Near-infrared quality control method for stir-frying degree of spina date seeds

Abstract

The invention discloses a near-infrared quality control method for the stir-frying degree of spina date seeds, and belongs to the technical field of traditional Chinese medicine processing quality control. The method mainly comprises the following steps: (1) processing; (2) collecting spectral information; (3) preprocessing spectral information; (4) selecting characteristic wavelengths: extracting the characteristic wavelengths of the spina date seed samples with different stir-frying degrees by using a CARS algorithm; and (5) classifying the spina date seed samples with different stir-frying degrees by utilizing a PLS-DA model, and dividing the spina date seed powder samples with different stir-frying degrees into four areas, namely a spina date seed crude product, a processed sample, a moderately processed sample and a processed sample. According to the method for rapidly identifying different stir-frying degrees of the spina date seeds, the near infrared spectrum technology, characteristic wavelength selection and other multivariate mathematical statistics are combined, and the stir-frying degree and quality of the spina date seeds can be rapidly, accurately and objectively evaluated through the method.

Description

Technical Field

[0001] The invention belongs to the technical field of detection of quality control of preparation of Chinese herbal medicine pieces, and particularly relates to a near-infrared quality control method for the frying degree of sour jujube seeds. Background Art

[0002] The dried mature seeds of the rhamnaceae plant Ziziphus jujuba Mill.var.spinosa (Bunge) Hu exH.F.Chou are sweet, sour and flat in nature. They are in the liver, gallbladder and heart meridians, and have the effects of nourishing the heart and liver, calming the mind and tranquilizing the nerves, and astringing sweat and producing body fluid. Modern pharmacological studies have shown that it has sedative, hypnotic, anti-anxiety, anti-depression, anti-cancer, anti-inflammatory, and anti-Alzheimer's disease (AD) effects. The seeds of the Chinese jujube are often used as medicine after being fried, and the effects of treating insomnia due to deficiency and restlessness, nourishing the heart and astringing sweat are enhanced after being fried. The traditional quality control method of the degree of frying of the seeds of the Chinese jujube is mainly based on the traditional identification of visual observation and taste and the determination of the content of the ingredients, such as thin layer chromatography, high performance liquid chromatography, ultra-high performance liquid chromatography tandem mass spectrometry, etc., but these methods are costly, complicated to operate, and time-consuming, and cannot be used for conventional analysis and online detection. Therefore, a fast and objective method is needed to distinguish the degree of frying of the seeds of the Chinese jujube. As an emerging rapid detection method, near-infrared spectroscopy analysis technology is widely used in the field of Chinese medicine analysis because of its advantages of being fast, convenient, accurate, and easy to achieve online control of the production process, and no sample pretreatment is required, the operation is simple, and it is suitable for real-time detection of large quantities of samples. In order to better control the quality of Chinese jujube seed slices, the present invention proposes a near-infrared quality control method to quickly identify Chinese jujube seed slices with different degrees of frying. Summary of the invention

[0003] Purpose of the invention: To solve the shortcomings of the existing technology and provide a near-infrared quality control method for the frying degree of Chinese jujube seeds. This detection method can quickly, conveniently and accurately identify Chinese jujube seed powders with different frying degrees, which is of great significance to the quality control of Chinese jujube seed decoction pieces and the guarantee of clinical efficacy.

[0004] Technical solution: To solve the above technical problems, the present invention provides the following technical solutions:

[0005] A near infrared quality control method for the frying degree of spinach seeds comprises the following steps:

[0006] (1) Processing: Grind the processed spiny jujube seed sample into powder and sieve;

[0007] (2) Collecting spectral information: The near-infrared spectral information of the jujube seed powder was collected by using an Antaris II FT-NIR spectrometer in diffuse reflectance mode;

