Method for rapidly determining gelatinization degree of fermented grains based on machine vision detection

Abstract

The invention mainly relates to the technical field of white spirit brewing, and provides a method for rapidly determining the gelatinization degree of fermented grains based on machine vision detection in order to improve the efficiency and accuracy of the detection of the gelatinization degree of the fermented grains of white spirit, which comprises the following steps: collecting a fermented grain sample image through a machine vision detection technology, and extracting a color with a large area ratio as a characteristic color number; through a spatial gradient value and gelatinization degree sensitivity calculation weight or based on partial least square discriminant analysis, screening out key characteristic color numbers significantly related to the gelatinization degree from the characteristic color numbers; and establishing a gelatinization degree prediction model by taking the proportion of the area of the key characteristic color number in the image as an independent variable and the gelatinization degree as a dependent variable so as to realize rapid, quantitative and nondestructive detection of the gelatinization degree of the fermented grains.

Description

Technical Field

[0001] This invention mainly relates to the field of liquor brewing technology, and in particular to a method for rapidly determining the gelatinization degree of mash based on machine vision detection. Background Technology

[0002] The degree of gelatinization of mash is a key quality indicator in the process of baijiu brewing. It directly affects the starch conversion efficiency, the fermentation effect of mash, and the final yield and quality of baijiu. Therefore, it is necessary to monitor it accurately and in a timely manner.

[0003] Currently, the main methods for determining the gelatinization degree of mash in the baijiu brewing industry are chemical analysis methods, including enzymatic hydrolysis and titration. These methods require complex sample pretreatment, consume large amounts of chemical reagents, have long testing cycles (usually several hours to tens of hours), low testing efficiency, and generate chemical waste that causes environmental pollution. Furthermore, chemical methods cannot achieve real-time online monitoring at the production site, resulting in significant delays in test results. This prevents timely adjustments to process parameters based on changes in gelatinization degree during production, affecting brewing efficiency and product stability.

[0004] While near-infrared spectroscopy, ultraviolet-visible spectroscopy, and Raman spectroscopy can achieve quantitative analysis of some substances in spectroscopic analysis techniques, they have significant limitations in industrial field testing of complex matrix samples such as fermented grains. Fermented grains have a complex composition, containing starch, water, cellulose, microbial metabolites, and other substances. Existing spectroscopic techniques are easily affected by sample matrix interference, requiring a large amount of basic data as support and complex chemometric algorithms for data correction. Furthermore, the correction effect is unstable, leading to significant fluctuations in detection accuracy. Simultaneously, existing spectroscopic techniques have extremely high requirements for sample homogeneity; differences in particle size and density of fermented grains can significantly affect the accuracy of spectral data.

[0005] Machine vision inspection systems (electronic eyes), as a rapid detection technology, have been widely used in the field of qualitative analysis of samples. They are mainly used to detect surface features such as color differences and appearance of samples, and have advantages such as fast detection speed, simple operation, and no contact contamination. However, existing electronic eye technology can only achieve qualitative differentiation between samples and cannot perform quantitative analysis of specific quality indicators, making it difficult to meet the precise quantification requirements of key parameters such as the degree of gelatinization of mash. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method for rapidly determining the gelatinization degree of fermented grains based on machine vision detection, with the aim of improving the efficiency and accuracy of detecting the gelatinization degree of fermented grains in liquor production.

[0007] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0008] A method for rapidly determining the gelatinization degree of fermented mash based on machine vision detection, the method comprising:

[0009] Samples of fermented mash with different degrees of gelatinization were prepared, and the original images of the samples were collected. The original images were preprocessed, and the color numbers with an area ratio greater than a set value were used as feature color numbers.

[0010] Key feature colors related to the degree of gelatinization were selected from the feature colors, and the area ratio of each key feature color was calculated.

[0011] A gelatinization degree prediction model was established based on the area ratio of key feature color numbers.

