A method for identifying high-temperature Daqu based on near-infrared spectroscopy technology

Abstract

The application discloses a method for identifying high-temperature Daqu based on near-infrared spectroscopy technology in the field of food detection technology, which comprises the following steps: 1) collecting Maotai-flavor high-temperature Daqu samples and collecting near-infrared spectrum data of the samples; 2) accurately quantifying the melanoidin content of the samples; 3) establishing a melanoidin identification model of near-infrared spectroscopy technology combined with a partial least squares method; 4) collecting near-infrared spectrum data of unknown samples, inputting the data into the model of step (3), obtaining melanoidin content data of the samples, and determining the Daqu type of the unknown samples according to the numerical range of the content data falling into which black, white or yellow Daqu. The application establishes a correlation model of near-infrared spectrum information and melanoidin information by the partial least squares method, and then predicts the melanoidin content of unknown samples by the near-infrared spectrum data, and further identifies the Daqu type (black Daqu, yellow Daqu and white Daqu).

Description

Technical Field

[0001] This invention relates to the field of food testing technology, specifically to a method for identifying high-temperature koji (a type of fermented rice) based on near-infrared spectroscopy. Background Technology

[0002] Maotai-flavor baijiu is one of the major aroma types of baijiu in my country, characterized by its mellow body, prominent Maotai aroma, long-lasting fragrance even in an empty glass, and lingering aftertaste. This unique flavor is closely related to its special brewing process, which is steeped in ancient traditions. Maotai-flavor baijiu uses high-temperature Daqu (a type of starter culture) as the saccharification and fermentation agent. The raw material for high-temperature Daqu is wheat, which is crushed, mixed with water and starter culture to form a starter culture. This starter culture is then kept warm indoors to allow microorganisms to grow and multiply, producing all the flavor components required for brewing. After air drying and storage, the high-temperature Daqu is complete. The spatial differences in environmental factors during the koji-making process lead to the formation of koji with different characteristics: During the stacking process, the local environment (such as temperature, moisture, oxygen concentration, etc.) varies in different spatial locations (such as the stacking position in the koji room, the inner and outer layers of the stack), resulting in finished koji with different characteristics. They also differ significantly in color, namely black koji, white koji, and yellow koji. The indicators such as microorganisms, amino acids, flavor substances, saccharification power, liquefaction power, and esterification power of these three types of koji are different. After storage, mixing, grinding, and use in winemaking, their quality and proportion have an important impact on winemaking.

[0003] Black koji has a prominent soy sauce aroma with a hint of caramel, but low saccharification power; white koji has higher saccharification power but insufficient soy sauce aroma; yellow koji has more suitable indicators. In production, if the amount of black koji is small, the wine has a pleasant caramel aroma; if the amount is large, the finished wine will have a strong burnt bitterness, resulting in a "burnt overcooked soy sauce" effect, affecting the style and quality of the wine. Currently, the standard for koji leaving the warehouse is yellow koji ≥ 80%, white koji + black koji < 20%, and white koji > black koji. In actual production, the black, white, and yellow koji used in brewing are mainly graded and proportioned by experienced koji masters based on sensory evaluation (appearance, cross-section, aroma, etc.). However, this sensory classification method is subjective because the color of the koji is gradual, and some atypical samples are difficult to evaluate objectively; in addition, some samples have different fermentation conditions, or a large number of koji are difficult to evaluate individually. In these two cases, a crushing method can be used to test the homogenized sample, and then calculate the melanoidin content or koji category. Currently, the detection of melanin levels in styraxes used in production typically involves extraction (water bath extraction, ethanol extraction, or ultrasonic extraction) followed by quantification using ultraviolet colorimetry. However, this process is usually time-consuming, generally taking 2–7 hours, and is cumbersome. Summary of the Invention

[0004] This invention addresses the shortcomings of existing technologies by designing a method for quickly and accurately identifying large batches of koji (a type of Chinese liquor).

[0005] One of the purposes of the present application is to provide a method for identifying high-temperature Daqu based on near-infrared spectroscopy technology, comprising the following steps:

[0006] 1) Collecting Jiangxiang-type high-temperature Daqu samples, collecting the near-infrared spectrum data of the samples;

[0007] 2) Accurately quantifying the melanoidin content of the above-mentioned samples;

[0008] 3) Establishing a melanoidin discrimination model combining near-infrared spectroscopy technology and partial least squares method;

[0009] 4) Collecting the near-infrared spectrum data of unknown samples, inputting them into the model of step (3), obtaining the melanoidin content data of the samples, and determining the Daqu type of the unknown samples according to which numerical range the content data falls into.

