Jasmine tea quality rapid scoring method

Abstract

The invention discloses a rapid jasmine tea quality scoring method, which comprises the following steps of: acquiring infrared detection data of a jasmine tea sample by using a Fourier infrared spectrometer, solving the weight of a wave number in a range of 4000-400cm <-1 > by using a variable coefficient method, optimizing and establishing a jasmine tea quality scoring BP neural network prediction model by using a genetic algorithm, and determining the quality of the jasmine tea according to a normalized wave number characteristic value. And inputting the constructed prediction model, and calculating the quality score of the sample. The method disclosed by the invention is short in time consumption, low in requirement on an experimental site, capable of saving manpower and material resources and remarkably improving the detection efficiency, and suitable for on-site rapid scoring prediction of the jasmine tea quality.

Description

Technical Field

[0001] The invention relates to the technical field of tea, and aims to provide a simple and rapid method for scoring the quality of jasmine tea. The method is not only applicable to jasmine tea, but also to other types of tea, and provides an effective means for controlling and evaluating the quality of tea. Background Art

[0002] Jasmine tea has a long history and is known as "the best fragrance in the world". It has a wide variety of varieties and is the largest and most valuable tea in my country. At present, the methods for evaluating tea quality can be roughly divided into three categories, namely sensory evaluation, physical and chemical testing, and emerging technologies. Sensory evaluation includes quantitative description analysis (QDA) and flavor wheel, physical and chemical testing includes ultraviolet-visible spectrophotometry (UV-vis), high performance liquid chromatography (HPLC), liquid chromatography-mass spectrometry (LC-MS), gas chromatography-mass spectrometry (GC-MS), nuclear magnetic resonance spectroscopy (NMR), etc. Emerging technologies mainly include computer vision imaging technology, spectral technology, electronic nose and electronic tongue technology, etc. Sensory evaluation is still the basic means of tea quality evaluation. The trained reviewers will score the quality of different tea samples. However, this method has high requirements for both the evaluator and the evaluation environment; although the quality analysis of jasmine tea through physical and chemical indicators can obtain more objective and valuable physical and chemical indicator information, it has the disadvantages of cumbersome operation, time-consuming, large consumption of organic reagents, and environmental unfriendliness. Spectroscopic technology, including near-infrared spectroscopy (NIRS), mid-infrared spectroscopy (MIRS), terahertz spectroscopy (THz), Raman spectroscopy (RS), hyperspectral imaging (HSI), etc., is a method of analysis and measurement using the dispersion characteristics of light, and is currently also used in tea quality grading and variety identification.

[0003] With the emergence of emerging technologies such as spectroscopy, tea quality evaluation has gradually shifted from traditional sensory evaluation and chemical analysis to non-destructive, fast, and efficient instrumental analysis. Therefore, it is necessary to develop a simple and fast jasmine tea quality scoring method based on spectroscopy technology, which is suitable for on-site rapid testing and can meet the scoring needs of most tea product industries and consumers for jasmine tea quality. Summary of the invention

[0004] The invention provides a method for quickly scoring the quality of jasmine tea. The method comprises the following steps: firstly drying a small amount of ground tea sample under an infrared lamp, drying and pressing the sample into a sheet, and then performing infrared detection under a Fourier infrared spectrometer to obtain infrared detection data of the sample, and obtaining the infrared detection data of the sample at 4000-400 cm by using a coefficient of variation method. -1The weight of the wave number within the range is selected, and the wave numbers with the top 15 weights from high to low are selected. The BP neural network prediction model for the quality score of jasmine tea is established by optimizing the genetic algorithm, and the normalized wave number eigenvalue is input into the constructed prediction model to calculate the sample quality score.

[0005] The method for quickly scoring the quality of jasmine tea of ​​the present invention comprises the following steps:

[0006] (1) 10-30 mg of jasmine dry tea sample was taken and ground and mixed evenly with dried potassium bromide in a weight ratio of 1:10-1:50. After drying for 10 minutes, the tablets were pressed. The infrared spectrum information of the sample was collected using a Nicolet iS10 Fourier transform infrared spectrometer. Each sample was measured three times in parallel and the average spectrum was taken.

