A green tea freshness quality evaluation method based on near-infrared spectroscopy and ingredient correlation
By using near-infrared spectroscopy and component correlation methods, a green tea freshness quality evaluation model was established, which solved the problems of complexity and subjectivity of traditional methods, and achieved efficient, rapid and objective evaluation of green tea freshness quality, ensuring the accuracy and continuity of tea quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTHWEST UNIV
- Filing Date
- 2026-01-05
- Publication Date
- 2026-05-19
AI Technical Summary
Existing methods for evaluating the freshness and quality of green tea are complex, time-consuming, require expensive equipment, and are highly subjective, making it difficult to achieve efficient, rapid, and objective quantitative evaluation.
A method based on near-infrared spectroscopy and component correlation was adopted to obtain the set of freshness components and grade thresholds of green tea samples, establish the relationship between freshness components and freshness quality grade, classify them using a logistic regression model, and determine the freshness quality grade of green tea by combining near-infrared spectroscopy technology, thus avoiding damage to the tea leaves themselves.
It enables efficient, rapid, and objective evaluation of the freshness and quality of green tea, ensuring the continuity and accuracy of grading standards, and preserving the original quality and appearance of the tea.
Smart Images
Figure CN121453714B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of green tea freshness and quality evaluation, and particularly relates to a method for evaluating green tea freshness and quality based on near-infrared spectroscopy and component correlation. Background Technology
[0002] Tea, as one of the world's most widely consumed beverages, has a significant impact on consumers' taste and quality. Among the many flavor characteristics of tea, umami, as one of the basic tastes, plays a significant role in enhancing the overall taste and quality of the tea soup. Green tea, as an important variety of tea, is favored by consumers for its unique flavor. Umami, as a hallmark flavor characteristic of green tea, manifests as the freshness, vibrancy, freshness, and vitality of the tea soup. Its presentation depends not only on umami substances such as amino acids but also on the combined action of appropriate concentrations of polyphenols to create a refreshing and harmonious sensory experience. Therefore, the umami quality of green tea is considered a core indicator for evaluating green tea quality and is also the most easily identifiable and representative product matching indicator for consumers. Currently, most green tea product quality labels use a comprehensive grade, which deviates from consumers' actual perception. Using umami as a directional indicator would help improve product recognition and market acceptance. In terms of evaluation methods, although traditional sensory evaluation includes umami elements, it is highly subjective and difficult to accurately quantify umami.
[0003] In green tea, the umami flavor is mainly contributed by various secondary metabolites such as amino acids, catechins, and free sugars. The content, ratio, and threshold of these substances will affect the presentation of the umami flavor in the tea soup. Therefore, it is feasible to identify and determine the umami flavor of green tea through chemical components.
[0004] Existing chemical detection methods such as quantitative descriptive analysis (QDA) or high-performance liquid chromatography (HPLC) can achieve a certain degree of evaluation, but they are generally limited by complex procedures, long processing times, expensive equipment, and high skill requirements for operators. Therefore, establishing an efficient, rapid, and objective technology for evaluating the freshness and quality of green tea is of great significance for accurately identifying and controlling the quality of tea. Summary of the Invention
[0005] To address the aforementioned shortcomings in existing technologies, this invention provides a method for evaluating the freshness and quality of green tea based on near-infrared spectroscopy and component correlation, which solves the problems of low applicability and efficiency of existing methods for measuring freshness and quality grades.
[0006] To achieve the aforementioned objectives, the technical solution adopted by this invention is: a method for evaluating the freshness and quality of green tea based on near-infrared spectroscopy and component correlation, comprising:
[0007] Obtain several green tea samples and the freshness and quality grades of each green tea sample;
[0008] Determine the set of freshness components and set the level threshold for each freshness component in the set;
[0009] Based on the threshold values of each freshness component, the freshness component grade of each green tea sample is determined. Then, based on the freshness component grade and freshness quality grade of each green tea sample, the final freshness component-freshness quality grade relationship is determined. The expression for the final freshness component-freshness quality grade relationship is as follows:
[0010]
[0011] in, The freshness quality grade corresponding to the highest posterior probability; To make the largest The value of ; For freshness quality grade parameters; A vector of freshness components; The freshness quality grade is numbered; This is the final model parameter vector; For the first The freshness component weight vectors corresponding to each grade; For the first Each level corresponds to a bias term; This represents the total number of freshness quality grades.
