Rapid Identification Method for the Quality of Daqu Based on Raman Spectroscopy
The samples of Daqu were separated and classified by Raman spectroscopy and machine learning models, and the problems of slow quality identification speed and low accuracy of Daqu were solved, and fast and accurate quality identification was achieved.
Patent Information
- Application Number
- CN202211323274.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-10-27
- Publication Date
- 2025-07-11
- Estimated Expiration
- 2042-10-27
AI Technical Summary
In the prior art, the identification speed of the mass of Daqu is slow, the accuracy is low, and the error is large depending on human experience.
Raman spectroscopy technology is used to combine machine learning and deep learning, and the Raman spectra of the Daqu samples are centrifuged and detected respectively by detecting the Raman spectra of substances and microbial communities, using machine learning support vector machines and deep learning 2D-CNN models for classification, and combining soft voting method to determine the Daqu quality category.
The rapid and accurate identification of the quality of the Daqu is achieved, the identification speed and accuracy are improved, and human error is reduced.
Smart Images
Figure CN115575380B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of quality control, and particularly to a method for rapid identification of the quality of Daqu based on Raman spectroscopy. Background Art
[0002] As a multi-enzyme and multi-bacteria composite active preparation, Daqu plays roles in saccharification, fermentation, grain input, and aroma generation during liquor brewing. It is usually made from raw materials such as wheat, barley, and peas through inoculation fermentation, natural temperature rise and fall, and then air-dried. The prepared Daqu needs to be stored. The stored Daqu directly participates in liquor brewing fermentation during the liquor fermentation process. The quantity, variety, and enzyme activity of its microorganisms play a decisive role in the liquor yield rate and premium rate of liquor production through the "secondary Daqu making" in the cellar pool.
[0003] The quality control of Daqu usually needs to be evaluated and classified into premium Daqu, first-class Daqu, second-class Daqu, and unqualified Daqu. The existing evaluation of Daqu quality relies on front-line technicians' decades of work experience. By combining the fragrance, cross-section, skin thickness, appearance, thoroughness of fermentation, and quality of bacteria cultivation of Daqu for judgment, there are human errors, resulting in slow identification speed and low accuracy of Daqu quality. Summary of the Invention
[0004] The technical problem to be solved by the present invention: Provide three methods for rapid identification of the quality of Daqu based on Raman spectroscopy to solve the problems of slow identification speed and low accuracy of the existing technology for identifying the quality of Daqu.
[0005] The technical solution adopted by the present invention to solve the above technical problems: First, a method for rapid identification of the quality of Daqu based on Raman spectroscopy, including the following steps:
[0006] S01. Sample the Daqu to be tested and grind it into a powder.
[0007] S02. Add sterile water and make the powdered Daqu evenly mixed with the sterile water to obtain a mixed solution.
[0008] S03. Centrifuge the mixed solution for the first time. The precipitate after the first centrifugation is the sample of the substance to be tested of Daqu, and centrifuge the supernatant for the second time. The precipitate after the second centrifugation is the sample of the microbial community to be tested of Daqu.
[0009] S04. Perform Raman spectroscopy detection on the sample of the substance to be tested and the sample of the microbial community to be tested respectively.
[0010] S05. Denoise the detected Raman spectroscopy.
[0011] S06. Classify the sample of the substance to be tested using the Raman model of Daqu substances to obtain the probability that the sample of the substance to be tested belongs to each quality category, and classify the sample of the microbial community to be tested using the Daqu microbial community model to obtain the probability that the sample of the microorganism to be tested belongs to each quality category;
[0012] S07. For the probability that the sample of the substance to be tested belongs to each quality category and the probability that the sample of the microorganism to be tested belongs to each quality category, use the soft voting method to obtain the quality category of the Daqu to be tested.
[0013] The second, a rapid method for identifying the quality of Daqu based on Raman spectroscopy, includes the following steps:
[0014] S101. Sample the Daqu to be tested and grind it into powder;
[0015] S102. Add sterile water and make the powdered Daqu and the sterile water evenly mixed to obtain a mixed solution;
[0016] S103. Centrifuge the mixed solution for the first time, and the precipitate after the first centrifugation is the sample of the substance to be tested of Daqu;
[0017] S104. Conduct Raman spectroscopy detection on the sample of the substance to be tested;
[0018] S105. Denoise the detected Raman spectroscopy;
[0019] S106. Classify the sample of the substance to be tested using the Raman model of Daqu substances to obtain the quality category of the Daqu to be tested.