[0007] Spectral acquisition range 4000-400cm -1 , scanning times 32 times, resolution 0.4cm -1 . The spectral data were collected in a laboratory environment with a temperature of 25 ± 1 °C and a humidity of 35% RH;

[0008] (2) In order to reduce the noise, light scattering and tea particles in the collection environment, the original spectra were preprocessed using a variety of methods such as baseline correction;

[0009] (3) Find 4000~400cm -1 The weight of the wavenumber in the range is used to determine the characteristic wavenumber for constructing the jasmine tea quality scoring model;

[0010] (4) Determine the number of nodes in the hidden layer of the BP neural network prediction model and construct a BP neural network prediction model for jasmine tea quality score.

[0011] Preferably, in step (1), 10 mg of jasmine dry tea sample is taken, and the mixture is ground and mixed with dried potassium bromide at a weight ratio of 1:10-20, and then tabletted;

[0012] Preferably, in step (2), the original spectrum is preprocessed using baseline correction, smoothing, and standard normal variable (SNV) methods.

[0013] Preferably, in step (3), the coefficient of variation method is used to obtain 4000-400 cm -1The weight of the wavenumber in the range is selected, and the wavenumbers with the top 15 weights from high to low are selected as the characteristic wavenumbers to construct the jasmine tea quality evaluation model; further preferably, the wavenumbers with the top 15 weights are: 3375, 3379, 3373, 3371, 3381, 3338, 3369, 3340, 3336, 3367, 3383, 3342, 3365, 3344, 3334cm -1 It should be understood that the wave number may fluctuate within an acceptable measurement range, such as ±10cm -1 , more preferably ±5cm -1 .

[0014] Preferably, in step (3), the wave numbers ranked in the top 15 by weight and the quality of jasmine tea are analyzed using Pearson correlation analysis; the quality of jasmine tea includes appearance score, soup color score, aroma score, taste score, leaf bottom score and sensory score, and the quality of jasmine tea is evaluated according to GB / T23776-2018.

[0015] Preferably, in step (3), based on origin 2022 software, Pearson correlation analysis is used to analyze the correlation between the spectral characteristic values ​​and the quality of jasmine tea; and principal component analysis (PCA) is used to perform dimensionality reduction processing on the characteristic data with greater correlation.

[0016] Preferably, in step (4), the input layer of the BP neural network prediction model is 1, which is the principal component 1 eigenvalue after principal component analysis of the transmittance at the top 15 wavenumbers ranked by weight, the output layer is 1, that is, the jasmine tea quality score, the hidden layer is 1, and the number of nodes is set in the range of 2-11; the optimal number of hidden layer nodes is 7, and the BP neural network prediction model is a BP neural network optimized based on a genetic algorithm (GN); the average absolute errors of the BP neural network prediction model before and after optimization based on the genetic algorithm (GN) are 3.92% and 1.47%, respectively, both less than 5%, and the model prediction accuracy is high.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] (1) The method of the present invention uses only one commonly used inorganic reagent, without the need to use other organic reagents, thus avoiding the cumbersome preparation of multiple reagents and being environmentally friendly;

[0019] (2) The method of the present invention can process a large number of samples at a time, without the need for sensory evaluation and physical and chemical index testing of the product. The operation is fast and simple, which greatly improves the efficiency of product quality scoring and can be used for on-site rapid testing;

[0020] (3) The average absolute error of the sample quality score of the method of the present invention is controlled within 5%, and the prediction accuracy is relatively high. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The infrared spectra of different jasmine tea samples are shown below;

[0022] Figure 2 This is the correlation analysis diagram between infrared spectrum characteristic wavenumber and tea quality;

[0023] Figure 3 Comparison chart of prediction deviation (A) and predicted sensory score (B) between BP neural network and GNBP neural network. DETAILED DESCRIPTION

[0024] The terms used in the present invention, unless otherwise specified, generally have the meanings commonly understood by those of ordinary skill in the art. The present invention will be further described in detail below in conjunction with specific embodiments and with reference to data. It should be understood that these embodiments are only intended to illustrate the present invention and are not intended to limit the scope of the present invention in any way.