[0012] Near-infrared spectra of each green tea sample were extracted to determine the wavelength characteristics of each freshness component at each grade.
[0013] The near-infrared spectrum of the green tea sample to be tested is obtained. Based on the near-infrared spectrum of the green tea sample to be tested and the wavelength characteristics of each freshness component at each grade, the grade of each freshness component of the green tea sample to be tested is obtained. Based on the final freshness component-freshness quality grade relationship, the freshness quality grade of the green tea sample to be tested is obtained.
[0014] Furthermore, the formula for determining the final freshness component-freshness quality grade relationship is as follows:
[0015] The content of each freshness component in each green tea sample was measured, and the grade of each freshness component in each green tea sample was obtained based on the grade threshold of each freshness component. The green tea sample information was then integrated to obtain a green tea sample dataset.
[0016]
[0017] in, For green tea sample datasets; For the first One green tea sample; For the first The freshness quality grade of each green tea sample; For the first Content of each freshness component; For the first Each freshness component grade; Index for green tea samples; A collection of green tea samples; Index of freshness components; It is a collection of freshness components;
[0018] Based on the green tea sample dataset, several subsets of green tea sample data were obtained by classifying them according to their freshness and quality grades.
[0019] For each subset of green tea sample data, sort them from best to worst based on freshness quality grade;
[0020] For the r-th green tea sample data subset, based on the freshness component-freshness quality grade relationship of the (r-1)-th green tea sample data subset, construct the freshness component-freshness quality grade relationship of the r-th green tea sample data subset, until the freshness component-freshness quality grade relationship of the last green tea sample data subset is obtained, which is taken as the final freshness component-freshness quality grade relationship.
[0021] Furthermore, a logistic regression model with parameter initialization was selected as the classification model to fit the relationship between freshness components and freshness quality grades.
[0022] Furthermore, when r=1, the expression for the relationship between freshness components and freshness quality grade is:
[0023]
[0024] in, Let be a binary variable used to indicate whether it corresponds to the freshness quality grade of the first green tea sample data subset. A value of 1 indicates yes, and a value of 0 indicates no. When r=1, it is determined as =1 posterior probability; To determine the threshold; For freshness quality grade parameters; A vector of freshness components; These are the model parameters when r=1; It is the sigmoid activation function; This is the weight vector of the freshness components when r=1; For transpose; This is the bias term when r=1.
[0025] Furthermore, when r is not equal to 1, the logistic regression model is extended to a multi-class classification model; the number of classes is equal to the value of r.
[0026] Furthermore, when r is not equal to 1, the expression for the initial parameters of the logistic regression model is:
[0027]
[0028] in, This is the initial freshness component weight vector of the logistic regression model when r is not equal to 1; The influence weights of the relationship between freshness components and freshness quality grades when the value is r-1; The freshness component weight vector at r-1; This is a random freshness component weight vector generated during training on the r-th green tea sample data subset; This is the initial bias term of the logistic regression model when r is not equal to 1; The bias term when it is r-1; This is a random bias term generated during training on the r-th green tea sample data subset.
[0029] The beneficial effects of this invention are as follows: This invention solves the problem of grade boundary shifts caused by differences in component distribution among samples of different grades in traditional methods; by transferring preceding subset parameters, this invention trains the current grade while preserving migration grade characteristics, ensuring that the judgment criteria for each grade gradually connect from "excellent" to "inferior," avoiding grade boundary fragmentation; combined with near-infrared spectroscopy, it determines the quantity / grade of freshness components through near-infrared spectroscopy. This ensures that the overall technical process does not damage the tea itself, preserving the original quality and appearance of the tea. Attached Figure Description
[0030] Figure 1 This is a flowchart of the method of the present invention.
[0031] Figure 2 This is a schematic diagram showing the average aspartic acid content of samples with different freshness in the embodiments of the present invention.
[0032] Figure 3 This is a schematic diagram showing the average glutamic acid content of samples with different freshness levels in an embodiment of the present invention.
[0033] Figure 4 This is a schematic diagram showing the average asparagine content of samples with different freshness in the embodiments of the present invention.