[0010] The working principle of the present application: melanoidin is a kind of brownish-brown, complex macromolecular compound formed in the later stage of Maillard reaction, which has hydrogen-containing groups such as C-H, N-H and O-H, which respond on near-infrared spectrum. Among white Daqu, yellow Daqu and black Daqu, the content of melanoidin shows an increasing rule, and the content of melanoidin is related to the type of Daqu. The present application establishes a correlation model between near-infrared spectrum information and melanoidin information by partial least squares method, and then predicts the melanoidin content of unknown samples with near-infrared spectrum data. The content of melanoidin is used to further identify the type of Daqu (black Daqu, yellow Daqu and white Daqu).

[0011] Beneficial effects: 1) The melanoidin detection method provided by the present application is fast, simple and safe, and the detection time is only 10-15 minutes, which is at least 8 times faster than the traditional method. The detection process is safe, non-toxic and harmless. The near-infrared detection method provided by the present application can predict the melanoidin content of Daqu samples, and the analysis error is less than 10%.

[0012] 2) The present application can quickly identify the types of black, white and yellow Daqu, and provide a basis for the proportion and content of production Daqu. The original sensory identification is improved to data-based, more scientific and accurate control of liquor production. The Daqu type identification method provided by the present application has an accuracy of 92%.

[0013] 3) The implementation of the present application ensures the accurate identification of Daqu, which can significantly improve the yield and quality of liquor throughout the year, and save labor and time costs after eliminating the tedious detection operation; the present application can be used as a general method for evaluating Daqu quality, which can be applied by other liquor enterprises and play a demonstration and promotion role in the industry.

[0014] Further, the typical sample of high-temperature Daqu has not less than 10 black koji, white koji and yellow koji. Preferably, the total amount of samples is not less than 50, and each sample is not less than 20 g. The more the total amount of samples, the more accurate the model. However, too many samples increase the time and cost of model establishment. Preferably, the number of samples is less than 100 (only 72 in the embodiment of the present application). After optimizing each step of modeling, the accuracy of the final discrimination is as high as 92%.

[0015] Further, step 1) is to pretreat the samples: crushing the typical sample of high-temperature Daqu to a particle size of 15-200 mesh. Preferably, the particle size is 30-60 mesh; more preferably, the particle size is 60 mesh.

[0016] Further, step 2) includes water bath extraction, ethanol extraction or ultrasonic extraction. In the water bath extraction, pure water is added to the crushed sample, heated and stirred, then centrifuged, and the supernatant is taken, diluted before detection, and then centrifuged to obtain the supernatant. The absorbance of the supernatant is measured at 470 nm, and the content of melanin in Daqu is calculated according to the formula. The calculation formula is: C = A / eb,

[0017] In the formula: C is the content of melanin (mmol / L);

[0018] A is the absorbance of the sample solution (470 nm);

[0019] e is the molar extinction coefficient of melanin, which is 0.64 L / mmol·cm;

[0020] b is the thickness of the cuvette (cm).

[0021] The specific method is as follows: 4 g of Daqu powder is weighed, 80 ml of distilled water is added, heated and stirred at 60°C for 6 h, then centrifuged twice (8000 rpm for 10 min at 20°C), the supernatant is taken, diluted 2.5 times before detection, and then centrifuged (8000 rpm for 10 min at 20°C). The absorbance of the supernatant is measured at 470 nm, and the content of melanin in Daqu is calculated.

[0022] Further, step 3) includes the following steps when modeling: a, sample grouping: the samples are divided into a modeling sample group and a prediction sample group. The grouping method is random grouping, K-S method or SPXY method;

[0023] b, model establishment: taking the absorbance value at each wave number of near-infrared spectrum as the independent variable, and taking the content data of melanin as the response value, a PLS model of the content of melanin is established by cross-validation. The root mean square error on each principal component is investigated, and the principal component number with the lowest root mean square error is selected as the optimal principal component number. The model is established on this PC number, and the determination coefficient of the calibration set, the determination coefficient of the validation set and the cross-validation mean square error are investigated.