[0025] Experimental Materials

[0026] ① Jasmine tea sample A, named Chunlun Jasmine Red Gold (Fuzhou Jasmine Tea), the tea base is roasted green tea, produced in Fuzhou, Fujian.

[0027] ② Jasmine tea sample B, named Longhao Jasmine (Jasmine tea), the tea base is roasted green tea, produced in Fuzhou, Fujian.

[0028] ③ Jasmine tea sample C, named Zhang Yiyuan Jasmine Tea (Longhao), the tea base is Fuding white tea, produced in Hengxian County, Guangxi.

[0029] ④ Jasmine tea sample D, named Chunlun Jasmine Dabaihao (Fuzhou Jasmine Tea), the tea base is roasted green tea, produced in Fuzhou, Fujian.

[0030] ⑤ Jasmine tea sample E, named Chunlun Jasmine Silver Snail (Jasmine Tea), the tea base is roasted green tea, produced in Fuzhou, Fujian.

[0031] ⑥Jasmine tea sample F, named Zhang Yiyuan Jasmine Tea (Xiangxue), the tea base is roasted green tea, produced in Tongzhou District, Beijing.

[0032] ⑦ Jasmine tea sample G, named Chunlun Jasmine Yinhao (Jasmine Tea), the tea base is roasted green tea, produced in Fuzhou, Fujian.

[0033] ⑧Jasmine tea sample H, named Chunlun Jasmine Yinhao (Jasmine tea), the tea base is roasted green tea, produced in Fuzhou, Fujian.

[0034] ⑨Jasmine tea sample I, named Chunlun Fuzhou Jasmine Tea, the tea base is roasted green tea, produced in Fuzhou, Fujian.

[0035] ⑩Jasmine tea sample J, named Chunlun Jasmine Dabaihao (Jasmine Tea), the tea base is roasted green tea, produced in Fuzhou, Fujian.

[0036] Jasmine tea sample K, named Chunlun Jasmine Tea (Longjing), is made of roasted green tea produced in Fuzhou, Fujian.

[0037] Jasmine tea sample L, named Fumingyuan Jasmine Silver Needle, is made from roasted green tea produced in Fuzhou, Fujian.

[0038] Jasmine tea sample M, named Monkey King Jasmine Tea (Chun Hao), is made of roasted green tea produced in Changsha, Hunan.

[0039] Jasmine tea sample N, named Zhang Yiyuan Jasmine Tea (Maojian), the tea base is roasted green tea, produced in Tongzhou, Beijing.

[0040] Jasmine tea sample O, named Monkey King Special Jasmine Tea No. 1, is made of roasted green tea produced in Changsha, Hunan.

[0041] Jasmine tea sample P is named Wu Yutai Special Jasmine Tea. The tea base is roasted green tea and is produced in Changping District, Beijing.

[0042] Sensory Evaluation Method of Jasmine Tea

[0043] The evaluation was conducted in accordance with GB / T23776-2018, and the review panel consisted of 9 screened and trained personnel. Review process: First, a small number of representative tea samples were presented to the reviewers for scoring of the tea appearance. After the evaluation, 3g of jasmine tea was placed in a set of transparent glass cups with lids of the same specifications, 150mL of boiling water was added, and the cups were immediately covered with lids and brewed for 3 minutes. The brewed tea leaves were then presented to the reviewers together with the review bowls. The reviewers unscrewed the cup lids and smelled to score the aroma, poured the tea liquid into a white ceramic tasting bowl to score the soup color and taste, and finally scored the tea leaf bottom. The sensory score of each tea sample was calculated according to formula (1). The final sensory score of each jasmine tea was the score obtained by adding the sum of the remaining scores after removing the highest and lowest scores and dividing it by the number of people.

