Daqu quality grading method and system based on multi-point spectral feature recombination

Abstract

The invention provides a Daqu quality grading method and system based on multi-point spectral feature recombination, and relates to the technical field of wine brewing. Based on spectral data of all point positions of Daqu, a recombination spectral feature vector of each point position is constructed; calculating a mean value, a standard deviation, a variable coefficient, a range and a center-surface difference value of the recombined spectral feature vector on each feature dimension, splicing the recombined spectral feature vector into a spatial statistical feature vector of the yeast for making hard liquor, and establishing an average saccharification power prediction model and a spatial uniformity prediction model by taking the spatial statistical feature vector of the yeast for making hard liquor as input; according to the method, the average saccharifying power and the saccharifying power space uniformity index to be measured are predicted, quality grading is carried out on the yeast according to the average saccharifying power and the saccharifying power space uniformity index of the yeast to be measured, the problem that existing quality grading of the yeast is inaccurate is solved, and the method is suitable for quality grading of the yeast.

Description

Technical Field

[0001] This invention relates to the field of brewing technology, and in particular to a method and system for grading the quality of Daqu (a type of starter culture) based on the recombination of multi-point spectral features. Background Technology

[0002] Daqu (a type of starter culture) is a key saccharifying and fermenting agent in the brewing process of Baijiu (Chinese liquor). Its saccharification power level directly affects starch conversion efficiency, fermentation process stability, and base liquor quality. Current methods for determining the saccharification power of Daqu mainly rely on chemical analysis and enzyme activity assays, which have problems such as complex operating procedures, long testing cycles, high sample destructiveness, and difficulty in meeting the needs of rapid on-site grading and process control.

[0003] Existing methods for grading the quality of Daqu (a type of starter culture) typically require the determination of saccharification power. However, this method suffers from problems such as complex procedures, long testing cycles, high sample destructiveness, and difficulty in meeting the needs of rapid on-site grading and process control. Furthermore, when grading the quality of Daqu, only the overall saccharification power is considered, which can lead to Daqu with uneven saccharification power being mistakenly classified as either qualified or high-grade products. This results in inaccurate Daqu quality grading and affects the stability of the brewing process. Summary of the Invention

[0004] The technical problem solved by this invention is to provide a method and system for classifying the quality of Daqu (a type of Chinese liquor) based on the recombination of multi-point spectral features, thereby solving the problem of inaccurate quality classification of existing Daqu.

[0005] The technical solution adopted by this invention to solve the above-mentioned technical problems is a method for classifying the quality of Daqu (a type of Chinese liquor) based on multi-point spectral feature reconstruction, comprising the following steps:

[0006] S1. Obtain spectral data of multiple points of Daqu, including surface points and center points, local saccharification power of each point, and average saccharification power of Daqu.

[0007] S2. Preprocess the spectral data of all points to obtain the preprocessed spectral data of each point;

[0008] S3. Based on the preprocessed spectral data of each point, construct a recombinant spectral feature vector for each point. The recombinant spectral feature vector includes spectral data of the saccharification power sensitive band, morphological characteristics of the absorption band, and the ratio of the average reflectance of the hydration band to the average reflectance of the scattering band.

[0009] S4. Calculate the mean, standard deviation, coefficient of variation, range, and center surface difference of the recombined spectral feature vector in each feature dimension, and splice them into a large-curve spatial statistical feature vector.

[0010] S5. Using spatial statistical feature vectors as input and average glycation power as output, a partial least squares regression model for predicting average glycation power is established. The mean and standard deviation of local glycation power at all points are calculated. Based on the mean and standard deviation of local glycation power at all points, a spatial uniformity index for glycation power is constructed. Using spatial statistical feature vectors as input and the spatial uniformity index for glycation power as output, a spatial uniformity prediction model is established using support vector regression.

[0011] S6. Using the spatial statistical feature vector of the Daqu to be tested as input, the average saccharification power and spatial uniformity index of the Daqu to be tested are obtained by using the average saccharification power prediction model and the spatial uniformity prediction model.

[0012] S7. The quality of the Daqu (a type of starter culture) is graded based on the average saccharification power and spatial uniformity of saccharification power.

