A digital evaluation method for the sensory quality of formulated products

Abstract

This invention discloses a digital evaluation method for the sensory quality of formulated products, comprising: establishing a training sample matrix X using measurement data of m formulated products as training samples; establishing an expert sensory evaluation vector Y by combining expert evaluation results of the training samples; and setting the test samples x... m+1 Add the training sample matrix X to form set X (m+1) The number of sensory indicators in the expert sensory evaluation vector Y is used as the number of test samples x. m+1 Assign the same number of sensory indicators; based on set X (m+1) The distance between samples is used to express the distributional differences between samples using a similarity metric. Based on the distributional differences between the test sample and the training sample, the sensory index evaluation value of the training sample with the highest distributional similarity is selected to predict the test sample x. m+1 The sensory quality assessment method of this invention is simple and effective, and the assessment results are not affected by subjective human factors, thus having high reliability.

Description

Technical Field

[0001] This invention belongs to the field of product quality testing technology, specifically, it relates to a method for digitally evaluating the sensory quality of formulated products. Background Technology

[0002] In traditional formulation industries that rely on sensory evaluation, product quality assessment primarily depends on the personal experience of tasters. For example, in industries such as beer, cigarettes, beverages, and food, tasters not only need extensive industry background knowledge and rich tasting experience, but they are also required to be able to capture and evaluate numerous characteristics of the product, such as color, aroma, taste, and foam, in a very short time.

[0003] However, organizing expert evaluations of product quality sensory characteristics is a cumbersome and costly process. In particular, expert-based evaluations are heavily influenced by subjective preferences, physiological differences, physical and psychological states, and can also harm the health of the experts. Furthermore, this method is unsuitable for sensory evaluation of large batches of samples. Therefore, designing a more objective, rapid, and digital sensory quality evaluation method has become an urgent need in the formulation industry.

[0004] In the current digital formulation design industry, which features sensory evaluation, researchers typically use product testing information on chemical components and sensory evaluation indicators to build digital models. For example, in the beer industry, companies first test the chemical components of beer samples, such as carbohydrates, nitrogenous substances, minerals, and trace elements, and then establish a relationship with sensory indicators such as color, aroma, taste, and foam to determine a mathematical model. However, in reality, due to limitations in testing conditions, speed, and cost, companies often only provide a few conventional testing indicators. The mathematical model built using these limited conventional indicators and sensory indicators has low accuracy and therefore cannot achieve the effect of expert evaluation. Summary of the Invention

[0005] The purpose of this invention is to provide a digital evaluation method for the sensory quality of formulated products, in order to solve the problems that traditional expert evaluation methods are heavily influenced by subjective factors and that existing digital evaluation methods have low accuracy.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0007] A digital evaluation method for the sensory quality of a formulated product includes:

[0008] Using measurement data from m formulation products as training samples, a training sample matrix X is established;

[0009] Based on the expert evaluation results, an evaluation value is assigned to the sensory index of each training sample in the training sample matrix X, and an expert sensory evaluation vector Y is established.

[0010] On test sample x m+1 When conducting sensory evaluation, the test sample x m+1 Add the training sample matrix X to form set X (m+1) ;

[0011] The number of sensory indicators in the expert sensory evaluation vector Y is the number of test samples x. m+1 Allocate the same number of sensory indicators;

[0012] According to set X (m+1) The distance between samples is used to express the distributional differences between samples using a similarity metric.

[0013] Based on the distribution differences between the test samples and the training samples, the sensory index evaluation value of the training sample with the highest distribution similarity is selected to predict the test sample x. m+1 Sensory evaluation values.

[0014] In some embodiments of this application, the process of expressing the distributional differences between samples using a similarity measurement method may include:

[0015] A training sample sequence Z is constructed using the training sample matrix X and the expert sensory evaluation vector Y. (m) :

[0016] Z (m) = (z1, z2, ..., z m );z i ={(x i ,y1),(x i ,y2),...(x i y k ), ..., (x i y s )};i=(1,2,...,m);

[0017] Where, x i Represents the i-th training sample; (x i y k y in ) k Let x represent the i-th training sample. i The k-th sensory indicator; s is the total number of sensory indicators;

[0018] Calculate the training sample sequence Z (m) Each sample Z i Similarity metric;

[0019] Build test sample x m+1 Test sample z m+1 :

[0020] z m+1 ={(x m+1 ,y1),(x m+1 ,y2),...,(x m+1 y s )};

[0021] Test sample z m+1 Add training sample sequence Z (m) The test sample sequence Z is formed. (m+1) :

[0022] Z (m+1) = (z1, z2, ..., z m , z m+1 );

[0023] Calculate the test sample sequence Z (m+1) In the middle, the test sample Z m+1 Similarity measure between the training samples and the training samples.

