Tea quality evaluation data set construction method based on multi-source cross-modal acquisition
By using multi-source cross-modal data acquisition and data standardization processing, a tea quality assessment dataset was constructed, which solved the problems of subjectivity and data fusion in tea quality assessment and achieved objective and high-precision assessment of tea quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-18
- Publication Date
- 2026-04-03
AI Technical Summary
Existing tea quality assessment technologies suffer from problems such as strong subjectivity, limited data dimensions, shallow integration, and poor data quality and consistency, making it difficult to achieve high-precision and robust digital quality assessment.
A multi-source cross-modal acquisition method is adopted to acquire multi-source data of tea through intelligent sensory devices such as electronic nose and electronic tongue, construct a sensory quality data quantification model, perform data standardization and deep fusion, remove noise data, and establish a mapping relationship between multi-source data and professional human evaluation.
This approach enables objective and quantifiable assessment of tea quality, improves data quality and fusion effects, and enhances the robustness and practicality of the tea quality assessment model.
Smart Images
Figure CN121787950A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of tea quality assessment technology, and more specifically, to a method for constructing a tea quality assessment dataset based on multi-source cross-modal acquisition. Background Technology
[0002] For a long time, tea quality assessment has relied heavily on the sensory evaluation of professional tea tasters. This method comprehensively evaluates the appearance, liquor color, aroma, taste, and infused leaves of tea through human vision, smell, taste, and touch, but it faces a series of insurmountable technical bottlenecks.
[0003] First, sensory evaluation is highly subjective, making it difficult to achieve objective quantification and stable assessment. Tea tasting results are easily influenced by the individual tea taster's experience, physical condition, psychological state, and environmental factors. This leads to significant fluctuations in the evaluation results of different tea tasters, and even the same tea taster at different times, for the same sample. The consistency, reproducibility, and comparability of the evaluation results are poor. This subjectivity makes it difficult to precisely quantify and define quality standards, hindering the refined control of production processes and the stable and uniform quality of products. Furthermore, it makes it difficult to effectively digitize and transfer valuable expert experience on a large scale.
[0004] To overcome the limitations of manual evaluation, existing technologies are trending towards digital and instrumental testing. For example, high-performance liquid chromatography (HPLC) is used to determine the content of specific biochemical components such as tea polyphenols and amino acids, or intelligent sensory devices such as electronic tongues and noses are used to acquire the physicochemical signals of tea. However, these technologies still have significant drawbacks: First, the data dimensions are limited, lacking effective integration of multi-source information. Most studies either focus on only a few biochemical components or analyze instrument signals in isolation, failing to systematically correlate the chemical composition, processing parameters, physical appearance characteristics, and sensory quality attributes of tea across modalities. Such single-source studies cannot construct a complete evaluation of the overall quality of tea and are insufficient to reflect the complex interactions between various quality elements. Second, data quality and consistency challenges are prominent. During multi-source data acquisition, outliers inevitably exist in the dataset due to equipment precision, operational errors, or environmental interference. Existing technologies often employ simple statistical elimination or mean imputation methods, failing to address the nonlinear characteristics of tea data with refined processing, leading to reduced data reliability and consequently affecting the accuracy of subsequent analytical models. Third, cross-modal data fusion technology is weak: data from different sources (such as concentration values of chemical components, resistance signals of electronic noses, and image pixel values) have differences in dimensions and magnitudes, and direct fusion will lead to bias. Although some studies have attempted to standardize data, they have failed to combine multi-source heterogeneous data with professional human evaluation to achieve an accurate mapping model of comprehensive quality level, thus limiting the predictive accuracy and practical application value of digital assessment models.
[0005] In summary, existing tea quality assessment technologies are either limited by human subjectivity or suffer from narrow data dimensions, crude processing, and superficial integration, failing to meet the high requirements for digital and intelligent tea quality control. Therefore, there is an urgent need in this field for a comprehensive method that can systematically collect multi-source data and effectively standardize and deeply integrate it, laying a solid and reliable data foundation for building a high-precision and robust digital tea quality assessment model. Summary of the Invention
[0006] To achieve the above objectives, this application provides a method for constructing a tea quality assessment dataset based on multi-source cross-modal acquisition, comprising the following steps: Determine the data categories for evaluating tea quality, including sensory quality data; The raw data corresponding to the data collection categories; among them, the raw data of sensory quality data includes the maximum resistivity data vector of aroma intensity, the maximum resistivity data vector of flavor thickness, and the maximum resistivity data vector of freshness. Load the sensory quality data quantification model. The sensory quality data quantification model is used to obtain the relationship between the original data of tea samples and the sample scores, and generate the original dataset for tea quality assessment. The original tea quality assessment dataset was optimized to generate a new tea quality assessment dataset. Before loading the sensory quality data quantification model, the sensory quality data quantification model is constructed, including: A standardized matrix Z for sensory quality is generated based on the raw data, including: quantifying the raw data to obtain a standardized matrix for aroma intensity. Standardized matrix of flavor thickness Standardized matrix of freshness ; A sensory quality data quantification model is generated based on the sensory quality standardization matrix Z.
