Discriminating method for processing degree of honey-fried astragalus membranaceus based on sensory technology and component analysis

Abstract

The invention discloses a method for judging the processing degree of honey-fried astragalus membranaceus based on a sensory technology and component analysis. The method comprises the following steps: firstly, collecting multi-source characteristic data of samples with different processing degrees through an electronic tongue, an electronic eye and high performance liquid chromatography; then, according to the RGB color characteristics, the sample is divided into three stages of processing failure, moderation and over-stood; secondly, extracting taste features which change cooperatively with the processing process by adopting local weighted principal component analysis in each stage, and establishing a prediction model of the sensory features and the content of core active components by utilizing partial least squares regression; and finally, fusing the dimension-reduced taste information, the predicted component information and the standardized color information, and accurately judging the processing degree of the to-be-detected sample by calculating the mahalanobis distance. According to the method, objectification, digitization and rapidness of the distinguishing process are achieved, the defects that a traditional experience method is high in subjectivity and poor in reproducibility are effectively overcome, and a reliable technical means is provided for distinguishing the processing quality of the radix astragali preparata with honey.

Description

Technical Field

[0001] This invention belongs to the field of traditional Chinese medicine processing degree discrimination technology, specifically involving a method for judging the processing degree of honey-processed Astragalus membranaceus based on sensory technology and component analysis. Background Technology

[0002] Astragalus (Astragali Radix), a commonly used traditional Chinese medicine, possesses the effects of tonifying qi and raising yang, consolidating the exterior and stopping sweating, promoting diuresis and reducing edema, and generating fluids and nourishing blood. Honey-processing is one of its mainstream processing methods. The sweet and mild nature of honey enhances the qi-tonifying, lung-moistening, spleen-strengthening, and stomach-harmonizing effects of astragalus. Modern research has confirmed that the honey-processing process affects the content and transformation of core active ingredients in astragaloside A, verbascoside, and gentianin, thus directly impacting its efficacy and clinical safety. Therefore, accurate and objective judgment of the degree of honey-processed astragalus (such as the heat level) is crucial to ensuring the uniformity of its processed slices and the stable and controllable efficacy.

[0003] Currently, the industry's judgment of the endpoint of honey-processed astragalus mainly relies on the traditional sensory experience of processing personnel. This involves observing color changes on the surface of the slices (such as from light yellow to bright yellow, dark yellow, or even yellowish-brown), smelling the caramelized aroma of honey, and tasting the balance of sweetness and bitterness. While this experiential method of "seeing, smelling, and tasting" contains valuable practical wisdom, it is highly subjective, has poor reproducibility, and depends heavily on individual experience and sensory states. Differences in judgment can easily arise between different operators, making it difficult to establish unified and quantifiable quality standards.

[0004] In recent years, some modern analytical techniques have been attempted for the quality evaluation of traditional Chinese medicine. For example, high-performance liquid chromatography (HPLC) can accurately determine the content of active ingredients, but this method is time-consuming, costly, and destructive, making it impossible to achieve rapid and non-destructive identification of processed medicinal materials on the production line. Electronic tongue technology, as an artificial taste analysis system, can convert the overall taste information of a sample (such as sour, sweet, bitter, salty, umami, etc.) into objective and quantifiable sensor response signals, showing advantages in the flavor evaluation of food and medicine. Electronic eyes (or high-resolution image analysis technology) can objectively and accurately quantify the color characteristics of samples, avoiding the subjective bias of human judgment. However, existing technologies often use these methods in isolation or only perform simple data listing and comparison, failing to deeply explore the profound relationship between "sensory attributes (color, taste)" and "intrinsic material basis (chemical composition)." Specifically, existing technologies still have the following shortcomings: 1. Single-source judgment criteria: Judging based solely on color or a few components ignores the fact that the processing of traditional Chinese medicine is a process of overall change in "nature, taste, and efficacy", and lacks a comprehensive evaluation system that integrates multi-dimensional information.

[0005] 2. Poor model universality: Processing is a continuous dynamic process. The physicochemical properties of samples change differently at different processing stages (such as early, middle and late stages). Existing methods that use a single global mathematical model (such as a single regression or classification model) to process data of the entire processing process are difficult to accurately characterize this nonlinear and stage-specific change characteristics, resulting in limited discrimination accuracy.

[0006] 3. Disconnected from empirical logic: The judgment logic of existing analytical methods is difficult for technicians in the traditional Chinese medicine processing field to understand and trust, and cannot effectively connect with and inherit the essence of traditional experience of "judging by color and distinguishing by taste". Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method for judging the degree of processing of honey-processed Astragalus membranaceus based on sensory technology and component analysis. Combining electronic tongue, electronic eye, and high-performance liquid chromatography, a three-in-one correlation analysis system of "color-taste-component" is constructed. First, RGB color space clustering analysis is used to objectively divide the continuously changing processing process into several processing stages with similar appearance characteristics. Then, within each stage, improved weighted principal component analysis (PCA) is used to extract taste features that change in tandem with the processing process, and partial least squares regression (PLS) is used to establish a predictive correlation between sensory features and core component content within that stage. Finally, the dimensionality-reduced taste information, predicted component information, and standardized color information are fused to construct a comprehensive feature vector, and a discrimination method based on Mahalanobis distance is used to achieve refined and high-precision discrimination of different degrees of processing of honey-processed Astragalus membranaceus.

[0008] A method for determining the degree of processing of honey-processed Astragalus membranaceus based on sensory technology and component analysis includes the following steps: Step S1: Multi-source feature data acquisition; preparation Different grades of honey-processed Astragalus membranaceus were pulverized to obtain... A set of honey-processed Astragalus membranaceus standard samples were analyzed by electronic tongue, electronic eye, and high-performance liquid chromatography (HPLC). The taste, surface appearance, and content of core active ingredients of honey-processed astragalus were analyzed in a group of standard samples to obtain results for honey-processed astragalus at different processing degrees. Taste response value data RGB three-channel color values ​​and core active ingredient content data; Preferably, the electronic tongue is a TS-5000Z type electronic tongue, the electronic eye is a Canon M50 type high-resolution camera, and the high-performance liquid chromatography is an e2695-2424 type HPLC; The electronic tongue is equipped with five specific taste sensors for sour, salty, bitter, umami, and astringent tastes, as well as three reference electrodes (Ag / AgCl) to measure the initial taste signal values ​​(sour, bitter, astringent, umami, and salty) and the three aftertaste signal values ​​(aftertaste bitterness, aftertaste astringency, and richness) of the five basic tastes.

[0009] The electronic eye shooting conditions are as follows: using a standard white background, adjusting the vertical distance between the camera and the sample, using the same light intensity and shooting parameters, and avoiding interference from factors such as ambient light and shooting angle on the color data.

[0010] The standardized processing of the sample images was performed using ImageJ image analysis software: First, image preprocessing was performed, and background areas and sample edge impurities were removed by cropping tools. A denoising algorithm was used to reduce image noise interference. Then, a uniform area on the sample surface was selected as the region of interest (ROI). The built-in color analysis function of the software was used to accurately extract the RGB three-channel color values ​​(R value, G value, B value) of each ROI and calculate the average value of each channel. The preparation method of the test samples of verbascoside and mangiferin for high performance liquid chromatography (HPLC) analysis is as follows: Take about 1g of honey-processed Astragalus membranaceus sample powder (passed through a No. 4 sieve), accurately weigh it, place it in a stoppered conical flask, accurately add 50ml of 80% methanol solution containing 4% concentrated ammonia test solution (take 4ml of concentrated ammonia test solution, add 80% methanol to 100ml, shake well), stopper tightly, weigh, heat under reflux for 1 hour, cool, weigh again, make up the lost weight with 80% methanol solution containing 4% concentrated ammonia test solution, shake well, filter, accurately measure 25ml of the subsequent filtrate, evaporate to dryness, dissolve the residue in 80% methanol, transfer to a 5ml volumetric flask, add 80% methanol to the mark, shake well, filter, and take the subsequent filtrate to obtain the product.

