Maotai-flavor liquor process category and authenticity identification model construction method, identification method and equipment

Abstract

The invention discloses a Maotai-flavor liquor process category and authenticity identification model construction method, an identification method, an identification device and equipment. The method comprises the following steps: acquiring data of liquor samples; preprocessing the sample data to obtain preprocessed sample data; the preprocessed sample data are input into an initial identification model constructed based on a graph neural network or a partial least square discrimination method, the initial identification model is trained, target identification models are obtained, and the target identification models comprise a gas chromatography identification model, an ion chromatography identification model and / or a three-dimensional fluorescence spectrum identification model. According to the method, multiple analysis technologies such as GC, IC and EEM and a chemometrics modeling analysis module are adopted, analysis results of different technologies are mutually verified, a target identification model with high prediction precision is constructed, and accurate identification of the process category and authenticity of the Maotai-flavor liquor is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of liquor identification, in particular to a Jiangxiang liquor process category and authenticity identification model construction method, identification method and equipment. BACKGROUND

[0002] Liquor has a long history in China as a drink for emotional communication and moral cultivation. Jiangxiang liquor is one of the important types of Chinese liquor. As a pure grain liquor, it has a unique and complex production process, a long production cycle, and a high cost. According to the different brewing processes, there are three types of Jiangxiang liquor on the market: Jiangxiang liquor (Daqu), Jiangxiang liquor (other), and fake Jiangxiang liquor (imitation Jiangxiang flavoring liquor). Jiangxiang liquor (Daqu) is a traditional process liquor with strict brewing process requirements, prohibiting the addition of exogenous aroma, color, and flavor substances, and having a long fermentation period. It is known for its prominent Jiangxiang flavor, elegant and delicate taste, full-bodied body, long aftertaste, and long-lasting aroma in an empty cup. Jiangxiang liquor (other) includes bran-curd Jiangxiang liquor and mixed-curd Jiangxiang liquor. Bran-curd Jiangxiang liquor is made from grains such as sorghum, which are ground and fermented with a sugar-fermentation agent prepared by inoculating pure microorganisms on a carrier such as wheat bran. Mixed-curd Jiangxiang liquor is made by blending Daqu Jiangxiang liquor and bran-curd Jiangxiang liquor in a certain proportion, or by using a mixture of high-temperature Daqu and bran-curd as a sugar-fermentation agent, or by using a mixture of high-temperature Daqu and exogenous microorganisms and enzymes as a sugar-fermentation agent. The blending ratio of Daqu Jiangxiang liquor is generally not less than 30% (volume fraction). Compared with Jiangxiang liquor (Daqu), Jiangxiang liquor (other) is more affordable. Imitation Jiangxiang flavoring liquor is a low-cost prepared liquor with certain Jiangxiang liquor flavor characteristics produced by adding flavoring agents to edible alcohol. It is a fake Jiangxiang liquor. In the sales market, the process of various types of Jiangxiang liquor is usually labeled according to relevant standards. For example, the Jiangxiang liquor national standard (GB / T 10781.4-2024) implemented since June 1, 2025 defines the above two types of Jiangxiang liquor as Jiangxiang liquor (Daqu) and Jiangxiang liquor (other). The national standard GB / T 17204-2021 "Beverage liquor terminology and classification" defines the last type of liquor as flavoring liquor.

[0003] In practice, the process labels of Maotai-flavor liquor on the market are relatively chaotic, and there are a large number of false labels and fake labels. In order to protect the authenticity of the traditional Daqu Maotai-flavor liquor process, avoid some unscrupulous businessmen using cheaper Maotai-flavor liquor (others) to pass off as good, even non-Maotai-flavor liquor (imitation sauce flavoring liquor) to confuse the public, and maintain the legitimate rights and interests of consumers, it is necessary to identify the process category and authenticity of Maotai-flavor liquor.

[0004] At present, the identification of the process category and authenticity of Maotai-flavor liquor is mainly carried out by sensory evaluation, infrared spectroscopy, ultraviolet spectroscopy and chromatography. Although the sensory evaluation method can quickly identify the process and quality of liquor to a certain extent, it requires higher sensory evaluation personnel, is greatly affected by the subjective consciousness of the evaluator, and the results are unstable. Moreover, the identification effect of complex mixed and blended liquor samples is poor. The accuracy of the results obtained by infrared spectroscopy is low and needs to be improved. Ultraviolet spectroscopy is mainly suitable for compounds with conjugated systems or aromatic structures. For compounds without such structures, they may not absorb or weakly absorb in the ultraviolet region, and the peak shape of ultraviolet spectroscopy is usually wide and the resolution is poor. The data collected by chromatography is large, and the cost and time required are high. In summary, the existing methods for identifying the process category and authenticity of Maotai-flavor liquor still have certain limitations. SUMMARY

[0005] The purpose of the present application is to provide a Maotai-flavor liquor process category and authenticity identification model construction method, identification method, device and equipment, so as to realize efficient and accurate identification of the process category and authenticity of Maotai-flavor liquor.

[0006] To achieve the above purpose, the present application provides the following solutions:

[0007] In a first aspect, the present application provides a Maotai-flavor liquor process category and authenticity identification model construction method, which comprises:

[0008] Obtaining sample data, the sample data comprising data collected by a gas chromatograph, an ion chromatograph and / or a three-dimensional fluorescence spectrometer on liquor samples, the liquor samples comprising Daqu Maotai-flavor liquor, non-Daqu Maotai-flavor liquor and / or imitation sauce flavoring liquor, and the collected data comprising chromatographic data and / or spectral data corresponding to each liquor sample;

[0009] Pretreating the sample data based on a chemometrics modeling analysis module to obtain pretreated sample data;

[0010] inputting the pretreated sample data into an initial discrimination model constructed based on a graph neural network or a partial least squares discriminant method, training the initial discrimination model, and obtaining a target discrimination model, the target discrimination model being used for process category and authenticity discrimination of an input wine sample to be discriminated and outputting a discrimination result, the discrimination result indicating the process category and authenticity of the wine sample to be discriminated, the target discrimination model comprising a gas chromatography discrimination model, an ion chromatography discrimination model and / or a three-dimensional fluorescence spectrum discrimination model, the process category comprising Daqu Maotai-flavor liquor, non-Daqu Maotai-flavor liquor and / or imitation Maotai-flavor liquor, and the imitation Maotai-flavor liquor indicating that the wine sample to be discriminated is a counterfeit Maotai-flavor liquor.

[0011] Optionally, the application provides a Maotai-flavor liquor process category and authenticity discrimination model construction method, the collected data comprises chromatographic data, the sample data is pretreated by the chemometrics modeling analysis module to obtain pretreated sample data, and the pretreated sample data comprises:

[0012] The chromatographic data is subjected to chromatographic peak integration, background deduction and baseline correction processing.

[0013] The chromatographic peak of the sample with a compound retention time drift in the sample data subjected to chromatographic peak integration, background deduction and baseline correction processing is subjected to retention time alignment.

[0014] Optionally, the application provides a Maotai-flavor liquor process category and authenticity discrimination model construction method, the collected data comprises spectrum data, the sample data is pretreated by the chemometrics modeling analysis module to obtain pretreated sample data, and the pretreated sample data comprises:

[0015] The spectrum data is corrected by using a standard sample.

[0016] The corrected spectrum data is subjected to scattering deduction by using a Spline spline interpolation algorithm.