[0008] (3) Spectral information preprocessing: Optimize the initial near-infrared spectrum of wild jujube seed powder, and preprocess the spectrum in Unscrambler X 10.4 software. The optimization methods include: multiplicative scatter correction (MSC), standard normal variate (SNV), Savitzky-Golay (SG) convolution smoothing, first derivative (1d), second derivative (2d), multiplicative scatter correction + first derivative (MSC + 1d), multiplicative scatter correction + second derivative (MSC + 2d), standard normal variate + first derivative (SNV + 1d), standard normal variate + second derivative (SNV + 2d), SG convolution smoothing + first derivative (SG + 1d), SG convolution smoothing + second derivative (SG + 2d);

[0009] (4) Selection of characteristic wavelengths: Use MATLAB 2021a software to select the characteristic wavelengths of the near-infrared spectrum of wild jujube seeds by competitive adaptive reweighted sampling (CARS), interval combination optimization (ICO) or random frog leaping (RF) algorithm to further improve the model accuracy;

[0010] (5) Classify wild jujube seed samples with different frying degrees using the PLS-DA model: The PLS-DA model divides the wild jujube seed powder samples with different frying degrees into four regions: raw wild jujube seeds, under-processed samples, moderately processed samples, and over-processed samples.

[0011] In step (1), the frying temperature of the wild jujube seed sample is 130 °C.

[0012] In step (1), the frying time of the wild jujube seed sample is 12 min.

[0013] In step (1), the sieving is through a 50-mesh sieve.

[0014] In step (2), the scanning range of the Antaris II FT-NIR spectrometer is 12000 - 4000 cm -1 , and the resolution is 16 cm -1 , and the number of scans is 32 times.

[0015] In step (2), each sample is measured in parallel 3 times to obtain average spectral data for subsequent analysis.

[0016] In step (3), the initial near-infrared spectrum of wild jujube seed powder is optimized using multiplicative scatter correction (MSC).

[0017] In step (4), the CARS algorithm is used to extract the characteristic wavelengths of wild jujube seed samples with different frying degrees.

[0018] Technical effects: The beneficial effects of the present invention are as follows:

[0019] The present invention discloses a rapid identification method for different frying degrees of wild jujube seeds by combining near-infrared spectroscopy technology with multivariate mathematical statistics such as characteristic wavelength selection. This method can rapidly, accurately and objectively evaluate the quality of the frying degree of wild jujube seeds. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Near-infrared original spectrogram.

[0021] Figure 2 PCA diagram, PLS-DA diagram, BP confusion matrix and model score diagram, BP prediction result accuracy result diagram based on near-infrared original data.

[0022] Figure 3 Near-infrared spectrogram (after MSC preprocessing).

[0023] Figure 4 Variation of sampling variables, RMSECV and regression coefficient path of CARS algorithm with sampling operation (A) and wavelength selection result of CARS algorithm (B).

[0024] Figure 5 PCA diagram, PLS-DA diagram, HCA model diagram of wild jujube seed samples with different frying degrees.

[0025] Figure 6 Scatter diagram of predicted values and reference values of PLSR model (A); difference between predicted values and reference values of calibration set of PLSR model (B); difference between predicted values and reference values of prediction set of PLSR model (C).

[0026] Figure 7 Typical photos of wild jujube seed slices and powders with different frying degrees. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0027] The present invention will be further described below in conjunction with the drawings and embodiments.

[0028] 1 Instruments and Materials

[0029] Instruments: Antaris II FT-NIR spectrometer, Thermo Fisher Scientific, USA;

[0030] C21-SN216 type multifunctional induction cooker, Guangdong Midea Life Appliance Manufacturing Co., Ltd.; BCE95I-10CN type ten-thousandth electronic balance, Sartorius, Germany; QE-300 type high-speed crusher, Zhejiang Yili Industry and Trade Co., Ltd.

[0031] Materials: The specific batch numbers and source information of 16 batches of wild jujube seed samples are shown in Table 1. They were identified by Professor Lu Tulín of the Key Laboratory of Traditional Chinese Medicine Processing of Nanjing University of Chinese Medicine as the dried mature seeds of Ziziphus jujuba Mill. var. spinosa (Bunge) Hu ex H. F. Chou of the Rhamnaceae family, meeting the standards specified in the Chinese Pharmacopoeia (2020 Edition).