[0024] c. External validation: the prediction sample set of step a is substituted into the PLS model, and the prediction set correlation coefficient, the prediction root mean square error or RPD are investigated;

[0025] d. Model optimization and determination: the model is optimized according to the model itself index and the external validation result, and the optimization method can be spectral pretreatment or waveband selection, and finally the discrimination model is determined.

[0026] The model optimization is not a necessary step, but usually the effect of the optimized model is better in most cases.

[0027] The spectral pretreatment mode can be one or a combination of several of the following modes: smoothing (Smooth), derivation (Derivatives), normalization (Normalize), baseline correction (Baseline), detrending (Detrend), multiplicative scatter correction (MSC), standard normalized variate (SNV), etc. Among them, the smoothing (Smooth) can be Moving Average or SG smoothing; the derivation (Derivatives) can be first derivative, second derivative, third derivative, fourth derivative, preferably first derivative or second derivative; the normalization (Normalize) can be Unit Vector Normalization, Area Normalization, Mean Normalization, preferably Unit Vector Normalization.

[0028] The waveband selection can adopt correlation coefficient method, variance analysis method, uninformative variable elimination method (UVE), genetic algorithm (GA), successive projections algorithm (SPA), interval partial least squares method (Interval PLS) or a combination method thereof.

[0029] The determination of the model, the established model (b) or the optimized model (d), determines the acceptable model parameters, and each parameter thereof should generally meet: R 2 Cal ≥ 0.9 (preferably, R 2 Cal ≥ 0.95), R 2 Val ≥ 0.85 (preferably, R 2 Val ≥ 0.90), R SEP≥0.9, RPD≥2.0 (preferably, RPD≥2.5, more preferably, RPD≥3). Meanwhile, R 2 Cal R 2 Val R SEP The higher the RPD, the better; the lower the RMSECV and RMSEP, the better. Refer to the following priority levels: R 2 Val R 2 Cal R SEP RPD, RMSECV, and RMSEP are used as the selection criteria. In other words, if multiple models meet the set requirements, the model with better parameters is preferred.

[0030] Furthermore, when acquiring near-infrared spectral data of the samples, an integrating sphere diffuse reflectance mode was used, with a scanning band of 10000–4000 cm⁻¹. -1 Resolution 1-16cm -1 2cm is preferred -1 4cm -1 8cm -1 4cm is preferred. -1 The number of scans ranged from 8 to 64, with each sample being scanned three times and the average spectrum taken.

[0031] Furthermore, and more specifically, during sample inspection: PCA processing is performed on the spectral data to remove Hotelling's T 2 Samples outside a certain confidence interval (such as 99%, 95%, 90%); and examine melanoidin data, paying attention to outlier points (outliers). Outliers can be identified using box plots or by data points deviating from three standard deviations (when they conform to a normal distribution). If there are no outliers, no action is required. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the overall technical roadmap of the present invention;

[0033] Figure 2 This is a flowchart of the PLS modeling process of the present invention;

[0034] Figure 3 The near-infrared spectrum of Daqu (a type of koji);

[0035] Figure 4 RMSE plots for different principal component numbers (PCs);

[0036] Figure 5 A quantitative model diagram of PLS;

[0037] Figure 6 The prediction results are shown in the image.

[0038] Figure 7 The content distribution of the melanoidins in various samples. DETAILED DESCRIPTION

[0039] The following is further described in detail through specific embodiments:

[0040] Example 1

[0041] 1. Sample collection: 75 samples of different types of Maotai-flavor warehouse starter from Guizhou Guotai Liquor Industry Group Co., Ltd. and Guizhou Guotai Winery Co., Ltd. were collected. The Daqu was crushed and then pulverized by a pulverizer (FW-100 high-speed universal pulverizer, Beijing Zhongxingweiye Instrument Co., Ltd.) and passed through a 60-mesh sieve. 20 g of the Daqu powder was reserved for later use.

[0042] 2. Near-infrared spectral data acquisition: spectral data were collected using a Thermo Antaris II Fourier transform near-infrared spectrometer (with an integrating sphere accessory) with a wavelength range of 10,000-4,000 cm -1 , a resolution of 4 cm -1 , and 32 acquisition times. The background was collected every 20 min. Each sample was loaded and the spectral data were collected three times to obtain an average spectrum.