[0013] Further, in S1, sampling points for Daqu are determined, and samples are taken from each sampling point. The samples are then crushed and passed through a 20-mesh sieve to obtain sample powder. A near-infrared spectrometer with a rotating stage is used to collect spectral data of the sample powder in the 80nm to 2526nm wavelength range using diffuse reflectance to obtain spectral data for each point. The saccharification power of the sample powder is measured to obtain the local saccharification power of each point. The remaining Daqu is then crushed and passed through a 20-mesh sieve to obtain Daqu powder. The saccharification power of the Daqu powder is measured to obtain the average saccharification power of the Daqu.

[0014] Furthermore, the preprocessing includes Savinger smoothing, first derivative, standard normal variable transformation, and removal of outlier spectra.

[0015] Furthermore, in S3, the process of acquiring spectral data of the glycation power sensitive band is as follows: calculate the correlation coefficient between the reflectance of each band and the average glycation power, take the band with the correlation coefficient greater than the first threshold as the glycation power sensitive band, and extract the spectral data of the glycation power sensitive band.

[0016] Furthermore, in S3, the morphological characteristics of the absorption band include envelope depth, full width at half maximum (FWHM), peak intensity, and band center position.

[0017] Furthermore, in S5, the formula for constructing the spatial uniformity index of glycation power based on the mean and standard deviation of local glycation power at all points is as follows: ,in, Indicators representing spatial uniformity of saccharification power This represents the standard deviation of local glycation power at all sites. This represents the mean of local glycosylation power at all points.

[0018] Furthermore, in S7, the quality of the Daqu (a type of starter culture) is graded based on its average saccharification power and spatial uniformity index. This includes setting a first and second target value for the average saccharification power, and a third and fourth target value for the spatial uniformity index. The first target value is greater than the second target value, and the third target value is greater than the fourth target value. If the average saccharification power of the Daqu is greater than the first target value and the spatial uniformity index is greater than the third target value, then the Daqu is grade A. If the average saccharification power of the Daqu is greater than the second target value and the spatial uniformity index is greater than the third target value, then the Daqu is grade B. Otherwise, the Daqu is grade C.

[0019] This invention also provides a quality grading system for Daqu (a type of Chinese liquor) based on multi-point spectral feature reconstruction, realizing the quality grading method for Daqu based on multi-point spectral feature reconstruction as described above. The system includes a spectral data acquisition unit, a spectral data preprocessing unit, a spectral feature vector reconstruction unit, a spatial statistical feature vector construction unit, an average saccharification power prediction unit, a saccharification power spatial uniformity index prediction unit, and a quality grading unit. The spectral data acquisition unit is used to acquire spectral data from multiple points of the Daqu to be tested. The spectral data preprocessing unit is used to preprocess the spectral data from all points to obtain preprocessed spectral data for each point. The spectral feature vector reconstruction unit is used to construct a reconstructed spectral feature vector for each point, the reconstructed spectral feature vector including spectral data of the saccharification power sensitive band. The morphological characteristics of the absorption band and the ratio of the average reflectance of the hydration band to the average reflectance of the scattering band are analyzed. The spatial statistical feature vector construction unit is used to calculate the mean, standard deviation, coefficient of variation, range, and central surface difference of the recombinant spectral feature vector in each feature dimension, and splices them into a spatial statistical feature vector of the koji to be tested. The average saccharification power prediction unit is used to predict the average saccharification power of the koji to be tested using the spatial statistical feature vector of the koji to be tested as input and the average saccharification power prediction model. The saccharification power spatial uniformity index prediction unit is used to predict the spatial uniformity index of the koji to be tested using the spatial statistical feature vector of the koji to be tested as input and the spatial uniformity prediction model. The quality grading unit is used to grade the koji to be tested based on the average saccharification power and the saccharification power spatial uniformity index.