[0032] Table 1 Information of Wild Jujube Seed Samples

[0033]

[0034]

[0035] 2 Methods

[0036] 2.1 Medicinal Material Treatment

[0037] According to the processing method described in the Chinese Pharmacopoeia (2020 Edition) for wild jujube seeds, the frying temperature was controlled at 130 °C, and the clean wild jujube seeds were fried in a pot for 3, 12, and 21 minutes. Each batch of wild jujube seed samples included 4 samples in total, including the raw product, for a total of 64 samples.

[0038] The samples fried for 3 minutes did not show the characteristics described in the pharmacopoeia and were classified as under-processed; the samples fried for 12 minutes had characteristics consistent with those described in the pharmacopoeia and were classified as moderately processed; the samples fried for 21 minutes showed obvious scorch marks and were classified as over-processed. Before analysis, all wild jujube seed samples were ground into powder, passed through a 50-mesh sieve, and stored in a dry, sealed, and light-proof environment.

[0039] 2.2 Near-Infrared Spectroscopy Collection

[0040] The spectral information of each sample powder was collected in diffuse reflection mode by an Antaris II FT-NIR spectrometer; the scanning range was 12000 - 4000 cm -1 , with a resolution of 16 cm -1 , and the number of scans was 32 times. Each sample was measured in parallel 3 times to obtain the average spectrum for subsequent analysis.

[0041] 2.3 Spectral Information Pretreatment

[0042] By adopting appropriate preprocessing methods for the near-infrared spectral data of samples, the noise information contained in the spectra is removed, the characteristic information is retained, and the performance of the established model is improved. There are various spectral preprocessing methods. In practical applications, it is usually necessary to examine different methods to optimize the best preprocessing method. Common ones include smoothing, derivative, standard normal variate transformation (SNV), multiplicative scatter correction (MSC), etc. In this study, a total of 11 algorithms were used to optimize the initial near-infrared spectra of wild jujube seed powder, including multiplicative scatter correction (MSC), standard normal variate (SNV), Savitzky-Golay (SG) convolution smoothing, first derivative (1d), second derivative (2d), and their combinations (MSC+1d, MSC+2d, SNV+1d, SNV+2d, SG+1d, SG+2d), and the spectra were preprocessed in Unscrambler X 10.4 software.

[0043] 2.4 Feature Wavelength Selection

[0044] The collinearity problem caused by too many variables in the near-infrared original spectra will affect the accuracy and stability of subsequent model predictions to varying degrees. Therefore, feature wavelength selection is crucial. After spectral preprocessing, competitive adaptive reweighted sampling (CARS), interval combination optimization (ICO), and random frog leaping (RF) algorithms are used to select the feature wavelengths of the near-infrared spectra of wild jujube seeds to further improve the model accuracy. The CARS algorithm uses the adaptive reweighted sampling technique to screen out the wavelength points with larger absolute values of regression coefficients in the PLS model, while removing the wavelength points with smaller weights, and selects the subset with the lowest RMSECV value through cross-validation, thus effectively identifying the optimal variable combination. The ICO algorithm divides the full spectral line into several equal-width intervals. Under the guidance of MPA, it iteratively searches for the optimal interval combination through soft shrinkage and further conducts local search to optimize the width of the selected intervals. The RF algorithm first iteratively processes the spectra, counts the spectral bands selected during the iterative process, calculates the probability of each spectral band being selected during the entire iterative process, and the higher the probability, the more important the spectral variables in this section. The application of the three wavelength selection algorithms can be carried out through MATLAB 2021a software.

