A rapid quantitative detection method for adulteration of traditional Chinese medicines

Abstract

The present invention relates to a rapid quantitative detection method for adulteration of traditional Chinese medicine. First, a certain number of adulterated traditional Chinese medicine samples are prepared, near-infrared spectra are collected, and the samples are divided into a training set and a prediction set; each spectral signal of the training set is first decomposed into K discrete modal components using variational mode decomposition; the modal components are reorganized into K new modes, and a support vector regression sub-model is established for each mode; then the spectra of the prediction set samples are subjected to the same decomposition and combination as the training set, and the corresponding sub-models are substituted into the support vector regression sub-model to obtain the prediction results of each sub-model, and these results are weighted averaged to obtain the final prediction value. The present invention makes full use of the advantages of fast near-infrared spectral analysis and variational mode decomposition in mining the embedded information of the signal, and can quickly and accurately predict the content of components in adulterated traditional Chinese medicine. The present invention is suitable for quantitative analysis of different traditional Chinese medicines in adulterated traditional Chinese medicines.

Description

Technical Field

[0001] The invention belongs to the technical field of drug safety detection, and relates to a method for quantitatively detecting adulteration of traditional Chinese medicines based on near-infrared spectroscopy. Background Art

[0002] With the booming development of the traditional Chinese medicine market, the demand for Chinese medicinal materials is increasing day by day, and they play an important role in medicine, health products, cosmetics, etc. Due to the wide variety of varieties, some medicinal materials look similar, but the price difference is very large. Adulteration of traditional Chinese medicine often occurs in the market, affecting the quality of traditional Chinese medicine. Therefore, it is urgent to establish a fast and effective detection method. Near-infrared spectroscopy is widely used in multi-component determination of traditional Chinese medicine due to its advantages of non-destructive, fast speed, low cost and high efficiency. It is an effective analytical method for quality control and monitoring of Chinese herbal medicine.

[0003] Due to the weak intensity of near-infrared absorption and the wide spectral peak, it requires multivariate correction methods, such as partial least squares (PLS). However, the original spectra of complex samples of Chinese medicinal materials usually contain a large amount of noise information, signal overlap and background that are irrelevant to the target, which often leads to a decrease in the prediction accuracy of the modeling method. In order to improve the modeling effect, most literature uses spectral preprocessing methods to remove the noise background and then establish a multivariate correction model. However, the use of inappropriate preprocessing methods may introduce new noise and reduce the accuracy of the model, and useful information may be embedded in the background and noise components, making it difficult to fully extract information from the spectrum. Better results can be obtained by mathematically decomposing the original signal before multivariate correction, such as wavelet transform (ZC Liu, WS Cai, XG Shao, A weighted multiscale regression for multivariate calibration of near infrared spectra, Analyst, 2009, 134: 261-266) and empirical mode decomposition (Bian Xihui, Tan Xiaoyao, Wei Junfu, Liu Luxian, Li Shujuan, Li Ming, A new method for rapid and accurate determination of aromatic content in fuel oil, Chinese invention patent, 2015, CN105136682B). These methods decompose the original spectral signal into a series of components, make full use of the information embedded in each component, and obtain effective prediction. However, wavelet transform cannot adaptively decompose the signal, requires multiple parameters to be preset, and optimization is difficult, and can only process linear non-stationary signals. Empirical mode decomposition is a good adaptive decomposition method for nonlinear non-stationary signals, but it will have mode aliasing and endpoint effects during the decomposition process.

[0004] Variational mode decomposition (VMD) is a novel signal decomposition method (K Dragomiretskiy, D Zosso, Variational mode decomposition, IEEE Transactions on Signal Processing, 2014, 62(3): 531-544). VMD can not only adaptively decompose into a series of modal components uk with frequencies according to signal characteristics, but also effectively suppress modal aliasing and endpoint effects. Due to its obvious advantages, the variational mode decomposition method has been successfully applied in many engineering fields. The present invention combines it with near-infrared spectroscopy and applies it to the quantitative detection of adulteration of traditional Chinese medicine for the first time. Summary of the invention

[0005] The purpose of the present invention is to develop a rapid and accurate quantitative analysis method for adulterated samples of traditional Chinese medicine based on near infrared spectroscopy. The adulterated samples of traditional Chinese medicine are scanned by a near infrared spectroscopy instrument, and the spectrum of each sample is subjected to variational mode decomposition to obtain a series of modal components and reorganized. A support vector regression (SVR) model is established for the reorganized modes, and the results of the model are weighted averaged to obtain the final result. The specific process is as follows: Figure 1 The technical solution provided for implementing the present invention includes the following steps:

[0006] (1) Prepare a certain number of adulterated Chinese medicine samples

[0007] Obtain a number of Chinese medicinal materials of different types and batches, place each medicinal material in a 60°C oven for 3 hours to dry, and remove the moisture contained in the medicinal material. Then, grind and crush the dried medicinal materials, pass through a 120-mesh sieve, and store in a sealed bag. Adulterate each Chinese medicine according to different mass percentages to ensure that the sum of the percentage content of each component medicinal material in each sample is 100%, thereby obtaining m Chinese medicine adulterated samples.