[0044] C=X1×a+X2×b+X3×c+X4×d+X5×e(1)

[0045] In the formula: C: total sensory score of tea; X1, X2, X3, X4, X5: evaluation scores of five quality factors; a, b, c, d, e: score proportion of five quality factors.

[0046] Table 1 Jasmine tea quality factors and proportions

[0047]

[0048] Fourier transform infrared spectroscopy

[0049] (1) Spectral acquisition

[0050] First, a small amount of finely ground tea sample was dried under an infrared lamp, and then ground and mixed evenly with dried potassium bromide at a weight ratio of 1:20, dried for 10 minutes, and then tableted. The infrared spectrum information of the sample was collected using a Nicolet iS10 Fourier infrared spectrometer. Each sample was measured three times in parallel and the average spectrum was taken. The spectrum collection range was 4000-400cm -1 , scanning times 32 times, resolution 0.4cm -1 The spectral data were collected in a laboratory environment with a temperature of 25 ± 1 °C and a humidity of 35% RH.

[0051] (2) Spectral preprocessing

[0052] In order to reduce the noise in the collection environment, the influence of light scattering and tea particles, this experiment used baseline correction, smoothing, standard normal variable (SNV) and other methods to pre-process the original spectrum.

[0053] (3) Spectral feature extraction

[0054] 4000~400cm was obtained by using the coefficient of variation method -1 The weights of the wave numbers within the range, select the top 15 wave numbers from high to low weights as the characteristic wave numbers.

[0055] Correlation analysis and data dimensionality reduction

[0056] Based on origin 2022 software, Pearson correlation was used to analyze the correlation between spectral feature values ​​and jasmine tea quality. Principal component analysis (PCA) was used to reduce the dimension of the characteristic data with high correlation, and the extracted principal components and characteristic variables were used as input variables of the discriminant model.

[0057] Evaluation model construction and verification

[0058] (1) Evaluation model construction

[0059] A jasmine tea quality score prediction model was created using a BP neural network, and the BP neural network was optimized based on a genetic algorithm (GN). The BP neural network is a multi-layer feedforward neural network based on error back propagation, including an input layer, a hidden layer, and an output layer. The number of input layer units is determined according to the dimension of the input information. In this experiment, the number of input layer units is determined according to the dimension of the input information, that is, the Pearson correlation analysis principle is used to screen the relevant eigenvalues ​​that are significantly correlated with the sensory score of tea quality. If there is one eigenvalue that is significantly correlated, the number of units is set to 1, and so on; the number of output layer units is 1, which is the quality of tea. The hidden layer is set to 1, and the number of nodes in the hidden layer will affect the accuracy of model recognition, so the present invention selects the number of nodes with the smallest root mean square error of the training set. The S-type function is used as the action function of the model; the learning factor and momentum factor in the training process are both set to 0.1.

[0060] This experiment uses the neural network toolbox in MATLAB to train the network. However, the BP neural network has problems such as taking local extreme values ​​(optimal solutions) and low training efficiency, while the genetic algorithm (GN) can make up for the problem of BP neural network finding the optimal solution. Therefore, this case uses the BP neural network and GNBP neural network to establish prediction models for comparison.

[0061] (2) Model verification

[0062] The present invention uses mean absolute error (MAE), mean square error (MSE), root mean square error (RMSE), and mean absolute percentage error (MAPE) as indicators for evaluating the prediction effect of the model. The smaller the MAE, MSE, RMSE, and MAPE, the more reliable the established model is. Accuracy is one of the basic indicators for evaluating model performance, and the error between the predicted value and the measured value can be used to reflect the accuracy.

[0063] Example 1: Sensory evaluation of different jasmine tea qualities

[0064] The sensory evaluation results of different jasmine tea samples are shown in Table 2.