[0045] 2.5 Regression Model Performance Evaluation

[0046] Using MATLAB 2021a software, predictions of wild jujube seed powder samples with different frying degrees were realized. The K-S algorithm was adopted to divide all samples into a calibration set and a prediction set at a ratio of 7:3, and a partial least squares regression (PLSR) model was established and its performance was predicted. The performance evaluation of the PLSR model is an objective basis for judging the quality of the model, and the accuracy and reliability of the model can be described by calculating relevant indicators. The PLSR regression algorithm was run in MATLAB 2021a software, and the performance of the established model was evaluated. The evaluation parameters in the PLSR model include the calibration determination coefficient (R2c), the prediction determination coefficient (R2p), the root mean square error of calibration (RMSEC), the root mean square error of prediction (RMSEP), the RMSEP / RMSEC ratio, and the relative deviation percentage (RPD). When R2c and R2p are closer to 1, and RMSE is closer to 0, it indicates that the PLSR model has a high prediction accuracy. When the RMSEP / RMSEC ratio is in the range of 0.8 - 1.2, appropriate fitting can be achieved. RPD is used to evaluate the stability performance of the PLSR model, where the larger the RPD value, the better the stability. When RPD > 2, the PLSR model can be used for practical applications.

[0047] 3 Results

[0048] 3.1 Original spectrum analysis of near-infrared spectroscopy

[0049] 3.1.1 Near-infrared spectrum characteristics

[0050] In the near-infrared spectrum, the main absorption peaks come from O-H, C-H, C-C, C═C, and C═O functional groups. As Figure 1 , the spectrum shows 6 characteristic absorption peaks. The absorption peak near 8300 cm -1 is the second-order overtone absorption from the C-H stretching vibration. The absorption peak near 6900 cm -1 is attributed to the first-order overtone absorption of the O-H stretching vibration. The absorption peak near 5650 cm -1 is caused by the first-order overtone absorption of the C-H stretching vibration. The absorption peak near 5180 cm -1 may be due to the second overtone absorption of the C═O stretching vibration or the combination of O-H stretching and bending vibrations. The peak near 4659 cm -1 is caused by the stretching vibrations of C-C and C═C. The absorption peak between 4000 cm -1 and 4400 cm -1 is mainly caused by the combined frequencies of C-H, C-H 2 and C-H 3 .

[0051] 3.1.2 Qualitative discrimination based on PCA, PLS-DA, and BPNN

[0052] As Figure 2 , the PCA and PLS-DA models created based on the original spectral data showed that there was a significant overlap in the distribution ranges among the wild jujube seed powder samples with different frying degrees, lacking clear separation. Therefore, it was necessary to further implement classification using a classification model. To better separate the wild jujube seed samples with different frying degrees, BPNN was used to deeply analyze and quantify the data, and the results were as Figure 2 . The classification accuracy of the BPNN model for the wild jujube seed samples with different frying degrees was only 70.4%. It can be seen that the wild jujube seed powder samples with different frying degrees could not be accurately classified based on the original spectral information. In the analysis of near-infrared spectra, due to the existence of numerous variables, these variables often contained instrument noise and irrelevant noise components, seriously affecting the accuracy of the classification and prediction models. Therefore, it was necessary to preprocess the original spectra and select characteristic wavelengths to optimize the spectral information and extract key variables, thereby improving the overall quality of model construction.

[0053] 3.2 Processing and Optimization of Near-Infrared Spectral Information

[0054] 3.2.1 Spectral Information Preprocessing

[0055] Eleven methods were used to process the original spectral maps of the wild jujube seed powder samples with different frying degrees, including SNV, MSC, SG, first derivative, second derivative, and their combinations (MSC + 1d, MSC + 2d, SNV + 1d, SNV + 2d, SG + 1d, SG + 1d). The prediction accuracy and performance parameters were used to evaluate the quality of the model, and the preprocessing results are shown in Table 2. The results showed that the MSC method was most suitable for the wild jujube seed powder samples with different frying degrees, R 2 C = 0.8210, R 2 p = 0.7087, RPD = 2.5664, and the ratio of RMSEP / RMSEC was 1.1736. As shown in the optimized Fourier transform near-infrared spectrum Figure 3 , it can be found that after spectral preprocessing, the originally complex spectrum became clearer and more concise, facilitating subsequent analysis.