[0020] The beneficial effects of this invention are as follows: This invention provides a method and system for quality grading of Daqu (a type of Chinese liquor) based on multi-point spectral feature recombination. Based on the spectral data of each point of Daqu, a recombined spectral feature vector is constructed for each point. The mean, standard deviation, coefficient of variation, range, and center surface difference of the recombined spectral feature vector are calculated in each feature dimension and then concatenated to form a spatial statistical feature vector of Daqu. Using the spatial statistical feature vector of Daqu as input, an average saccharification power prediction model and a spatial uniformity prediction model are established to predict the average saccharification power and spatial uniformity index of the Daqu to be tested. Based on the average saccharification power and spatial uniformity index of the Daqu to be tested, the quality of Daqu is graded, thus solving the problem of inaccurate quality grading of existing Daqu. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating a method for classifying the quality of Daqu (a type of Chinese liquor) based on multi-point spectral feature recombination provided by the present invention.

[0022] Figure 2 This is a schematic diagram of the spectral data of 6 points from 60 pieces of Daqu (a type of Chinese liquor) in a Daqu quality classification method based on multi-point spectral feature recombination provided by the present invention.

[0023] Figure 3 This is a schematic diagram of the preprocessed spectral data in a method for classifying the quality of Daqu (a type of Chinese liquor) based on multi-point spectral feature reconstruction provided by the present invention.

[0024] Figure 4 This is a schematic diagram of the principal component space in a method for classifying the quality of Daqu (a type of Chinese liquor) based on multi-point spectral feature recombination provided by the present invention.

[0025] Figure 5 This is a schematic diagram of the Poulson correlation coefficient curve in a method for classifying the quality of Daqu (a type of Chinese liquor) based on multi-point spectral feature recombination provided by the present invention.

[0026] Figure 6 This is a schematic diagram of the half-width at half-maximum (WHM) search results in a large-batch quality classification method based on multi-point spectral feature recombination provided by the present invention. Detailed Implementation

[0027] This invention addresses the problem that existing methods for grading Daqu (a type of starter culture) only consider the overall saccharification power of the Daqu, leading to the misclassification of Daqu with uneven saccharification power into qualified or high-grade products and resulting in inaccurate Daqu quality grading. It provides a Daqu quality grading method based on multi-point spectral feature recombination, such as... Figure 1 As shown, it includes the following steps:

[0028] S1. Obtain spectral data of multiple points of Daqu, including surface points and center points, local saccharification power of each point, and average saccharification power of Daqu.

[0029] Specifically, 60 samples of high-temperature Daqu (a type of starter culture) from different batches in the same production workshop were selected. All samples were approximately rectangular in shape and had a moisture content of 10%–14% at room temperature. Each Daqu sample was air-dried for at least 24 hours to determine sampling points, including four surface points, one center point, and one point near the center point. Samples were taken from each sampling point, pulverized, and passed through a 20-mesh sieve to obtain sample powder. A near-infrared spectrometer with a rotating stage was used to collect the spectral data of the sample powder in the 80nm–2526nm wavelength range using diffuse reflectance, obtaining the spectral data for each of the 60 Daqu samples at six points. ,like Figure 2 As shown. To improve data accuracy, spectral data of the sample powder at the sampling point can be collected multiple times and the average value can be calculated as the spectral data of the sampling point; the saccharification power of the sample powder at the sampling point can be measured to obtain the local saccharification power of the sampling point; the remaining Daqu can be crushed and passed through a 20-mesh sieve to obtain Daqu powder, and the saccharification power of the Daqu powder can be measured to obtain the average saccharification power of the Daqu. The saccharification power can be determined by referring to the existing technology "QB / T 4257-2011 General Analytical Method for Brewing Daqu".

[0030] S2. Preprocess the spectral data of all points to obtain the preprocessed spectral data of each point.

[0031] Specifically, for Preprocessing is performed, including Savinger smoothing, first derivative, standard normal transformation, and outlier removal. Savinger smoothing and first derivative are used to reduce high-frequency noise and highlight spectral peak shapes, such as... Figure 3 As shown; standard normal variable transformation is used to eliminate scattering differences caused by particle size and packing density; principal component analysis (PCA) is used to project the preprocessed spectra of all points onto the principal component space, as shown. Figure 4 As shown. Abnormal samples were identified. After verification, the samples that deviated from the overall distribution were mostly caused by local contamination of the samples or measurement spot offset. A total of 14 abnormal samples were removed.