[0017] Optionally, the application provides a Maotai-flavor liquor process category and authenticity discrimination model construction method, the pretreated sample data is input into an initial discrimination model constructed based on a graph neural network, the initial discrimination model is trained, and a target discrimination model is obtained, and the target discrimination model comprises:

[0018] The pretreated sample data is input into an initial feature processing module, and initial feature representations of each data node are output, the initial feature processing module comprises a message function, an aggregation function and an update function.

[0019] Parameters of each function in the initial feature processing module are updated according to the initial feature representations, and target feature representations are output, and the feature processing module after the parameter update is used as the target discrimination model.

[0020] Optionally, the application provides a Jiangxiang Baijiu process category and authenticity identification model construction method, wherein the preprocessed sample data is input into an initial feature processing module, and the initial feature representation of each data node is output, which includes:

[0021] extracting feature data from the preprocessed sample data;

[0022] constructing an adjacency matrix of each data node according to the feature data, obtaining a neighbor node set of each data node, and the adjacency matrix representing the connection relationship between each data node;

[0023] determining the initial feature representation of each data node based on each feature processing function in the initial feature processing module and the connection relationship.

[0024] Optionally, the application provides a Jiangxiang Baijiu process category and authenticity identification model construction method, wherein determining the initial feature representation of each data node based on each feature processing function in the initial feature processing module and the connection relationship includes:

[0025] determining the neighbor node message of each data node based on the message passing function;

[0026] determining the aggregate message of each data node based on the aggregation function according to the neighbor node message;

[0027] updating the feature data based on the update function and the aggregate message to obtain the initial feature representation corresponding to each data node.

[0028] Optionally, the application provides a Jiangxiang Baijiu process category and authenticity identification model construction method, wherein when the collected data includes three-dimensional fluorescence spectrum data, and before the preprocessed sample data is input into an initial identification model constructed based on a partial least squares discrimination method, the method further includes:

[0029] factorizing the three-dimensional fluorescence spectrum data.

[0030] Optionally, the application provides a Jiangxiang Baijiu process category and authenticity identification model construction method, wherein the preprocessed sample data is input into an initial identification model constructed based on a partial least squares discrimination method, the initial identification model is trained to obtain a target identification model, which includes:

[0031] extracting a feature matrix from the preprocessed or factorized sample data as an independent variable matrix;

[0032] performing principal component analysis on the independent variable matrix and the category matrix to obtain a corresponding score matrix;

[0033] The score matrix is input into an initial linear regression function to perform multivariate linear regression to determine a target regression coefficient matrix, and a linear regression function corresponding to the target regression coefficient matrix is used as the target discrimination model.

[0034] In a second aspect, the application provides a Jiang-flavor liquor process category and authenticity discrimination method, which is discriminated by the target discrimination model constructed in the Jiang-flavor liquor process and authenticity discrimination model construction method of the first aspect. The discrimination method comprises the following steps:

[0035] Collecting data of a sample to be discriminated is collected by a gas chromatograph, an ion chromatograph and / or a three-dimensional fluorescence spectrometer, and the sample to be discriminated comprises Daqu Jiang-flavor liquor, non-Daqu Jiang-flavor liquor and / or imitation Jiang-flavor liquor. The collecting data comprises chromatographic data and / or spectral data corresponding to the sample to be discriminated.

[0036] The collecting data is preprocessed based on a chemometrics modeling analysis module to obtain preprocessed collecting data.

[0037] The preprocessed collecting data is input into a pre-constructed target prediction model to output a discrimination result, which indicates the process category and authenticity of the sample to be discriminated. The process category comprises Daqu Jiang-flavor liquor, non-Daqu Jiang-flavor liquor and / or imitation Jiang-flavor liquor. The imitation Jiang-flavor liquor indicates that the sample to be discriminated is a counterfeit Jiang-flavor liquor. The target model comprises a gas chromatographic discrimination model, an ion chromatographic discrimination model and / or a three-dimensional fluorescence spectral discrimination model.

[0038] In a third aspect, the application provides a computer device, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the Jiang-flavor liquor process category and authenticity discrimination model construction method of the first aspect or the Jiang-flavor liquor process category and authenticity discrimination method of the second aspect.

[0039] According to the embodiments of the application, the following technical effects are achieved:

[0040] The application provides a Maotai-flavor liquor process category and authenticity identification model construction method, identification method, device and equipment. The application adopts GC, IC and / or EEM methods to collect sample data including chromatographic data and / or spectral data of Maotai-flavor liquor samples and counterfeit samples produced by various processes, then pre-processes the sample data based on a chemometrics modeling analysis module, and finally inputs the pre-processed data into an initial identification model constructed based on a graph neural network or a partial least squares discrimination method to train the initial identification model as training data, so as to finally construct a target identification model corresponding to each sample data. When the collected data of a liquor sample to be identified is input into the corresponding target identification model, the identification result indicating the process category and authenticity can be quickly and accurately output.

[0041] That is, in the application, GC, IC and EEM analysis techniques are adopted, a chemometrics modeling analysis module is combined, the analysis results of different techniques are verified with each other, the accuracy of the prediction result is effectively improved, the visualization display of the differences between Maotai-flavor liquors of different process categories and authenticity is realized, and the accurate discrimination of Maotai-flavor liquors of different process categories and authenticity is realized through a machine recognition model. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative effort.

[0043] Figure 1 A flowchart of a Maotai-flavor liquor process category and authenticity identification model construction method of some embodiments of the application;

[0044] Figure 2 A flowchart of a Maotai-flavor liquor process category and authenticity identification model construction method of some embodiments of the application;

[0045] Figure 3 A gas chromatogram of a Maotai-flavor liquor (Daqu) sample, a Maotai-flavor liquor (other) sample and a pseudo-Maotai-flavor liquor sample;

[0046] Figure 4 An ion chromatogram of a Maotai-flavor liquor (Daqu) sample, a Maotai-flavor liquor (other) sample and a pseudo-Maotai-flavor liquor sample;

[0047] Figure 5 A three-dimensional fluorescence spectrum and factor decomposition of a Maotai-flavor liquor (Daqu) sample, a Maotai-flavor liquor (other) sample and a pseudo-Maotai-flavor liquor sample;

[0048] Figure 6 (a) is a schematic diagram of three-dimensional fluorescence spectrum before pretreatment;

[0049] Figure 6 (b) is a schematic diagram of three-dimensional fluorescence spectrum after pretreatment;

[0050] Figure 7 (a)-(b) are schematic diagrams of graph neural network analysis of gas chromatography data of liquor samples of different process categories;

[0051] Figure 8 (a)-(b) are schematic diagrams of graph neural network analysis of ion chromatography data of liquor samples of different process categories;

[0052] Figure 9 (a)-(b) are schematic diagrams of graph neural network analysis of three-dimensional fluorescence spectrum data of liquor samples of different process categories;

[0053] Figure 10 (a)-(d) are schematic diagrams of partial least squares discriminant algorithm analysis of gas chromatography data of liquor samples of different process categories;

[0054] Figure 11 (a)-(d) are schematic diagrams of partial least squares discriminant algorithm analysis of ion chromatography data of liquor samples of different process categories;

[0055] Figure 12 (a)-(d) are schematic diagrams of partial least squares discriminant algorithm analysis of three-dimensional fluorescence spectrum data of liquor samples of different process categories;

[0056] Figure 13 is a flowchart of a Maotai-flavor liquor process category and authenticity identification method of some embodiments of the present application;

[0057] Figure 14 is a structural schematic diagram of a computer device provided by some embodiments of the present application. DETAILED DESCRIPTION

[0058] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0059] The above purposes, features and advantages of the present application can be more obvious and easy to understand. The present application will be further described in detail below with reference to the drawings and specific embodiments.