[0011] The preparation method of the astragaloside A test sample for high performance liquid chromatography (HPLC) analysis is as follows: Take about 1g of honey-processed astragalus sample powder (passed through a No. 4 sieve), accurately weigh it, place it in a round-bottom flask, accurately add 50ml of methanol, weigh it, heat under reflux for 4 hours, cool it, weigh it again, make up the weight loss with methanol, shake well, filter it, accurately measure 25ml of the filtrate, recover the solvent to dryness, dissolve the residue in methanol, transfer it to a 5ml volumetric flask, add methanol to the mark, shake well, and the test sample is obtained.

[0012] The chromatographic conditions for the verbascoside and mangiferin were as follows: gradient elution with acetonitrile as mobile phase A and 0.2% formic acid solution as mobile phase B; detection wavelength of 260 nm, flow rate of 0.8 mL / min, column temperature of 35 ℃, and injection volume of 10 μL.

[0013] The chromatographic conditions for astragaloside A were as follows: acetonitrile-water (32:68) as the mobile phase; evaporative light scattering detector was used for detection; the flow rate was 0.8 mL / min; the column temperature was 35℃; the gain was 80; and the injection volume was 20 μL.

[0014] Step S2: Sample pre-segmentation based on multi-source feature data; Based on the RGB three-channel color values, the average intensity values ​​of each group of honey-processed Astragalus standard samples in the R, G, and B channels were extracted, and the honey-processed Astragalus with different degrees of processing were divided into three stages: under-processed, moderately processed, and over-processed. Step S3, Construction Phase Enhanced fusion feature matrix; Step S31: Extraction of stage-specific sensory features based on local weighted principal component analysis (PCA); Step S32: Constructing a stage component prediction model based on partial least squares regression (PLS); Step S33, Construction Phase The enhanced fusion feature matrix is ​​calculated to the stage. Mahalanobis distances between the centers of each category; Step S4: Determination of the degree of processing of the sample to be tested; Calculate the Mahalanobis distance from the fused feature vector of the sample to each category center, and classify the sample to the degree of processing corresponding to the stage with the smallest Mahalanobis distance.

[0015] Specifically, as described in step S1 Taste response value data , , This represents the total number of compound taste features corresponding to all sensors of the electronic tongue; The taste profile includes at least sour, bitter, astringent, bitter aftertaste, astringent aftertaste, umami, richness, and saltiness; the content data of the core active ingredients includes astragaloside A. Versicolor isoflavone glucoside and awn stalk flower plain .

[0016] Specifically, the sample pre-segmentation based on multi-source feature data in step S2 is as follows: Based on the RGB three-channel color values, the average intensity values ​​of each honey-processed astragalus standard sample in the R, G, and B channels are extracted to construct the color feature vector of that standard sample. ; Based on the empirical pattern of color change from light yellow to dark yellow and then to yellowish-brown during the processing of honey-processed Astragalus membranaceus, and combined with the color feature vectors of n sets of standard samples of honey-processed Astragalus membranaceus... Honey-processed astragalus, with different degrees of processing, is divided into three stages: Phase I: The color corresponding to light yellow indicates that the processing level is insufficient. Phase II: The corresponding color is bright yellow, and the degree of processing is moderate. Phase III: The corresponding color is dark yellow or yellowish-brown, indicating that the processing degree is over-processed. Let the sample size for each stage be as follows: , , ,and .

[0017] Specifically, the process of extracting stage-specific sensory features based on local weighted principal component analysis in step S31 is as follows: Step S31: Extraction of stage-specific features based on local principal component analysis; Step S311, for the stage Assuming this stage includes Samples, construction phase The original feature matrix ; ; In the above formula, ; For the stage Inner Sample Taste response value data, ; Step S312, Calculation Stage Inner One original feature Compared with this stage Average RGB intensity value of the sample The absolute value of the Pearson correlation coefficient is used as the stage correlation weight. : , ; In the above formula, For the stage Inner The first sample Each electronic tongue taste response value; for Standard sample number The average value of each electronic tongue taste response value ; For the stage Inner The average RGB intensity value of the sample , Stages Inner The intensity values ​​of the R, G, and B channels of a sample; For the stage Inside The average RGB intensity values ​​of the standard samples ; Step S313: Process the original feature matrix By performing weighting, we obtain the weighted feature matrix. After standardization, the standardized feature matrix is ​​obtained. For the standardized feature matrix Perform principal component analysis (PCA); First, calculate the normalized feature matrix. covariance matrix , for transpose; Then, the covariance matrix... Perform eigenvalue decomposition to solve the eigenvalue problem: ; In the above formula, For eigenvalues; For feature vectors; Finally, by solving this eigenvalue problem, the eigenvalues ​​are obtained. and the corresponding feature vector ; The principal components are sorted according to their eigenvalues, and the top-ranked components are selected. One principal component, making the cumulative variance contribution rate ≥ 85%; Stage of obtaining Sensory principal component score matrix: ; In the above formula, For the stage forward A matrix composed of eigenvectors .

[0018] Specifically, the construction of the stage component prediction model based on partial least squares regression in step S32 is as follows: Step S321, Collection Phase Inside Content data of the core active ingredients verbascoside AC, astragaloside A CG, and verbascoside FO in the sample, during the construction phase. Component content matrix : ; In the above formula, , , Each represents a stage Inner The contents of verbascoside AC, astragaloside A CG and verbascoside FO in the samples; Because the contents of the three core active ingredients have different dimensions and orders of magnitude, it is necessary to... Standardization was performed to obtain a standardized component content matrix. ; ; In the above formula, for The mean vector of each column; for The standard deviation vector of each column; For all elements equal to 1 3D column vector; This represents the transpose of a matrix; Step S322: Establish a partial least squares regression (PLS) model, in order to Let X be the independent variable. Let Y be the dependent variable, and latent variables be extracted through iteration. Until the preset stopping condition is reached; make , , for the One latent variable, , To determine the optimal number of latent variables, calculate the weight vector. : ; In the above formula, The residual matrix is ​​the independent variable. for transpose; Let Y be the score vector, initially taken as The first column, during the iteration process, is... Updated to convergence. The residual matrix of the dependent variable is... The Y-load vector; ; Calculate the X score vector The score vector is the projection of the residuals of the independent variables onto the weight direction, expressed as: ; The X-load vector is obtained through least squares estimation. and Y load vector ; ; ; By removing the extracted latent variable information from the current independent variable residual matrix and dependent variable residual matrix, we obtain the first... The residual matrices after each iteration are expressed as follows: , ; Determining the optimal number of latent variables through cross-validation. Number of latent variables for each candidate Calculate the sum of squared predicted residuals : ; In the above formula, For the first Predicted values ​​for each sample; To exclude the first The component prediction model established after the first sample is used for the second sample. Predicted values ​​for each sample; Based on the calculated sum of squared predicted residuals, the root mean square error of the prediction is further calculated. ; ; Select to make Minimum value As the optimal number of latent variables , or when Time selection As the optimal number of latent variables ; extract After identifying the latent variables, the final regression coefficient matrix is ​​obtained. : ; In the above formula, This is the weight matrix. ; Let X be the load matrix. ; Let Y be the load matrix. ; Stage of obtaining The component prediction model is .