[0034] Figure 5 This is a schematic diagram showing the average serine content of samples with different freshness levels in an embodiment of the present invention.
[0035] Figure 6 This is a schematic diagram showing the average theanine content of samples with different freshness in the embodiments of the present invention.
[0036] Figure 7 This is a schematic diagram showing the average cysteine content of samples with different freshness in the embodiments of the present invention.
[0037] Figure 8 This is a schematic diagram showing the average methionine content of samples with different freshness levels in an embodiment of the present invention.
[0038] Figure 9 This is a schematic diagram showing the average L-leucine content of samples with different freshness in the embodiments of the present invention.
[0039] Figure 10 This is a schematic diagram showing the average L-phenylalanine content of samples with different freshness in the embodiments of the present invention.
[0040] Figure 11 This is a schematic diagram showing the average histidine content of samples with different freshness in the embodiments of the present invention.
[0041] Figure 12 This is a schematic diagram showing the average ornithine content of samples with different freshness in the embodiments of the present invention.
[0042] Figure 13 This is a schematic diagram showing the average lysine content of samples with different freshness levels in an embodiment of the present invention.
[0043] Figure 14 This is a schematic diagram illustrating the average tyrosine content of samples with different freshness levels in an embodiment of the present invention.
[0044] Figure 15 This is a schematic diagram showing the average gallic acid content of samples with different freshness in the embodiments of the present invention.
[0045] Figure 16 This is a schematic diagram showing the average content of gallic catechin gallate in samples of different freshness in this invention.
[0046] Figure 17 This is a schematic diagram illustrating the average content of gallic catechin in samples of different freshness in this invention.
[0047] Figure 18 This is a schematic diagram showing the average catechin content of samples with different freshness in the embodiments of the present invention.
[0048] Figure 19 This is a schematic diagram illustrating the average caffeine content of samples with different freshness levels in an embodiment of the present invention.
[0049] Figure 20 This is a schematic diagram showing the average content of epicatechin in samples of different freshness in this invention.
[0050] Figure 21 This is a schematic diagram showing the average content of catechin gallate in samples of different freshness in this invention.
[0051] Figure 22 This is a schematic diagram showing the average content of gallic catechin gallate in samples of different freshness in this invention.
[0052] Figure 23 This is a schematic diagram showing the average content of epigallocatechin in samples of different freshness in this invention.
[0053] Figure 24 This is a schematic diagram showing the average succinic acid content of samples with different freshness levels in an embodiment of the present invention. Detailed Implementation
[0054] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.
[0055] like Figure 1 As shown, in one embodiment of the present invention, a method for evaluating the freshness and quality of green tea based on near-infrared spectroscopy and component correlation includes:
[0056] Obtain several green tea samples and the freshness and quality grades of each green tea sample;
[0057] Determine the set of freshness components and set the level threshold for each freshness component in the set;
[0058] Based on the threshold values of each freshness component, the freshness component grade of each green tea sample is determined. Then, based on the freshness component grade and freshness quality grade of each green tea sample, the final freshness component-freshness quality grade relationship is determined. The expression for the final freshness component-freshness quality grade relationship is as follows:
[0059]
[0060] in, The freshness quality grade corresponding to the highest posterior probability; To make the largest The value of ; For freshness quality grade parameters; A vector of freshness components; The freshness quality grade is numbered; This is the final model parameter vector; For the first The freshness component weight vectors corresponding to each grade; For the first Each level corresponds to a bias term; This represents the total number of freshness quality grades.
[0061] Near-infrared spectra of each green tea sample were extracted to determine the wavelength characteristics of each freshness component at each grade.
[0062] The near-infrared spectrum of the green tea sample to be tested is obtained. Based on the near-infrared spectrum of the green tea sample to be tested and the wavelength characteristics of each freshness component at each grade, the grade of each freshness component of the green tea sample to be tested is obtained. Based on the final freshness component-freshness quality grade relationship, the freshness quality grade of the green tea sample to be tested is obtained.
[0063] In this embodiment, by seeking the relationship between various umami components and umami quality grades, a formula for determining umami quality grades through umami components is established. Based on this, to avoid damaging the tea leaves during the determination of umami components, near-infrared spectroscopy technology is used to determine the quantity / grade of umami components (umami component grades are based on content). This ensures that the overall technical process does not damage the tea leaves themselves, preserving their original quality and appearance.
