Method and system for discriminating years of raw Pu'er tea in different storage aging periods

Abstract

The invention relates to a method and a system for discriminating years of raw Pu'er tea in different storage aging periods. The method comprises the following steps: S1, preparing a sample; s2, analyzing components to determine the content of non-volatile compounds; s31, dividing the sample into three storage aging periods; s32, screening the number and the type of the non-volatile compounds by adopting LASSO regression to obtain a key predictive factor; s33, taking the key predictive factor as an input variable, and combining linear discriminant analysis to construct a year discriminant model; s4, outputting a result; the method has the advantages that based on LASSO regression of non-volatile compounds, linear discriminant analysis and other frontier machine learning technologies, an accurate model for Pu'er raw tea year discrimination is constructed, a matched visual module is combined, an efficient and visual tool is provided for Pu'er raw tea year discrimination, and the Pu'er raw tea year discrimination method is suitable for popularization and application. And scientific technical support is provided for quality evaluation and market transaction of annual Pu'er raw tea.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tea identification, and particularly relates to a Pu'er raw tea year identification method and system with different storage aging periods. BACKGROUND

[0002] Pu'er raw tea is a compacted tea made of sun-dried green tea of Yunnan large-leaf variety by steaming and pressing, and its quality characteristics are pure and lasting aroma and mellow and sweet taste. Pu'er raw tea shows significant quality conversion characteristics within a scientific storage aging time, i.e., its taste attributes gradually evolve to positive indexes such as alcohol, sweetness, thickness and smoothness. At the same time, the aroma changes regularly, evolving from initial fragrance to woodsy aroma, sweet aroma and typical aged aroma. The storage aging of Pu'er raw tea is a process in which the internal substances are subjected to a series of complex biochemical conversions through the synergistic effect of multiple factors such as microbial activity, enzymatic reaction, wet heating and oxidation process under specific environmental conditions, and finally the unique quality characteristics are formed. Thus, the quality of Pu'er tea changes with the storage time, which increases the unique enjoyment of drinking and makes it have good collection value.

[0003] The research results show that with the extension of the storage time, the soup color of Pu'er raw tea gradually deepens, the aroma characteristics change from fragrance to aged aroma, the taste changes from mellow to thick and mellow, and the color of the leaf bottom also deepens and darkens. After storage for more than 10 years, the quality of the tea is improved, but after storage for 18 years, the quality begins to decline, which is mainly caused by the changes of the internal components in different degrees.

[0004] However, in the market, consumers usually have difficulty in accurately determining the storage year of the sold Pu'er tea. In the field of tea year identification, the traditional method mainly includes sensory evaluation method, which highly depends on the experience and subjective judgment of the evaluators, and is significantly affected by personal preferences, emotions and external environmental factors, so its objectivity is obviously insufficient. In addition, the results of sensory evaluation often lack clear quantitative standards, have poor repeatability, and are difficult to maintain consistency between different evaluators, which leads to great limitations in the accuracy and reliability of year identification. This situation leads to the frequent invasion of bad products into the space of good products in the year Pu'er tea market.

[0005] Therefore, it is urgent to establish an efficient and reliable Pu'er tea year identification model to promote the standardized management of the Pu'er tea market, and based on this, the present application is proposed. SUMMARY

[0006] One of the purposes of the present application is to provide a Pu'er raw tea year identification method with different storage aging periods, which solves the above-mentioned defects.

[0007] To achieve the above object, the technical scheme of the present application is as follows:

[0008] A Pu'er raw tea year identification method for different storage aging periods, comprising the following steps:

[0009] S1. Sample preparation, obtaining Pu'er raw tea samples of different storage aging periods;

[0010] S2. Component analysis, performing component analysis on the samples to determine the content of non-volatile compounds;

[0011] S3. Model construction;

[0012] S31. The samples are divided into three storage aging periods, namely LV1, LV2 and LV3, wherein LV1 represents a storage aging period of 1-3 years, LV2 represents a storage aging period of 4-8 years, and LV3 represents a storage aging period of 9 years or more;

[0013] S32. LASSO regression is used to screen the number and type of non-volatile compounds to obtain key prediction factors;

[0014] S33. The key prediction factors are used as input variables, and a linear discriminant analysis is combined to construct a year identification model;

[0015] S4. Result output, including the predicted probability distribution of each sample in the three storage aging periods, and the storage aging period with the highest probability is selected as the final prediction value.