[0012] The key characteristic color area ratios of the mash sample to be tested are extracted, and the degree of gelatinization of the mash sample is calculated based on the gelatinization degree prediction model.

[0013] Furthermore, the method also includes: after extracting the feature color codes with an area ratio greater than a set value, performing principal component analysis on the extracted feature color codes to verify the feasibility of calculating the gelatinization degree of the fermented mash based on the feature color codes.

[0014] Furthermore, the principal component analysis performed on the extracted characteristic color codes to verify the feasibility of calculating the gelatinization degree of the fermented mash based on the characteristic color codes includes:

[0015] The extracted feature color proportion data are standardized.

[0016] Calculate the correlation matrix of the standardized data, perform eigenvalue decomposition on the correlation matrix, and obtain the eigenvalues ​​and variance contribution rates of each principal component;

[0017] Effective principal components are selected based on the set eigenvalue threshold and cumulative variance contribution rate threshold.

[0018] Based on the scores of the effective principal components, a principal component score map is generated to verify the feasibility of calculating the gelatinization degree of fermented mash based on the feature color number.

[0019] Furthermore, when preparing mash samples with different degrees of gelatinization, the range of gelatinization degrees covered was 50%-95%.

[0020] Furthermore, the key feature colors related to the degree of gelatinization are selected from the feature colors by using partial least squares method.

[0021] Furthermore, key characteristic color codes related to the degree of gelatinization were selected from the characteristic color codes, including:

[0022] Acquire multiple mash sample images with known gelatinization degrees, and extract the feature color numbers and their area proportions in each sample that are greater than a set threshold.

[0023] Calculate the spatial gradient value of the feature color number ,in and These are the characteristic color numbers. The average area ratio in the high-gelatinization sample group and the low-gelatinization sample group. For characteristic color number The standard deviation of the area proportion in all samples;

[0024] Calculate the sensitivity of the feature color to gelatinization Calculate the local sensitivity between adjacent gelatinized mash samples. ,Will The median as a sensitivity to gelatinization ;in, and For characteristic color number In the Area percentage and degree of gelatinization in each sample and For the characteristic color number i in the first... Area percentage and degree of gelatinization in each sample;

[0025] Based on the spatial gradient value of the characteristic color number and gelatinization sensitivity Calculate the characteristic color number Weights: ,in, For characteristic color number The normalized value of the frequency of occurrence, , and These are the preset weighting coefficients. and These represent the maximum values ​​of the spatial gradient of the characteristic color number and the sensitivity to gelatinization, respectively, across all samples.

[0026] Based on characteristic color code All feature colors are sorted by weight, and the top M feature colors or those with a weight higher than a set threshold T are selected as key feature colors.

[0027] Furthermore, the high gelatinization degree sample group has a gelatinization degree range of 85%-95%, and the low gelatinization degree sample group has a gelatinization degree range of 50%-60%.

[0028] Furthermore, the gelatinization degree prediction model based on the area proportion of key characteristic color codes includes: establishing a multiple linear regression model for gelatinization degree with gelatinization degree of the fermented mash as the dependent variable and the area proportion of key characteristic color codes as the independent variable. The model expression is as follows: ,in For the degree of gelatinization, This represents the area percentage of each key characteristic color number. These are the weighting coefficients corresponding to each key feature color number. This is a constant term.

[0029] The beneficial effects of this invention are as follows: By acquiring images of fermented mash using machine vision and extracting characteristic color codes, key characteristic color codes are selected based on weighted scoring of spatial gradient and sensitivity to gelatinization or partial least squares discriminant analysis. A quantitative regression model between the area ratio of key characteristic color codes and gelatinization degree is established, enabling rapid, accurate, and non-destructive online detection of fermented mash gelatinization degree without the need for chemical reagents or complex pretreatment. This significantly improves the efficiency and environmental friendliness of gelatinization degree detection while reducing operating costs, providing reliable technical support for real-time quality monitoring and process optimization in the liquor brewing process. Attached Figure Description

[0030] Figure 1 This is a flowchart of the method for rapidly determining the gelatinization degree of fermented mash based on machine vision detection according to the present invention.