[0020] The third, a rapid method for identifying the quality of Daqu based on Raman spectroscopy, includes the following steps:
[0021] S201. Sample the Daqu to be tested and grind it into powder;
[0022] S202. Add sterile water and make the powdered Daqu and the sterile water evenly mixed to obtain a mixed solution;
[0023] S203. Centrifuge the mixed solution for the first time and centrifuge the supernatant for the second time. The precipitate after the second centrifugation is the sample of the microbial community to be tested of Daqu;
[0024] S204. Conduct Raman spectroscopy detection on the sample of the microbial community to be tested;
[0025] S205. Denoise the detected Raman spectroscopy;
[0026] S206. Classify the sample of the microbial community to be tested using the Daqu microbial community model to obtain the quality category of the Daqu to be tested.
[0027] Further, sampling of the Daqu to be measured is carried out by the five-point sampling method.
[0028] Further, uniform mixing is achieved by a homogenizer or a vortex mixer.
[0029] Further, the rotation speed of the first centrifugation is 300 - 500 rpm.
[0030] Further, the rotation speed of the second centrifugation is not less than 7000 rpm.
[0031] Further, the denoising process includes removing one or more of cosmic rays, baseline correction, Savitzky-Golay smoothing, and normalization.
[0032] Further, the Raman model of Daqu substances is obtained by training the spectral data of Daqu substance samples with input quality categories using machine learning support vector machines.
[0033] Further, the Daqu microbial community model is obtained by training the spectral data of Daqu microbial community samples to be measured with quality categories using deep learning 2D-CNN.
[0034] Advantages of the present invention: The rapid Daqu quality identification method based on Raman spectroscopy of the present invention extracts the substance samples and microbial community samples of Daqu, performs Raman spectroscopy detection, and uses machine learning support vector machines to train the spectral data of Daqu substance samples with input quality categories to obtain a Raman model of Daqu substances. The Daqu microbial community model is obtained by training the Daqu microbial community samples to be measured with quality categories using deep learning 2D-CNN. The substance samples to be measured of the Daqu to be measured are classified through the Raman model of Daqu substances to obtain the probability that the substance samples to be measured belong to each quality category. The microbial community samples of the Daqu to be measured are classified through the Daqu microbial community model to obtain the probability that the microbial samples to be measured belong to each quality category. The quality category of the Daqu to be measured is obtained using the soft voting method, solving the problems of slow identification speed and low accuracy of the existing technology for Daqu quality identification. Compared with the existing technology, the present invention combines the substances and microorganisms of Daqu, resulting in higher accuracy and faster model operation speed. Description of the Drawings
[0035] Attached Figure 1 is a schematic flow chart of the rapid Daqu quality identification method based on Raman spectroscopy of the present invention. Detailed Embodiments
[0036] The first rapid Daqu quality identification method based on Raman spectroscopy of the present invention, as shown in the attached Figure 1 figure, includes the following steps:
[0037] S01. Sample the Daqu to be tested and grind it into powder;
[0038] Specifically, sample the Daqu to be tested by the five - point sampling method and grind it into powder, which is convenient for extracting substance samples and microbial community samples.
[0039] S02. Add sterile water and uniformly mix the powdered Daqu with the sterile water to obtain a mixed solution;
[0040] Specifically, use a homogenizer to rotate for no less than 10 minutes at a speed of 3 - 12 times per second, or use a vortex mixer to vortex - oscillate for no less than 15 minutes to uniformly mix the powdered Daqu and sterile water, so that the microorganisms are fully suspended in the sterile water.
[0041] S03. Conduct the first centrifugation on the mixed solution. The precipitate after the first centrifugation is the sample of the substance to be tested in the Daqu, and conduct the second centrifugation on the supernatant. The precipitate after the second centrifugation is the sample of the microbial community to be tested in the Daqu;
[0042] Specifically, use a centrifuge to centrifuge the mixed solution. The purpose of the first centrifugation of the mixed solution is to separate the microorganisms and the raw materials of the Daqu. Therefore, the rotation speed of the first centrifugation should not be too high. Usually, the rotation speed used is 300 - 500 rpm. Other rotation speeds can also be used as long as the microorganisms and the raw materials of the Daqu can be separated. Thus, the precipitate after the first centrifugation is the sample of the substance to be tested in the Daqu, and the supernatant is the bacterial suspension containing the microbial community. Conduct the second centrifugation on the supernatant. The purpose is to precipitate the microbial community. Therefore, the rotation speed of the second centrifugation should be high. Usually, the rotation speed is not less than 7000 rpm. However, the rotation speed can be selected according to the actual situation and can also be lower than 7000 rpm as long as the microbial community can be precipitated. The precipitate after the second centrifugation is the sample of the microbial community to be tested in the Daqu. Particularly, for the sample of the microbial community to be tested, it can be repeatedly washed to remove impurities.