[0060] It can be understood that Chinese liquor (Maotai-flavor liquor) is one of the world's three major distillate spirits, together with Cognac from France and Scotch Whisky from the United Kingdom, due to its unique brewing process, regional characteristics and flavor.

[0061] For example, the traditional brewing process of Maotai-flavor liquor represented by Chishui River Basin is "12987", where "1" represents a one-year production cycle, "2" represents two feedings (of sorghum), "9" represents nine times of cooking (distillation) of raw materials, "8" represents eight times of fermentation, and "7" represents seven times of liquor taking, which guarantees the unique flavor of Maotai-flavor liquor. In practice, Maotai-flavor liquor can be divided into Maotai-flavor liquor (Daqu) and Maotai-flavor liquor (other) according to different brewing processes. In addition, some counterfeit Maotai-flavor liquor, imitation Maotai-flavor liquor, is still circulating in the market. During the marketing process, the process category of Maotai-flavor liquor can be marked in accordance with the relevant standards.

[0062] For example, the Maotai-flavor liquor national standard (GB / T 10781.4-2024) implemented on June 1, 2025 stipulates that "Maotai-flavor liquor (Daqu)" is Maotai-flavor liquor brewed with waxy sorghum and wheat as raw materials and completely using high-temperature Daqu as saccharifying and fermenting agent; "Maotai-flavor liquor (other)" is Maotai-flavor liquor brewed with grains as raw materials and not completely using or not using high-temperature Daqu as saccharifying and fermenting agent. Imitation Maotai-flavor liquor is "flavoring liquor" defined in the national standard GB / T 17204-2021 "Beverage Wine Terms and Classification" - prepared with solid-state, liquid-state, solid-liquid liquor or edible alcohol as the base, adding food additives, and having some Maotai-flavor liquor flavor characteristics.

[0063] At present, the process category of Maotai-flavor liquor on the market is quite chaotic, with a lot of false labeling and other phenomena. In order to protect the authenticity of the traditional Daqu Maotai-flavor process, prevent some unscrupulous businessmen from using lower-cost Maotai-flavor liquor (other) to pass off as good liquor, even counterfeit Maotai-flavor liquor to deceive consumers, and safeguard the legitimate rights and interests of consumers, it is necessary to identify the authenticity of the process category of Maotai-flavor liquor.

[0064] In practice, the traditional methods and related identification technologies for identifying the process category and authenticity of Maotai-flavor liquor have the problems of low identification efficiency and accuracy.

[0065] The application provides a method for identifying the process category and authenticity of Maotai-flavor liquor by using gas chromatography (GC), ion chromatography (IC) and excitation emission matrix (EEM) combined with chemometrics.

[0066] That is, in the application, the differences between various process-made Maotai-flavor liquor and counterfeit Maotai-flavor liquor are analyzed based on GC, IC and EEM technology, the chromatographic and spectral fingerprint of different process categories and authentic Maotai-flavor liquor (Maotai-flavor liquor (Daqu), Maotai-flavor liquor (other) and imitation Maotai-flavor liquor) is constructed, and an artificial intelligence machine learning model is established by using a chemometrics modeling and analysis module to perform pattern recognition analysis, thereby realizing accurate identification of the process category and authenticity of Maotai-flavor liquor.

[0067] It can be understood that, for the convenience of description and understanding, the Maotai-flavor liquor (Daqu) in the application is Daqu Maotai-flavor liquor, and the Maotai-flavor liquor (other) is non-Daqu Maotai-flavor liquor.

[0068] In order to better understand the method for constructing the identification model of the process category and authenticity of Maotai-flavor liquor and the identification method provided by the application, the following will be described in detail with reference to the drawings.

[0069] As shown in Figure 1 , the method for constructing the identification model of the process category and authenticity of Maotai-flavor liquor provided by some embodiments of the application includes the following steps: Figure 1

[0070] S110, obtaining sample data, the sample data including data collected by using a gas chromatograph, an ion chromatograph and / or a three-dimensional fluorescence spectrometer on a liquor sample, the liquor sample including Daqu Maotai-flavor liquor, non-Daqu Maotai-flavor liquor and / or imitation Maotai-flavor liquor, and the collected data including chromatographic data and / or spectral data corresponding to each Maotai-flavor liquor sample.

[0071] S120, pre-processing the sample data based on a chemometrics modeling and analysis module to obtain pre-processed sample data.

[0072] ​S130, input the pretreated sample data into an initial discrimination model based on a graph neural network or a partial least squares discrimination method, train the initial discrimination model, and obtain a target discrimination model. The target discrimination model is used for process and authenticity category discrimination of an input wine sample to be discriminated, and outputs a discrimination result. The discrimination result represents the process category and authenticity of the wine sample to be discriminated. The target discrimination model includes a gas chromatography discrimination model, an ion chromatography discrimination model, and / or a three-dimensional fluorescence spectrum discrimination model. The process category includes Daqu Maotai-flavor liquor, non-Daqu Maotai-flavor liquor, and / or imitation Maotai-flavor liquor. The imitation Maotai-flavor liquor indicates that the wine sample to be discriminated is a counterfeit Maotai-flavor liquor.

[0073] Specifically, in the embodiments of the present application, in order to improve the efficiency and accuracy of the Maotai-flavor liquor process category and authenticity discrimination, sample data can be collected first. For example, the chromatographic and spectral information of Maotai-flavor liquor produced by various processes and counterfeit products is collected by gas chromatography, ion chromatography and / or three-dimensional fluorescence spectroscopy technology as sample data. The sample data includes data collected by gas chromatography, ion chromatography and / or three-dimensional fluorescence spectroscopy on wine samples.

[0074] For example, the chromatographic data and / or spectral data of Daqu Maotai-flavor liquor (i.e. Maotai-flavor liquor (Daqu)), non-Daqu Maotai-flavor liquor (i.e. Maotai-flavor liquor (other)) and imitation Maotai-flavor liquor are collected as sample data.

[0075] Further, the collected sample data is pretreated. Specifically, the sample data can be pretreated by using a chemometrics modeling analysis module, i.e. a chemometrics modeling analysis software capable of running on a computer device.

[0076] For example, for chromatographic data, data preprocessing such as chromatographic integration, background subtraction, smoothing and baseline correction can be performed, and retention time correction can be performed by chromatographic peak alignment.

[0077] For the obtained complete three-dimensional fluorescence spectrum information, i.e. three-dimensional fluorescence spectrum data, the three-dimensional fluorescence spectrum of the sample can be corrected by using a standard sample to eliminate the influence of the instrument and the environment on the sample spectrum. Spline interpolation algorithm is used for background subtraction to eliminate the influence of Rayleigh and Raman peaks, and the pretreated three-dimensional fluorescence spectrum data, i.e. sample data, is obtained.

[0078] Finally, the pretreated sample data is input into an initial discrimination model based on a graph neural network (GNN) or partial least squares discriminant analysis (PLS-DA), and the initial discrimination model is then trained, and finally a target discrimination model is constructed, which is used to output a discrimination result of the wine sample to be discriminated, indicating the process category and authenticity of the wine sample to be discriminated, such as indicating Daqu Maotai-flavor liquor, non-Daqu Maotai-flavor liquor, and / or imitation Maotai-flavor liquor, i.e., Maotai-flavor liquor (Daqu), Maotai-flavor liquor (other), and / or imitation Maotai-flavor liquor.

[0079] When the output process category is imitation Maotai-flavor liquor, it indicates that the wine sample to be discriminated is a counterfeit Maotai-flavor liquor.