[0019] Specifically, in step S33, the stage Sensory principal component score matrix and component prediction model Concatenate the columns to obtain the fused feature matrix. Enhance the fusion feature matrix to the stage The Mahalanobis distance calculation process for the category centers is as follows: Calculation phase The mean and standard deviation of the three channels R, G, and B, respectively, for the stage Color feature vector of each honey-processed Astragalus membranaceus standard sample Normalization is performed to obtain the stage. Normalized RGB matrix within : ; In the above formula, , , Stages Inner The intensity values ​​of the R, G, and B channels of each sample; , , and , , Stages The mean and standard deviation of the three channels R, G, and B; Will As a new column added to the fusion feature matrix, the enhanced fusion feature matrix is ​​obtained. ; Calculation phase Category Center : ; In the above formula, For the stage The number of all samples in the sample; To enhance the fusion feature matrix The row vectors ; Calculation phase Global merged covariance matrix : ; Calculate the enhanced fusion feature matrix To the stage Category Center Mahalanobis distance : ; In the above formula, For the fusion feature vector of the sample, , For the sample in the stage The molecular vector is obtained from the principal components of the sensory data. This represents the predicted content sub-vectors of the three core active ingredients in the sample. This is the normalized RGB color sub-vector of the sample.

[0020] Specifically, the process for determining the degree of processing of the sample to be tested in step S4 is as follows: Step S41: Collect the honey-processed astragalus sample to be tested. Taste response data and RGB three-channel color values; Step S42: Determine the stage of the honey-processed astragalus sample to be tested based on the obtained RGB three-channel color values. ; Step S43: Calculate the sensory principal component score and predicted component content of the honey-processed astragalus sample to be tested, and concatenate them with the normalized RGB color features to obtain the enhanced fusion feature vector of the honey-processed astragalus sample to be tested. ; Step S44: Calculate the enhanced fusion feature vector of the honey-processed Astragalus membranaceus sample to be tested. To the category center of each stage Mahalanobis distance ; Step S45: Determine the degree of processing of the honey-processed Astragalus membranaceus sample to be tested, corresponding to the stage with the smallest Mahalanobis distance.

[0021] Compared with the prior art, the present invention has the following beneficial effects: 1. The method of this invention breaks away from the limitations of traditional methods that rely on human visual color judgment and taste experience. By accurately capturing RGB color parameters with an electronic eye and quantifying the response value of the taste sensor with an electronic tongue, combined with the quantitative data of core components from HPLC, an objective quantitative evaluation system integrating "color-taste-component" is constructed. This system replaces human senses with data, realizing the digital characterization of the quality of honey-processed astragalus at different processing degrees. It fundamentally eliminates the inconsistency in judgment caused by individual experience differences, sensory state and environmental factors, and establishes a reliable foundation for the unification of quality standards.

[0022] 2. The method of this invention utilizes an electronic tongue to sensitively capture subtle dynamic changes in taste characteristics such as sweetness and bitterness during the processing, and an electronic eye to accurately quantify the objective evolution of color from light to dark. Then, the content of active ingredients such as astragaloside determined by HPLC is used as a chemical benchmark for correlation calibration. Through the construction of a segmented PCA-PLS fusion feature matrix, the intrinsic correlation between multi-source features and processing degree is deeply explored and strengthened, thereby significantly improving the accuracy of distinguishing different processing grades (such as light roasting, medium roasting, and heavy roasting) and the scientific reliability of the judgment conclusion.

[0023] 3. The method of this invention can complete both electronic tongue and electronic eye detection within minutes, realizing rapid and non-destructive acquisition of the appearance and taste characteristics of a large number of samples. Although HPLC analysis takes a certain amount of time, the overall process forms a highly efficient analysis mode that combines rapid sensory screening with precise component verification. This mode is significantly better than the traditional method that relies on time-consuming physicochemical detection throughout the entire process. It can meet the needs of large-scale production quality control on the actual production line for rapid monitoring, timely feedback and adjustment of the processing temperature of honey-processed astragalus.

[0024] 4. The method of this invention simultaneously covers three major quality dimensions of honey-processed Astragalus membranaceus: appearance and color, overall taste, and core medicinal components. It systematically and completely reveals the synergistic change pattern between "color change - taste migration - component transformation" during the processing. Compared with existing technologies that only focus on a single indicator (such as only looking at color or only measuring individual components), this invention provides a more comprehensive and three-dimensional panoramic information on chemical and sensory changes. This provides solid and multi-dimensional data support and decision-making basis for optimizing the processing parameters of honey-processed Astragalus membranaceus, monitoring the quality of the whole process, and scientifically formulating comprehensive quality standards. Attached Figure Description

[0025] To provide a more intuitive understanding of the technical implementation of this invention, the accompanying drawings involved in the embodiments of this invention are briefly described below. These drawings are used to assist in illustrating the implementation methods and are not intended to limit the invention. Those skilled in the art can make derivative designs based on the drawings without creative effort.

[0026] Figure 1 This is a flowchart of a method for determining the degree of processing of honey-processed Astragalus membranaceus based on sensory technology and component analysis according to the present invention; Figure 2 This is a PCA scatter plot of the electronic tongue taste response values ​​in an embodiment of the present invention; Figure 3 This is a PCA scatter plot of the RGB values ​​of the electronic eye in an embodiment of the present invention; Figure 4 This is the PLSR model of verbascoside in the embodiments of the present invention; Figure 5 This is a PLSR model of astragaloside A in the embodiments of the present invention; Figure 6 This is the PLSR model of gentianin in the embodiments of the present invention; Figure 7 This is a schematic diagram of the pathological tissue of the spleen sections of rats in each group in the embodiments of the present invention. Detailed Implementation

[0027] To facilitate understanding and implementation of the present invention by those skilled in the art, the various steps of the method proposed in this invention are described in detail below. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various modifications or alterations to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0028] Example 1 like Figure 1 As shown, this embodiment discloses a method for determining the degree of processing of honey-processed Astragalus membranaceus based on sensory technology and component analysis, including the following steps: Step S1: Multi-source feature data acquisition; preparation Different grades of honey-processed Astragalus membranaceus were pulverized to obtain... A set of honey-processed Astragalus membranaceus standard samples were analyzed by electronic tongue, electronic eye, and high-performance liquid chromatography (HPLC). The taste, surface appearance, and content of core active ingredients of honey-processed astragalus were analyzed in a group of standard samples to obtain results for honey-processed astragalus at different processing degrees. Taste response value data RGB three-channel color values ​​and core active ingredient content data; In this embodiment, the electronic tongue is a TS-5000Z type electronic tongue, the electronic eye is a Canon M50 type high-resolution camera, and the high-performance liquid chromatography is an e2695-2424 type HPLC. The electronic tongue is equipped with five specific taste sensors for sour, salty, bitter, umami, and astringent tastes, as well as three reference electrodes (Ag / AgCl) to measure the initial taste signal values ​​(sour, bitter, astringent, umami, and salty) and the three aftertaste signal values ​​(aftertaste bitterness, aftertaste astringency, and richness) of the five basic tastes.