[0024] In some embodiments of this application, the calculation of the training sample sequence Z (m) Each sample Z i The process of calculating similarity metrics may include:

[0025] From the training sample sequence Z (m) In the selection of sensory index y k = The training samples of u form a sequence of samples of the same category, and the sensory index y is selected. k Training samples ≠ u form sequences of samples of different categories; where u is the sensory index y. k The evaluation value;

[0026] Combining the same sensory index y k For each training sample, a similarity metric is calculated using sequences of samples from the same category and sequences of samples from different categories. The calculation method is as follows:

[0027]

[0028] in, Indicates training sample x i In its k-th sensory index y k The evaluation value is the similarity measure between y and other training samples when u is the value; k = 1, 2, ..., s; u = 1, 2, ..., h; v is the similarity measure between y and other training samples. k = the total number of training samples of u, and v < m; in the formula, the numerator represents the number of training samples of u, and v < m; ic training samples of the same class with the shortest distance to z i The sum of distances; the denominator represents the sum of distances to z. i c training samples of different classes with the shortest distance to z i The sum of distances; c is an integer, which can be set by the user.

[0029] In some embodiments of this application, the calculation of the test sample sequence Z (m+1) In the middle, the test sample Z m+1 The process of measuring similarity between training samples can include:

[0030] The test samples x are respectively m+1 Each sensory indicator is assigned a different evaluation value;

[0031] Calculate the test sample x respectively m+1 When each sensory index is equal to a different evaluation value, test sample z m+1 The similarity metric between the sample and the training sample is calculated as follows:

[0032]

[0033] in, Indicates test sample x m+1 In its k-th sensory index y k The evaluation value is a similarity measure between the sample and the training sample when u is used; in the formula, the molecule represents the similarity between z and the training sample. m+1 c training samples of the same class with the shortest distance to z m+1 The sum of distances; the denominator represents the sum of distances to z. m+1 c training samples of different classes with the shortest distance to z m+1 The sum of distances.

[0034] In some embodiments of this application, the test sample x m+1 Methods for predicting sensory index evaluation values ​​may include:

[0035] Calculate the test sample z m+1 ={(x m+1 ,y1),(x m+1 ,y2),...,(x m+1 y s Randomness test Q for each combination in )} k value:

[0036]

[0037] Wherein, the numerator represents the number of elements in the set that satisfy the requirements;

[0038] Select Q kThe evaluation value u corresponding to the maximum value among the values ​​is used as the k-th sensory index y of the test sample. k The evaluation value is expressed by the formula:

[0039] CL K =MAX(Q) k );

[0040] Among them, CL K That is, the test sample x m+1 The kth sensory index y k The evaluation value.

[0041] In some embodiments of this application, when forming the set X (m+1) At that time, first for set X (m+1) Preprocessing is performed to reduce interference; then, principal component analysis is used to perform dimensionality reduction mapping on the preprocessed matrix to reduce the matrix space dimension, generating the dimensionality-reduced principal component score matrix B. (m+1) :

[0042]

[0043] Where n is the number of measurement data for each sample; p is the number of principal components, and p < n; B i W is the score vector; W is the loading matrix, representing the degree of correlation between the principal components and their corresponding original variables, calculated using the following formula:

[0044]

[0045] Where, λ i Let a be the eigenvalue of the i-th sample. ij Let be the j-th feature vector of the i-th sample.

[0046] Using the dimensionality-reduced principal component score matrix B (m+1) Calculate the set X (m+1) The distance between samples can solve the problem that it is difficult to effectively predict their sensory quality using general methods.

[0047] In some embodiments of this application, the Mahalanobis distance method is preferably used to calculate the set X. (m+1) The distance between samples.

[0048] In some embodiments of this application, in order to obtain rich structural and compositional information of the formulated product, it is preferable to use a near-infrared spectrometer, mass spectrometer, or chromatogram to acquire spectral data, mass spectrometry data, or chromatographic data of training samples and test samples to form sample measurement data.

[0049] In some embodiments of this application, the set X(m+1) Preprocessing may include first derivative, Norris smoothing filter, and PCA principal component preprocessing.