[0080] It can be understood that in the embodiments of the present application, the sample data including chromatographic data and / or spectral data of Maotai-flavor liquors and counterfeits produced by various processes are collected by using GC, IC, and / or EEM methods, and when the process category and authenticity discrimination model is constructed, the corresponding target discrimination model can be constructed based on the sample data collected in different ways to generate a gas chromatography discrimination model, an ion chromatography discrimination model, and / or a three-dimensional fluorescence spectrum discrimination model.

[0081] It can be understood that in the embodiments of the present application, the sample data including chromatographic data and / or spectral data of Maotai-flavor liquors and counterfeits produced by various processes are collected by using GC, IC, and / or EEM methods, and then the sample data is pretreated based on ModelLab software, and finally the pretreated data is input into an initial discrimination model based on a graph neural network or partial least squares discriminant analysis to train the initial discrimination model, and finally a target discrimination model corresponding to each sample data is constructed, so that after the collected data of the wine sample to be discriminated is input into each target discrimination model, a discrimination result indicating the process category and authenticity attribute can be quickly and accurately output.

[0082] That is, in the present application, GC, IC, and EEM multiple analysis techniques are used in combination with chemometric modeling analysis modules, which can effectively improve the accuracy of the prediction results by verifying the analysis results of different techniques with each other, and realize the visualization display of the differences between Maotai-flavor liquors of different process categories and authenticity.

[0083] Optionally, in combination with Figure 2As shown, in some embodiments of the present application, in S110, for the acquisition of sample data, the gas chromatograph, ion chromatograph and three-dimensional fluorescence spectrometer can be used respectively to collect the wine sample data during data collection.

[0084] For example, the Nexis GC-2030 gas chromatograph, Dionex Integrion HPIC high-pressure ion chromatograph and Ultra Fluor 1000 three-dimensional fluorescence spectrometer can be used to collect data of the wine sample.

[0085] Specifically, the following methods can be used:

[0086] That is, during data collection, sample preparation can be performed first.

[0087] For example, 10.0 mL of the wine sample to be tested can be accurately taken into a 20 mL test tube, 0.1 mL of 2% (v / v) tert-amyl alcohol, n-pentyl acetate and 2-ethylbutyric acid 3-component internal standard mixed 50% (v / v) ethanol aqueous solution is added, and the mixture is sealed, shaken, mixed, and allowed to stand and balance, and then used for GC analysis.

[0088] 1.0 mL of the wine sample to be tested can be accurately taken into a 50 mL volumetric flask, 10 μL of 2% (v / v) 2-ethylbutyric acid internal standard 50% (v / v) ethanol aqueous solution is added, and the volume is adjusted with secondary deionized water, and the mixture is shaken, stabilized and then used for IC analysis.

[0089] 3.0 mL of the wine sample to be tested can be taken and placed in a cuvette, and the sample temperature is controlled at room temperature (25℃), and then used for EEM analysis.

[0090] Further, the detection conditions can be determined.

[0091] For example, for gas chromatography detection (GC): DB-Wax UI chromatographic column (30 m x 0.25 mm x 0.25 μm); temperature program: initial temperature 28℃, hold for 12 min, increase to 125℃, hold for 8 min, increase to 240℃, hold for 8 min; injection port temperature 250℃; FID detector temperature 250℃; carrier gas flow rate 0.5 mL / min; carrier flow mode is constant flow; split injection (split ratio, 45:1); injection volume 1 μL.

[0092] For ion chromatography detection (IC): Dionex IonPac AS11-HC 4x250mm anion chromatographic column; EGC500KOH eluent automatic generator; AERS 4mm self-circulation suppressor, suppression current 104mA; flow rate: 0-40min flow rate 1.0mL / min, 40-47min flow rate 0.8mL / min, 47-50min flow rate 1.0mL / min; injection volume 25.0μL, dilution factor 50.0; detector: conductivity detector, detection cell temperature 35℃.

[0093] For three-dimensional fluorescence detection (EEM): excitation wavelength λex: 200-600nm, bandwidth 5nm; emission wavelength λem: 200-600nm, bandwidth 5nm; photomultiplier tube (PMT) voltage: 600V; sample temperature 25℃.

[0094] Then after the preparation of the samples of different process categories and authentic and fake Maotai-flavor liquor as described above, the prepared liquor samples can be subjected to data acquisition by using a gas chromatograph, an ion chromatograph and a three-dimensional fluorescence spectrometer, for example, a Nexis GC-2030 gas chromatograph, a Dionex Integrion HPIC high-pressure ion chromatograph and an Ultra Fluor 1000 three-dimensional fluorescence spectrometer, respectively.

[0095] For example, the representative sample chromatograms of the samples of each process and fake Maotai-flavor liquor acquired by the gas chromatograph are shown in FIG. 1. Figure 3 Among them, Figure 3 are the gas chromatograms of the samples of Maotai-flavor liquor (Daqu), Maotai-flavor liquor (other) and imitation Maotai-flavor liquor.

[0096] For example, the representative sample chromatograms of the samples of each process and fake Maotai-flavor liquor acquired by the gas chromatograph are shown in FIG. 1. Figure 4 Among them, are the ion chromatograms of the samples of Maotai-flavor liquor (Daqu), Maotai-flavor liquor (other) and imitation Maotai-flavor liquor.

[0097] Figure 5 For example, the representative sample chromatograms of the samples of each process and fake Maotai-flavor liquor acquired by the gas chromatograph are shown in FIG. 1.

[0098] It can be understood that the sample data corresponding to the samples of different process categories and authentic and fake Maotai-flavor liquor acquired by the above-mentioned manner can specifically include chromatographic data and spectral data, and then the acquired sample data can be preprocessed by using a chemometrics modeling analysis module, i.e. a chemometrics modeling analysis software capable of running on a computer device, such as ModelLab Chorman and ModelLabSpecman3D.

[0099] That is, in some embodiments, in combination withFigure 2 As shown in S120, for the sample data collected by GC and IC, the pretreatment based on the chemometrics modeling analysis software, i.e. ModelLab Chorman, can specifically include the following steps:

[0100] S121, the chromatographic peak integration, background deduction and baseline correction processing are performed on the chromatographic data.

[0101] S122, the retention time alignment is performed on the chromatographic peaks of the sample with the drift of the retention time of the compounds in the sample data after the chromatographic peak integration, background deduction and baseline correction processing.

[0102] Specifically, for the sample data collected by GC and IC, the ModelLab Chorman general scientific big data chemometrics modeling software can be respectively batch imported, the respective solutions are established, and the data processing is performed. The chromatographic peak integration, background deduction and baseline correction and other data preprocessing operations are performed on the GC and IC chromatographic data, and the retention time alignment is performed on the chromatographic peaks of the sample with the drift of the retention time of the compounds.

[0103] Optionally, in some embodiments, for the three-dimensional fluorescence spectral data collected by the three-dimensional fluorescence spectrometer, the pretreatment can be performed by the following way:

[0104] S123, the spectral data is corrected by using the standard sample.

[0105] S124, the corrected spectral data is scattered deducted by using the Spline spline interpolation algorithm.

[0106] Specifically, all the collected EEM data can be batch imported into the ModelLab Specman3D chemometrics modeling software for data processing. The three-dimensional fluorescence spectrum of the sample is corrected by using the QC sample, i.e. the standard sample, to eliminate the influence of the instrument and the environment on the sample spectrum. The Spline spline interpolation algorithm is used for scattering deduction to eliminate the influence of Rayleigh and Raman peaks, while retaining the complete three-dimensional fluorescence spectral data.