[0029] The electronic eye shooting conditions are as follows: using a standard white background, adjusting the vertical distance between the camera and the sample, using the same light intensity and shooting parameters, and avoiding interference from factors such as ambient light and shooting angle on the color data.

[0030] The standardized processing of the sample images was performed using ImageJ image analysis software: First, image preprocessing was performed, and background areas and sample edge impurities were removed by cropping tools. A denoising algorithm was used to reduce image noise interference. Then, a uniform area on the sample surface was selected as the region of interest (ROI). The built-in color analysis function of the software was used to accurately extract the RGB three-channel color values ​​(R value, G value, B value) of each ROI and calculate the average value of each channel. The preparation method of the test samples of verbascoside and mangiferin for high performance liquid chromatography (HPLC) analysis is as follows: Take about 1g of honey-processed Astragalus membranaceus sample powder (passed through a No. 4 sieve), accurately weigh it, place it in a stoppered conical flask, accurately add 50ml of 80% methanol solution containing 4% concentrated ammonia test solution (take 4ml of concentrated ammonia test solution, add 80% methanol to 100ml, shake well), stopper tightly, weigh, heat under reflux for 1 hour, cool, weigh again, make up the lost weight with 80% methanol solution containing 4% concentrated ammonia test solution, shake well, filter, accurately measure 25ml of the subsequent filtrate, evaporate to dryness, dissolve the residue in 80% methanol, transfer to a 5ml volumetric flask, add 80% methanol to the mark, shake well, filter, and take the subsequent filtrate to obtain the product.

[0031] The preparation method of the astragaloside A test sample for high performance liquid chromatography (HPLC) analysis is as follows: Take about 1g of honey-processed astragalus sample powder (passed through a No. 4 sieve), accurately weigh it, place it in a round-bottom flask, accurately add 50ml of methanol, weigh it, heat under reflux for 4 hours, cool it, weigh it again, make up the weight loss with methanol, shake well, filter it, accurately measure 25ml of the filtrate, recover the solvent to dryness, dissolve the residue in methanol, transfer it to a 5ml volumetric flask, add methanol to the mark, shake well, and the test sample is obtained.

[0032] The chromatographic conditions for the verbascoside and mangiferin are as follows: acetonitrile as mobile phase A, 0.2% formic acid solution as mobile phase B, gradient elution as specified in Table 1; detection wavelength of 260 nm, flow rate of 0.8 mL / min, column temperature of 35 ℃, and injection volume of 10 μL.

[0033] Table 1. Gradient elution parameters for verbascoside and gentianin ; The chromatographic conditions for astragaloside A were as follows: acetonitrile-water (32:68) as the mobile phase; evaporative light scattering detector was used for detection; the flow rate was 0.8 mL / min; the column temperature was 35℃; the gain was 80; and the injection volume was 20 μL.

[0034] Step S2: Sample pre-segmentation based on multi-source feature data; Based on the RGB three-channel color values, the average intensity values ​​of each group of honey-processed Astragalus standard samples in the R, G, and B channels were extracted, and the honey-processed Astragalus with different degrees of processing were divided into three stages: under-processed, moderately processed, and over-processed. Step S3, Construction Phase Enhanced fusion feature matrix; Step S31: Extraction of stage-specific sensory features based on local weighted principal component analysis (PCA); Step S32: Constructing a stage component prediction model based on partial least squares regression (PLS); Step S33, Construction Phase The enhanced fusion feature matrix is ​​calculated to the stage. Mahalanobis distances between the centers of each category; Step S4: Determination of the degree of processing of the sample to be tested; Calculate the Mahalanobis distance from the fused feature vector of the sample to each category center, and classify the sample to the degree of processing corresponding to the stage with the smallest Mahalanobis distance.

[0035] Specifically, as described in step S1 Taste response value data , , This represents the total number of compound taste features corresponding to all sensors of the electronic tongue; The taste profile includes at least sour, bitter, astringent, bitter aftertaste, astringent aftertaste, umami, richness, and saltiness; the content data of the core active ingredients includes astragaloside A. Versicolor isoflavone glucoside and awn stalk flower plain .

[0036] Specifically, the sample pre-segmentation based on multi-source feature data in step S2 is as follows: Based on the RGB three-channel color values, the average intensity values ​​of each honey-processed astragalus standard sample in the R, G, and B channels are extracted to construct the color feature vector of that standard sample. ; The color feature vectors of n groups of honey-processed astragalus standard samples Based on the empirical pattern of color change from light yellow to dark yellow and then to yellowish-brown during the processing of honey-processed Astragalus membranaceus, and combined with the color feature vectors of n sets of standard samples of honey-processed Astragalus membranaceus, Honey-processed astragalus, with different degrees of processing, is divided into three stages: Phase I: The color corresponding to light yellow indicates that the processing level is insufficient. Phase II: The corresponding color is bright yellow, and the degree of processing is moderate. Phase III: The corresponding color is dark yellow or yellowish-brown, indicating that the processing degree is over-processed. Let the sample size for each stage be as follows: , , ,and .

[0037] Specifically, the process of extracting stage-specific sensory features based on local weighted principal component analysis in step S31 is as follows: Step S311, for the stage Assuming this stage includes Samples, construction phase The original feature matrix ; ; In the above formula, ; For the stage Inner Sample Taste response value data, ; Step S312, Calculation Stage Inner One original feature Compared with this stage Average RGB intensity value of the sample The absolute value of the Pearson correlation coefficient is used as the stage correlation weight. : , ; In the above formula, For the stage Inner The first sample Each electronic tongue taste response value; for Standard sample number The average value of each electronic tongue taste response value ; For the stage Inner The average RGB intensity value of the sample , Stages Inner The intensity values ​​of the R, G, and B channels of a sample; For the stage Inside The average RGB intensity values ​​of the standard samples ; Step S313: Process the original feature matrix By performing weighting, we obtain the weighted feature matrix. After standardization, the standardized feature matrix is ​​obtained. For the standardized feature matrix Perform principal component analysis (PCA); First, calculate the normalized feature matrix. covariance matrix ; Then, the covariance matrix... Eigenvalue decomposition requires solving the eigenvalue problem. ; In the above formula, For eigenvalues; For feature vectors; Finally, by solving this eigenvalue problem, the eigenvalues ​​are obtained. and the corresponding feature vector ; The principal components are sorted according to their eigenvalues, and the top-ranked components are selected. One principal component, making the cumulative variance contribution rate ≥ 85%; Stage of obtaining Sensory principal component score matrix: ; In the above formula, For the stage forward A matrix composed of eigenvectors .