[0050] In some embodiments of this application, the process of establishing the expert sensory evaluation vector Y may include:

[0051] Experts were organized to evaluate m training samples, and an expert sensory evaluation vector Y was established for each training sample.

[0052]

[0053] Assign evaluation values ​​1, 2, ..., h to each sensory index of each training sample in a grade or level manner.

[0054] Compared with the prior art, the advantages and positive effects of the present invention are as follows: The present invention utilizes the intrinsic relationship between the test data of the formulated product and its sensory quality to establish a digital evaluation model; it uses the distance between samples and adopts a similarity measurement method to express the distribution differences between samples; based on the distribution differences between the test samples and the training samples, it selects the sensory index evaluation value of the training sample with the highest distribution similarity to predict the sensory index evaluation value of the test samples. The method is simple and effective, the evaluation results are not affected by human subjective factors, and the reliability is high.

[0055] Other features and advantages of the present invention will become clearer after reading the detailed description of the embodiments of the present invention in conjunction with the accompanying drawings. Attached Figure Description

[0056] Figure 1 This is a flowchart illustrating an embodiment of the digital evaluation method for the sensory quality of formulated products proposed in this invention. Detailed Implementation

[0057] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0058] Formulated products with sensory evaluation characteristics mostly use agricultural and sideline products as their main raw materials rather than standard industrial products. Therefore, their intrinsic attribute information is difficult to describe clearly digitally. Similarly, for formulated products, expert sensory evaluation is difficult to quantify, and therefore, it is impossible to describe all sensory characteristics one by one.

[0059] This embodiment utilizes the inherent relationship between raw material / product attribute information and their sensory quality, and employs a scientific solution method to establish a mapping relationship, thereby realizing the mapping from the input space to the output space:

[0060]

[0061] This is to obtain a digital assessment of the sensory quality of the product.

[0062] Because near-infrared spectroscopy contains abundant structural and compositional information, it is highly suitable for representing the compositional properties of hydrocarbon organic substances. Furthermore, the sample spectra contain thousands of samples, and 80%–90% of the useful information in a substance sample can be expressed in near-infrared spectroscopy. Therefore, this embodiment abandons the traditional method of establishing a digital model between product chemical composition information and sensory indicators. Instead, it fully utilizes the rich feature information in the sample's near-infrared spectrum to establish a digital expression relationship between the product's physicochemical characteristics and sensory indicators. That is, a more accurate relationship model is constructed between the two, and the reliability of the model's prediction results is evaluated, providing experts with a more comprehensive reference.

[0063] Of course, for companies whose sample testing data is high-dimensional testing data such as mass spectrometry / chromatography, the modeling method and evaluation technology of this embodiment can also be used to obtain effective sensory evaluation results.

[0064] When the input space dimension is high, such as when the raw material detection data is high-dimensional data with thousands of dimensions such as mass spectrometry / chromatography / near-infrared spectroscopy, it is difficult to effectively predict its sensory quality using general methods. Therefore, this embodiment uses principal component analysis to perform dimensionality reduction space mapping for high-dimensional detection data in order to achieve digital evaluation of the sensory quality of the formulated product.

[0065] The following is combined Figure 1 The specific design process of the digital evaluation method for the sensory quality of the formulated products in this embodiment is described in detail.

[0066] S101. Construct the training sample matrix X.

[0067] In this embodiment, m product samples can be extracted from the sample library as training samples. Spectral, mass spectrometric, or chromatographic data of these m training samples are acquired using a near-infrared spectroscopy analyzer, mass spectrometer, or chromatogram, collectively referred to as measurement data. This measurement data is typically high-dimensional and can number in the thousands.

[0068] Next, the acquired measurement data is preprocessed, such as by performing first derivative, Norris smoothing filtering, and PCA principal component preprocessing, to form the training sample matrix X:

[0069]

[0070] Where n is the number of measurement data points for each training sample, i.e., the dimension of spectral points, mass spectrometry points, or chromatographic points, etc., and the value of n is usually around 2000 to 3000. The training sample matrix X formed in this way is a high-dimensional space matrix of m×n.

[0071] S102. Assign evaluation values ​​to the sensory indicators of each training sample in the training sample matrix X, and establish an expert sensory evaluation vector Y.

[0072] Experts were organized to evaluate m training samples, and an expert sensory evaluation vector Y was established for each training sample.

[0073]

[0074] Among them, y k Let represent the k-th sensory indicator, and s be the total number of sensory indicators.