[0031] Figure 2 (a), 2(b), 2(c), 2(d), and 2(e) are images of fermented mash with different degrees of gelatinization, respectively.

[0032] Figure 3 for Figure 2 (a) Spectral data of the corresponding fermented mash image;

[0033] Figure 4 for Figure 2 (b) Corresponding spectral data of fermented mash image;

[0034] Figure 5 for Figure 2 (c) Corresponding spectral data of fermented mash image;

[0035] Figure 6 for Figure 2 (d) Corresponding spectral data of fermented mash image;

[0036] Figure 7 for Figure 2 (e) Corresponding spectral data of fermented mash image;

[0037] Figure 8 PCA score chart for color number of fermented mash sample;

[0038] Figure 9 PLS-DA score chart for the color number of the fermented mash sample;

[0039] Figure 10 A verification diagram of color number replacement for fermented mash samples;

[0040] Figure 11 This is a VIP value chart for the color code of the fermented mash sample. Detailed Implementation

[0041] The core technical solution of this invention lies in acquiring images of fermented mash samples using machine vision inspection technology, extracting colors with a large area proportion as feature colors; calculating weights based on spatial gradient values ​​and gelatinization sensitivity, or using partial least squares discriminant analysis, selecting key feature colors that are significantly related to gelatinization from the feature colors; and establishing a gelatinization prediction model with the proportion of the area of ​​the key feature colors in the image as the independent variable and gelatinization as the dependent variable, thereby achieving rapid, quantitative, and non-destructive detection of the gelatinization degree of fermented mash.

[0042] like Figure 1 As shown, the method for rapidly determining the gelatinization degree of fermented mash based on machine vision detection according to the present invention includes the following steps:

[0043] Step S1: Prepare mash samples with different degrees of gelatinization, collect the original images of the samples, preprocess the original images, and use the color numbers with an area ratio greater than a set value as the feature color numbers.

[0044] The prepared mash samples should cover different technological stages in the baijiu brewing process. This invention prepares a series of samples with a gelatinization degree ranging from 50% to 95% by precisely controlling the gelatinization conditions, thus covering the gelatinization degree range commonly encountered in actual production.

[0045] In this embodiment, samples of mash that had not been distilled after mixing were collected. Under laboratory conditions, the mash samples were cooked, and different cooking times and temperatures were controlled to prepare 15 samples with gelatinization degrees of 51.2%, 53.5%, 56.5%, 60.1%, 62%, 63.5%, 74.1%, 76.5%, 74.6%, 81%, 84.6%, 87.5%, 91.2%, 93.6%, and 94.5%, respectively.

[0046] After the electronic eye is powered on, it operates stably until the lighting stabilizes to ensure a consistent testing environment. Instrument calibration is then performed by turning on the upper and lower backlights to eliminate background interference, and using a 24-color calibration plate and a 5mm aperture to ensure testing accuracy. Each sample of fermented grains is placed in a glass petri dish, then sequentially placed on the sample tray, and the electronic eye acquires images of each sample individually.

[0047] The original images of the fermented mash samples with different degrees of gelatinization collected in this embodiment are as follows: Figure 2 As shown ( Figure 2 (a), 2(b), 2(c), 2(d), and 2(e) are the original images of 5 out of the 15 samples of fermented mash with different degrees of gelatinization collected by the electronic eye. After background subtraction processing of the original images, the color of the samples was analyzed. Figure 2The color spectral data of the original spectra of each sample corresponding to (a), 2(b), 2(c), 2(d), and 2(e) are as follows: Figure 3 , Figure 4 , Figure 5 , Figure 6 and Figure 7 As shown.

[0048] Through electronic eye detection, a total of 25 color codes were extracted from all samples in this embodiment. Background subtraction was performed on the collected raw image data to remove irrelevant interference information. The color code distribution of all samples was statistically analyzed, and color codes with an area ratio greater than 1% were selected as feature color codes.