[0038] Specifically, the construction of the stage component prediction model based on partial least squares regression in step S32 is as follows: Step S321, Collection Phase Inside Content data of the core active ingredients verbascoside AC, astragaloside A CG, and verbascoside FO in the sample, during the construction phase. Component content matrix : ; In the above formula, , , Each represents a stage Inner The contents of verbascoside AC, astragaloside A CG and verbascoside FO in the samples; Because the contents of the three core active ingredients have different dimensions and orders of magnitude, it is necessary to... Standardization was performed to obtain a standardized component content matrix. ; ; In the above formula, for The mean vector of each column; for The standard deviation vector of each column; For all elements equal to 1 3D column vector; This represents the transpose of a matrix; Step S322: Establish a partial least squares regression (PLS) model, in order to Let X be the independent variable. Let Y be the dependent variable, and latent variables be extracted through iteration. Until the preset stopping condition is reached; make , , for the One latent variable, , To determine the optimal number of latent variables, calculate the weight vector. : ; In the above formula, The residual matrix is ​​the independent variable. for transpose; Let Y be the score vector, initially taken as The first column, during the iteration process, is... Updated to convergence. The residual matrix of the dependent variable is... The Y-load vector; ; Calculate the X score vector The score vector is the projection of the residuals of the independent variables onto the weight direction, expressed as: ; The X-load vector is obtained through least squares estimation. and Y load vector ; ; ; By removing the extracted latent variable information from the current independent variable residual matrix and dependent variable residual matrix, we obtain the first... The residual matrices after each iteration are expressed as follows: , ; Determining the optimal number of latent variables through cross-validation. Number of latent variables for each candidate Calculate the sum of squared predicted residuals : ; In the above formula, For the first Predicted values ​​for each sample; To exclude the first The component prediction model established after the first sample is used for the second sample. Predicted values ​​for each sample; Based on the calculated sum of squared predicted residuals, the root mean square error of the prediction is further calculated. ; ; Select to make Minimum value As the optimal number of latent variables , or when Time selection As the optimal number of latent variables ; extract After identifying the latent variables, the final regression coefficient matrix is ​​obtained. : ; In the above formula, This is the weight matrix. ; Let X be the load matrix. ; Let Y be the load matrix. ; Stage of obtaining The component prediction model is .

[0039] Specifically, in step S33, the stage Sensory principal component score matrix and component prediction model Concatenate the columns to obtain the fused feature matrix. Enhance the fusion feature matrix to the stage The Mahalanobis distance calculation process for the category centers is as follows: Calculation phase The mean and standard deviation of the three channels R, G, and B, respectively, for the stage Color feature vector of each honey-processed Astragalus membranaceus standard sample Normalization is performed to obtain the stage. Normalized RGB matrix within : ; In the above formula, , , Stages Inner The intensity values ​​of the R, G, and B channels of a sample; , , and , , Stages The mean and standard deviation of the three channels R, G, and B; Will As a new column added to the fusion feature matrix, the enhanced fusion feature matrix is ​​obtained. ; Calculation phase Category Center : ; In the above formula, For the stage The number of all samples in the sample; To enhance the fusion feature matrix The row vectors ; Calculation phase Global merged covariance matrix : ; Calculate the enhanced fusion feature matrix To the stage Category Center Mahalanobis distance : ; In the above formula, For the fusion feature vector of the sample, , For the sample in the stage The molecular vector is obtained from the principal components of the sensory data. This represents the predicted content sub-vectors of the three core active ingredients in the sample. This is the normalized RGB color sub-vector of the sample.

[0040] Specifically, the process for determining the degree of processing of the sample to be tested in step S4 is as follows: Step S41: Collect the honey-processed astragalus sample to be tested. Taste response data and RGB three-channel color values; Step S42: Determine the stage of the honey-processed astragalus sample to be tested based on the obtained RGB three-channel color values. ; Step S43: Calculate the sensory principal component score and predicted component content of the honey-processed astragalus sample to be tested, and concatenate them with the normalized RGB color features to obtain the enhanced fusion feature vector of the honey-processed astragalus sample to be tested. ; Step S44: Calculate the enhanced fusion feature vector of the honey-processed Astragalus membranaceus sample to be tested. To the category center of each stage Mahalanobis distance ; Step S45: Determine the degree of processing of the honey-processed Astragalus membranaceus sample to be tested, corresponding to the stage with the smallest Mahalanobis distance.

[0041] The technical effects of the method of the present invention will be further explained below through an example of distinguishing honey-processed astragalus with different degrees of processing.

[0042] Materials and Instruments: The electronic eye was a Canon M50 high-resolution camera; the electronic tongue was a TS-5000Z electronic tongue (Insent, Japan); the high-performance liquid chromatography (HPLC) was an e2695-2424 model (Waters, USA); verrucoside isoflavone glucoside (batch number CHB231028, 20mg), astragaloside (batch number CHB231022, 20mg), astragaloside A (batch number CHB230803, 20mg), and reference standards were all purchased from Chengdu Kloma Biotechnology Co., Ltd., with a purity ≥98%. Astragalus membranaceus slices (batch number: 20240701) were purchased from Jiangxi Hongkang Traditional Chinese Medicine Slices Co., Ltd.; honey (batch number: 20240801) was purchased from Hunan Baochun Pharmaceutical Co., Ltd.; acetonitrile, formic acid, and methanol were all chromatographic grade, all other reagents were analytical grade, and water was purified water.

[0043] I. Determination of the degree of processing of honey-processed Astragalus membranaceus samples; Step 1: Preparation of standard samples of honey-processed Astragalus membranaceus; Take raw Astragalus membranaceus slices, add diluted refined honey, mix thoroughly, let it soak until fully hydrated, and stir-fry at 180℃ for 15 min, 18 min, 21 min, and 24 min respectively. Remove and cool, then pulverize and pass through a No. 4 sieve to obtain standard samples of honey-processed Astragalus membranaceus with different degrees of processing. In this example, for every 100 kg of raw Astragalus membranaceus, 25 kg of refined honey is used. Stir-frying for 15 min corresponds to the under-processing stage, stir-frying for 18 min corresponds to the moderately processed stage, and stir-frying for 21 min and 24 min corresponds to the over-processing stage. Step 2: Multi-source feature data acquisition; 1) Electronic tongue detection; Uses the TS-5000Z electronic tongue, equipped with acidity ( ),bitterness( ), astringent taste ( Bitter aftertaste ), astringent aftertaste ( ), umami ( ), richness ( ), salty ( There are a total of 8 sensors; Sample pretreatment: Mix 2g of honey-processed Astragalus membranaceus standard sample with 100mL of boiling water, let stand for 10min, then filter the supernatant. Take 80mL of the filtrate and pour it into sample cups, label them, and wait for testing. The results are obtained by electronic tongue detection. ( The taste response data are shown in Table 2 below. The electronic tongue sensor was initially cleaned by soaking in 30 mmol / L KCl solution and 0.3 mmol / L tartaric acid solution for 90 seconds, then soaked in the reference solution for 120 seconds, and finally immersed in the filtrate for measurement. Each sample was measured three times and the average value was taken.

[0044] Table 2. Electronic tongue response values ​​of honey-processed Astragalus membranaceus at different processing degrees ; 2) Electronic eye detection; Using a Canon M50 camera, adjust the vertical distance (30cm) between the camera and the standard sample, and acquire images of the standard sample under a standard white background, fixed lighting (5000K) and shooting parameters (shutter speed 1 / 30s, aperture 6.3, exposure compensation +2.0, ISO 125) to avoid interference from ambient light, shooting angle, etc. on color data; ImageJ software was used to standardize the images of all standard samples. First, image preprocessing was performed, and background areas and sample edge impurities were removed by cropping tools. A denoising algorithm was used to reduce image noise interference. Then, a uniform area on the surface of the standard sample was selected as the region of interest (ROI). The built-in color analysis function of the software was used to accurately extract the RGB three-channel color values ​​(R value, G value, B value) of each ROI, and the average value of each channel was calculated as the characteristic color parameter of the standard sample. The calculation results are shown in Table 3 below.

[0045] Table 3. RGB color characteristic values ​​of honey-processed Astragalus membranaceus at different processing degrees ; 3) Determination of the content of core active ingredients; The core active components of honey-processed Astragalus membranaceus standard samples with different processing degrees were detected by high performance liquid chromatography, including verbascoside, astragaloside A and gentianin. The results are shown in Table 4 below.