[0056] Table 2 Preprocessing Results of Wild Jujube Seed Powder Spectral Data

[0057]

[0058]

[0059] 3.2.2 Selection of Characteristic Wavelengths

[0060] After preprocessing, a feature wavelength selection algorithm was further used to extract key spectral information. In this study, ICO, CARS, and RF wavelength selection algorithms were used to improve the performance of the prediction model. The number of latent variables is an important indicator affecting the prediction performance of the PLSR model. Generally, the number of LVs should not exceed 10 to avoid overfitting, while too few LVs will lead to underfitting and reduced accuracy. As shown in Table 3, compared with Full-PLS, the prediction performance of the model was improved for the samples of wild jujube kernel powder with different frying degrees under CARS-PLS and RF-PLS treatments, while ICO-PLS did not show obvious improvement, and even the prediction parameters of the calibration set decreased, indicating that the key wavelengths extracted by the CARS and RF wavelength selection algorithms are effective. For the wild jujube kernel samples with different frying degrees, when using the CARS algorithm, the model achieved R 2 c = 0.9735, R 2 p = 0.9673, RMSEP / RMSEC = 1.0648, RPD = 6.2251; when using the RF algorithm, it achieved R 2 c = 0.8859, R 2 p = 8964, RMSEP / RMSEC = 0.8879, RPD = 3.1243. These results indicate that the model is reliable and shows good prediction performance and appropriate fitting degree. And for the wild jujube kernel samples with different frying degrees after being processed by the CARS algorithm, the R 2 c, R 2 p, and RPD parameters are all better than those of the RF algorithm. Therefore, using the CARS algorithm to extract the characteristic wavelengths of wild jujube kernel samples with different frying degrees can obtain the most ideal results.

[0061] Table 3 Results of characteristic wavelength selection

[0062]

[0063]

[0064] 3.2.3 Results of CARS algorithm

[0065] The CARS algorithm is a classical wavelength selection algorithm and has been applied to the field of identifying different frying degrees. Figure 4 respectively represent the changing trends of the number of sample variables, RMSECV, and the regression coefficient path with the number of sample runs. When the number of samplings increases from 0 to 10, the number of selected wavelengths decreases rapidly and then tends to be smooth, which reflects that the CARS algorithm can not only quickly extract spectral data but also refine spectral information on this basis. In the early stage, the wrong characteristic spectral information was eliminated by the CARS algorithm. The RMSECV value in the figure shows a trend of first decreasing and then increasing. When the number of iterations reaches 69 times (corresponding to Figure 4When the blue line in the third figure), the RMSECV value reaches the minimum. At this time, a total of 16 characteristic wavelengths are extracted, such as Figure 4 B, and the red squares represent the characteristic wavelengths extracted from the original spectral information by the CARS algorithm.

[0066] 3.3 Near-infrared spectroscopy analysis after pretreatment and wavelength selection

[0067] 3.3.1 Qualitative discrimination based on multivariate statistical methods

[0068] After pretreatment and characteristic wavelength selection, the background noise and irrelevant interference of the near-infrared original spectrum are eliminated, and the key spectral information is extracted. On this basis, the PCA and PLS-DA models are used again to classify the wild jujube seed samples with different frying degrees, intuitively demonstrating the effectiveness and necessity of spectral pretreatment and wavelength selection. From Figure 5 , it can be observed that the PLS-DA model divides the wild jujube seed powder samples with different frying degrees into four regions, and the classification between the raw wild jujube seeds (blue dots), under-fried samples (purple dots), moderately fried samples (orange dots) and over-fried samples (red dots) is clear and definite. On this basis, hierarchical cluster analysis (HCA) is further carried out, as Figure 5 , and the results show that the wild jujube seed samples with different frying degrees can be divided into 4 categories, which also verifies the results of the PLS-DA model. The above results show that the pretreatment and wavelength selection of near-infrared spectroscopy help to extract valuable bands from complex spectral information, thereby improving the classification accuracy of wild jujube seed samples with different frying degrees.