[0049] Preferably, before screening key characteristic color codes, step S2 involves performing principal component analysis (PCA) on the extracted characteristic color codes to verify the feasibility of characterizing the gelatinization degree of the fermented mash based on characteristic color codes. Specifically, this includes:

[0050] (1) Data preprocessing: The color codes corresponding to the feature color codes with an area ratio > 1% are standardized to eliminate the weight bias caused by the difference in the numerical range of the area ratio of different color codes; the standardization formula is: ,in, This is the original area percentage data for a single color number. This represents the average area percentage of the corresponding color number across all samples. This represents the standard deviation of the area percentage of the corresponding color number.

[0051] (2) Construction of correlation matrix: Since the area ratio of all color numbers has the same dimension (all are %), the correlation matrix is ​​used instead of the covariance matrix for analysis to avoid feature extraction bias caused by dimension interference;

[0052] (3) Eigenvalue and eigenvector solution: The correlation matrix is ​​solved by eigenvalue decomposition to obtain the eigenvalues ​​and eigenvectors corresponding to each principal component. The eigenvalues ​​represent the information carrying capacity of the corresponding principal component.

[0053] (4) Determination of effective principal components: Effective principal components are extracted by following the dual screening criteria of eigenvalue λ>1 and cumulative variance contribution rate ≥85%;

[0054] (5) Sample discriminability verification: PCA score plots were drawn, and samples from different gelatinization ranges were clustered in different regions with no overlap, proving that the feature color area ratio data can accurately represent the differences in different gelatinization levels of fermented mash, providing an effective data foundation for subsequent modeling.

[0055] In this embodiment, the principal component analysis results of mash samples with different degrees of gelatinization are as follows: Figure 8As shown in the figure, the sum of PC1 and PC2 reaches 95.475%, representing most of the information of the sample. The five groups of mash samples with different degrees of gelatinization are located in different areas without overlap, and the color characteristics are very obvious. The recognition index is 97, indicating that the electronic eye can distinguish the color differences between the five groups of mash samples.

[0056] Step S2: Select key feature colors related to the degree of gelatinization from the feature colors, and calculate the area ratio of each key feature color.

[0057] This invention provides two methods for filtering key feature color codes:

[0058] (1) Screening of key feature colors based on partial least squares method (PLS-DA): Partial least squares discriminant analysis was performed on the feature colors, and the stability of the model was verified by 200 permutation tests. The key feature colors related to the degree of gelatinization were determined with a VIP value greater than 1.0 as the screening condition.

[0059] Partial least squares discriminant analysis (PLS-DA) can determine the relationships between samples based on the characteristic variables of each group, thus better obtaining information on inter-group differences. To accurately identify the characteristic color differences of mash with different degrees of gelatinization, PLS-DA analysis was performed on the spectral data of the characteristic color codes. The results are as follows: Figure 9 As shown. After 200 permutation tests to assess whether the PLS-DA model overfitted the data, the results are as follows. Figure 10 As shown, the dependent variable explained a rate of 0.996 (R²), the predictive power was 0.991 (Q²), and the intercept R² > Q², with Q² being a negative intercept. This indicates that the model is not overfitting, the model is more stable and reliable, and has better predictive power.

[0060] Using a VIP value greater than 1.0 as the filter criterion, such as... Figure 11 As shown, a total of 7 key feature color codes were selected. The key difference colors, ranked from highest to lowest VIP value, are 1877, 2165, 2711, 2150, 2695, 2438, and 1892.

[0061] (2) Another method for screening key feature color codes provided in this invention includes:

[0062] Acquire multiple mash sample images with known gelatinization degrees, and extract the feature color numbers and their area proportions in each sample that are greater than a set threshold.