[0032] S3. Based on the preprocessed spectral data of each point, construct a recombinant spectral feature vector for each point. The recombinant spectral feature vector includes spectral data of the saccharification power sensitive band, morphological characteristics of the absorption band, and the ratio of the average reflectance of the hydration band to the average reflectance of the scattering band.

[0033] Specifically, the process of acquiring spectral data for the glycation power sensitive bands is as follows: calculate the correlation coefficient between the reflectance of each band and the average glycation power; identify bands with correlation coefficients greater than a first threshold as glycation power sensitive bands; and extract the spectral data of these sensitive bands. The correlation coefficient mentioned is the Paulson correlation coefficient, and the Paulson correlation coefficient curve is shown in Figure 1. Figure 5As shown, the first threshold is 0.4.

[0034] The morphological characteristics of the absorption band are denoted as... The absorption bands include envelope depth, full width at half maximum (FWHM), peak intensity, and band center position. The absorption bands include those related to the vibrations of water and organic groups in Daqu (a type of Chinese liquor), corresponding to wavelengths in the ranges of 1900 nm to 1950 nm and 2100 nm to 2200 nm.

[0035] Envelope depth is denoted as , ,in, Indicates the reflectivity of the envelope baseline. This indicates the maximum reflectivity within the absorption band.

[0036] Half-height and width are denoted as According to the formula: Calculate the reflectivity at half-height, where, This represents the reflectivity at half-maximum (HHM). We find the wavelengths on either side of the absorption band where the reflectivity equals that at HHM. and ,but ,like Figure 6 As shown.

[0037] Peak intensity is denoted as , .

[0038] The position of the heart is recorded as Within the absorption band region, the weighted average wavelength is calculated based on absorption intensity. ,in, Indicates the wavelength points within the absorption band range. This represents the weight of the reflectivity corresponding to the wavelength point within the absorption band.

[0039] The morphological characteristics of the absorption band, the ratio of the average reflectance of the hydration band to the average reflectance of the scattering band, are denoted as... The formula is: ,in, This represents the average reflectance of the hydration absorption band calculated in the 1900nm to 1950nm wavelength range. This represents the average reflectivity of the scattering band calculated in the 1100nm to 1250nm wavelength range. This is used to reduce the impact of sample water content and scattering differences on modeling, and to enhance absorption information related to saccharification mechanisms.

[0040] The recombined spectral eigenvectors of the sites are: , Indicates the first The recombined spectral feature vectors at each point, for a large piece of curvature, are as follows: .

[0041] S4. Calculate the mean, standard deviation, coefficient of variation, range, and center surface difference of the recombined spectral feature vector in each feature dimension, and splice them into a large-scale spatial statistical feature vector.

[0042] Specifically, for all recombination features Xre at 6 locations within the same large curvature block, the mean, standard deviation, coefficient of variation, range, and center-surface difference are calculated for each feature dimension. The mean represents the average level of the recombination feature at each location; the standard deviation and coefficient of variation characterize the degree of fluctuation within the large curvature block; the coefficient of variation is equal to the quotient of the standard deviation and the mean; the range characterizes the difference between the maximum and minimum values ​​within the block; and the center-surface difference represents the difference between the center point and the surface points. The mean, standard deviation, coefficient of variation, range, and center-surface difference are then concatenated to form the spatial statistical feature vector of the large curvature. .

[0043] S5. Using spatial statistical feature vectors as input and average glycation power as output, a partial least squares regression model for predicting average glycation power is established. The mean and standard deviation of local glycation power at all points are calculated. Based on the mean and standard deviation of local glycation power at all points, a spatial uniformity index for glycation power is constructed. Using spatial statistical feature vectors as input and the spatial uniformity index for glycation power as output, a spatial uniformity prediction model is established using support vector regression.