[0107] It can be understood that the standard sample can be a solid control product with strong stability, such as polystyrene. The present application does not make any limitation.

[0108] As shown in Figure 6 (a) and Figure 6 (b), the results before and after the three-dimensional fluorescence background and scattering processing are shown. As shown in Figure 6(a)It can be seen that there are first-order Rayleigh scattering, second-order Rayleigh scattering and Raman scattering peaks in the three-dimensional fluorescence spectrum before processing. As can be seen from 6(b), after processing by the scattering subtraction algorithm, a very "clean" spectrum result is obtained.

[0109] It can be understood that by the pre-processing mode of each of the above embodiments, interference factors can be eliminated to improve the training speed of the identification model, and the target identification model constructed can output accurate identification results.

[0110] It can also be understood that the graph neural network is a deep learning algorithm specially used for processing graph structure data, which mainly learns the feature representation of the nodes in the graph and the relationship of the edges, so as to perform node classification, link prediction, graph classification and other tasks.

[0111] Correspondingly, the initial identification model constructed based on the graph neural network can include an initial feature processing module and an initial classifier. The target identification model includes a corresponding target feature processing module and a target classifier.

[0112] Among them, the feature processing module can include a plurality of feature processing functions, such as message functions, aggregation functions, update functions, and readout functions.

[0113] It can also be understood that for the pre-processed sample data, such as the chromatographic data and three-dimensional fluorescence spectrum data of each sample, each sample data node is included. Each data node can correspond to multiple features, such as the data node of the chromatographic data of each sample corresponding to the retention time, peak area and other features, and the data node of the three-dimensional fluorescence spectrum data corresponding to the excitation wavelength, emission wavelength and fluorescence intensity.

[0114] In practice, the training of the constructed initial identification model can be specifically a process of iterating the parameter update of the feature processing module and the classifier by processing the features of each data node through the feature processing functions in the feature processing module and outputting the processing results.

[0115] Therefore, in some embodiments of the present application, when the pre-processed sample data is used to perform iterative training of the initial feature processing module, the following steps can be used to achieve the training:

[0116] S131, input the pre-processed sample data into the initial feature processing module, and output the initial feature representation of each data node. The initial feature processing module includes message functions, aggregation functions and update functions.

[0117] S132, update the parameters of each function in the initial feature processing module according to the initial feature representation, and output the target feature representation. The feature processing module with updated parameters is used as the target identification model.

[0118] Specifically, a graph neural network is used to construct a process category and authenticity identification model of Maotai-flavor liquor, and the model identification rate, i.e., the modeling quality, can be evaluated.

[0119] In practice, in the training and construction process of the model, the preprocessed sample data can be first divided into two groups, such as in a 7:3 ratio, into a training set and a test set.

[0120] Further, the training set can be used as training data and input into the initial identification model for model training. That is, the training data is input into the initial feature processing module in the initial identification model, so that each function in the initial feature processing module processes each feature of the data node, and finally outputs the initial feature representation of the data node.

[0121] Further, the output initial feature representation can be used to adjust the parameters of each function in the initial feature processing module to iteratively train each initial feature processing module until the iteration condition is met, such as reaching the number of iterations, and finally output the target feature representation.

[0122] It can be understood that when the iteration condition is met, the corresponding feature processing module is the target identification model, which can be used for actual process category and authenticity identification of Maotai-flavor liquor.

[0123] It can also be understood that the identification model also includes a classifier, so that after inputting the feature representation of each data node output by the feature processing module into the classifier, the final identification result can be output.

[0124] That is, in the construction process of the identification model, the parameter updating process of the classifier can also be included.

[0125] It can be understood that in the embodiments of the present application, through the algorithm process based on the graph neural network, the node classification, link prediction, and graph classification tasks are mainly performed by learning the feature representation of the nodes in the graph and the relationship of the edges.

[0126] Correspondingly, one training of the identification model in the above embodiments, such as the first training in S131, can include the following steps:

[0127] S01, extracting feature data in sample data.

[0128] S02, constructing an adjacency matrix of each data node according to the feature data to obtain a neighbor node set of each data node, the adjacency matrix representing the connection relationship between the data nodes.

[0129] S03, determining an initial feature representation of each data node based on each function in the initial feature processing module and the connection relationship.

[0130] Specifically, after the pre-processing of the sample data, the sample data can include feature data, and then feature extraction can be performed on the feature data.

[0131] For example, assuming that the sample data set is

[0132] wherein, X i is the data of the i-th sample, y i is the corresponding process label, indicating the identification result.

[0133] As can be set, y i = 1 represents Maotai-flavor liquor (Daqu), y i = 2 represents Maotai-flavor liquor (other), y i = 3 represents imitation Maotai-flavor liquor, i.e. pseudo Maotai-flavor liquor.

[0134] Then, for n sample data, each sample i has j data nodes, and each data node ij corresponds to multiple features.

[0135] It can be understood that the data node is the collected sample data, i.e. each sample data can be taken as a data node.

[0136] For example, as shown in Figure 3 to 4 , the collected chromatographic data can include the intensity corresponding to each retention time, i.e. the intensity corresponding to each retention time as a data node. Figure 3 to 4 It can be known that the intensity has multiple peak values at different retention times.

[0137] Then, correspondingly, each data node of each chromatographic data can include the retention time t ij , peak area a ij and the like.

[0138] Similarly, for the three-dimensional fluorescence spectrum data as shown in the figure, the excitation wavelength corresponding to each emission wavelength can be included, i.e. the excitation wavelength corresponding to each emission wavelength as a data node.

[0139] Correspondingly, the features corresponding to the data nodes of each three-dimensional fluorescence spectrum can include the excitation wavelength λ ex,ij , emission wavelength λ em,ij and fluorescence intensity I ij .

[0140] Correspondingly, the data node feature vector x 0 ij of each sample is Correspondingly, the node feature matrix of all samples is

[0141] The connection relationship between the data nodes can be represented by the constructed adjacency matrix A.

[0142] For example, A ij,kl = 1 can represent that the nodes ij and kl are connected by an edge, otherwise 0, representing no connection.

[0143] Further, in S03, according to the extracted feature data of each data node, the feature representation of each data node is determined based on the feature processing function and the connection relationship, which can specifically include the following steps:

[0144] S001, based on the message passing function, determining the neighbor node message of each data node.

[0145] S002, according to the neighbor node message, based on the aggregation function, determining the aggregation message of each data node.

[0146] S003, based on the update function and the aggregation message, updating the feature data to obtain the initial feature representation corresponding to each data node.

[0147] Specifically, after determining the connection relationship of each data node, the data nodes can be processed by setting various functions.

[0148] For example, for the data node ij, if the neighbor data node set is N(ij), then the message m 1 kl received from the neighbor data node kl, can be calculated by the message function M 1 , as shown in the following formula:

[0149]

[0150] Where W 1 m is the weight matrix of the message function.

[0151] Then the data node ij aggregates the received messages, for example, using the summation aggregation function, the aggregated message is:

[0152]

[0153] The data node ij updates its own feature according to the aggregated message, and the update function U 1 may be a linear transformation including an activation function, which is represented as follows:

[0154]

[0155] Where W 1 u is the weight matrix of the update function, and σ is the activation function, x 0 ij and a 0 ij The concatenation is performed.

[0156] It can be understood that for a multi-layer GNN, the data node feature is updated by iterating the above one-layer process, and after L-layer update, the final data node feature representation x L ij .

[0157] That is, in S132, through iterative training, the parameters of each function are updated synchronously in the process of continuously updating the feature representation of the data node, so as to realize the construction of the discrimination model.