[0075] Taking tobacco leaves as a training sample as an example, the sensory indicators of tobacco leaves typically include five, i.e., s=5, namely aroma, harmony, off-flavors, irritation, and aftertaste. Experts assign x to each training sample. i Five sensory indicators are assigned evaluation values ​​to form the expert sensory evaluation vector Y.

[0076] In this embodiment, the evaluation value of each sensory indicator can be set to 1, 2, ..., h, that is, the evaluation value of the sensory indicator is defined in the form of a level or grade.

[0077] S103. Construct a training sample sequence Z using the training sample matrix X and the expert sensory evaluation vector Y. (m) :

[0078] Z (m) = (z1, z2, ..., z m );z i ={(x i ,y1),(x i ,y2),...(x i y k ), ..., (x i y s )};i=(1,2,...,m);

[0079] Where, x i Represents the i-th training sample; (x i y k y in ) k Let x represent the i-th training sample. i The kth sensory index.

[0080] S104, When there is a test sample x m+1 When inputting, the test sample x m+1 Add the training sample matrix X to form set X (m+1) :

[0081] X (m+1)= (x1, x2, ..., x m x m+1 ).

[0082] S105, Construct test sample x m+1 Test sample z m+1 :

[0083] z m+1 ={(x m+1 ,y1),(x m+1 ,y2),...,(x m+1 y s )}.

[0084] This example uses test sample x. m+1 S sensory indicators are also assigned. As for the evaluation value of each sensory indicator, the sensory indicator evaluation value of the training sample with the highest distribution similarity can be selected to predict the sensory indicator evaluation value of the test sample based on the distribution difference between the test sample and the training sample.

[0085] S106, Test sample z m+1 Add training sample sequence Z (m) The test sample sequence Z is formed. (m+1) :

[0086] Z (m+1 ) = (z1, z2, ..., z m , z m+1 ).

[0087] S107, Regarding set X (m+1 The matrix is ​​preprocessed, and then principal component analysis is used to perform dimensionality reduction mapping on the preprocessed matrix to generate the dimensionality-reduced principal component score matrix B. (m+1) :

[0088]

[0089] Where p is the number of principal components, and p < n, p is usually less than 100, and in this embodiment p can be configured as 10; B i Let W be the score vector; W is the loading matrix, representing the degree of correlation between the principal components and their corresponding original variables, which can be calculated using the following formula:

[0090]

[0091] Where, λ i Let a be the eigenvalue of the i-th sample. ij Let be the j-th feature vector of the i-th sample.

[0092] S108. Based on the principal component score matrix B (m+1)Calculate the distance between samples.

[0093] In this embodiment, the Mahalanobis distance method can be used to calculate the distance between samples, and its calculation formula is as follows:

[0094]

[0095] Where, d ij Let represent the distance between the i-th sample and the j-th sample, where i, j = (1, 2, ..., m+1). This allows us to form the distance matrix D:

[0096]

[0097] S109. Based on the distance between samples, use a similarity metric to express the test sample sequence Z. (m+1) The distributional differences among the samples.

[0098] First, calculate the training sample sequence Z. (m) Each sample Z i The similarity metric is calculated using the following method:

[0099] From the training sample sequence Z (m) In the selection of sensory index y k = The training samples of u form a sequence of samples of the same category, and the sensory index y is selected. k Training samples ≠u form different categories of sample sequences.

[0100] For example, we can first start from the training sample sequence Z (m) In this process, training samples with the first sensory indicator y1 = 1 are selected to form a sequence of samples of the same class; training samples with the first sensory indicator y1 ≠ 1 are selected to form a sequence of samples of a different class. Similarly, training samples with the first sensory indicator y1 = 2 are selected to form a sequence of samples of the same class; training samples with the first sensory indicator y1 ≠ 2 are selected to form a sequence of samples of a different class. This process continues until training samples with the first sensory indicator y1 = h are selected to form a sequence of samples of the same class, and training samples with the first sensory indicator y1 ≠ h are selected to form a sequence of samples of a different class. Then, training samples with the second sensory indicator y2 = 1 are selected to form a sequence of samples of the same class, and training samples with the second sensory indicator y2 ≠ 1 are selected to form a sequence of samples of a different class; and so on. This continues until the s-th sensory indicator y1 = h is selected. s Training samples with a value of 1 form a sequence of samples of the same class. The s-th sensory index y is selected. sTraining samples ≠ 1 constitute a sequence of samples from different categories; ...; select the s-th sensory index y s The training samples of h constitute a sequence of samples of the same class, and the s-th sensory index y is selected. s The training samples ≠h constitute a sequence of samples of different classes.