[0063] Calculate the spatial gradient value of the feature color number ,in and They are color numbers The average area ratio in the high-gelatinization sample group and the low-gelatinization sample group. For color number The standard deviation of the area proportion in all samples;

[0064] Calculate the sensitivity of the feature color to gelatinization Calculate the local sensitivity between adjacent gelatinized mash samples. ,Will The median as a sensitivity to gelatinization ;in, and For characteristic color number In the Area percentage and degree of gelatinization in each sample and For the characteristic color number i in the first... Area percentage and degree of gelatinization in each sample;

[0065] Based on the spatial gradient value of the characteristic color number and gelatinization sensitivity Calculate the characteristic color number Weights: ,in, For color number The normalized value of the frequency of occurrence, , and These are the preset weighting coefficients. and These represent the maximum values ​​of the spatial gradient and the sensitivity to gelatinization for all feature color numbers, respectively.

[0066] Based on characteristic color code All feature colors are sorted by weight, and the top M feature colors or those with a weight higher than a set threshold T are selected as key feature colors.

[0067] Step S3: Establish a gelatinization degree prediction model based on the area ratio of key feature color numbers.

[0068] The area occupancy of key feature colors corresponding to samples with different degrees of gelatinization in this embodiment is shown in Table 1.

[0069] Table 1. Area percentage of extracted differential feature colors corresponding to different degrees of gelatinization

[0070]

[0071] Multiple linear fitting was performed with the degree of gelatinization as the dependent variable and the area ratio of key feature color numbers as the independent variable. The fitting results are shown in Table 2.

[0072] Table 2. Results of Multiple Linear Fitting

[0073]

[0074] The final fitted degree of gelatinization of the fermented mash is:

[0075] Y=148.94049+1.10888×C1(1877)+2.00927×C2(1892)+1.40978×C3(2150)-6.5 2839×C4(2165)-11.81345×C5(2438)-86.67281×C6(2695)+50.69678×C7(2711)

[0076] The parameters of the fitted multivariate linear model of the degree of gelatinization of mash and key characteristic color, the results of the analysis of variance, and the comparison with the measured degree of gelatinization are shown in Tables 3, 4, and 5, respectively. This shows that the model has a good fit and can accurately predict the degree of gelatinization through the key difference color of mash.

[0077] Table 3 Model Parameters

[0078]

[0079] Table 4. Model Variance Analysis

[0080]

[0081] Table 5 Comparison of Actual and Predicted Values ​​from the Model

[0082]

[0083] Step S4: Extract the area ratio of key characteristic color numbers of the mash sample to be tested, and calculate the degree of gelatinization of the mash sample based on the gelatinization degree prediction model.

[0084] In this embodiment, a sample of fermented mash from a certain production batch was taken and pretreated according to the above sample preparation method. The gelatinization degree image of the sample to be tested was obtained by electronic eye. The relevant values ​​of the key characteristic color obtained by the above processing were substituted into the gelatinization degree calculation model to calculate the gelatinization degree as follows: Y=148.94049+1.10888×C1(1877)+2.00927×C2(1892)+1.40978×C3(2150)-6.52839×C4(2165)-11.81345×C5(2438)-86.67281×C6(2695)+50.69678×C7(2711)=82.1.

[0085] The gelatinization degree result measured in this embodiment was compared with the result of 80.3% obtained by the traditional chemical reagent hydrolysis-iodine colorimetric method. The results showed that the relative error between the two was less than 5%, indicating that the method of the present invention has good accuracy and reliability.

Claims

1. A method for rapidly determining the gelatinization degree of fermented mash based on machine vision inspection, characterized in that, The method includes: Samples of fermented mash with different degrees of gelatinization were prepared, and the original images of the samples were collected. The original images were preprocessed, and the color numbers with an area ratio greater than a set value were used as feature color numbers. Key feature colors related to the degree of gelatinization were selected from the feature colors, and the area ratio of each key feature color was calculated. A gelatinization degree prediction model was established based on the area ratio of key feature color numbers. The key characteristic color area ratios of the mash sample to be tested are extracted, and the degree of gelatinization of the mash sample is calculated based on the gelatinization degree prediction model.