[0043] 3. Melanoidin content determination: 4 g of Daqu powder was added to 80 mL of pure water. After heating and stirring at 60°C for 6 h, the mixture was centrifuged twice (8000 rpm for 10 min at 20°C). The supernatant was diluted 2.5 times and then centrifuged again (8000 rpm for 10 min at 20°C). The absorbance of the supernatant was measured at 470 nm, and the Daqu melanoidin content was calculated.

[0044] The calculation formula is C = A / eb

[0045] In the formula, C represents the content of melanoidins (mmol / L);

[0046] A represents the absorbance of the sample solution (470 nm);

[0047] e represents the molar extinction coefficient of melanoidins, which is 0.64 L / mmol·cm;

[0048] b represents the thickness of the cuvette (cm).

[0049] 4. Establishment of a melanoidin discrimination model based on near-infrared spectroscopy combined with partial least squares

[0050] (1) Data inspection: the spectral data were analyzed by Hotelling’s T 2 , and no abnormal sample points were found

[0051] (2) Sample grouping: K-S method was used to group the samples, 60 samples for modeling and 12 samples for external validation.

[0052] Table 1 Sample distribution

[0053]

[0054] (3) Spectrum processing: The data was imported into The Unscrambler 10.4 software (CAMO, Norway), and spectrum processing was performed on the 10000-4000 cm -1 band, including the following (alone or superimposed): no processing (RAW), smoothing (Smoothing), normalization (Normalize), first derivative (1st Derivative), second derivative (2nd Derivative), baseline correction (Baseline), SNV, Spectroscopic, De-trending, MSC.

[0055] (4) Model establishment: The absorbance value at each wave number of near infrared spectrum was used as the independent variable (X), and the melanin-like content data was used as the response value (Y), and cross-validation was used to establish PLS models under different processing methods (including the original spectrum without processing). The R 2 Cal , R 2 Val of the model were investigated at the optimal principal component number (cumulative contribution rate greater than 85%, RMSE at the minimum or tending to be stable). The acceptable R 2 Cal = 0.90, R 2 Val = 0.85 of the model was set. If there are multiple models meeting the requirements, the one with higher R 2 Val is preferred.

[0056] Table 2 Model effect under different processing methods

[0057]

[0058] From the above table, it can be seen that the spectrum processed by "MSC + first derivative" has the best PLS model effect, with higher R 2 and lower RMSECV, representing better fitting effect and stability. Therefore, "MSC + first derivative" is determined as the best processing method. On the residual plot Figure 4 , PC = 8 is lower and tends to be stable, so the principal component number (PC) of the model is determined to be 8.

[0059] (5) External validation:

[0060] Twelve prediction samples were put into the PLS model as external validation Figure 5 ). The correlation coefficient of external validation was 0.97, and RPD>3.0. The data predicted by the model was compared with the data detected by traditional method (Table 3), Figure 6 The results were shown in the figure of prediction, and there was no significant difference (paired t test p>0.05). The model reached the practical level and could be used as a rapid quantitative method of melanoidins.

[0061] Table 3 External validation data

[0062]

[0063]

[0064] Example 2

[0065] Typical samples of different types of Jiangxiang-type starter were collected from Guizhou Guotai Liquor Group Co., Ltd. and Guizhou Guotai Winery Co., Ltd. There were 18 samples of black koji, yellow koji and white koji respectively. The content of melanoidins in the samples was detected by traditional method Figure 7 ), and the range of melanoidins in each type of Daqu was defined according to the range of melanoidins in typical samples: white koji (0, 4.5), yellow koji [4.5, 6.0), and black koji [6.0, +∞). According to the above standard, the content of melanoidins predicted by near infrared spectroscopy in Example 1 was put into the range, and the prediction type was compared with the result of the type determined by the koji maker (recorded as the real type). The accuracy was 92% (calculation method: number of correct prediction / total number of prediction).

[0066] Table 4 Prediction results of Daqu types

[0067]

[0068]

[0069] The above is only an embodiment of the present application, and the common knowledge of specific structures and properties in the scheme is not described in detail. It should be noted that for those skilled in the art, without departing from the structure of the present application, some modifications and improvements can be made, which should be regarded as the protection scope of the present application. The protection scope of the present application should be subject to the content of its claims, and the specific implementation mode and the like in the specification can be used to explain the content of the claims.