[0101] Combining the same sensory index y k For each training sample sequence, both those of the same and different categories, calculate the similarity metric. The calculation method is as follows:

[0102]

[0103] in, Indicates training sample x i The similarity measure between the sample and other training samples when the evaluation value of its k-th sensory index is u; k = 1, 2, ..., s; u = 1, 2, ..., h; v is the similarity measure between the sample and other training samples. k = the total number of training samples for u, and v < m, that is, the evaluation value of the k-th sensory index is the total number of training samples for u; c is an integer, which can be set by the user; in the above similarity measure In the formula, the numerator represents z. i c training samples of the same class with the shortest distance to z i The sum of distances (these training samples can be obtained from y) k =Selected from the same category of sample sequences of u), reflecting the degree of similarity between training sample data of the same category; the denominator represents the similarity with z. i c training samples of different classes with the shortest distance to z i The sum of distances (these training samples can be obtained from y) k (≠u) is selected from different categories of sample sequences, reflecting the degree of dispersion among training samples of different categories.

[0104] Secondly, calculate the test sample sequence Z. (m+1) The middle test sample Z m+1 The similarity metric is calculated using the following method:

[0105] (1) are the test samples x m+1 Each sensory indicator is assigned a different evaluation value;

[0106] For example, for test sample x m+1 Assign evaluation values ​​of 1, 2, and h to its first sensory index y1; assign evaluation values ​​of 1, 2, and h to its second sensory index y2; and so on, assigning evaluation values ​​of 1, 2, and h to its s-th sensory index y1. s The evaluation values ​​are assigned to 1, 2, and h, respectively.

[0107] (2) Calculate the test sample x respectively m+1 When each sensory index is equal to a different evaluation value, test sample z m+1 The similarity metric between the sample and the training sample is calculated as follows:

[0108]

[0109] in, Indicates test sample x m+1 The similarity measure between the sample and the training sample when the evaluation value of its k-th sensory metric is u. In the above similarity measure... In the formula, the numerator represents z. m+1 c training samples of the same class with the shortest distance to z m+1 The sum of distances reflects the similarity between training and test samples of the same class; the denominator represents the sum of distances between z and z. m+1 c training samples of different classes with the shortest distance to z m+1 The sum of the distances reflects the degree of dispersion between training samples and test samples of different categories.

[0110] S110, Calculate the test sample z m+1 ={(x m+1 ,y1),(x m+1 ,y2),...,(x m+1 y s Randomness test Q for each combination in )} k value.

[0111] Q required for randomness test k A value function can be defined as:

[0112]

[0113] Where the numerator represents the number of elements in the set that satisfy the requirement. That is, the above formula represents the sensory index y. k When =u, the similarity measure between training samples is greater than or equal to the ratio of the number of similarity measures between the test sample and the training sample to v+1.

[0114] Thus, for test sample x m+1 In this context, each sensory index corresponds to h ratios; that is, Q1 includes h ratios; Q2 includes h ratios; ...; Q s It also includes h ratios.

[0115] Q k The magnitude of the value reflects the result of testing the sample z. m+1 The constructed test sample sequence Z (m+1)The probability that the sample z conforms to an independent and identically distributed distribution also reflects the test sample z. m+1 Belongs to y k = The probability of u. Q k The smaller the value, the greater the difference in distribution between the test samples and the training samples; Q k The larger the value, the higher the similarity in distribution between the test sample and the training sample. From another perspective, it can be determined whether the test sample belongs to y. k = probability of u.

[0116] S111, according to Q k The evaluation value of each sensory indicator of the test sample is determined using a credibility evaluation method.

[0117] In this embodiment, Q can be selected. k The evaluation value u corresponding to the largest ratio among the values ​​is used as the k-th sensory index y of the test sample. k The evaluation value can be expressed by the formula as follows:

[0118] CL K =MAX(Q) k ), k = 1, 2, ..., s;

[0119] Among them, CL K That is, the test sample x m+1 The kth sensory index y k The evaluation value.

[0120] That is, the sensory index evaluation value of the training sample with the highest distribution similarity is selected to predict the sensory index evaluation value of the test sample.

[0121] This embodiment employs digital sensory evaluation based on product analysis spectral / mass spectrometry / chromatographic information, which is more accurate, more stable, and less costly than traditional expert experience evaluation or digital sensory evaluation based on the detection of a few key chemical components.