[0065] According to the sensory score calculation formula in the national standard "GB / T23776-2018", the quality factors (appearance, aroma, soup color, taste, leaf bottom) and the overall sensory evaluation scores of the 16 types of jasmine tea can be calculated. Then, a one-way ANOVA was performed on the five quality factors and the overall sensory evaluation scores of the 16 types of jasmine tea in SPSS software. The results showed that:

[0066] a. There are differences in sensory scores and quality factor scores of different varieties of jasmine tea;

[0067] b. Among the 16 jasmine tea samples, sample B had the highest sensory score, which was 89.510 points, and sample G had the lowest sensory score, which was 79.988 points;

[0068] c. According to the sensory scores and one-way analysis of variance, the sensory quality of the 16 types of jasmine tea can be ranked as follows: B>A, C, D, K, L, M, N>F, I, J, O, P>E>H, G.

[0069] d. According to the sensory scores and one-way ANOVA, the 16 types of jasmine tea can be divided into 5 levels (characterizing the sensory quality of jasmine tea). Sample B is the first level, samples A, C, D, K, L, M, and N are the second level, samples F, I, J, O, and P are the third level, sample E is the fourth level, and samples G and H are the fifth level.

[0070] Table 2 Sensory evaluation scores of jasmine tea

[0071]

[0072] Note: The abcd in the upper right corner of the data represents the significant difference between different samples after variance analysis. The same letters represent insignificant differences, and different letters represent significant differences (p<0.05).

[0073] Example 2: Fourier transform infrared spectroscopy analysis

[0074] (1) Fourier transform infrared spectroscopy peak analysis

[0075] The results of Fourier transform infrared spectra of 16 kinds of jasmine tea are as follows Figure 1 As shown in the figure, the common absorption peaks of the 16 tea samples are: 3420~3360cm -1 -OH vibration absorption peak in the range of 2930-2920cm -1 The range is attributed to -CH absorption, 2860 to 2850 cm -1 -CH2 absorption, 1650~1640cm -1 It is caused by the stretching vibration of the C=C double bond, 1450cm -1 The peaks near 1370 cm are due to the stretching vibration of the C=C double bond on the benzene ring. -1 The vicinity is usually attributed to the bending vibration of the CH bond, 1240-1230cm -1 The stretching vibration absorption peak of CO, 1148 cm -1 The nearby peaks may also be caused by the vibration of the CO bond. In addition, there are some variant peaks, such as 825cm -1 、763cm -1 There is also a CH absorption peak near 453cm -1In order to further explore the similarities between these 16 types of tea, this paper also conducted a simple comparative analysis of the internal structures of these 16 types of jasmine tea based on FT-IR combined with correlation analysis and cluster analysis.

[0076] (2) Correlation analysis between Fourier transform infrared spectra and sensory quality scores of 16 tea samples

[0077] 16 kinds of tea were measured at different wave numbers (4000~400cm -1 ) were statistically analyzed and the coefficient of variation method was used to obtain the transmittance of 4000-400 cm -1 The weights of the wavenumbers in the range are calculated, and the top 15 wavenumbers are extracted as characteristic wavenumbers. After analysis, the top 15 wavenumbers in the final weight ranking are: 3375, 3379, 3373, 3371, 3381, 3338, 3369, 3340, 3336, 3367, 3383, 3342, 3365, 3344, 3334. The correlation between the above Fourier transform infrared spectroscopy characteristic wavenumbers and the quality of jasmine tea is analyzed, and the results are as follows Figure 2 As shown. Figure 2 It can be seen that the 15 characteristic wavenumbers extracted based on FT-IR of jasmine tea are highly significantly correlated with the quality grade of tea (based on the total sensory quality score), sensory quality score and aroma sensory score (P <= 0.01), among which it is negatively correlated with the quality grade of tea, with a correlation coefficient of -0.63 or -0.64; it is positively correlated with the overall sensory quality of tea, with a correlation coefficient of 0.63; it is also positively correlated with the sensory quality of tea aroma, with a correlation coefficient of 0.76. In summary, the 15 characteristic wavenumbers extracted in this paper are all helpful for the quality assessment of jasmine tea.