[0007] Among them, the standardized matrix for obtaining aroma intensity. Includes the following steps; Set the aroma intensity response matrix , representing the aroma intensity response value of the maximum resistance data vector A across k sensors; A professional sensory evaluation team was set up to complete the evaluation. The sample scores form the aroma intensity evaluation matrix corresponding to the original data. ; Z-score normalization data preprocessing method is used for data preprocessing. Matrix and Matrix data quantization to generate response matrix and evaluation matrix The response matrix and evaluation matrix Constructing a standardized matrix .
[0008] A standardized matrix for obtaining aroma intensity was used. Using the same method, obtain the standardized matrix of flavor thickness. Standardized matrix of freshness The normalized matrix of the aroma intensity Standardized matrix of flavor thickness Standardized matrix of freshness This forms a standardized matrix Z.
[0009] Among them, the aroma intensity response matrix Represented as: ,in, Let be the aroma intensity response value of the i-th tea sample. A represents the aroma intensity response values of K sensors in a single sample; Aroma Intensity Evaluation Matrix Represented as: .
[0010] Generate a normalized matrix Includes the following steps: Aroma intensity response matrix Calculate the first Normalized values of aroma intensity response of each sensor The same method was used to calculate the standardized value of the flavor thickness response. Freshness response standardized value ; Aroma intensity evaluation matrix Calculate the first Standardized value of aroma intensity for each evaluator Using the same method, standardized values for flavor thickness were obtained. Standardized value of freshness ; Aroma intensity response standardized value Flavor thickness response standardized value and freshness response standardized value Constructing the response matrix Aroma intensity standardized value Standardized values of flavor thickness and freshness standardized value Constructing an evaluation matrix .
[0011] Among them, the aroma intensity response standardized value The calculation method is expressed as follows: ,in, For the first The first under the same sensor Standardized value of aroma intensity of tea leaves For the first Liede Aroma intensity response value of tea leaves For the first Under the same sensor The average aroma intensity of this type of tea For the first Under the same sensor Standard deviation of aroma intensity of tea varieties; Among them, the List the average of all tea sample data The calculation method is as follows: , No. List the standard deviation of all tea sample data The calculation method is as follows: .
[0012] Furthermore, the standardized value of aroma intensity The calculation method is expressed as follows: ,in, For the first The standardized value of the i-th type of tea aroma intensity for the evaluator. For the first Rank Aroma intensity score for tea varieties For the first The ratings The average aroma intensity of this type of tea For the first The appraiser Standard deviation of aroma intensity of tea varieties; Among them, the The ratings Average aroma intensity of tea The calculation method is as follows: ; No. The standard deviation of the scores given by the evaluators for all tea samples The calculation method is as follows: .
[0013] Furthermore, generating a quantitative model for sensory quality data includes the following steps: calculate weight vector According to the weight vector calculate score vector ;calculate weight vector According to the weight vector calculate score vector ; Build and The regression model is expressed as: ,in, For regression coefficients, For residuals; Using the sum of squared residuals Solve the problem; calculate load vector And update the residuals; set up The decomposition and prediction relationships are used to generate a quantitative model of sensory quality data.
[0014] set up The decomposition and prediction relationships include: set up The decomposition relationship is as follows: ,in, , , For residuals, ; set up The predictive relationship is as follows: ,in, , ; By combining the decomposition and prediction relationships, a quantitative model for sensory quality data is generated, which is represented as follows: .
[0015] Optimizing the original dataset for tea quality assessment includes the following steps: Calculate the average value of the sample data in the original dataset for tea quality assessment. with standard deviation ; , , in, For the collection of biochemical composition data, sensory quality data, processing parameter data, and spectral image data of tea samples; The statistic G is calculated as follows: ; Remove sample data from the original dataset for tea quality assessment where G is greater than a specified value.