[0158] Further, the readout function R X converts the node features of the graph into a graph-level representation, such as using a summation operation:

[0159]

[0160] where N X is the total number of data nodes.

[0161] It is input into the classifier for process category and authenticity discrimination. Assuming that the weight matrix of the classifier is W c Y , and the bias is b c Y , then the prediction probability is:

[0162]

[0163]

[0164] It can be understood that through the above embodiments, a gas chromatography discrimination model for gas chromatography data, an ion chromatography discrimination model for ion chromatography data, and a three-dimensional fluorescence spectrum discrimination model for three-dimensional fluorescence spectrum data can be constructed based on a graph neural network.

[0165] For example, as shown in Figure 7 (a) and Figure 7 (b), for the collected gas chromatography data of wine samples, a graph neural network (GNN) is used to model and cross-validation analysis of three different processes and authentic and fake Maotai-flavor liquor samples. The cross-validation classification accuracy of the constructed model is 100%.

[0166] For example, as shown in Figure 8 (a) and Figure 8(b) As shown, for ion chromatography data, GNN is used to model and cross-validation analysis of Maotai-flavor liquor samples of three different processes and authenticity. The cross-validation classification accuracy of the model is 100%.

[0167] For example, as shown in Figure 9 (a) and Figure 9 (b), for three-dimensional fluorescence spectrum data, GNN is used to model and cross-validation analysis of Maotai-flavor liquor samples of three different processes and authenticity. The cross-validation classification accuracy of the model is 100%.

[0168] Alternatively, in some embodiments, the discriminant model can also be constructed based on partial least squares discriminant method.

[0169] It can be understood that the partial least squares discriminant method is a multivariate data analysis method, mainly used to solve the data modeling problem of multiple collinearity between variables. The partial least squares discriminant method is a special form of partial least squares regression, which is realized when the classification data is used as the response variable, and it uses the binary Y matrix for regression modeling, which can be used for pattern recognition.

[0170] In some embodiments, based on the partial least squares discriminant method for constructing the discriminant model, factor decomposition can also be performed for three-dimensional fluorescence spectrum data.

[0171] Specifically, the data represented by three-dimensional fluorescence has high-dimensional collinearity, and a sample is described by an I×J×K data array, and the high-dimensional matrix composed of all components can be expressed as follows, that is, a high-order linear component model:

[0172]

[0173] Where x ijk is a matrix element in X(I×J×K), corresponding to the fluorescence intensity of the i-th excitation wavelength and the j-th emission wavelength at k; N is the number of load matrix array, that is, the number of factors; a in , b jn , c kn are elements in excitation matrix A, emission matrix B, and relative concentration matrix C, respectively; e ijk is an element of high-dimensional residual number array E, with a size of I×J×K. The high-order linear component model includes three parts: known (target) analyte, unknown (or uncorrected) substance, and background interference (noise).

[0174] Further, the high-order loop factor decomposition (HILOFAC) algorithm modeling:

[0175] In order to make full use of A IN , B JN and C KNThree matrices are accurately factorized, and the above three linear models are linearly decomposed based on SVD decomposition (HILOFAC), as follows:

[0176] FM(:,:,n)=A(:,n)*C(k,n)*B(:,n) T

[0177] a (i) T =((C+X i.. B). / (B T B)+(B+X i.. C). / (C T C))

[0178] c (k) T =(B+X ..k A). / (A T A)+(A+X ..k B). / (B T B))

[0179] b (j) T =(A+X .j. C). / (C T C)+(C+X .j. A). / (A T A))

[0180] In the formula, FM k (I×J×N) three-dimensional matrix, N represents the number of components, k represents the sample, and the subscript of all rows is referred to as the current matrix, that is, all rows are involved in the calculation. Unlike the past parallel factorization algorithm (PARAFAC), to avoid the appearance of imaginary numbers and sensitivity to the number of factors, the algorithm introduces tensor singular value decomposition (tensor SVD) to perform high-order cyclic decomposition on three linear data, so that a (i) , b (j) , c (k) can be obtained by cyclically minimizing the loss function. Generally, after a limited number of cycles (usually <10), the matrix decomposition result that satisfies the residual error condition can be obtained.

[0181] It can be understood that when the discriminant model is constructed based on the partial least squares discriminant method, for three-dimensional fluorescence spectrum data, the factor decomposition of the preprocessed three-dimensional fluorescence spectrum data can be performed first, that is, the A, B and C matrix results after factor decomposition can be directly used for subsequent model training or qualitative discrimination, that is, process category and authenticity discrimination.

[0182] For gas chromatography data and ion chromatography data, the pre-processed sample data can be directly used for model training and construction.

[0183] Optionally, in some embodiments, constructing the discrimination model based on the partial least squares discrimination method can specifically include the following steps:

[0184] S133, extracting a feature matrix in the pre-processed sample data as an independent variable matrix;

[0185] S134, performing principal component analysis on the independent variable matrix and the category matrix to obtain a corresponding score matrix;

[0186] S135, inputting the score matrix into an initial linear regression function to perform multiple linear regression to determine a regression coefficient matrix, and a linear regression function corresponding to the target regression coefficient matrix is taken as a target discrimination model.

[0187] Specifically, for the pre-processed sample data such as gas chromatography data, ion chromatography data and three-dimensional fluorescence spectrum data, feature extraction can be performed first to obtain a corresponding feature matrix as an independent variable matrix in the algorithm, which can be represented as X.

[0188] Correspondingly, for the prediction result, i.e., the process category and authenticity matrix of Maotai-flavor liquor, i.e., the dependent variable matrix, it can be represented as Y.

[0189] Further, the independent variable matrix and the dependent variable matrix can be subjected to principal component analysis to determine a corresponding score matrix.

[0190] In practice, the basic model of PLS-DA can be represented as:

[0191] X = TP + E

[0192] Y = UQ + F

[0193] Wherein, X is the feature matrix of the sample data, i.e., the independent variable matrix described above, Y is the process category and authenticity matrix of Maotai-flavor liquor, i.e., the dependent variable matrix, T and U are the score matrices obtained by principal component decomposition of X and Y, P and Q are the load matrices of the principal component decomposition of X and Y, and E and F are error matrices.

[0194] Further, after the above steps, multiple linear regression can be performed on the obtained score matrix, i.e., the regression coefficient matrix is calculated by using a linear regression function.

[0195] The linear regression function is represented as:

[0196] U = TB

[0197] Wherein, the regression coefficient matrix can be represented as:

[0198] B=(X T X) -1 X T Y

[0199] That is, after obtaining the score matrix, a discriminant function can be established to use the extracted PLS components as new feature variables. The discriminant function is usually in the form of a linear combination, that is, each component is multiplied by a regression coefficient, and then the contributions of all components are accumulated to obtain the final discriminant score.

[0200] It can be understood that the corresponding linear regression function at this time is the target identification model.

[0201] It can also be understood that through the above execution process, the target identification model constructed can be used to identify the process category and authenticity of the wine sample to be identified based on the constructed target identification model. That is, the score matrix T' of the fluorescence intensity matrix X' of the wine sample to be identified can be calculated based on the matrix P, thereby obtaining the predicted category Y' of the wine sample to be identified, and Y' satisfies the following equation:

[0202] Y′=T′BQ

[0203] Among them, Y' is the identification result, which indicates the process category and authenticity of the wine sample to be identified, such as sauce-flavored liquor (Daqu), sauce-flavored liquor (others) or imitation sauce-flavored liquor (indicating counterfeit sauce-flavored liquor).