[0008] (2) Collect near-infrared spectra of samples

[0009] Set the measurement parameters of the near infrared spectrum, sampling wave number range 12000-4000cm -1 , scanning times 64 times. The spectrum of each sample was collected three times, and the average of the three spectra was taken as the final spectrum of each sample.

[0010] (3) Using a certain grouping method, the m adulterated Chinese medicine samples are divided into a training set containing m1 samples and a prediction set containing m2 samples. The grouping method may be KS, Duplex, SPXY or the like.

[0011] (4) First, each sample i (i = 1, 2, ..., m1) in the training set is subjected to variational mode decomposition, and the spectrum of each sample is obtained as a matrix ui containing K discrete mode components: K×n , m1 samples get m1 modal component matrix u1 K×n , u2 K×n ,...,um1 K×n The number of modal components K is determined according to the change of the root mean square error of prediction (RMSEP) with the number of decompositions, and the number of decompositions K corresponding to the minimum RMSEP is the optimal number of decompositions.

[0012] (5) Place the modal component ui of the i-th (i=1, 2, ..., m1) spectral signal in the i-th row of the corresponding modal Uk matrix in the order of k. In this way, the modal component matrix uu of m1 samples is K×n (i=1,2,...,m1) are recombined into K new modal matrices Uk m1×n (k=1,2,...,K), each Uk contains the same number of samples (m1) and variables (n) as the training set. Then for each modal matrix Uk m×n Support vector regression sub-models were established separately.

[0013] (6) Perform the same variational mode decomposition and combination on the spectral signals of the m2 samples in the prediction set as in the training set to obtain the K mode matrices Uk corresponding to the training set. m2×n (k=1, 2, ..., K). These K modal matrices are respectively substituted into the corresponding K support vector regression sub-models established in step (5) to obtain the prediction results of each sub-model.

[0014] (7) The prediction results of the K sub-models are weighted averaged to obtain the final prediction result, where the weight is the inverse of the fourth power of the root mean square error of cross validation (RMSECV) of each support vector regression sub-model.

[0015] The main advantage of the present invention is that a weighted multi-scale support vector regression model (VMD-WMR) based on variational mode decomposition is established for the near-infrared spectra of complex adulterated traditional Chinese medicine samples, making full use of the frequency information of the spectral signal for better prediction. When the present invention is actually applied, it is only necessary to collect the near-infrared spectrum data of the traditional Chinese medicine sample and substitute it into the model to perform quantitative analysis on the target object, thereby achieving simple operation, green environmental protection, low cost, fast detection speed and accurate detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is the principle diagram of the weighted multi-scale support vector regression method based on variational mode decomposition (VMD-WMR).

[0017] Figure 2This is the near-infrared spectrum of adulterated Chinese medicine samples.

[0018] Figure 3 This is a graph showing how the RMSEP of VMD-WMR modeling of adulterated Chinese medicine samples changes with the number of modal components.

[0019] Figure 4 It is the modal component uk diagram obtained by variational mode decomposition of the 17th adulterated Chinese medicine sample in the training set.

[0020] Figure 5 This is the relationship between the predicted value and the true value of adulterated Chinese medicine samples using VMD-WMR modeling. DETAILED DESCRIPTION

[0021] For a better understanding of the present invention, the present invention is further described in detail below in conjunction with embodiments, but the scope of protection claimed by the present invention is not limited to the scope represented by the embodiments.

[0022] Example:

[0023] This example uses near infrared spectroscopy data analysis to determine the content of galangal in adulterated Panax notoginseng. The specific steps are as follows:

[0024] (1) Preparation of 75 adulterated Chinese medicine samples

[0025] Four kinds of Chinese medicine, namely Panax notoginseng, Curcuma zedoaria, Curcuma longa and Alpinia officinalis, were purchased from different pharmacies. They were uniformly dried, and the oven temperature was set to 60°C. Each kind of medicinal material was placed in the oven and taken out after 3 hours. The dried medicinal materials were ground into powder and passed through a 120-mesh sieve. The content range of Panax notoginseng, Curcuma zedoaria, Curcuma longa and Alpinia officinalis was 0-100% (g / g), with an interval of 5%. The four kinds of Chinese medicine were mixed at different mass percentages to obtain 75 adulterated samples of Chinese medicine. Each sample was numbered and stored in a dryer for future use.

[0026] (2) Collect near-infrared spectra of samples

[0027] The near-infrared spectrometer VERTEX70 (Bruker, Germany) was used to collect near-infrared spectral data of 75 adulterated Chinese medicine samples, with a sampling wavenumber range of 12000-4000cm -1 , with an interval of 2cm -1 , containing 4148 variables in total. The number of scans was 64. The spectrum of each sample was collected three times, and the average of the three spectra was taken as the final spectrum of each sample. The near infrared spectra of 75 samples are shown in Figure 2 shown.

[0028] (3) The KS grouping method was used to group the 75 samples, with 50 samples as the training set and 25 samples as the prediction set.