[0016] According to the present invention, comprehensive fusion of multi-source data can be achieved, accurately reflecting the cross-modal correlation of the chemical composition, processing process, physical appearance and sensory attributes of tea. By establishing a mapping relationship between the detection results of intelligent sensory devices such as electronic nose and electronic tongue and the results of human tea evaluation, the objectivity and quantification of sensory quality assessment can be realized, completely solving the problem that sensory evaluation is greatly affected by individual experience and state. In terms of data processing, noisy data can be effectively identified and removed, improving the stable output of tea in complex environments, enhancing data quality and fusion effect, and effectively enhancing the robustness and practicality of the comprehensive tea data model. Attached Figure Description
[0017] Figure 1 This is a flowchart illustrating the steps of constructing a tea quality assessment dataset according to an embodiment of the present invention. Figure 2 This is a flowchart of the sensory quality data quantification model construction process provided by an embodiment of the present invention. Detailed Implementation
[0018] The specific implementation of the present invention will now be described in detail with reference to the accompanying drawings.
[0019] The method for constructing a tea quality assessment dataset based on multi-source cross-modal acquisition provided by this invention is as follows: Figure 1 As shown, it includes the following steps: Step S100: Determine the data classification for evaluating tea quality; Generally, factors to be considered when evaluating tea quality include the content of elements such as tea polyphenols, taste, aroma, visual appeal, and control of processing techniques.
[0020] In this step, the data categories for evaluating tea quality are determined, including: biochemical component data, including the content of elements such as tea polyphenols, amino acids, catechins, and caffeine; sensory quality data, including aroma intensity, flavor richness, and freshness; process parameter data, including fixation temperature, rolling pressure and time, and drying temperature and time; and spectral image data, including tea color and shape.
[0021] Step S110 collects the raw data corresponding to the data classification; among which, the raw data of sensory quality data includes the maximum resistance data vector of aroma intensity, the maximum resistance data vector of flavor thickness, and the maximum resistance data vector of freshness. Different data collection methods are used for different data categories when collecting raw data.
[0022] For biochemical component data, the content of tea polyphenols was collected using the Folin-Ciocalteu reagent method, the content of amino acids was collected using the ninhydrin solution colorimetric method, and the content of catechins and caffeine was collected using high performance liquid chromatography. For sensory quality data, an electronic nose system was used to collect aroma intensity, and an electronic tongue system was used to collect flavor thickness and freshness. For the process parameter data, the temperature of blanching is collected by the temperature sensor inside the blanching pan, the kneading pressure is collected by the pressure sensor on the kneading machine pressurizing device, the kneading time is obtained by the recorded data of the automatic control system, the drying temperature is obtained by the temperature sensor inside the dryer, and the drying time is obtained by the recorded data of the drying equipment. Spectral image data is acquired by scanning with a hyperspectral camera.
[0023] Because different data categories use different data collection methods, the structure of the raw data for each category differs. For example, biochemical component data are percentage values, while process parameter data are numerical values.
[0024] The data collected from the electronic nose and electronic tongue system are data matrices, namely, the maximum resistance data vector A for aroma intensity, the maximum resistance data vector B for flavor thickness, and the maximum resistance data vector C for freshness.
[0025] Step S120: Load the sensory quality data quantification model. The sensory quality data quantification model is used to obtain the relationship between the original data of tea samples and the sample scores, and generate the original dataset for tea quality assessment. Since the output of the electronic nose and electronic tongue system is a numerical matrix corresponding to resistance values, it cannot directly represent the relative scores of specific indicators. Therefore, sensory quantification of the numerical matrix is required to obtain the relative scores of aroma intensity, flavor body, and freshness of tea. In this step, the resistance values are quantified into specific sensory scores through a sensory quality data quantification model.
[0026] Before loading the sensory quality data quantification model, construct the sensory quality data quantification model; the construction process is as follows: Figure 2 As shown, it includes the following steps: Step S200: Generate a standardized matrix Z of sensory quality based on the raw data; The standardized matrix Z for generating sensory quality refers to: quantizing the original data to obtain a standardized matrix of aroma intensity. Standardized matrix of flavor thickness Standardized matrix of freshness ; Next, a standardized matrix of aroma intensity is obtained. For example, a detailed explanation will be provided: Step S201: Set the aroma intensity response matrix , representing the aroma intensity response value of the maximum resistance data vector A across k sensors; Aroma Intensity Response Matrix Represented as: ,in, Let be the aroma intensity response value of the i-th tea sample. A represents the aroma intensity response values of K sensors in a single sample; In the same way, set the flavor thickness response matrix. Freshness response matrix , representing the response values of the maximum resistance data vector B for flavor thickness and the maximum resistance data vector C for freshness.