[0204] For example, Figure 10 As shown in Figures (a)-10(d), using gas chromatography data as an example, supervised partial least squares discriminant analysis (PLS-DA) was used to model and cross-validate three different process categories of Maotai-flavor liquor: Maotai-flavor liquor (Daqu), Maotai-flavor liquor (other), and Maotai-flavored liquor. The developed model achieved a 100% classification accuracy rate.

[0205] Considering that too many sample groups will lead to poor model training results, a two-class classification model is established for each group and the other two groups of samples. The results show that the accuracy of the two-class classification model for each category of wine samples can reach 100%.

[0206] in, Figure 10 (a) Displays the statistics of the model's prediction accuracy for samples in each classification using the cross-validation method. The samples located on the diagonal of the matrix in the figure are correctly predicted samples, while samples that are misclassified by the model or cannot be attributed are displayed in a position that deviates from the diagonal. This allows for intuitive observation of the specific statistics of the accuracy rate, rejection rate (false negative), and false positive rate of each classification;

[0207] PLS discriminant analysis score plot ( Figure 10(b) shows the distribution trend of samples of different categories in the score plot composed of the first two PLS latent variables;

[0208] Figure 4 shows the predicted quality trend graph of the partial least squares regression model Figure 10 (c) shows the trend of the predicted variance explanation rate of the independent variable X and the dependent variable Y of the model with the number of latent variables used for modeling, and the number of latent variables corresponding to the position where the explanation rate is close to 1 (the highest point of the curve) is the optimal parameter for modeling;

[0209] Figure 5 shows the predicted classification probability distribution graph of the partial least squares discriminant model Figure 10 (d) shows the probability density function of the predicted value of the model for different categories, and the sharper and less overlapping the peak shape is, the higher the prediction accuracy of the model for the grouped samples of the category.

[0210] For example, as shown in Figure 2, for ion chromatography data, PLS-DA was used to model and cross-validation analysis of three different process categories and true and false Maotai-flavor liquor samples. The cross-validation classification accuracy of the model is 90.74%, but the two-class model accuracy of each process category of liquor samples can be improved to 100%. Figure 11 For example, as shown in Figure 3, for three-dimensional fluorescence spectrum data, PLS-DA was used to model and cross-validation analysis of three different process categories and true and false Maotai-flavor liquor samples. The cross-validation classification accuracy of the model is 92.59%, but the two-class model accuracy of each process category of liquor samples can be improved to 100%.

[0211] Figure 12 It can be understood that for the identification model constructed by the above steps, the test set in the collected data can also be used for verification.

[0212] In practice, the data of three different process categories and true and false Maotai-flavor liquor samples can be used as a training set to establish a process category and true and false identification model (A, B, and C processes represent Maotai-flavor liquor (Daqu), Maotai-flavor liquor (other), and imitation Maotai-flavor liquor, respectively), and 10 known process liquor samples are randomly taken as a test set, which are respectively substituted into six models to output results, and the reliability of the model is verified.

[0213] Table 1 shows the prediction results of the partial least squares discriminant model (GC, IC, and EEM data)

[0214] Table 1 shows the prediction results of the partial least squares discriminant model (GC, IC, and EEM data)

[0215]

[0216] Table 2 shows the prediction results of the graph neural network model (GC, IC, and EEM data)

[0217] ​

[0218] It can be understood that the process category and authenticity identification model construction method of Maotai-flavor liquor provided in the application adopts GC, IC and / or EEM method to collect sample data of liquor samples including chromatographic data and / or spectral data, and then respectively preprocesses various sample data based on a chemometrics modeling analysis module such as ModelLab, and finally inputs the preprocessed data into an initial identification model based on a graph neural network or a partial least squares discrimination method to train the initial identification model as training data, and finally constructs a target identification model corresponding to each type of sample data.

[0219] After the target identification model is constructed by the above method, the constructed target identification model can be used to identify the process category and authenticity of the actual liquor sample to be identified.

[0220] That is, on the other hand, as shown in Figure 13 The application also provides a Maotai-flavor liquor process category and authenticity identification method, which can specifically include:

[0221] S210, collecting data of a liquor sample to be identified, the data being collected by a gas chromatograph, an ion chromatograph and / or a three-dimensional fluorescence spectrometer, the liquor sample to be identified including Daqu Maotai-flavor liquor, non-Daqu Maotai-flavor liquor and / or imitation Maotai-flavor liquor, and the collected data including chromatographic data and / or spectral data corresponding to the liquor sample to be identified.

[0222] S220, preprocessing the collected data based on a chemometrics modeling analysis module to obtain preprocessed collected data.

[0223] S230, inputting the preprocessed collected data into a pre-constructed target prediction model to output an identification result, the identification result indicating the process category and authenticity of the liquor sample to be identified, the process category including Daqu Maotai-flavor liquor, non-Daqu Maotai-flavor liquor and / or imitation Maotai-flavor liquor, the imitation Maotai-flavor liquor indicating that the liquor sample to be identified is a counterfeit Maotai-flavor liquor, and the target model including a gas chromatograph identification model, an ion chromatograph identification model and / or a three-dimensional fluorescence spectrometer identification model.

[0224] Specifically, in the embodiments of the application, in order to accurately identify the process category and authenticity of Maotai-flavor liquor, for the liquor sample to be identified, the collected data of each liquor sample can be collected by a gas chromatograph, an ion chromatograph and / or a three-dimensional fluorescence spectrometer, and the collected data includes chromatographic data and / or spectral data.

[0225] Further, the collected data can be preprocessed.

[0226] It can be understood that the preprocessing of the collected data in this embodiment is similar to the preprocessing process in the construction of the above-mentioned identification model, and will not be repeated here.

[0227] Finally, the preprocessed collected data can be input into the target identification model constructed in the above-mentioned embodiments.

[0228] For example, the data collected by the gas chromatograph can be input into the gas chromatograph identification model, the data collected by the ion chromatograph can be input into the ion chromatograph identification model, and the data collected by the three-dimensional fluorescence spectrometer can be input into the three-dimensional fluorescence spectrum identification model to analyze the collected data and output the final identification result.

[0229] For example, taking GC data as an example, a graph neural network (GNN) is used to model and cross-validation analysis of different process categories and true and false Maotai-flavor liquor. The classification accuracy of the constructed model is 100%.

[0230] Based on IC data, GNN is used to model and cross-validation analysis of different process categories and true and false Maotai-flavor liquor samples. The classification accuracy of the constructed model is 100%.

[0231] Based on EEM data, GNN is used to model and cross-validation analysis of different process categories and true and false Maotai-flavor liquor samples. The classification accuracy of the constructed model is 100%.

[0232] It can be understood that by using GC, IC and / or EEM methods to collect sample data including chromatographic data and / or spectral data of different process categories and true and false Maotai-flavor liquor, and then based on ModelLab software, each sample data is preprocessed, and finally the preprocessed data is input into the initial identification model based on graph neural network or partial least squares discriminant method to train the initial identification model as training data, and finally the target identification model corresponding to each sample data is constructed, so that after the collected data of the wine sample to be identified is input into each target identification model, the identification result representing the process category and true and false characteristics can be quickly and accurately output.

[0233] That is, in this application, GC, IC and EEM multiple analysis techniques are used, and the chemometrics modeling analysis module can verify each other through different technical analysis results, effectively improving the accuracy of the prediction results, realizing the visualization display of the differences between different process categories and true and false Maotai-flavor liquor. And through the machine recognition model, the accurate discrimination of the process category and true and false of Maotai-flavor liquor is realized.