Claims

1. A method for identifying high-temperature Daqu based on near-infrared spectroscopy technology, characterized in that, It comprises the following steps: 1) Collecting Maotai-flavor high-temperature Daqu samples, collecting near-infrared spectrum data of the samples; 2) Accurately quantifying the melanoidin content of the above samples; 3) Establishing a melanoidin discriminant model of near-infrared spectroscopy combined with partial least squares method, defining the melanoidin range of black Daqu, white Daqu and yellow Daqu, the melanoidin range of white Daqu is (0, 4.5), the melanoidin range of yellow Daqu is [4.5, 6.0], and the melanoidin range of black Daqu is [6.0, +∞); 4) Collecting the near-infrared spectrum data of unknown samples, substituting the model of step (3), obtaining the melanoidin content data of unknown samples, and determining the Daqu type of unknown samples according to which numerical range the content data falls into. In the step 1), the near-infrared spectrum detection range is 10000~4000cm -1 ; The quantification method of step 2) is water bath extraction, adding pure water to the crushed sample, heating and stirring, centrifuging, taking the supernatant, diluting before detection, and then centrifuging to obtain the supernatant, measuring the absorbance at 470 nm, and calculating the melanoidin content of Daqu according to the formula.

2. The method for identifying high-temperature Daqu based on near-infrared spectroscopy according to claim 1, characterized in that: There are not less than 10 samples of each type of black Daqu, white Daqu and yellow Daqu in the high-temperature Daqu sample.

3. The method for identifying high-temperature Daqu based on near-infrared spectroscopy according to claim 2, characterized in that: Step 1) First, pretreat the sample: crush the high-temperature Daqu sample to a particle size of 15-200 mesh.

4. The method for discriminating high-temperature Daqu based on near-infrared spectroscopy according to claim 1, characterized in that, Step 3) When modeling, it specifically comprises the following steps: a, sample grouping: group the samples into modeling sample group and prediction sample group, the grouping method is random grouping, K-S method or SPXY method; b, model establishment: taking the absorbance value at each wave number of near-infrared spectrum as the independent variable, and taking the content data of melanoidin as the response value, a PLS model of melanoidin content is established by using cross-validation method, the root mean square error on each principal component is investigated, the principal component number with the lowest root mean square error is selected as the optimal principal component number, and the model is established on this PC number, the determination coefficient of the calibration set, the determination coefficient of the validation set and the cross-validation mean square error of the model are investigated; c, external validation: substituting the prediction sample group of step a into the PLS model, investigating the prediction set correlation coefficient, prediction root mean square error or RPD; d, model optimization and determination: optimizing the model according to the model's own indicators and external validation results, the optimization methods include spectrum pretreatment or wave band selection, and finally determining the discriminant model.

5. The method for identifying high-temperature Daqu based on near-infrared spectroscopy according to claim 4, characterized in that: The near infrared spectrum data of the sample was collected by using the integral sphere diffuse reflection mode, the scanning wave band was: 10000-4000 cm -1 , the resolution was 1-16 cm -1 , the scanning times were 8-64, and each sample was scanned three times to obtain the average spectrum.

6. The method for identifying high-temperature Daqu based on near-infrared spectroscopy according to any one of claims 1-5, characterized in that: The spectrum pretreatment includes one or a combination of several of the following: smoothing, derivation, standardization, baseline correction, detrending, multivariate scatter correction, and variable standardization.

7. The method for identifying high-temperature Daqu based on near-infrared spectroscopy according to claim 6, characterized in that: The spectrum pretreatment method is MSC+first derivative.

8. The method for identifying high-temperature Daqu based on near-infrared spectroscopy according to any one of claims 1-5, characterized in that: The wave band selection includes one or a combination of several of the following: correlation coefficient method, variance analysis method, uninformative variable elimination method, genetic algorithm, continuous projection algorithm, and interval partial least squares method.

9. The method for identifying high-temperature Daqu according to claim 8, characterized in that: Step 3) also included sample inspection when modeling: PCA on spectral data, elimination of samples outside of Hotelling's T 2 a certain confidence interval.

Images