[0043] S04. Conduct Raman spectroscopy detection on the sample of the substance to be tested and the sample of the microbial community to be tested respectively;
[0044] Specifically, a certain amount of the substance sample to be tested is placed in a quartz cuvette on the stage, and the Raman spectrometer is used to capture the Raman spectrum of the substance sample to be tested. The excitation wavelength of the Raman spectrum is 785 nm, the scanning spectrum range is 500 - 3750 cm-1, the laser power range is 5 mw, and the scanning time is 10 s. A certain amount of the microbial community sample to be tested is placed on a glass slide, dried, and the Raman spectrometer is used to collect the single-cell Raman spectra in the microbial community sample. The excitation wavelength is 532 nm, the scanning spectrum range is 500 - 3750 cm-1, the laser power range is 3 mw, and the scanning time is 5 s. N single-cell Raman spectra are combined into a two-dimensional matrix. The rows of the two-dimensional matrix represent the Raman spectral data of single cells, and the columns represent the Raman bands. The N is not less than 300, and the two-dimensional matrix is the Raman spectrum of the microbial community sample.
[0045] S05. Denoise the detected Raman spectrum;
[0046] Specifically, the denoising process includes one or more of removing cosmic rays, baseline correction, Savitzky-Golay smoothing, and normalization. The baseline correction includes iterative adaptive weighted penalty least squares and least squares polynomial fitting.
[0047] S06. Classify the substance sample to be tested using the Daqu substance Raman model to obtain the probability that the substance sample to be tested belongs to each quality category. Classify the microbial community sample to be tested using the Daqu microbial community model to obtain the probability that the microbial sample to be tested belongs to each quality category.
[0048] Specifically, the Daqu substance Raman model is obtained by training the spectral data of the Daqu substance samples with known quality categories using the machine learning support vector machine. For example, 500 Daqu with known quality categories are taken, and operations are carried out according to S01 - S05 to obtain the Raman spectra of the substance samples of 500 Daqu. The Raman spectra of the substance samples of 500 Daqu are divided into a training set and a validation set. Training is carried out using the training set, and verification is carried out using the validation set to obtain the quality category that can accurately identify the substance samples of the Daqu in the validation set, that is, the training of the model is completed, and the Daqu substance Raman model is obtained. The output of the Daqu substance Raman model is the probability of each quality category. For example, if the Daqu is divided into four quality categories, namely premium Daqu, first-class Daqu, second-class Daqu, and unqualified Daqu, then the output after classifying the Raman spectrum of the substance to be tested by the Daqu substance Raman model is: premium Daqu A1, first-class Daqu A2, second-class Daqu A3, unqualified Daqu A4;
[0049] The daqu microbial community model is obtained by training samples of daqu microbial communities to be tested with quality categories using deep learning 2D-CNN. For example, 30,000 single cells of the microbial communities of daqu with different quality categories are taken, and every 300 single-cell Raman spectra are combined into a two-dimensional matrix. Then, for daqu of the same quality category, there are 100 two-dimensional matrices. When using deep learning 2D-CNN for learning, Resnet50 can be adopted, setting 49 convolutional layers and one fully connected layer, using categorical cross-entropy as the loss function, with the standard parameter initial learning rate of 0.001. Each time 1 epoch is trained, the parameters are updated, and the update multiplication factor is 0.95. Adam is used as the optimizer. Finally, the classifier outputs the probability of each quality category, such as premium daqu B1, first-class daqu B2, second-class daqu B3, and unqualified daqu B4.
[0050] S07. For the probability that the sample to be tested belongs to each quality category and the probability that the microbial sample to be tested belongs to each quality category, the quality category of the daqu to be tested is obtained by using the soft voting method.
[0051] Specifically, the soft voting method can be implemented using Voting. Its principle is: calculate the average value of the probabilities of each quality category in the output results of the two models respectively, and the quality category corresponding to the largest average value is the quality category of the daqu. For example, the average probability of premium daqu is: (A1 + B1) / 2; the average probability of first-class daqu is: (A2 + B2) / 2; the average probability of second-class daqu is: (A3 + B3) / 2; the average probability of unqualified daqu is: (A4 + B4) / 2; compare the sizes of these four values. If (A2 + B2) / 2 is the largest, then the daqu to be tested is first-class daqu; if (A4 + B4) / 2 is the largest, then the daqu to be tested is unqualified daqu.
[0052] The second method for rapid identification of daqu quality based on Raman spectroscopy in the present invention includes the following steps:
[0053] S101. Sample the daqu to be tested and grind it into a powder.
[0054] S102. Add sterile water and make the powdered daqu and the sterile water evenly mixed to obtain a mixed solution.
[0055] S103. Centrifuge the mixed solution for the first time, and the precipitate after the first centrifugation is the sample of the substance to be tested in the daqu.
[0056] S104. Perform Raman spectroscopy detection on the sample of the substance to be tested.