[0046] Table 4. Content of core active ingredients in honey-processed Astragalus membranaceus at different processing degrees ; As can be seen from the data in Table 4, the contents of the three core active ingredients of raw Astragalus membranaceus all showed a decreasing trend after processing, but rebounded in the later stage of processing. Among them, the contents of verbascoside and astragaloside A rebounded to the peak value after 24 minutes of honey processing, while the contents of gentianin reached the highest value after 21 minutes of honey processing.

[0047] Step 3: Pre-segmentation of standard samples; Based on the empirical pattern of color change from light yellow to bright yellow to dark yellow / yellowish brown during the processing of honey-processed astragalus, and combined with the measured RGB values, the samples were divided into three stages: Phase I (Inadequate processing): The corresponding light yellow color is (e.g., honey-roasted sample for 15 minutes: R=195~197, G=159~165, B=108~112). Stage II (Appropriate Processing): The corresponding bright yellow color is (e.g., honey-roasted sample for 18 min: R=180~183, G=136~144, B=91~96). Phase III (Over-processing): The corresponding color is deep yellow or yellowish-brown (e.g., honey-roasted for 21 min: R=141~150, G=86~94, B=64~66; honey-roasted for 24 min: R=116~119, G=79~82, B=61~67). Step 4: Constructing the enhanced fusion feature matrix; For each stage ( Perform the following processing respectively: Stage-specific sensory feature extraction from locally weighted principal component analysis; stage Weighted PCA is performed on the original feature matrix to ensure that the cumulative variance contribution rate is ≥85%, thus obtaining the stage. Sensory principal component score matrix; In this embodiment, as Figure 2 The image shows the analysis results after directly extracting sensory features using principal component analysis. Figure 2 It can be seen that the first two principal components contribute 78% and 14% to the variance, respectively, and together explain 92% of the effective information in the honey-processed Astragalus membranaceus sample; for example Figure 3 The image shows the analysis results after sensory feature extraction using the locally weighted principal component analysis method provided by this invention. Figure 3 It can be seen that the contribution rates of the first two principal components to the variance are 98% and 2%, respectively, which can explain 100% of the effective information of the honey-processed astragalus sample. Moreover, compared with the traditional principal component analysis method, when the first two principal components are selected, the method of the present invention can extract the effective information of the honey-processed astragalus sample more completely.

[0048] Construct a stage component prediction model for partial least squares regression (PLS); like Figures 4-6 As shown, based on the electronic tongue and electronic eye response values ​​and core active ingredient content data obtained in Tables 2-4, a partial least squares regression (PLSR) model was constructed, as follows: Figures 4-6Slope / offset / RMSE / R-square in the table represent four data labels (slope / offset / root mean square error / coefficient of determination). The correlation coefficient Rp (Slope / R-square) and the root mean square error of prediction RMSEP (offset / RMSE) are two core evaluation indicators that can be used to comprehensively evaluate the model's fitting performance and predictive ability. The closer the correlation coefficient Rp is to 1 and the smaller the RMSEP value, the better the model's stability and reliability.

[0049] like Figure 4 As shown, in the PLSD model of verbascoside isoflavone glucoside, as Figure 4 As shown in (a) above, the model prediction correlation coefficient using only the electronic tongue sensor response results is 0.9223, and the RMSEP value is 0.0272; Figure 4 As shown in (b) above, the correlation coefficient of the model prediction using only the electronic eye response results is 0.8859, and the RMSEP value is 0.0329; Figure 4 As shown in (c), the model prediction correlation coefficient based on the response values ​​of the electronic tongue and electronic eye is 0.8849, and the RMSEP value is 0.0331.

[0050] like Figure 5 As shown, in the PLSD model of astragaloside A, as Figure 5 As shown in (a) above, the model prediction correlation coefficient using only the electronic tongue sensor response results is 0.9668, and the RMSEP value is 0.1083; Figure 5 As shown in (b) above, the model prediction correlation coefficient using only the electronic eye response results is 0.9516, and the RMSEP value is 0.1308; Figure 5 As shown in (c), the model prediction correlation coefficient based on the response values ​​of the electronic tongue and electronic eye is 0.9518, and the RMSEP value is 0.1306.

[0051] like Figure 6 As shown, in the PLSD model of styrax linoleum, as Figure 6 As shown in (a) above, the model prediction correlation coefficient using only the electronic tongue sensor response results is 0.9361, and the RMSEP value is 0.039; Figure 6 As shown in (b) above, the correlation coefficient of the model prediction using only the electronic eye response results is 0.8446, and the RMSEP value is 0.0038; Figure 6 As shown in (c), the model prediction correlation coefficient based on the response values ​​of the electronic tongue and electronic eye is 0.8447, and the RMSEP value is 0.0038.

[0052] The above results indicate that in the partial least squares discriminant (PLSD) models for verbenafil glucoside and astragaloside A, the model constructed using only an electronic tongue exhibits higher reliability. For the PLSD model of mangiferin, the modeling scheme using only an electronic eye or a combination of electronic tongue and electronic eye is significantly more reliable than the model using only an electronic tongue. Comprehensive analysis shows that regardless of whether an electronic tongue, electronic eye, or a fusion of both detection strategies is used, the PLSD models constructed based on these detection signals all have prediction correlation coefficients (Rp) greater than 0.80, and the differences in the root mean square error of prediction (RMSEP) are small. This result confirms that the detection methods using only an electronic tongue, only an electronic eye, or a combination of both can effectively quantify and characterize the target active ingredients (vernafil glucoside, astragaloside A, and mangiferin) in honey-processed astragalus, providing reliable technical support for the accurate determination of the degree of processing of honey-processed astragalus and verifying the feasibility of applying electronic sensory technology in the rapid quantitative analysis of active ingredients in traditional Chinese medicine.

[0053] The quantitative prediction model obtained based on the above calculation results is as follows: Verbena isoflavone glucoside = 0.771 - 0.019 * acidity (standardized) - 0.092 * bitterness (standardized) - 0.036 * astringency (standardized) + 0.075 * bitter aftertaste (standardized) + 0.06 * astringency aftertaste (standardized) - 0.065 * umami (standardized) + 0.023 * saltiness (standardized) + 0.087 * richness (standardized) - 0.022 * R (standardized) + 0.017 * G (standardized) + 0.07 * B (standardized); Astragaloside A = 1.642 + 0.033 * Sourness (standardized) - 0.438 * Bitterness (standardized) - 0.065 * Astringency (standardized) + 0.115 * Bitter Aftertaste (standardized) + 0.076 * Astringency Aftertaste (standardized) - 0.308 * Umami (standardized) + 0.008 * Saltiness (standardized) + 0.237 * Richness (standardized) - 0.194 * R (standardized) - 0.047 * G (standardized) + 0.153 * B (standardized); Mangiferin = 0.34 - 0.005 * Sourness (standardized) - 0.043 * Bitterness (standardized) - 0.011 * Astringency (standardized) - 0.002 * Bitter Aftertaste (standardized) + 0.034 * Astringency Aftertaste (standardized) + 0.007 * Umami (standardized) + 0.015 * Saltiness (standardized) + 0.007 * Richness (standardized) + 0.001 * R (standardized) - 0.006 * G (standardized) - 0.002 * B (standardized); The center vectors for each stage are calculated as follows: Phase I (Honey-roasted for 15 minutes): =[0.4735,0.6505,0.4907,0.9315,0.4601,0.3305,0.7568,0.2017,0.4631,0.8144,0.1069,196,162,110]; Phase II Center (Honey-roasted for 18 min): =[0.4938,0.6752,0.8648,0.2143,0.5282,0.8830,0.9747,0.6112,0.3859,1.1470,0.1152,181.5,140,93.5]; Phase III Center (average of honey-roasted for 21 min and honey-roasted for 24 min): =[0.3788,0.0841,0.34875,0.79715,0.9248,0.6839,0.8312,0.46935,0.47605,1.8002,0.12345,131.5,85.25,64.5]; Step 4: Feature fusion and Mahalanobis distance calculation; The sensory principal component scores, predicted component contents, and normalized RGB features of the honey-processed astragalus sample were concatenated into an enhanced fusion feature matrix. Calculate the enhanced fusion feature vector of the honey-processed astragalus sample to be tested. To the category center of each stage Mahalanobis distance The degree of processing of honey-processed Astragalus membranaceus samples was determined to be at the stage with the smallest Mahalanobis distance.