[0029] (4) The number of modal components K is determined according to the change of the root mean square error of prediction (RMSEP) with the number of decompositions. The number of decompositions K corresponding to the minimum RMSEP is the optimal number of decompositions. Figure 3 The variation of the predicted root mean square error of adulterated Panax notoginseng samples with the number of modal components is shown, where the number of modal components corresponding to the minimum value of the predicted root mean square error is 5, which is determined to be the optimal parameter for the number of modal components. Variational modal decomposition is performed on each sample i (i = 1, 2, ..., 50) in the training set, and the spectrum of each sample is obtained as a matrix ui containing 5 discrete modal components 5×4148 , Figure 4 The figure shows the 5 modal components decomposed from sample No. 17. 50 samples give 50 modal component matrices u1 5×4148 , u2 5×4148 , ..., u50 5×4148 .

[0030] (5) The modal component matrix ui of 50 samples 5×4148 (i=1,2,...,50) are recombined into 5 new modal matrices Uk 50×4148 (k = 1, 2, ..., 5), each Uk contains the same number of samples (50) and variables (4148) as the training set. Then for each modal matrix Uk 50×4148 Support vector regression sub-models were established separately.

[0031] (6) The spectral signals of the 25 samples in the prediction set are subjected to the same variational mode decomposition and combination as the training set, and the 25 mode matrices Uk corresponding to the training set are obtained. 25×4148 (k=1, 2, ..., 5). These five modal matrices are respectively substituted into the corresponding five support vector regression sub-models established in step (5) to obtain the prediction results of each sub-model.

[0032] (7) The prediction results of the five sub-models are weighted averaged to obtain the final prediction result, where the weight is the inverse of the fourth power of the root mean square error (RMSECV) of the cross-validation of each support vector regression sub-model.

[0033] In order to verify the superiority of the multiscale support vector regression modeling method based on variational mode decomposition (VMD-WMR) for adulterated traditional Chinese medicine samples, the root mean square error of prediction (RMSEP) and correlation coefficient (R) were compared with those of PLS ​​and SVR. Figure 5The relationship between the predicted value and the true value of the adulterated Chinese medicine sample data using VMD-WMR modeling is shown. The RMSEP and R of VMD-WMR are 2.0034 and 0.9972, while the RMSEP and R of PLS ​​alone are 6.2844 and 0.9785, and the RMSEP and R of SVR are 5.5861 and 0.9828. Therefore, using VMD-WMR modeling for adulterated Chinese medicine samples can obtain more accurate prediction results.

Claims

1. A rapid quantitative detection method for adulteration of traditional Chinese medicine, characterized in that: 1) Prepare a certain number of adulterated Chinese medicine samples: obtain a number of Chinese medicines of different types and batches, place each of the Chinese medicines in a 60°C oven for 3 hours to dry, and remove the moisture contained in the medicines; grind and crush the dried medicines, pass through a 120-mesh sieve, and store them in a sealed bag; adulterate each Chinese medicine according to different mass percentages to ensure that the sum of the percentage content of each component medicine in each sample is 100%, thereby obtaining m adulterated Chinese medicine samples; 2) Collect near-infrared spectra of the samples: set the measurement parameters of the near-infrared spectrum, the sampling wavenumber range is 12000-4000cm-1, and the number of scans is 64 times; collect the spectrum of each sample three times, and take the average of the three spectra as the final spectrum of each sample; (3) Using a certain grouping method, m adulterated Chinese medicine samples are divided into a training set containing m1 samples and a prediction set containing m2 samples, wherein the grouping method is one of the three methods: KS, Duplex, and SPXY; 4) First, each sample i (i = 1, 2, ..., m1) in the training set is subjected to variational mode decomposition, and the spectrum of each sample is obtained as a matrix ui containing K discrete mode components K×n , m1 samples get m1 modal component matrix u1 K×n， u2 K×n ,...,um1 K×n ; 5) The modal component matrix ui of m1 samples K×n (i=1,2,...,m1) are recombined into K new modal matrices Uk m1×n (k=1,2,...,K), each Uk contains the same number of samples (m1) and variables (n) as the training set; then for each modal matrix Uk m×n Establish support vector regression sub-models respectively; 6) Perform variational mode decomposition and combination on the spectral signals of the prediction set m2 samples in the same way as the training set to obtain K mode matrices Uk corresponding to the training set m2×n (k=1, 2, ..., K); these K modal matrices are substituted into the corresponding K support vector regression sub-models established in step 5) to obtain the prediction results of each sub-model; 7) The prediction results of the K sub-models are weighted averaged to obtain the final prediction result, where the weight is the inverse of the fourth power of the cross-validation root mean square error of each support vector regression sub-model.

2. A rapid quantitative detection method for adulteration of Chinese herbal medicines according to claim 1, characterized in that: The number of modal components K is determined according to the change of the predicted root mean square error with the number of decompositions. The number of decompositions K corresponding to the minimum value of the predicted root mean square error is the optimal number of decompositions.

3. A rapid quantitative detection method for adulteration of Chinese herbal medicines according to claim 1, characterized in that: The process of reorganizing the mode Uk in the training set is to put the modal component ui of the i-th (i=1, 2, ..., m1) spectral signal into the i-th row of the corresponding mode Uk matrix in the order of k.

Images