[0016] Further, the key prediction factors include water extract, theabrownin, epicatechin, epicatechin gallate, caffeine, quercetin, rutin and gallic acid.

[0017] Further, the LASSO regression formula is as follows:

[0018]

[0019] In the formula, J(θ) represents the target function of LASSO regression, i.e. the loss function that needs to be minimized; h θ (x (i) ) is the predicted value of the model, representing the model output based on the input features x (i) and the parameter vector θ; y (i) The true label or target value of the i-th sample; m represents the number of training samples; λ is the regularization parameter that controls the weight of the regularization term, and the complexity of the LASSO model is mainly regulated by λ, as the value of λ increases, the number of variables selected by the model will decrease accordingly, thereby effectively reducing the complexity of the model, thereby reducing the risk of overfitting to a certain extent; denotes the mean square error, which measures the difference between the predicted value and the true value of the model; denotes the L1 regularization term, i.e., the sum of the absolute values of all parameters, which helps some parameters to become zero, thereby realizing feature selection.

[0020] Further, the year discrimination model finds the optimal projection direction w by maximizing the ratio of the inter-class scatter to the intra-class scatter, and the specific formula is as follows:

[0021] Intra-class scatter matrix

[0022] Inter-class scatter matrix

[0023] Fisher criterion

[0024] Solve the generalized eigenvalue problem: Get the optimal projection direction w;

[0025] In the formula:

[0026] S w is used to measure the dispersion degree between samples in the same category; c represents the number of categories, i.e., 3 warehouse aging periods; i is the set of all samples belonging to the i-th category; x is a sample vector; μ i is the mean vector of the i-th category sample; T represents the transpose operation of a matrix or a vector;

[0027] S b measures the difference between samples of different categories; N i denotes the number of samples of the i-th category; μ is the overall mean vector of all samples;

[0028] J(w) is used to find the optimal projection direction w that maximizes the inter-class distance and minimizes the intra-class distance.

[0029] Further, the result output further includes a visual report to display the year prediction result of the warehouse aging period in the form of an image.

[0030] The second purpose of the present application is to provide a Pu'er raw tea year discrimination system based on the above-mentioned Pu'er raw tea year discrimination method for different warehouse aging periods, comprising:

[0031] An input module is used to import the component analysis data of the Pu'er raw tea sample;

[0032] A key predictor extraction module is used to screen out key predictors;

[0033] A year discrimination module is used to give the year discrimination result of the Pu'er raw tea.

[0034] Further, a visualization module is included to provide a user interface to display the year prediction probability curve, predicted value and confidence interval of different storage aging cycle Puerh raw tea samples, and to summarize the prediction results of different storage aging cycles in numerical form.

[0035] A third object of the present application is to provide a computer-readable storage medium comprising the Puerh raw tea year identification method of different storage aging cycles.

[0036] The advantages of the present application are that: based on the frontier machine learning technologies such as LASSO regression and linear discriminant analysis (LDA) of non-volatile compounds, a precise model for Puerh raw tea year identification is constructed, and combined with the supporting visualization module, an efficient and intuitive tool is provided for Puerh raw tea year identification, which provides scientific technical support for the quality evaluation and market transaction of Puerh raw tea of different years. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 The flowchart of the Puerh raw tea year identification method of different storage aging cycles in the embodiment. DETAILED DESCRIPTION

[0038] The present application will be further described in detail below in conjunction with the embodiments, and it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer" and the like in the text indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application.