[0054] II. Verification through rat experiments; 1. Experimental objective; This experiment aims to verify that the "appropriately processed" sample identified by the method of the present invention is superior to the "insufficiently processed" and "overly processed" samples in improving the symptoms of spleen deficiency, thereby proving the scientificity and reliability of the identification method of the present invention.

[0055] 2. Experimental grouping and treatment; Fifty-six rats that had completed adaptive feeding were randomly divided into a blank control group (n=8) and a spleen deficiency model group (n=48). After successful model replication, the model group rats were randomly divided into a model group, a Buzhong Yiqi Wan (a traditional Chinese medicine) administration group (positive control group, raw drug dosage 0.945 g / kg), a raw Astragalus membranaceus group (6.3 g / kg), and a differential group of honey-processed Astragalus membranaceus (under-processed, moderate-processed, and over-processed, raw drug dosage 6.3 g / kg), with eight rats in each group. The spleen deficiency model was established in the 56 rats in the model group using a three-factor composite modeling method involving dietary indiscretion, diarrhea, and fatigue: rats were fasted on one day and given free access to water; on two days, they were fed 75 g·kg⁻¹·d and given free access to water. Rats were administered 20 mL of a 1 g·mL⁻¹ rhubarb decoction via gavage at noon each day. In the afternoon, they were placed in a swimming tank at (25±1)℃ and a water depth of 50 cm, and were forced to swim under a weight (a fuse with a mass equal to 10% of the rat's body weight was wrapped around the base of the rat's tail) until exhaustion (exhaustion was defined as the rat's nose submerged, its body sinking and unable to float, and its movements becoming significantly incoordination). After exhaustion, the rats were immediately removed and dried. This model was maintained for 15 days. The control group was fed a standard diet with free access to water for 15 consecutive days.

[0056] 3. Detection indicators; Organ index: After sacrificing the rats, the spleen and thymus were weighed, and the spleen index and thymus index were calculated (organ weight / body weight × 100%).

[0057] Immune inflammatory factors: Serum IL-2 and IgA levels were detected by ELISA.

[0058] Digestion and absorption indicators: Serum gastrin levels were detected by radioimmunoassay.

[0059] Spleen pathology: HE staining to observe spleen tissue morphology.

[0060] 4. Test results; a. Effects on spleen index and thymus index in rats; As shown in Table 5 below, compared with the blank group, the spleen weight, thymus weight, spleen index, and thymus index of the model group rats were significantly decreased (P<0.01); compared with the model group, all drug-treated groups significantly increased spleen weight, thymus weight, spleen index, and thymus index (P<0.05 or P<0.01); among the groups of insufficiently processed, moderately processed, and excessively processed honey-processed astragalus, the moderately processed honey-processed astragalus group showed better recovery of spleen weight, thymus weight, spleen index, and thymus index (P<0.01).

[0061] Table 5. Spleen index and thymus index in rats ; b. Effects on immune and inflammatory factor-related indicators in rats; As shown in Table 6 below, compared with the blank group, the serum IL-2 of rats in the model group was significantly increased (P<0.01), and the IgA level was significantly decreased (P<0.01). There were significant differences in each index between the model group and the blank group; compared with the model group, each administration group had different degrees of differences in the effects on rat serum IL-2 and IgA (P<0.05 or P<0.01). Among the groups judged as under-roasted, moderate and over-roasted of honey-fried Astragalus membranaceus, the moderately roasted honey-fried Astragalus membranaceus group had better effects on regulating immune inflammation-related indexes such as IL-2 and IgA (P<0.01).

[0062] Table 6. Immune inflammation-related indexes of rats in each group ; c. Effects on related indexes of rat digestion and absorption; As shown in Table 7 below, compared with the blank group, the serum level of Gastrin in rats in the model group was significantly decreased (P<0.01); compared with the model group, except that the raw Astragalus membranaceus group and the under-roasted honey-fried Astragalus membranaceus group had no statistical significance on the effect of Gastrin, the other administration groups could increase the serum level of Gastrin in rats to varying degrees (P<0.05 or P<0.01). Among the groups judged as under-roasted, moderate and over-roasted of honey-fried Astragalus membranaceus, the serum level of Gastrin in the moderately roasted Astragalus membranaceus group > over-roasted group > under-roasted group.

[0063] Table​​​​​​​​​​​​​​The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for determining the degree of processing of honey-processed Astragalus membranaceus based on sensory technology and component analysis, characterized in that, Includes the following steps: Step S1: Multi-source feature data acquisition; preparation Different grades of honey-processed Astragalus membranaceus were pulverized to obtain... A set of honey-processed Astragalus membranaceus standard samples were analyzed by electronic tongue, electronic eye, and high-performance liquid chromatography (HPLC). The taste, surface appearance, and content of core active ingredients of honey-processed astragalus were analyzed in a group of standard samples to obtain results for honey-processed astragalus at different processing degrees. Taste response value data RGB three-channel color values ​​and core active ingredient content data; Step S2: Sample pre-segmentation based on multi-source feature data; Based on the RGB three-channel color values, the average intensity values ​​of each group of honey-processed Astragalus standard samples in the R, G, and B channels were extracted, and the honey-processed Astragalus with different degrees of processing were divided into three stages: under-processed, moderately processed, and over-processed. Step S3, Construction Phase Enhanced fusion feature matrix; Step S31: Extraction of stage-specific sensory features based on local weighted principal component analysis (PCA); Step S32: Constructing a stage component prediction model based on partial least squares regression (PLS); Step S33, Construction Phase The enhanced fusion feature matrix is ​​calculated to the stage. Mahalanobis distance of the category center; Step S4: Determination of the degree of processing of the sample to be tested; Calculate the Mahalanobis distance from the enhanced fusion feature matrix of the sample to be tested to the category center of each stage, and classify the sample to be tested as belonging to the stage with the smallest Mahalanobis distance, corresponding to the degree of processing.

2. The method for determining the degree of processing of honey-processed Astragalus membranaceus based on sensory technology and component analysis according to claim 1, characterized in that, The steps described in step S1 Taste response value data , , This represents the total number of compound taste features corresponding to all sensors of the electronic tongue; The taste characteristics include at least sourness, bitterness, astringency, bitter aftertaste, astringent aftertaste, umami, richness, and saltiness; the content data of the core active ingredients include astragaloside A. Versicolor isoflavone glucoside and awn stalk flower plain .