[0064] In this embodiment, the process for obtaining the freshness quality grade of each green tea sample is as follows:
[0065] The umami intensity of green tea samples was evaluated using the QDA method. A sensory evaluation team was established to complete the QDA evaluation of umami intensity in this study. The evaluation team members underwent multiple training sessions before the formal evaluation began. Panel check software analysis showed that the evaluation team's discrimination ability, stability, and consistency all met the standards.
[0066] In the formal experiment, monosodium glutamate (MSG) was used as the standard reference for umami flavor. A 10-point scale method was used to descriptively analyze the umami intensity of green tea samples. The MSG equivalent umami intensity evaluation scale is shown in Table 1. The tea brewing conditions were a tea-to-water ratio of 1:50 and a brewing time of 4 minutes. The brewed tea was kept warm in a 50℃ water bath to ensure the tea temperature was between 45±2℃ when the evaluators tasted it. 50 mL of tea was collected, labeled with a random three-digit code, and distributed to the evaluation team. The evaluation team determined the umami intensity of the tea based on the standard reference solution. A 5-minute interval was required between two sample evaluations, and the participants were encouraged to rinse their mouths with water to reduce flavor interactions and taste fatigue.
[0067] Table 1 Umami Intensity and Corresponding Standard Reference Sample Concentration
[0068]
[0069] The QDA evaluation results of 1000 green tea samples showed that the umami intensity ranged from 2.0 to 4.25. Based on the evaluation results, the samples could be divided into three groups with different umami intensities: X1 was the high umami group, with 270 samples having an umami intensity above 3.6; X2 was the medium umami group, with 513 samples having an umami intensity between 2.8 and 3.6; and X3 was the low umami group, with 217 samples having an umami intensity below 2.8. It can be seen that the umami intensity of this batch of green tea covered most of the range from excellent to poor, and had good distinguishability.
[0070] The free amino acid content of tea samples with different freshness was determined. The results are as follows: Figures 2-14 As shown, in green tea samples of different freshness, theanine was the most abundant amino acid, followed by Aspartic acid, Asparagine, Glutamic acid, and Serine. Theanine, Aspartic acid, Asparagine, and Glutamic acid are all umami amino acids, while Serine is a sweet amino acid. In addition, Glutamic acid, Ala, and Threonine are sweet-tasting, while Tyr, L-leucine, and Theanine are bitter-tasting and present in lower amounts. With the decrease in the umami intensity of green tea, the contents of theanine, Aspartic acid, Glutamic acid, Asparagine, Cys, Met, L-leucine, Histidine, Orn, and Lysine all gradually decreased. Free amino acids in tea, especially theanine, are the main components contributing to the fresh and crisp taste of green tea. When the content of free amino acids such as theanine decreases, the fresh and crisp taste of green tea weakens significantly, and the tea soup becomes relatively bland. Furthermore, the flavor of tea is the result of the combined effects of multiple components, among which different free amino acids may exhibit different flavor characteristics. When the content of free amino acids decreases, the freshness of green tea decreases, possibly accompanied by changes in other flavor qualities such as bitterness, astringency, sourness, and sweetness, leading to a poorer balance of flavors. Comparing the differences in free amino acid content among different freshness levels, theanine, Glu (glutamic acid), and Serine (serine) showed significant differences among samples of different freshness levels, indicating that they may be key substances affecting the freshness of green tea. Phe refers to L-phenylalanine.
[0071] Comparison results of catechin components and caffeine content among different grades are as follows: Figures 15-23As shown in the figure, comparing the test results of tea samples with different freshness levels reveals that as the umami intensity decreases, the contents of GA (gallic acid), GC (gallocatechin), C (catechin), and GCG (gallocatechin gallate) decrease, while the contents of EGC (epigallocatechin), EC (epigallocatechin), and EGCG (epigallocatechin gallate) increase. The contents of EGC (epigallocatechin) and CAF (caffeine) remain relatively stable. Analyzing the differences in catechin and caffeine contents among green tea samples with different freshness levels, significant differences were found in the contents of GA (gallic acid), EGC (epigallocatechin), C (catechin), EC (epigallocatechin), EGCG (epigallocatechin gallate), and GCG (glucocatechin gallate). Among these, EGCG (epigallocatechin gallate) showed significant differences among all three groups of samples. EGCG (epigallocatechin gallate) is one of the main bitter and astringent components in green tea. Because it has a low bitter and astringent threshold, as the content of EGCG increases, the bitterness and astringency of green tea will become more obvious, which will have a significant impact on the overall taste of the tea soup and may further weaken the umami intensity of green tea.
