability distribution was obtained as the criterion for selection of the conversion set.The experimental results show that the spectral correction rates of the three models are 91.1%,94.7%and 94.2%respectively.This study provides technical support for the application of NIRS in drug composition detection and the model sharing between different in-strument measurements.