2. The method for rapidly determining the gelatinization degree of fermented mash based on machine vision detection according to claim 1, characterized in that, The method further includes: after extracting the feature color codes with an area ratio greater than a set value, performing principal component analysis on the extracted feature color codes to verify the feasibility of calculating the gelatinization degree of the fermented mash based on the feature color codes.

3. The method for rapidly determining the gelatinization degree of fermented mash based on machine vision detection according to claim 2, characterized in that, The principal component analysis of the extracted characteristic color codes to verify the feasibility of calculating the gelatinization degree of the fermented mash based on the characteristic color codes includes: The extracted feature color proportion data are standardized. Calculate the correlation matrix of the standardized data, perform eigenvalue decomposition on the correlation matrix, and obtain the eigenvalues ​​and variance contribution rates of each principal component; Effective principal components are selected based on the set eigenvalue threshold and cumulative variance contribution rate threshold. Based on the scores of the effective principal components, a principal component score map is generated to verify the feasibility of calculating the gelatinization degree of fermented mash based on the feature color number.

4. The method for rapidly determining the gelatinization degree of fermented mash based on machine vision detection according to claim 1, characterized in that, When preparing samples of fermented mash with different degrees of gelatinization, the range of gelatinization degrees covered is 50%-95%.

5. The method for rapidly determining the gelatinization degree of fermented mash based on machine vision detection according to claim 1, characterized in that, The selection of key feature colors related to the degree of gelatinization from feature colors includes: selecting key feature colors related to the degree of gelatinization from feature colors using partial least squares method.

6. The method for rapidly determining the gelatinization degree of fermented mash based on machine vision detection according to claim 1, characterized in that, Key characteristic color codes related to gelatinization degree were selected from the characteristic color codes, including: Acquire multiple mash sample images with known gelatinization degrees, and extract the feature color numbers and their area proportions in each sample that are greater than a set threshold. Calculate the spatial gradient value of the feature color number ,in and These are the characteristic color numbers. The average area ratio in the high-gelatinization sample group and the low-gelatinization sample group. For characteristic color number The standard deviation of the area proportion in all samples; Calculate the sensitivity of the feature color to gelatinization Calculate the local sensitivity between adjacent gelatinized mash samples. ,Will The median as a sensitivity to gelatinization ;in, and For characteristic color number In the Area percentage and degree of gelatinization in each sample and For the characteristic color number i in the first... Area percentage and degree of gelatinization in each sample; Based on the spatial gradient value of the characteristic color number and gelatinization sensitivity Calculate the characteristic color number Weights: ,in, For characteristic color number The normalized value of the frequency of occurrence, , and These are the preset weighting coefficients. and These represent the maximum values ​​of the spatial gradient of the characteristic color number and the sensitivity to gelatinization, respectively, across all samples. Based on characteristic color code All feature colors are sorted by weight, and the top M feature colors or those with a weight higher than a set threshold T are taken as key feature colors.

7. The method for rapidly determining the gelatinization degree of fermented mash based on machine vision detection according to claim 6, characterized in that, The high-gelatinization-degree sample group has a gelatinization-degree range of 85%-95%, and the low-gelatinization-degree sample group has a gelatinization-degree range of 50%-60%.

8. The method for rapidly determining the gelatinization degree of fermented mash based on machine vision inspection according to any one of claims 1-6, characterized in that, The model for predicting the degree of gelatinization based on the area ratio of key characteristic colors includes: using the degree of gelatinization of fermented mash as the dependent variable and the area ratio of key characteristic colors as the independent variable, a multiple linear regression model for the degree of gelatinization is established. The model expression is as follows: ,in For the degree of gelatinization, This represents the area percentage of each key characteristic color number. These are the weighting coefficients corresponding to each key feature color number. This is a constant term.

Images