[0072] The results of the comparison of succinic acid content among different grades are as follows: Figure 24 As shown, the succinic acid content decreases with the decrease in umami intensity, and the succinic acid content in sample X1 differs significantly from that in samples X2 and X3. Succinic acid itself contributes primarily to the umami flavor, but as a component of the tea infusion, its content changes may indirectly affect the overall flavor of the tea. When the succinic acid content decreases, its synergistic effect with amino acids in the tea infusion may weaken, leading to a decrease in the umami intensity of the tea infusion.
[0073] The formula for determining the final freshness component-freshness quality grade relationship is as follows:
[0074] The content of each freshness component in each green tea sample was measured, and the grade of each freshness component in each green tea sample was obtained based on the grade threshold of each freshness component. The green tea sample information was then integrated to obtain a green tea sample dataset.
[0075]
[0076] in, For green tea sample datasets; For the first One green tea sample; For the first The freshness quality grade of each green tea sample; For the first Content of each freshness component; For the first Each freshness component grade; Index for green tea samples; A collection of green tea samples; Index of freshness components; It is a collection of freshness components;
[0077] Based on the green tea sample dataset, several subsets of green tea sample data were obtained by classifying them according to their freshness and quality grades.
[0078] For each subset of green tea sample data, sort them from best to worst based on freshness quality grade;
[0079] For the r-th green tea sample data subset, based on the freshness component-freshness quality grade relationship of the (r-1)-th green tea sample data subset, construct the freshness component-freshness quality grade relationship of the r-th green tea sample data subset, until the freshness component-freshness quality grade relationship of the last green tea sample data subset is obtained, which is taken as the final freshness component-freshness quality grade relationship.
[0080] In this embodiment, the quantitative values of freshness components are converted into grade descriptions, and the joint influence relationship between the grades of each freshness component and the freshness quality grade is sought. The complex numerical calculations are transformed into simple grade relationship seeking, which reduces the difficulty and complexity of relationship finding.
[0081] We selected a logistic regression model with parameter initialization as the classification model to fit the relationship between freshness components and freshness quality grades.
[0082] When r=1, the expression for the relationship between freshness components and freshness quality grade is:
[0083]
[0084] in, Let be a binary variable used to indicate whether it corresponds to the freshness quality grade of the first green tea sample data subset. A value of 1 indicates yes, and a value of 0 indicates no. When r=1, it is determined as =1 posterior probability; To determine the threshold; For freshness quality grade parameters; A vector of freshness components; These are the model parameters when r=1; It is the sigmoid activation function; This is the weight vector of the freshness components when r=1; For transpose; This is the bias term when r=1.
[0085] When r is not equal to 1, the logistic regression model is extended to a multi-class model; the number of classes is equal to the value of r.
[0086] When r is not equal to 1, the expression for the initial parameters of the logistic regression model is:
[0087]
[0088] in, This is the initial freshness component weight vector of the logistic regression model when r is not equal to 1; The influence weights of the relationship between freshness components and freshness quality grades when the value is r-1; The freshness component weight vector at r-1; This is a random freshness component weight vector generated during training on the r-th green tea sample data subset; This is the initial bias term of the logistic regression model when r is not equal to 1; The bias term when it is r-1; This is a random bias term generated during training on the r-th green tea sample data subset.
[0089] In this embodiment, the calculation begins with the dataset containing the highest freshness quality grade. As the grade decreases, the number of categories in the logistic regression model increases. The model parameters from the previous freshness quality grade are retained, and the model parameters obtained for the current freshness quality grade are adjusted based on the influence of these parameters. This ensures that the final relational expression is applicable to teas of all freshness quality grades.