[0069] 3.3.2 PLSR model prediction results

[0070] After pretreatment and characteristic wavelength selection, the optimized near-infrared spectra of wild jujube seed powder samples with different frying degrees are obtained. On this basis, the PLSR model is used to predict and analyze the data. Generally, if the samples are closely distributed near the regression line, it means that the establishment of the regression model is successful and has good prediction performance. In the prediction model based on near-infrared spectroscopy, such as Figure 6 A, the sample points are closely clustered on the regression line, indicating that the regression model has good prediction performance; Figure 6 In B and C, the red line and the blue line overlap highly, R 2 p = 0.9673, R 2 c = 0.9735, indicating that the predicted values are close to the reference values. This further confirms the reliability of the PLSR prediction model based on near-infrared spectroscopy, indicating that it has strong performance in predicting the frying degree and provides positive technical support for distinguishing different frying degrees of wild jujube seeds.

Claims

1. A near infrared quality control method for the frying degree of spinach seeds, characterized in that: The steps include: (1) Processing: Grind the processed spiny jujube seed sample into powder and sieve; (2) Collecting spectral information: The near-infrared spectral information of the jujube seed powder was collected by using an Antaris II FT-NIR spectrometer in diffuse reflectance mode; (3) Spectral information preprocessing: The initial near-infrared spectrum of the Ziziphus jujuba seed powder was optimized and preprocessed in the Unscrambler X10.4 software. The optimization methods included: multiplicative scatter correction (MSC), standard normal variate (SNV), SG convolution smoothing (Savitzky-Golay, SG), first-order derivative (1d), second-order derivative (2d), multiplicative scatter correction + first-order derivative (MSC+1d), multiplicative scatter correction + second-order derivative (MSC+2d), standard normal variate + first-order derivative (SNV+1d), standard normal variate + second-order derivative (SNV+2d), SG convolution smoothing + first-order derivative (SG+1d), SG convolution smoothing + second-order derivative (SG+2d); (4) Selection of characteristic wavelengths: Using MATLAB 2021a software, competitive adaptive weighted sampling (CARS), interval combination optimization (ICO), or random frog leaping (RF) algorithm was used to select the characteristic wavelength of the near-infrared spectrum of the jujube kernel to further improve the model accuracy; (5) The PLS-DA model was used to classify the samples of Chinese jujube kernel with different degrees of frying: The PLS-DA model divided the Chinese jujube kernel powder samples with different degrees of frying into four areas: raw Chinese jujube kernel, samples that were not processed enough, samples that were processed moderately, and samples that were over-processed.

2. The near-infrared quality control method for the frying degree of Chinese date seeds according to claim 1, characterized in that: In step (1), the frying temperature of the spinach seed sample is 130°C.

3. The near-infrared quality control method for the frying degree of Chinese date seeds according to claim 1, characterized in that: In step (1), the frying time of the spinach seed sample is 12 minutes.

4. The near-infrared quality control method for the frying degree of Chinese date seeds according to claim 1, characterized in that: In step (1), the sieving is through a 50-mesh sieve.

5. The near-infrared quality control method for the frying degree of Chinese date seeds according to claim 1, characterized in that: In step (2), the scanning range of the Antaris II FT-NIR spectrometer is 12000~4000cm -1 , resolution 16cm -1 , scan times 32 times.

6. The near-infrared quality control method for the frying degree of Chinese date seeds according to claim 1, characterized in that: In step (2), each sample was measured three times in parallel to obtain average spectral data for subsequent analysis.

7. The near-infrared quality control method for the frying degree of Chinese date seeds according to claim 1, characterized in that: In step (3), the initial near infrared spectrum of the Ziziphus jujuba seed powder is optimized using multiplicative scatter correction (MSC).

8. The near-infrared quality control method for the frying degree of Chinese date seeds according to claim 1, characterized in that: In step (4), the CARS algorithm is used to extract the characteristic wavelengths of the jujube kernel samples with different degrees of frying.

Images