3. The method for determining the degree of processing of honey-processed Astragalus membranaceus based on sensory technology and component analysis according to claim 1, characterized in that, The sample pre-segmentation based on multi-source feature data in step S2 is as follows: Based on the RGB three-channel color values, the average intensity values ​​of each honey-processed astragalus standard sample in the R, G, and B channels are extracted to construct the color feature vector of that standard sample. ; Based on the empirical pattern of color change from light yellow to dark yellow and then to yellowish-brown during the processing of honey-processed Astragalus membranaceus, and combined with the color feature vectors of n sets of standard samples of honey-processed Astragalus membranaceus... Honey-processed astragalus, with different degrees of processing, is divided into three stages: Phase I: The color corresponding to light yellow indicates that the processing level is insufficient. Phase II: The corresponding color is bright yellow, and the degree of processing is moderate. Phase III: The corresponding color is dark yellow or yellowish-brown, indicating that the processing degree is over-processed. Let the sample size for each stage be as follows: , , ,and .

4. The method for determining the degree of processing of honey-processed Astragalus membranaceus based on sensory technology and component analysis according to claim 3, characterized in that, Step S31, which involves the extraction of stage-specific sensory features based on local weighted principal component analysis, is as follows: Step S311, for the stage Assuming this stage includes Samples, construction phase The original feature matrix ; ； In the above formula, ; For the stage Inner Sample Taste response value data, ; Step S312, Calculation Stage Inner One original feature Compared with this stage Average RGB intensity value of the sample The absolute value of the Pearson correlation coefficient is used as the stage correlation weight. : ， ； In the above formula, For the stage Inner The first sample Each electronic tongue taste response value; for Standard sample number The average value of each electronic tongue taste response value ; For the stage Inner The average RGB intensity value of the sample , Stages Inner The intensity values ​​of the R, G, and B channels of a sample; For the stage Inside The average RGB intensity values ​​of the standard samples ; Step S313: Process the original feature matrix By performing weighting, we obtain the weighted feature matrix. The standardized feature matrix is ​​then obtained by standardization. For the standardized feature matrix Perform principal component analysis (PCA); First, calculate the normalized feature matrix. covariance matrix , for transpose; Then, the covariance matrix... Perform eigenvalue decomposition to solve the eigenvalue problem: ； In the above formula, For eigenvalues; For feature vectors; By solving this eigenvalue problem, the eigenvalues ​​are obtained. and the corresponding feature vector ; The principal components are sorted according to their eigenvalues, and the top-ranked components are selected. One principal component, making the cumulative variance contribution rate ≥ 85%; Stage of obtaining Sensory principal component score matrix: ； In the above formula, For the stage forward A matrix composed of eigenvectors .

5. The method for determining the degree of processing of honey-processed Astragalus membranaceus based on sensory technology and component analysis according to claim 4, characterized in that, The construction of the stage component prediction model based on partial least squares regression in step S32 is as follows: Step S321, Collection Phase Inside Content data of the core active ingredients verbascoside AC, astragaloside A CG, and verbascoside FO in the sample, during the construction phase. Component content matrix : ； In the above formula, , , Each represents a stage Inner The contents of verbascoside AC, astragaloside A CG, and verbascoside FO in the samples were determined. ; Because the contents of the three core active ingredients have different dimensions and orders of magnitude, it is necessary to... Standardization was performed to obtain a standardized component content matrix. ; ； In the above formula, for The mean vector of each column; for The standard deviation vector of each column; For all elements equal to 1 3D column vector; This represents the transpose of a matrix; Step S322: Establish a partial least squares regression (PLS) model, in order to Let X be the independent variable. Let Y be the dependent variable, and latent variables be extracted through iteration. Until the preset stopping condition is reached; make , , for the One latent variable, , To determine the optimal number of latent variables, calculate the weight vector. : ； In the above formula, The residual matrix is ​​the independent variable. for transpose; Let Y be the score vector, initially taken as The first column, during the iteration process, is... Updated to convergence. The residual matrix of the dependent variable is... The Y-load vector; ; Calculate the X score vector The score vector is the projection of the residuals of the independent variables onto the weight direction, expressed as: ； The X-load vector is obtained through least squares estimation. and Y load vector ; ； ； By removing the extracted latent variable information from the current independent variable residual matrix and dependent variable residual matrix, we obtain the first... The residual matrices after each iteration are expressed as follows: ， ； Determining the optimal number of latent variables through cross-validation. Number of latent variables for each candidate Calculate the sum of squared predicted residuals : ； In the above formula, For the first Predicted values ​​for each sample; To exclude the first The component prediction model established after the first sample is used for the second sample. Predicted values ​​for each sample; Based on the calculated sum of squared predicted residuals, the root mean square error of the prediction is further calculated. ; ； Select to make Minimum value As the optimal number of latent variables , or when Time selection As the optimal number of latent variables ; extract After identifying the latent variables, the final regression coefficient matrix is ​​obtained. : ； In the above formula, This is the weight matrix. ; Let X be the load matrix. ; Let Y be the load matrix. ; Stage of obtaining The component prediction model is .

6. The method for determining the degree of processing of honey-processed Astragalus membranaceus based on sensory technology and component analysis according to claim 5, characterized in that, In step S33, the stage Sensory principal component score matrix and component prediction model Concatenate the columns to obtain the fused feature matrix. Enhance the fusion feature matrix to the stage The Mahalanobis distance calculation process for the category centers is as follows: Calculation phase The mean and standard deviation of the three channels R, G, and B, respectively, for the stage Color feature vector of each honey-processed Astragalus membranaceus standard sample Normalization is performed to obtain the stage. Normalized RGB matrix within : ； In the above formula, , , Stages Inner The intensity values ​​of the R, G, and B channels of each sample; , , and , , Stages The mean and standard deviation of the three channels R, G, and B; Will As a new column added to the fusion feature matrix, the enhanced fusion feature matrix is ​​obtained. ; Calculation phase Category Center : ； In the above formula, For the stage The number of all samples in the sample; To enhance the fusion feature matrix The row vectors ; Calculation phase Global merged covariance matrix : ； Calculate the enhanced fusion feature matrix To the stage Category Center Mahalanobis distance : ； In the above formula, For the fusion feature vector of the sample, , For the sample in the stage The molecular vector is obtained from the principal components of the sensory data. This represents the predicted content sub-vectors of the three core active ingredients in the sample. This is the normalized RGB color sub-vector of the sample.

7. The method for determining the degree of processing of honey-processed Astragalus membranaceus based on sensory technology and component analysis according to claim 6, characterized in that, The process for determining the degree of processing of the sample to be tested in step S4 is as follows: Step S41: Collect the honey-processed astragalus sample to be tested. Taste response data and RGB three-channel color values; Step S42: Determine the stage of the honey-processed astragalus sample to be tested based on the obtained RGB three-channel color values. ; Step S43: Calculate the sensory principal component score and predicted component content of the honey-processed astragalus sample to be tested, and concatenate them with the normalized RGB color features to obtain the enhanced fusion feature vector of the honey-processed astragalus sample to be tested. ; Step S44: Calculate the enhanced fusion feature vector of the honey-processed Astragalus membranaceus sample to be tested. To the category center of each stage Mahalanobis distance ; Step S45: Determine the degree of processing of the honey-processed Astragalus membranaceus sample to be tested, corresponding to the stage with the smallest Mahalanobis distance.

Images