[0044] For the average saccharification power prediction model, partial least squares regression (PLSR) is used to analyze the spatial statistical eigenvectors of large-grain koji. An average glycation end power (AUP) prediction model was established, and the number of principal components was determined using 10-fold cross-validation. Performance was evaluated on an external validation set, and the coefficient of determination, root mean square error (RMSE), and residual prediction bias were calculated. On the modeling set, the AUP prediction model had a coefficient of determination of 0.96, an RMSE of 2.0, and a residual prediction bias of 4.0. On the validation set, the coefficient of determination was 0.91, the RMSE was 3.0, and the residual prediction bias was 3.2.

[0045] Comparative Example 1: Only the spectral data of the center point of Daqu were selected. The spectral feature vector of the saccharification power sensitive band of the center point was reconstructed as input, and the average saccharification power was output. Partial least squares regression (PLSR) was used to build the model. Its coefficient of determination was 0.82, root mean square error was 3.8, and residual prediction bias was 2.0. The coefficient of determination of the external validation set was 0.74, root mean square error was 7.2, and residual prediction bias was 1.3. It can only achieve a rough quantitative prediction of the saccharification power of Daqu, and the prediction accuracy is lower than that of the average saccharification power prediction model.

[0046] Comparative Example 2: Using the spectral feature vector reconstructed from the spectral data of the saccharification power sensitive band of the whole block of Daqu as input and the average saccharification power as output, a partial least squares regression (PLSR) model was established. Its coefficient of determination was 0.92, root mean square error was 3.6, and residual prediction bias was 3.1. The coefficient of determination of the external validation set was 0.83, root mean square error was 4.9, and residual prediction bias was 2.4. The prediction performance was better than that of Comparative Example 1, but the prediction accuracy was still lower than that of the low-cost average saccharification power prediction model.

[0047] The comparison shows that introducing the morphological characteristics of the absorption band, the ratio of the average reflectance of the hydration band to the average reflectance of the scattering band, and combining the spatial statistical characteristics of the same large koji block with multi-point spectra can significantly improve the quantitative ability and cross-batch robustness of the saccharification power prediction model of koji, providing a more reliable technical means for rapid evaluation of koji quality.

[0048] For the spatial uniformity prediction model, the formula for constructing the spatial uniformity index of glycation power based on the mean and standard deviation of local glycation power at all points is as follows: ,in, Indicators representing spatial uniformity of saccharification power This represents the standard deviation of local glycation power at all sites. This represents the mean of local glycation power at all sites. The closer the value is to 1, the more uniform the local saccharification rate and the more even the distribution within the large koji block. (60 large koji blocks) The distribution range is approximately 0.65 to 0.98, among which... The sample was considered to have highly uniform internal saccharification forces. The sample is considered to be moderately homogeneous. The sample is considered to be a spatially non-uniformly distributed sample.

[0049] Using spatial statistical feature vectors as input and the spatial homogeneity index of saccharification power as output, a spatial homogeneity prediction model was established using Support Vector Regression (SVR). The SVR employed a radial basis function kernel, with the penalty factor and kernel function parameters as the main optimization parameters. The performance metrics of the optimal model on the modeling and validation sets were as follows: In spatial homogeneity prediction, the coefficient of determination for the modeling set was 0.90, the root mean square error was 0.03, and the residual prediction bias was 3.0; the coefficient of determination for the external validation set was 0.85, the root mean square error was 0.05, and the residual prediction bias was 2.5. The results were obtained from measured values ​​on the external validation set. Compared with the prediction The comparison shows that the high homogeneity samples and the low homogeneity samples are clearly separated on the scatter plot, indicating that the spatial homogeneity prediction model of the present invention has good quantitative characterization ability and generalization ability.

[0050] S6. Using the spatial statistical feature vector of the koji to be tested as input, the average saccharification power and spatial uniformity index of the koji to be tested are obtained by using the average saccharification power prediction model and the spatial uniformity prediction model.

[0051] S7. The quality of the Daqu (a type of starter culture) is graded based on the average saccharification power and spatial uniformity of saccharification power.