[0078] In view of the large amount of data for constructing the model with the transmittance at 15 characteristic wavenumbers of infrared spectra of 16 kinds of jasmine tea as the eigenvalue, the data was reduced in dimension based on the principal component analysis method in the origin 2022 version of the software, and the transmittance of the above infrared spectral characteristic wavenumbers was subjected to principal component analysis. The results of the principal component analysis are shown in Table 3. The eigenvalue of principal component 1 (PC1) is 15.00, and its contribution rate has reached 99.8%. The eigenvalue of principal component 2 (PC2) is 0.00238, and its contribution rate is only 0.02%. This shows that principal component 1 already contains most of the information of these highly correlated original data. Therefore, a principal component 1 can be selected to characterize the information of these 15 characteristic wavenumbers.

[0079] Table 3 Principal component contribution rate

[0080]

[0081] Example 3: Construction and verification of jasmine tea quality score prediction model

[0082] (1) Determination of the optimal number of hidden layer nodes in the neural network model

[0083] The sample set was randomly divided into a training set (70%) and a prediction set (30%) by MATLAB software. The neural network model for jasmine tea quality score: the input layer is 1, which is the principal component 1 eigenvalue after principal component analysis of the transmittance under 15 infrared spectral characteristic wavenumbers, the output layer is 1, that is, the jasmine tea quality score, the hidden layer is 1, and the number of nodes is set to 2-11. The number of hidden layer nodes corresponding to the minimum mean square error (MSE) of each quality index in the training set is the smallest is the best. The optimization results of the number of hidden layer nodes are shown in Table 4. It can be seen from Table 4 that when the number of hidden layer nodes is 7, the mean square error of the predicted quality score is the smallest, that is, 0.042205, so the optimal number of hidden layer nodes for constructing the model is determined to be 7.

[0084] Table 4 Mean square error (MSE) of the number of hidden layer nodes in the training set

[0085]

[0086] (2) Determination of jasmine tea quality scoring model

[0087] The optimized GNBP model is compared with the constructed BP model, and the results are shown in Table 5. The MAE, MASE, and RMSE of the optimized GNBP model are all reduced by 50%, and the mean absolute percentage error (MAPE) is slightly increased.

[0088] Table 5 Performance evaluation indicators of BP and GN-BP models

[0089]

[0090] The BP model and GNBP model were used to predict the five samples in the prediction set, and the constructed tea quality sensory score prediction model was verified. Figure 3 As shown. Figure 3 It can be seen from A that the prediction deviation of the sensory score of tea quality by the GNBP model has a relatively small fluctuation range, while the prediction deviation of the sensory score of tea quality by the BP model has a large fluctuation range. Figure 3 B can be seen that the predicted value of the sensory score of tea quality by the GNBP model is closest to the sensory score of tea quality, and the change trend is relatively consistent; in contrast, the predicted value of the sensory score of tea quality by the BP model is quite different from the sensory score of tea quality, and the change trend is somewhat different. From the above experimental results, it can be seen that the average absolute errors of the BP and GNBP neural network prediction models are 3.92% and 1.47% respectively, both less than 5%, and the model prediction accuracy is high.

[0091] Based on the above examples, the time taken by this method can be compared with that of the sensory evaluation method, and the results are shown in Table 6.

[0092] Table 6 Comparison of time consumption between the method of the present invention and the national standard sensory scoring

[0093]

[0094] As shown in Table 6, the national standard sensory scoring method takes at least 90 minutes and requires at least 2 staff members and 9 trained sensory analysts. 2 The method of the present invention takes about 25 minutes and only requires 1-2 staff members, without the need for trained sensory analysts, and is not limited to the experimental site. The detection time of the method of the present invention is only 27.78% of the time taken by the traditional national standard method, which significantly improves the detection efficiency and is suitable for on-site rapid scoring and prediction of jasmine tea quality.