[0122] Of course, the above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A digital evaluation method for the sensory quality of a formulated product, characterized in that, include: Using measurement data from m formulation products as training samples, a training sample matrix X is established; Based on the expert evaluation results, an evaluation value is assigned to the sensory index of each training sample in the training sample matrix X, and an expert sensory evaluation vector Y is established. On the test samples When conducting sensory evaluation, the test samples Add the training sample matrix X to form a set ; The test sample is based on the number of sensory indicators in the expert sensory evaluation vector Y. Allocate the same number of sensory indicators; According to the set The distance between samples is expressed using similarity measures to represent the distributional differences between samples, including: --Construct a training sample sequence using the training sample matrix X and the expert sensory evaluation vector Y. : ； ; ; in, This represents the i-th training sample; In Represents the i-th training sample The k-th sensory indicator; s is the total number of sensory indicators; --From training sample sequences In the selection of sensory indicators The training samples form a sequence of samples of the same category, and sensory indicators are selected. The training samples form different class sample sequences; among them, Sensory indicators The evaluation value; --Combining the same sensory indicators For each training sample, a similarity metric is calculated using sequences of samples from the same category and sequences of samples from different categories. The calculation method is as follows: ； in, Indicates training samples In its k-th sensory index The evaluation value is the similarity measure between the sample and other training samples when the value is u. ; ; for The total number of training samples, and In the formula, the numerator represents and c training samples of the same class with the shortest distance to The sum of distances; the denominator represents the sum of distances to the given value. c training samples of different classes with the shortest distance to The sum of distances; c is an integer, set by the user; --Build test samples Test samples : ； --Test samples Add training sample sequence , constitute the test sample sequence : ； --These are the test samples Each sensory indicator is assigned a different evaluation value; --Calculate the test samples respectively When each sensory index is equal to a different evaluation value, the test sample The similarity metric between the sample and the training sample is calculated as follows: ； in, Indicates test sample In its k-th sensory index The evaluation value is u, which is a measure of similarity between the sample and the training sample; in the formula, the molecule represents and c training samples of the same class with the shortest distance to The sum of distances; the denominator represents the sum of distances to the given value. c training samples of different classes with the shortest distance to The sum of distances; According to the test sample The similarity metric between the test sample and the training sample is used to select the sensory index evaluation value of the training sample with the highest distributional similarity to predict the test sample. Sensory evaluation values.

2. The digital evaluation method for the sensory quality of formulated products according to claim 1, characterized in that, The test sample Methods for predicting sensory index assessment values ​​include: Calculate the test sample Randomness test for each combination value: ； ； Wherein, the numerator represents the number of elements in the set that satisfy the requirements; choose The evaluation value corresponding to the maximum value among the values , as the kth sensory indicator of the test sample The evaluation value is expressed by the formula: ； in, That is, the test sample The kth sensory index The evaluation value.

3. The digital evaluation method for the sensory quality of formulated products according to claim 1 or 2, characterized in that, In forming the set At that time, for the set Preprocessing is performed, and principal component analysis is used to perform dimensionality reduction mapping on the preprocessed matrix to generate the dimensionality-reduced principal component score matrix. : ； Where n is the number of measurement data for each sample; p is the number of principal components, and p <n; W is the score vector; W is the loading matrix, representing the degree of correlation between the principal components and their corresponding original variables, calculated using the following formula: ； in, Let be the eigenvalues ​​of the i-th sample. Let j be the feature vector of the i-th sample; Based on the principal component score matrix Calculate the set The distance between samples.

4. The digital evaluation method for the sensory quality of the formulated product according to claim 3, characterized in that, The set is calculated using the Mahalanobis distance method. The distance between samples.

5. The digital evaluation method for the sensory quality of the formulated product according to claim 3, characterized in that, The measurement data for each sample are spectral data, mass spectrometry data, or chromatographic data obtained using a near-infrared spectrometer, mass spectrometer, or chromatogram.

6. The digital evaluation method for the sensory quality of the formulated product according to claim 3, characterized in that, For the set The preprocessing performed included first derivative, Norris smoothing filter, and PCA principal component preprocessing.

7. The digital evaluation method for the sensory quality of a formulated product according to claim 1 or 2, characterized in that, The process of establishing the expert sensory evaluation vector Y includes: Experts were organized to evaluate m training samples, and an expert sensory evaluation vector Y was established for each training sample. ; Assign evaluation values ​​1, 2, ..., h to each sensory index of each training sample in a grade or level manner.

Images