[0052] Specifically, in S7, the quality of the Daqu (a type of starter culture) is graded based on its average saccharification power and spatial uniformity index. This includes setting a first and second target value for the average saccharification power, and a third and fourth target value for the spatial uniformity index. The first target value is greater than the second target value, and the third target value is greater than the fourth target value. If the average saccharification power of the Daqu is greater than the first target value and the spatial uniformity index is greater than the third target value, then the Daqu is grade A. If the average saccharification power of the Daqu is greater than the second target value and the spatial uniformity index is greater than the third target value, then the Daqu is grade B. Otherwise, the Daqu is grade C. The third target value can be 0.90, and the fourth target value can be 0.80.

[0053] This invention also provides a quality grading system for Daqu (a type of Chinese liquor) based on multi-point spectral feature reconstruction, realizing the quality grading method for Daqu based on multi-point spectral feature reconstruction as described above. The system includes a spectral data acquisition unit, a spectral data preprocessing unit, a spectral feature vector reconstruction unit, a spatial statistical feature vector construction unit, an average saccharification power prediction unit, a saccharification power spatial uniformity index prediction unit, and a quality grading unit. The spectral data acquisition unit is used to acquire spectral data from multiple points of the Daqu to be tested. The spectral data preprocessing unit is used to preprocess the spectral data from all points to obtain preprocessed spectral data for each point. The spectral feature vector reconstruction unit is used to construct a reconstructed spectral feature vector for each point, the reconstructed spectral feature vector including spectral data of the saccharification power sensitive band. The morphological characteristics of the absorption band and the ratio of the average reflectance of the hydration band to the average reflectance of the scattering band are analyzed. The spatial statistical feature vector construction unit is used to calculate the mean, standard deviation, coefficient of variation, range, and central surface difference of the recombinant spectral feature vector in each feature dimension, and splices them into a spatial statistical feature vector of the koji to be tested. The average saccharification power prediction unit is used to predict the average saccharification power of the koji to be tested using the spatial statistical feature vector of the koji to be tested as input and the average saccharification power prediction model. The saccharification power spatial uniformity index prediction unit is used to predict the spatial uniformity index of the koji to be tested using the spatial statistical feature vector of the koji to be tested as input and the spatial uniformity prediction model. The quality grading unit is used to grade the koji to be tested based on the average saccharification power and the saccharification power spatial uniformity index.

Claims

1. A method for classifying the quality of Daqu (a type of Chinese liquor) based on multi-point spectral feature reconstruction, characterized in that: Includes the following steps: S1. Obtain spectral data of multiple points of Daqu, including surface points and center points, local saccharification power of each point, and average saccharification power of Daqu. S2. Preprocess the spectral data of all points to obtain the preprocessed spectral data of each point; S3. Based on the preprocessed spectral data of each point, construct a recombinant spectral feature vector for each point. The recombinant spectral feature vector includes spectral data of the saccharification power sensitive band, morphological characteristics of the absorption band, and the ratio of the average reflectance of the hydration band to the average reflectance of the scattering band. S4. Calculate the mean, standard deviation, coefficient of variation, range, and center surface difference of the recombined spectral feature vector in each feature dimension, and splice them into a large-curve spatial statistical feature vector. S5. Using spatial statistical feature vectors as input and average glycation power as output, a partial least squares regression model for predicting average glycation power is established. The mean and standard deviation of local glycation power at all points are calculated. Based on the mean and standard deviation of local glycation power at all points, a spatial uniformity index for glycation power is constructed. Using spatial statistical feature vectors as input and the spatial uniformity index for glycation power as output, a spatial uniformity prediction model is established using support vector regression. S6. Using the spatial statistical feature vector of the Daqu to be tested as input, the average saccharification power and spatial uniformity index of the Daqu to be tested are obtained by using the average saccharification power prediction model and the spatial uniformity prediction model. S7. The quality of the Daqu (a type of starter culture) is graded based on the average saccharification power and spatial uniformity of saccharification power.