[0027] Specifically, the flavor thickness response matrix Represented as: ; Freshness response matrix Represented as: .
[0028] Step S202: Construct the dataset: A professional sensory evaluation team was set up to complete the evaluation. , , The sample scores constitute the aroma intensity evaluation matrix corresponding to the original data. Flavor thickness evaluation matrix And freshness evaluation matrix , respectively represented as: , and .
[0029] , and constitute matrix, , and constitute matrix.
[0030] Step S203: Use the Z-score normalization data preprocessing method to perform... Matrix and Matrix data quantization to generate standardized matrices ; During the raw data acquisition process, due to the different sensitivities and ranges of different sensors, the sampled data have different response value ranges. For example, the response value range of sensor 1 is 0-1000, while that of sensor 2 is 0-1. Therefore, standardization processing is required to eliminate the differences between the sensors.
[0031] The standardization process specifically includes the following steps: 1) Through aroma intensity response matrix Calculate the first Normalized values of aroma intensity response of each sensor ; Aroma intensity response standardized value The calculation method is expressed as follows: ,in, For the first The first under the same sensor Standardized value of aroma intensity of tea leaves For the first Liede Aroma intensity response value of tea leaves For the first Under the same sensor The average aroma intensity of this type of tea For the first Under the same sensor Standard deviation of aroma intensity of tea varieties.
[0032] Among them, the List the average of all tea sample data The calculation method is as follows: ,in, For the first Under the same sensor Average aroma intensity of tea leaves.
[0033] Similarly, the average thickness of the tea flavor can be calculated. Average freshness of tea leaves .
[0034] No. List the standard deviation of all tea sample data The calculation method is as follows: .
[0035] Similarly, the standard deviation of the thickness of tea flavor can be calculated. Standard deviation of tea freshness .
[0036] The same method can be used to obtain the standardized value of the flavor thickness response. Freshness response standardized value .
[0037] 2) Evaluation matrix based on aroma intensity Calculate the first Standardized value of aroma intensity for each evaluator ; Aroma intensity standardized value The calculation method is expressed as follows: ,in, For the first The standardized value of the i-th type of tea aroma intensity for the evaluator. For the first Rank Aroma intensity score for tea varieties For the first The ratings The average aroma intensity of this type of tea For the first The appraiser Standard deviation of aroma intensity of tea varieties; Among them, the The ratings Average aroma intensity of tea The calculation method is as follows: ; Similarly, the average flavor thickness can be calculated. Average freshness of tea leaves ; Among them, the The standard deviation of the scores given by the evaluators for all tea samples The calculation method is as follows: ; Similarly, the standard deviation of the flavor profile of tea can also be calculated. Standard deviation of tea freshness .
[0038] The same method can be used to obtain the standardized value of the taste thickness for the evaluator. Standardized value of freshness .
[0039] 3) Aroma intensity response standardized value Flavor thickness response standardized value and freshness response standardized value Constructing the response matrix Aroma intensity standardized value Standardized values of flavor thickness and freshness standardized value Constructing an evaluation matrix Response matrix and evaluation matrix Constructing a standardized matrix .
[0040] Furthermore, this can be achieved through a matrix. and evaluation matrix To find the relationship between the original data of tea samples and the sample scores.
[0041] Step S210: Generate a sensory quality data quantification model based on the sensory quality standardization matrix Z; The sensory quality data quantification model is constructed by training a partial least squares regression model, including the following steps: Step S211: Calculation weight vector According to the weight vector calculate score vector ;calculate weight vector According to the weight vector calculate score vector ; Through the calculations in this step, the original sensor signal can be weighted using a weight vector. Mixed into a new comprehensive indicator and make It can maximize the expression The changing trend.
[0042] The specific calculation process includes: 1) Settings and covariance matrix , represented as: ; For covariance matrix Perform singular value decomposition, represented as: ,in, It is a left singular vector. It is a right singular vector. , It is a singular value; To make the covariance matrix If the projection variance is maximized in a certain direction, then the weight vector is taken. equal The left singular vector corresponding to the maximum singular value in the SVD decomposition ,Right now ,in, This represents the number of components extracted by partial least squares. 2) Calculation score vector , represented as: ,in, yes The sample at the th Projection onto each component; 3) Calculation weight vector : according to score vector Construct a regression equation, expressed as: Then, the univariate regression coefficients are solved using the least squares method, and expressed as: , , ; And based on this, calculate score vector , represented as: ,in, yes The sample at the th Projections onto each component.