[0039] The present embodiment proposes a Puerh raw tea year identification method of different storage aging cycles, as shown in Figure 1 The method includes four steps of sample preparation, component analysis, model construction and result output.

[0040] In the sample preparation, the present embodiment selects Puerh raw tea from Dehong Mangshi production area in 2023, 2022, 2021, 2020, 2019, 2018, 2017, 2016, 2015, 2014 and 2013, Puerh Jingmai Mountain production area in 2020, 2019, 2018, 2016, 2015, 2014, 2011, 2010 and 2008, and Banna Mountain production area in 2020, 2018, 2016, 2014, 2012, 2011 and 2008, all of which use local Yunnan large leaf population one bud two or three leaves as fresh leaf raw materials. The processing steps are: fresh leaves → withering → fixation → rolling → drying → pressing.

[0041] In the component analysis, the non-volatile compound content is determined by analyzing the sample.

[0042] In this step, the non-volatile compound content includes the following processes:

[0043] Sample pretreatment method: The sample is ground and sieved through a 40-mesh sieve, then sealed and stored in a -20°C refrigerator for standby use.

[0044] The content of tea polyphenols is determined by GB / T 8313-2018 "Determination of tea polyphenols and catechins in tea", the content of free amino acids is determined by GB / T 8314-2013 "Determination of total amount of free amino acids in tea", the content of soluble sugar is determined by anthrone-sulfuric acid colorimetry, the content of water extract is determined by GB / T 8305-2013 "Determination of tea water extract", the content of theaflavins, thearubigins and theabrownins is determined by the system analysis method (reference: Wang Q, Peng C, Gong J. Effects of enzymatic action on the formation of theabrownin during solid state fermentation of Pu-erh tea [J]. Journal of the Science of Food and Agriculture, 2011, 91(13): 2412-2418), and the content of catechins, flavonoids and purine alkaloids is determined by the method in the reference [Nian B, Chen L, Yi C, et al. A high performance liquid chromatography method for simultaneous detection of 20 bioactive components in tea extracts [J]. Electrophoresis, 2019, 40(21): 2837-2844] by high performance liquid chromatography (HPLC), in which mobile phase A is 0.261% phosphoric acid and 5% acetonitrile, and mobile phase B is 80% methanol; the elution gradient program is as follows: 0 to 16 min, mobile phase B linearly increases from 10% to 45%; 16 min to 22 min, mobile phase B linearly increases from 45% to 65%; from 22 min to 25.9 min, mobile phase B remains 65%; from 25.9 min to 29.0 min, mobile phase B linearly increases to 100%; from 29 min to 30 min, mobile phase B remains 100%. The column temperature is 35°C. Each sample is extracted and detected in triplicate.

[0045] Through component analysis, the contents of 19 non-volatile compounds can be obtained.

[0046] The construction process of the model is as follows:

[0047] The samples are divided into three warehouse aging periods, LV1, LV2 and LV3, where LV1 represents a warehouse aging period of 1-3 years, LV2 represents a warehouse aging period of 4-8 years, and LV3 represents a warehouse aging period of 9 years or more.

[0048] In this study, LASSO regression is used to optimize the construction of the prediction model, aiming to accurately select the number and type of variables in the early stage of model construction. LASSO regression is a linear regression analysis technique that introduces an L1 regularization penalty term, enhancing the sparsity of the model and helping to effectively reduce the complexity of the model by selecting key variables, thereby improving the performance parameter performance of the model and optimizing the model structure. The adjustment of the complexity of the model is mainly realized by controlling the relevant parameters, so as to avoid the occurrence of overfitting phenomenon.