[0057] S105. Denoise the detected Raman spectroscopy.
[0058] S106. Classify the sample of the substance to be tested using the daqu substance Raman model to obtain the quality category of the daqu to be tested.
[0059] Specifically, for the second method for rapid identification of the quality of Daqu based on Raman spectroscopy in the present invention, in the first method for rapid identification of the quality of Daqu based on Raman spectroscopy, the consideration of the microbial community is removed, and only the Raman model of Daqu substances is used to classify the substance samples of the Daqu to be tested, and the probability that the substance sample to be tested belongs to each quality category is obtained. The quality category with the highest probability is taken as the quality category of the Daqu to be tested. The training of the model and the collection of sample data are the same as those of the Raman model of Daqu substances in the first method for rapid identification of the quality of Daqu based on Raman spectroscopy.
[0060] The third method for rapid identification of the quality of Daqu based on Raman spectroscopy in the present invention includes the following steps:
[0061] S201. Sample the Daqu to be tested and grind it into a powder.
[0062] S202. Add sterile water and uniformly mix the powdered Daqu with the sterile water to obtain a mixed solution.
[0063] S203. Centrifuge the mixed solution for the first time and centrifuge the supernatant for the second time. The precipitate after the second centrifugation is the sample of the microbial community to be tested of the Daqu.
[0064] S204. Perform Raman spectroscopy detection on the sample of the microbial community to be tested.
[0065] S205. Denoise the detected Raman spectroscopy.
[0066] S206. Use the Raman model of the Daqu microbial community to classify the sample of the microbial community to be tested of the Daqu, and obtain the quality category of the Daqu to be tested.
[0067] Specifically, for the third method for rapid identification of the quality of Daqu based on Raman spectroscopy in the present invention, in the first method for rapid identification of the quality of Daqu based on Raman spectroscopy, the consideration of the Daqu substances is removed, and only the Raman model of the Daqu microbial community is used to classify the sample of the microbial community of the Daqu to be tested, and the probability that the substance sample to be tested belongs to each quality category is obtained. The quality category with the highest probability is taken as the quality category of the Daqu to be tested. The training of the model and the collection of sample data are the same as those of the Raman model of the Daqu microbial community in the first method for rapid identification of the quality of Daqu based on Raman spectroscopy.
Claims
1. A rapid identification method for the quality of Daqu based on Raman spectroscopy, characterized in that, It includes the following steps: S01. Sample the Daqu to be tested and grind it into powder; S02. Add sterile water and make the powdered Daqu and the sterile water evenly mixed to obtain a mixed solution; S03. Centrifuge the mixed solution for the first time. The precipitate after the first centrifugation is the sample of the substance to be tested in the Daqu, and centrifuge the supernatant for the second time. The precipitate after the second centrifugation is the sample of the microbial community to be tested in the Daqu; S04. Perform Raman spectroscopy detection on the sample of the substance to be tested and the sample of the microbial community to be tested respectively; S05. Denoise the detected Raman spectrum; S06. Classify the sample of the substance to be tested by using the Raman model of Daqu substances to obtain the probability that the sample of the substance to be tested belongs to each quality category, and classify the sample of the microbial community to be tested by using the Raman model of the Daqu microbial community to obtain the probability that the sample of the microorganism to be tested belongs to each quality category. The Raman model of Daqu substances is obtained by training the spectral data of the Daqu substance samples with quality categories input by using the machine learning support vector machine, and the Raman model of the Daqu microbial community is obtained by training the Daqu microbial community samples to be tested with quality categories by using the deep learning 2D-CNN; S07. For the probability that the sample of the substance to be tested belongs to each quality category and the probability that the sample of the microorganism to be tested belongs to each quality category, use the soft voting method to obtain the quality category of the Daqu to be tested.
2. The rapid identification method for the quality of Daqu based on Raman spectroscopy according to claim 1, wherein Sample the Daqu to be tested by the five-point sampling method.
3. The rapid identification method for the quality of Daqu based on Raman spectroscopy according to claim 1, characterized in that, The even mixing is achieved by a homogenizer or a vortex mixer.
4. The rapid identification method for the quality of Daqu based on Raman spectroscopy according to claim 1, characterized in that, The rotation speed of the first centrifugation is 300 - 500 rpm.
5. The rapid identification method for the quality of Daqu based on Raman spectroscopy according to claim 1, wherein The rotation speed of the second centrifugation is not less than 7000 rpm.
6. The rapid identification method for the quality of Daqu based on Raman spectroscopy according to claim 1, wherein The denoising process includes one or more of removing cosmic rays, baseline correction, Savitzky-Golay smoothing, and normalization.
Citation Information
Patent Citations
Method for rapidly detecting three thermoactinomycetes and application of method in fermentation product
CN115236060A