[0043] Step S212: Construction and The regression model is expressed as: ,in, For regression coefficients, This is the residual.
[0044] Step S213: Calculate the RSS solution coefficients; In this step, the residual sum of squares (RSS) is used to... The solution is as follows: , right Taking the derivative and setting it to 0, we can solve for: , .
[0045] Step S214: Calculation Cut-off vector, update residual; In order to constitute The contribution weight of each raw sensor data is defined during the process, and is calculated in this step. load vector And update the residuals; Load vector The calculation method is as follows: ; The method for updating residuals is as follows: , , in, For load vector, For updates The residual, i.e. China has not yet been Explanation of information, Then it has been a component The explanation Information, For updates The residual, i.e. China has not yet been The predicted information, Then it is based on ,right Best prediction .
[0046] Step S215: Setting Based on the decomposition and prediction relationships, a quantitative model for sensory quality data is generated: First, set The decomposition relationship is as follows: ,in, , , For residuals, ; set up The predictive relationship is as follows: ,in, , ; By combining the decomposition and prediction relationships, a quantitative model for sensory quality data is generated, which is represented as follows: .
[0047] In practical applications, biochemical component data, process parameter data, and spectral image data can all be processed in the same way as sensory quality data to generate original datasets for tea quality assessment corresponding to various data categories.
[0048] This invention provides specific embodiments, showing that the raw data of aroma intensity of six tea varieties obtained through an electronic nose system are shown in Table 1, and the expert scoring results are shown in Table 2: Table 1. Raw data on aroma intensity collected by electronic nose from multiple tea varieties.
[0049] Table 2. Expert scoring data on the aroma intensity of various tea varieties.
[0050] The original dataset for tea quality assessment generated by the sensory quality data quantification model is shown in Table 3. Table 3. Data on the mapping between expert scores for aroma intensity of various tea varieties and electronic nose sensor.
[0051] S130 optimizes the original tea quality assessment dataset to generate a tea quality assessment dataset.
[0052] Generally, data from different sources have differences in units and magnitudes, and direct fusion can lead to bias. Therefore, this step involves outlier handling of the original tea quality assessment dataset, including the following steps: 1) Calculate the mean and standard deviation of the sample data in the original dataset for tea quality assessment; , , in, For the collection of biochemical composition data, sensory quality data, processing parameter data, and spectral image data of tea samples; 2) Calculate the Grubbs statistic G. The calculation method is as follows: ; In the formula, the larger the value of G, the more likely the point is to be an outlier; 3) Remove sample data from the original dataset for tea quality assessment where G is greater than the specified value.
[0053] Select significance level =0.05 and sample size n, find the corresponding value from the Grubbs test critical value table. Perform conditional checks to remove outliers, as follows: .
[0054] The tea quality assessment dataset construction method provided by this invention enables comprehensive fusion of multi-source data. Through a complete assessment of the overall quality of tea, it can more accurately reflect the cross-modal correlation of tea's chemical composition, processing, physical appearance, and sensory attributes. By establishing a mapping relationship between the detection results of intelligent sensory devices such as electronic noses and electronic tongues and the results of human tea tasting, the objectivity and quantifiability of sensory quality assessment are achieved, completely solving the problem of sensory evaluation being greatly influenced by individual experience and state. Simultaneously, this invention employs a data processing flow combining Z-score standardization and Grubbs' test, effectively identifying and eliminating noisy data, improving the stable output of tea data in complex environments, enhancing data quality and fusion effects, and effectively strengthening the robustness and practicality of the comprehensive tea data model.
[0055] The above-disclosed embodiments are merely a few specific examples of the present invention. However, the present invention is not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.
Claims
1. A method for constructing a tea quality assessment dataset based on multi-source cross-modal acquisition, characterized in that, Includes the following steps: Determine the data categories for assessing tea quality, including sensory quality data; The raw data corresponding to the data collection categories; among them, the raw data of sensory quality data includes the maximum resistivity data vector of aroma intensity, the maximum resistivity data vector of flavor thickness, and the maximum resistivity data vector of freshness. Load the sensory quality data quantification model, which is used to obtain the relationship between the original data of tea samples and the sample scores, and generate the original dataset for tea quality evaluation. The original tea quality assessment dataset is optimized to generate a new tea quality assessment dataset. Before loading the sensory quality data quantification model, the sensory quality data quantification model is constructed, including: Generating a standardized matrix Z of sensory quality based on raw data includes: quantizing the raw data to obtain a standardized matrix of aroma intensity. Standardized matrix of flavor thickness Standardized matrix of freshness ; A sensory quality data quantification model is generated based on the sensory quality standardization matrix Z.