[0049] The LASSO regression formula is as follows:

[0050]

[0051] In the formula:

[0052] J(θ) represents the objective function of LASSO regression, i.e. the loss function that needs to be minimized;

[0053] h θ (x (i) ) is the predicted value of the model, representing the model output based on the input features x (i) and the parameter vector θ;

[0054] y (i) is the true label or target value of the i-th sample;

[0055] m represents the number of training samples;

[0056] λ is the regularization parameter that controls the weight of the regularization term. The complexity of the LASSO model is mainly regulated by λ. As the value of λ increases, the number of variables selected by the model will decrease accordingly, thereby effectively reducing the complexity of the model and reducing the risk of overfitting to a certain extent;

[0057] MSE represents the mean squared error, which measures the difference between the predicted value of the model and the true value;

[0058] L1 regularization term, i.e. the sum of the absolute values of all parameters, which helps some parameters to become zero, thereby realizing feature selection.

[0059] According to the regression function, finally, 8 prediction factors including water extract (WE), theabrownin (TB), epigallocatechin (EGC), epigallocatechin gallate (ECG), caffeine (CA), quercetin (Que), rutin (Rut), and gallic acid (GA) were screened out to establish the prediction model.

[0060] In this embodiment, the key prediction factors were used as input variables, and a linear discriminant analysis (LDA) was combined to construct a year discriminant model. As a typical supervised linear dimension reduction algorithm, the linear discriminant analysis (LDA) has wide application value in the field of feature extraction. The goal of LDA is to find a projection direction w, so that the interval between different storage aging periods in the new space is maximized, and the data of the same storage aging period is as close as possible. Specifically, LDA achieves this goal by maximizing the ratio of the between-class scatter and the within-class scatter. The specific formula is as follows:

[0061] Within-class scatter matrix

[0062] Between-class scatter matrix

[0063] Fisher criterion

[0064] Solve the generalized eigenvalue problem: Get the optimal projection direction w;

[0065] In the formula:

[0066] S w Used to measure the dispersion degree between samples in the same category; c represents the number of categories, i.e., 3 storage aging periods; X i is the set of all samples belonging to the ith category; x is a sample vector; μ i is the mean vector of the ith category sample; T represents the transpose operation of a matrix or a vector;

[0067] S b Measure the difference between samples of different categories; N i represents the number of samples of the ith category; μ is the overall mean vector of all samples;

[0068] J(w) is used to find the optimal projection direction w that maximizes the between-class distance and minimizes the within-class distance.

[0069] When evaluating the test set, the LASSO-LDA model has good performance. The accuracy of the model reaches 0.913, indicating that 91.3% of the samples in the test set are correctly classified. At the same time, the 95% confidence interval is (0.7196, 0.9893), which further illustrates the reliability of the model performance. In addition, the no information rate is 0.4348, and the p value of the model accuracy significantly higher than the no information rate is only 2.196x10- 6 This result has statistical significance, proving the effectiveness of the model in the classification task. The Kappa value of the model is 0.8631, reflecting the difference between the classification results of the model and random classification. A higher Kappa value indicates that the model has strong classification ability. The above results show that the LASSO-LDA model constructed in this study is feasible.

[0070] The result output includes the predicted probability distribution of each sample in the three warehouse aging periods, and by selecting the warehouse aging period with the highest probability as the final prediction value, it also includes a visualization report to display the year prediction results of the warehouse aging period in the form of an image.

[0071] Based on the above method, the embodiment also proposes a Pu'er raw tea year discrimination system, which includes:

[0072] An input module for importing component analysis data of Pu'er raw tea samples;

[0073] A key predictor extraction module for filtering out key predictors;

[0074] A year discrimination module for giving the year discrimination result of Pu'er raw tea;

[0075] A visualization module for providing a user interface to display the year prediction probability curve, prediction value and confidence interval of Pu'er raw tea samples in different warehouse aging periods, and summarize the prediction results of different warehouse aging periods in numerical form.

[0076] Finally, the embodiment also proposes a computer readable storage medium having a computer program stored thereon, which can implement the above method when executed by a processor.

[0077] The above embodiments are only used to explain the concept of the present application, and are not limited to the protection of the present application. Any non-substantial modification of the present application shall fall within the scope of protection of the present application.