[0095] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A method for rapid scoring of jasmine tea quality, characterized in that The method comprises the following steps: firstly drying a small amount of ground tea sample under an infrared lamp, drying and pressing the sample into a sheet, and then performing infrared detection under a Fourier infrared spectrometer to obtain infrared detection data of the sample, and obtaining the infrared detection data of the sample at 4000-400 cm by using the coefficient of variation method. -1 The weight of the wave number within the range, select the top 15 wave numbers from high to low weight; use genetic algorithm to optimize and establish the jasmine tea quality score BP neural network prediction model, input the normalized wave number eigenvalue into the constructed prediction model, and calculate the sample quality score.

2. The method according to claim 1, characterized in that The method comprises the following steps: (1) 10-30 mg of jasmine dry tea sample was taken and ground and mixed evenly with dried potassium bromide in a weight ratio of 1:10-1:

50. After drying for 10 minutes, the sample was pressed into tablets. The infrared spectrum information of the sample was collected using a Nicolet iS10 Fourier transform infrared spectrometer. Each sample was measured three times in parallel and the average spectrum was taken. The spectrum collection range was 4000-400 cm -1 , scanning times 32 times, resolution 0.4cm -1 ;The spectral data were collected in a laboratory environment with a temperature of 25±1℃ and a humidity of 35%RH; (2) In order to reduce the noise, light scattering and tea particle effects in the acquisition environment, the original spectrum was preprocessed using a baseline correction method; (3) Find 4000~400cm -1 The weight of the wavenumber in the range is used to determine the characteristic wavenumber for constructing the jasmine tea quality scoring model; (4) Determine the number of nodes in the hidden layer of the BP neural network prediction model and construct a BP neural network prediction model for jasmine tea quality score.

3. The method according to claim 2, characterized in that As described in step (1), 10 mg of jasmine dry tea sample is taken, and it is ground and mixed with dried potassium bromide at a weight ratio of 1:10-20, and then tableted.

4. The method according to claim 2, characterized in that As described in step (2), the original spectrum is preprocessed using baseline correction, smoothing, and standard normal variable (SNV) methods.

5. The method according to claim 2, characterized in that In step (3), the coefficient of variation method is used to obtain the value of 4000-400 cm -1 The weights of the wavenumbers in the range were calculated, and the top 15 wavenumbers with the highest to lowest weights were selected as the characteristic wavenumbers to construct the jasmine tea quality evaluation model.

6. The method according to claim 2, wherein in step (3), based on origin 2022 software, Pearson correlation analysis is used to analyze the correlation between the spectral eigenvalues ​​and the quality of jasmine tea; and principal component analysis (PCA) is used to perform dimensionality reduction processing on the characteristic data with greater correlation.

7. The method as claimed in claim 6, wherein the input layer of the BP neural network prediction model is 1, which is the principal component 1 eigenvalue after principal component analysis of the transmittance under the wave numbers ranked top 15 by weight, the output layer is 1, i.e., the jasmine tea quality score, the hidden layer is 1, and the number of nodes is set to a range of 2-11; the wave numbers ranked top 15 by weight are: 3375, 3379, 3373, 3371, 3381, 3338, 3369, 3340, 3336, 3367, 3383, 3342, 3365, 3344, 3334cm -1 .

8. The method according to claim 6, wherein the quality of the jasmine tea comprises an appearance score, a soup color score, an aroma score, a taste score, a leaf bottom score and a sensory score, and the quality of the jasmine tea is evaluated according to GB / T23776-2018.

9. The method according to claim 2, characterized in that As described in step (4), the optimal number of hidden layer nodes is 7, and the BP neural network prediction model is a BP neural network optimized based on a genetic algorithm (GN); the average absolute errors of the BP neural network prediction model before and after the optimization based on the genetic algorithm (GN) are 3.92% and 1.47% respectively, both less than 5%, and the model prediction accuracy is high.

Images