Claims
1. A method for evaluating the freshness and quality of green tea based on near-infrared spectroscopy and component correlation, characterized in that, include: Obtain several green tea samples and the freshness and quality grades of each green tea sample; Determine the set of freshness components and set the level threshold for each freshness component in the set; The content of each freshness component in each green tea sample was measured, and the freshness component grade in each green tea sample was obtained based on the grade threshold of each freshness component. Based on the freshness component grade and freshness quality grade of each green tea sample, the final freshness component-freshness quality grade relationship was determined; the expression of the final freshness component-freshness quality grade relationship is as follows: in, The freshness quality grade corresponding to the highest posterior probability; To make the largest The value of ; For freshness quality grade parameters; This is a vector of freshness components; The freshness quality grade is numbered; This is the final model parameter vector; For the first The freshness component weight vectors corresponding to each grade; For the first Each level corresponds to a bias term; This represents the total number of freshness quality grades. Near-infrared spectra of each green tea sample were extracted to determine the wavelength characteristics of each freshness component at each grade. The near-infrared spectrum of the green tea sample to be tested is obtained. Based on the near-infrared spectrum of the green tea sample to be tested and the wavelength characteristics of each freshness component at each grade, the grade of each freshness component of the green tea sample to be tested is obtained. Based on the final freshness component-freshness quality grade relationship, the freshness quality grade of the green tea sample to be tested is obtained.
2. The method for evaluating the freshness and quality of green tea based on near-infrared spectroscopy and component correlation according to claim 1, characterized in that, The formula for determining the final freshness component-freshness quality grade relationship is as follows: The content of each freshness component in each green tea sample was measured, and the grade of each freshness component in each green tea sample was obtained based on the grade threshold of each freshness component. The green tea sample information was then integrated to obtain a green tea sample dataset. in, For green tea sample datasets; For the first One green tea sample; For the first The freshness quality grade of each green tea sample; For the first Content of each freshness component; For the first Each freshness component grade; Index for green tea samples; A collection of green tea samples; Index of freshness components; It is a collection of freshness components; Based on the green tea sample dataset, several subsets of green tea sample data were obtained by classifying them according to their freshness and quality grades. For each subset of green tea sample data, sort them from best to worst based on freshness quality grade; For the r-th green tea sample data subset, based on the freshness component-freshness quality grade relationship of the (r-1)-th green tea sample data subset, construct the freshness component-freshness quality grade relationship of the r-th green tea sample data subset, until the freshness component-freshness quality grade relationship of the last green tea sample data subset is obtained, which is taken as the final freshness component-freshness quality grade relationship.
3. The method for evaluating the freshness and quality of green tea based on near-infrared spectroscopy and component correlation according to claim 2, characterized in that, We selected a logistic regression model with parameter initialization as the classification model to fit the relationship between freshness components and freshness quality grades.
4. The method for evaluating the freshness and quality of green tea based on near-infrared spectroscopy and component correlation according to claim 3, characterized in that, When r=1, the expression for the relationship between freshness components and freshness quality grade is: in, Let be a binary variable used to indicate whether it corresponds to the freshness quality grade of the first green tea sample data subset. A value of 1 indicates yes, and a value of 0 indicates no. When r=1, it is determined as =1 posterior probability; To determine the threshold; For freshness quality grade parameters; This is a vector of freshness components; These are the model parameters when r=1; It is the sigmoid activation function; This is the weight vector of the freshness components when r=1; For transpose; This is the bias term when r=1.
5. The method for evaluating the freshness and quality of green tea based on near-infrared spectroscopy and component correlation according to claim 3, characterized in that, When r is not equal to 1, the logistic regression model is extended to a multi-class model; the number of classes is equal to the value of r.
6. The method for evaluating the freshness and quality of green tea based on near-infrared spectroscopy and component correlation according to claim 5, characterized in that, When r is not equal to 1, the expression for the initial parameters of the logistic regression model is: in, This is the initial freshness component weight vector of the logistic regression model when r is not equal to 1; The influence weights of the relationship between freshness components and freshness quality grades when the value is r-1; The freshness component weight vector at r-1; This is a random freshness component weight vector generated during training on the r-th green tea sample data subset; This is the initial bias term of the logistic regression model when r is not equal to 1; The bias term when it is r-1; This is a random bias term generated during training on the r-th green tea sample data subset.