Claims

1. A method for distinguishing the age of Pu'er raw tea with different storage and aging periods, characterized in that: The following steps are involved: S1. Sample preparation: obtaining raw Pu'er tea samples from different storage and aging periods; S2. Composition analysis: perform composition analysis on the sample to determine the content of non-volatile compounds; S3. Model construction; S31. The samples were divided into three storage and aging periods: LV1, LV2, and LV3, where LV1 represents a storage and aging period of 1-3 years, LV2 represents a storage and aging period of 4-8 years, and LV3 represents a storage and aging period of 9 years or more. S32. LASSO regression was used to screen the number and type of non-volatile compounds and obtain key predictors; S33. Using key predictors as input variables and combining them with linear discriminant analysis, a year discrimination model was constructed. S4. Result output includes the predicted probability distribution of each sample in the three storage and aging cycles, and the storage and aging cycle with the highest probability is selected as the final predicted value.

2. The method for distinguishing the age of Pu'er raw tea with different storage and aging periods according to claim 1, characterized in that: The key predictors included aqueous extract, theabrownin, epigallocatechin, epigallocatechin gallate, caffeine, quercetin, rutin, and gallic acid.

3. The method for distinguishing the age of Pu'er raw tea with different storage and aging periods as claimed in claim 1, characterized in that: The LASSO regression formula is as follows: Where: J(θ) represents the objective function of LASSO regression, that is, the loss function that needs to be minimized; h θ (x (i) ) is the predicted value of the model, indicating the value of the input feature x (i) and the model output of the parameter vector θ; y (i) The true label or target value of the i-th sample; m represents the number of training samples; λ is the regularization parameter that controls the weight of the regularization term. The complexity of the LASSO model is mainly regulated by λ. As the value of λ increases, the number of variables screened by the model will decrease accordingly, thereby effectively reducing the complexity of the model and thus reducing the risk of overfitting to a certain extent. Represents the mean square error, which measures the difference between the model prediction value and the true value; represents the L1 regularization term, which is the sum of the absolute values ​​of all parameters. This regularization term helps to make some parameters become zero, thereby achieving feature selection.

4. The method for distinguishing the age of Pu'er raw tea with different storage and aging periods according to claim 1, wherein: The year discrimination model finds the optimal projection direction w by maximizing the ratio of inter-class divergence to intra-class divergence. The specific formula is as follows: Intra-class scatter matrix Between-class scatter matrix Fisher Criterion Solve the generalized eigenvalue problem: Get the optimal projection direction w; Where: S w It is used to measure the degree of dispersion between samples within the same category; c represents the number of categories, i.e., 3 storage and aging cycles; X i is the set of all samples belonging to the i-th class; x is a sample vector; μ i is the mean vector of the i-th class sample; T represents the transpose operation of the matrix or vector; S b Measures the differences between samples of different categories; N i represents the number of samples in the i-th class; μ is the overall mean vector of all samples; J(w) is used to find the optimal projection direction w that maximizes the inter-class distance and minimizes the intra-class distance.

5. The method for distinguishing the age of Pu'er raw tea with different storage and aging periods as claimed in claim 1, characterized in that: The result output also includes a visual report that displays the annual forecast results of the storage maturity cycle in the form of images.

6. A Pu'er raw tea age identification system based on the Pu'er raw tea age identification method with different storage and aging periods according to any one of claims 1 to 5, characterized in that: include: Input module, used to import component analysis data of Pu'er raw tea samples; Key predictor extraction module, used to screen out key predictor factors; The year identification module provides the year identification results of Pu'er raw tea.

7. The system for distinguishing the age of Pu'er raw tea with different storage and aging periods according to claim 6, characterized in that: It includes a visualization module that provides a user interface to display the year prediction probability curve, prediction value and confidence interval of Pu'er raw tea samples with different storage and aging periods, and summarizes the prediction results of different storage and aging periods in numerical form.

8. A computer-readable storage medium, characterized in that A method for distinguishing the age of Pu'er raw tea with different storage and aging periods as described in any one of claims 1 to 5.

Images