2. The method for constructing a tea quality assessment dataset according to claim 1, characterized in that, The standardized matrix for obtaining aroma intensity Includes the following steps; Set the aroma intensity response matrix , representing the aroma intensity response value of the maximum resistance data vector A across k sensors; A professional sensory evaluation team was set up to complete the evaluation. The sample scores form the aroma intensity evaluation matrix corresponding to the original data. ; Z-score normalization data preprocessing method is used for data preprocessing. Matrix and Matrix data quantization to generate response matrix and evaluation matrix The response matrix and evaluation matrix Constructing a standardized matrix .
3. The method for constructing a tea quality assessment dataset according to claim 1, characterized in that, A standardized matrix for obtaining aroma intensity was used. Using the same method, obtain the standardized matrix of flavor thickness. Standardized matrix of freshness .
4. The method for constructing a tea quality assessment dataset according to claim 2, characterized in that, The aroma intensity response matrix Represented as: ,in, Let be the aroma intensity response value of the i-th tea sample. A represents the aroma intensity response values of K sensors in a single sample; The aroma intensity evaluation matrix Represented as: .
5. The method for constructing a tea quality assessment dataset according to claim 2, characterized in that, The generated normalization matrix Includes the following steps: Aroma intensity response matrix Calculate the first Normalized values of aroma intensity response of each sensor The same method was used to calculate the standardized value of the flavor thickness response. Freshness response standardized value ; Aroma intensity evaluation matrix Calculate the first Standardized value of aroma intensity for each evaluator Using the same method, standardized values for flavor thickness were obtained. Standardized value of freshness ; Aroma intensity response standardized value Flavor thickness response standardized value and freshness response standardized value Constructing the response matrix Aroma intensity standardized value Standardized values of flavor thickness and freshness standardized value Constructing an evaluation matrix .
6. The method for constructing a tea quality assessment dataset according to claim 5, characterized in that, The aroma intensity response normalized value The calculation method is expressed as follows: ,in, For the first The first under the same sensor Standardized value of aroma intensity of tea leaves For the first Liede Aroma intensity response value of tea leaves For the first Under the same sensor The average aroma intensity of this type of tea For the first Under the same sensor Standard deviation of aroma intensity of tea varieties; Among them, the List the average of all tea sample data The calculation method is as follows: ， No. List the standard deviation of all tea sample data The calculation method is as follows: 。 7. The method for constructing a tea quality assessment dataset according to claim 5, characterized in that, The aroma intensity standardized value The calculation method is expressed as follows: ,in, For the first The standardized value of the i-th type of tea aroma intensity for the evaluator. For the first Rank Aroma intensity score for tea varieties For the first The ratings The average aroma intensity of this type of tea For the first The appraiser Standard deviation of aroma intensity of tea varieties; Among them, the The ratings Average aroma intensity of tea The calculation method is as follows: ； No. The standard deviation of the scores given by the evaluators for all tea samples The calculation method is as follows: 。 8. The method for constructing a tea quality assessment dataset according to claim 2, characterized in that, The generation of the sensory quality data quantification model includes the following steps: calculate weight vector According to the weight vector calculate score vector ;calculate weight vector According to the weight vector calculate score vector ; Build and The regression model is expressed as: ,in, For regression coefficients, For residuals; Using the sum of squared residuals Solve the problem; calculate load vector And update the residuals; set up The decomposition and prediction relationships are used to generate a quantitative model of sensory quality data.
9. The method for constructing a tea quality assessment dataset according to claim 8, characterized in that, The settings The decomposition and prediction relationships include: set up The decomposition relationship is: ,in, , , For residuals, ; set up The predictive relationship is as follows: ,in, , ; By combining the decomposition and prediction relationships, a quantitative model for sensory quality data is generated, which is represented as follows: 。 10. The method for constructing a tea quality assessment dataset according to claim 1, characterized in that, The optimization of the original dataset for tea quality assessment includes the following steps: Calculate the average value of the sample data in the original dataset for tea quality assessment. with standard deviation ; ， ， in, For the collection of biochemical composition data, sensory quality data, processing parameter data, and spectral image data of tea samples; The statistic G is calculated as follows: ; Remove sample data from the original dataset for tea quality assessment where G is greater than a specified value.