2. The method for classifying the quality of large-batch koji based on multi-point spectral feature reconstruction according to claim 1, characterized in that, In S1, sampling points for Daqu (a type of starter culture) are determined. Samples are taken from each sampling point, pulverized, and passed through a 20-mesh sieve to obtain sample powder. A near-infrared spectrometer with a rotating stage is used to collect spectral data of the sample powder in the 80nm to 2526nm wavelength range using diffuse reflectance to obtain spectral data for each point. The saccharification power of the sample powder is measured to obtain the local saccharification power of each point. The remaining Daqu is pulverized and passed through a 20-mesh sieve to obtain Daqu powder. The saccharification power of the Daqu powder is measured to obtain the average saccharification power of the Daqu.

3. The method for classifying the quality of large-batch koji based on multi-point spectral feature reconstruction according to claim 1, characterized in that, The preprocessing includes Savinger smoothing, first derivative, standard normal variable transformation, and removal of outlier spectra.

4. The method for classifying the quality of large-batch koji based on multi-point spectral feature reconstruction according to claim 1, characterized in that, In S3, the process of acquiring spectral data of the glycation power sensitive band is as follows: calculate the correlation coefficient between the reflectance of each band and the average glycation power, take the band with the correlation coefficient greater than the first threshold as the glycation power sensitive band, and extract the spectral data of the glycation power sensitive band.

5. The method for classifying the quality of large-batch koji based on multi-point spectral feature reconstruction according to claim 1, characterized in that, In S3, the morphological characteristics of the absorption band include envelope depth, full width at half maximum (FWHM), peak intensity, and band center position.

6. The method for classifying the quality of large-batch koji based on multi-point spectral feature reconstruction according to claim 1, characterized in that, In S5, the formula for constructing the spatial uniformity index of glycation power based on the mean and standard deviation of local glycation power at all points is as follows: ,in, Indicators representing spatial uniformity of saccharification power This represents the standard deviation of local glycation power at all sites. This represents the mean of local glycosylation power at all points.

7. The method for classifying the quality of large-batch koji based on multi-point spectral feature reconstruction according to claim 1, characterized in that, In S7, the quality of the Daqu (a type of starter culture) is graded based on its average saccharification power and spatial uniformity index. This includes setting a first and second target value for the average saccharification power, and a third and fourth target value for the spatial uniformity index. The first target value is greater than the second target value, and the third target value is greater than the fourth target value. If the average saccharification power of the Daqu is greater than the first target value and the spatial uniformity index is greater than the third target value, then the Daqu is grade A. If the average saccharification power of the Daqu to be tested is greater than the second target value and the spatial uniformity index is greater than the third target value, then the Daqu to be tested is grade B; otherwise, the Daqu to be tested is grade C.

8. A quality grading system for large-batch koji (a type of Chinese liquor) based on multi-point spectral feature recombination, characterized in that: To implement the method for quality grading of Daqu (a type of Chinese liquor) based on multi-point spectral feature reconstruction as described in claim 1, the system includes a spectral data acquisition unit, a spectral data preprocessing unit, a spectral feature vector reconstruction unit, a spatial statistical feature vector construction unit, an average saccharification power prediction unit, a saccharification power spatial uniformity index prediction unit, and a quality grading unit. The spectral data acquisition unit is used to acquire spectral data from multiple points of the Daqu to be tested. The spectral data preprocessing unit is used to preprocess the spectral data from all points to obtain preprocessed spectral data for each point. The spectral feature vector reconstruction unit is used to construct a reconstructed spectral feature vector for each point, wherein the reconstructed spectral feature vector includes spectral data of the saccharification power sensitive band, morphological features of the absorption band, and hydration band. The ratio of the average reflectance to the average reflectance of the scattering band; the spatial statistical feature vector construction unit is used to calculate the mean, standard deviation, coefficient of variation, range, and central surface difference of the recombinant spectral feature vector in each feature dimension, and splice them into the spatial statistical feature vector of the koji to be tested; the average saccharification power prediction unit is used to predict the average saccharification power of the koji to be tested using the spatial statistical feature vector of the koji to be tested as input and the average saccharification power prediction model; the saccharification power spatial uniformity index prediction unit is used to predict the spatial uniformity index of the koji to be tested using the spatial statistical feature vector of the koji to be tested as input and the spatial uniformity prediction model; the quality grading unit is used to grade the koji to be tested based on the average saccharification power and the saccharification power spatial uniformity index.

Images