[0234] On the other hand, the application also provides a Maotai-flavor liquor process category and true and false identification model construction device, comprising:

[0235] The first acquisition module is configured to acquire sample data, wherein the sample data comprises collected data of liquor samples by using a gas chromatograph, an ion chromatograph and / or a three-dimensional fluorescence spectrometer, the liquor samples comprise Daqu Maotai-flavor liquor, non-Daqu Maotai-flavor liquor and / or imitation Maotai-flavor liquor, and the collected data comprises chromatographic data and / or spectral data corresponding to each Maotai-flavor liquor sample.

[0236] The first preprocessing module is configured to preprocess the sample data based on a chemometrics modeling analysis module to obtain preprocessed sample data.

[0237] The construction module is configured to input the preprocessed sample data into an initial discrimination model constructed based on a graph neural network or a partial least squares discrimination method, train the initial discrimination model to obtain a target discrimination model, and output a discrimination result, wherein the target discrimination model is used for process category and authenticity discrimination of an inputted liquor sample to be discriminated, the discrimination result represents the process category and authenticity of the liquor sample to be discriminated, the target discrimination model comprises a gas chromatographic discrimination model, an ion chromatographic discrimination model and / or a three-dimensional fluorescence spectral discrimination model, and the process category comprises Daqu Maotai-flavor liquor, non-Daqu Maotai-flavor liquor and / or imitation Maotai-flavor liquor.

[0238] Optionally, the application provides a Daqu Maotai-flavor liquor process category and authenticity discrimination model construction device, the collected data comprises chromatographic data, and the first preprocessing module is specifically configured to:

[0239] perform chromatographic peak integration, background deduction and baseline correction processing on the chromatographic data;

[0240] perform retention time alignment processing on chromatographic peaks of a sample with a compound retention time drift in the sample data after the chromatographic peak integration, background deduction and baseline correction processing.

[0241] Optionally, the application provides a Daqu Maotai-flavor liquor process category and authenticity discrimination model construction device, the collected data comprises three-dimensional fluorescence data, and the first preprocessing module is specifically configured to:

[0242] correct the three-dimensional fluorescence spectral data by using a standard sample;

[0243] perform scatter deduction on the corrected three-dimensional fluorescence spectral data by using a Spline spline interpolation algorithm.

[0244] Optionally, the application provides a Daqu Maotai-flavor liquor process category and authenticity discrimination model construction device, and the construction module is specifically configured to:

[0245] The preprocessed sample data is input into an initial feature processing module for processing, and initial feature representations of each data node are output, the initial feature processing module comprising a message function, an aggregation function and an update function;

[0246] Parameters of each function in the initial feature processing module are updated according to the initial feature representations, and a target feature representation is output, and the feature processing module after updating the parameters is used as a target discrimination model.

[0247] Optionally, the application provides a Jiangxiang Baijiu process category and authenticity discrimination model construction device, and the construction module is specifically configured to:

[0248] Feature data in the preprocessed sample data is extracted;

[0249] According to the feature data, an adjacency matrix of each data node is constructed, and a neighbor node set of each data node is obtained, and the adjacency matrix represents the connection relationship between each data node;

[0250] Based on each function in the initial feature processing module and the connection relationship, initial feature representations of each data node are determined.

[0251] Optionally, the application provides a Jiangxiang Baijiu process category and authenticity discrimination model construction device, and the construction module is specifically configured to:

[0252] Based on the message passing function, neighbor node messages of each data node are determined;

[0253] Based on the aggregation function, aggregation messages of each data node are determined according to the neighbor node messages; and based on the update function and the aggregation messages, feature data is updated to obtain initial feature representations corresponding to each data node.

[0254] Optionally, the application provides a Jiangxiang Baijiu process category and authenticity discrimination model construction device, and the construction module is specifically configured to:

[0255] The decomposition module is configured to perform factor decomposition on the three-dimensional fluorescence spectrum data.

[0256] Optionally, the application provides a Jiangxiang Baijiu process category and authenticity discrimination model construction device, and the construction module is specifically configured to:

[0257] A feature matrix in the preprocessed or factor-decomposed sample data is extracted as an independent variable matrix;

[0258] Principal component analysis is performed on the independent variable matrix and the category matrix to obtain a corresponding score matrix;

[0259] The score matrix is input into an initial linear regression function to perform multiple linear regression to determine a target regression coefficient matrix, and a linear regression function corresponding to the target regression coefficient matrix is taken as the target discrimination model.

[0260] In another aspect, the application also provides a Jiang-flavor liquor process category and authenticity discrimination device, comprising:

[0261] The second acquisition module is configured to acquire collection data of a to-be-discriminated liquor sample, wherein the collection data is collected by a gas chromatograph, an ion chromatograph, and / or a three-dimensional fluorescence spectrometer on the to-be-discriminated liquor sample, the to-be-discriminated liquor sample includes Daqu Jiang-flavor liquor, non-Daqu Jiang-flavor liquor, and / or imitation Jiang-flavor liquor, and the collection data includes chromatographic data and / or spectral data corresponding to the to-be-discriminated liquor sample.

[0262] The second preprocessing module is configured to preprocess the collection data based on a chemometrics modeling analysis module to obtain preprocessed collection data.

[0263] The discrimination module is configured to input the preprocessed collection data into a pre-constructed target prediction model to output a discrimination result, wherein the discrimination result represents a process category and authenticity of the to-be-discriminated liquor sample, the process category includes Daqu Jiang-flavor liquor, non-Daqu Jiang-flavor liquor, and / or imitation Jiang-flavor liquor, the imitation Jiang-flavor liquor indicates that the to-be-discriminated liquor sample is a counterfeit Jiang-flavor liquor, and the target model includes a gas chromatograph discrimination model, an ion chromatograph discrimination model, and / or a three-dimensional fluorescence spectrometer discrimination model.

[0264] In some exemplary embodiments, a computer device, which can be a server or a terminal, is provided, and an internal structure diagram of the computer device can be as shown in FIG. 1. Figure 14As shown in the figure. The computer device includes a processor, a memory, an input / output interface (I / O for short) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through the system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capability. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the model algorithm used in the identification of the category and authenticity of the Jiangxiang Baijiu process. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement the Jiangxiang Baijiu process category and authenticity identification model construction method or the Jiangxiang Baijiu process category and authenticity identification method.

[0265] Those skilled in the art can understand that, Figure 14 The skilled in the art can understand that,

[0266] In an exemplary embodiment, a computer device is also provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in the above method embodiments.

[0267] In an exemplary embodiment, a computer readable storage medium is provided, storing a computer program, and the computer program is executed by the processor to implement the steps in the above method embodiments.

[0268] In an exemplary embodiment, a computer program product is provided, including a computer program, and the computer program is executed by the processor to implement the steps in the above method embodiments.

[0269] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods, or the functions of each module in the above-mentioned determination apparatus, can be completed by a computer program instructing relevant hardware. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. Any combination of the technical features of the above embodiments can be made, and for the sake of brevity, not all possible combinations of the technical features in the above embodiments are described, however, as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0270] The principles and implementation modes of the present application are described by using specific examples in the present application, and the above-mentioned embodiments are only used to help understand the method and its core idea of the present application; at the same time, for those skilled in the art, according to the idea of the present application, the specific implementation mode and application range will be changed. In conclusion, the content of the present application should not be understood as a limitation.

Claims

1. A method for constructing a model for identifying the process classification and authenticity of Maotai-flavor liquor, characterized in that: The method comprises: Acquiring sample data, the sample data including data collected from liquor samples using a gas chromatograph, an ion chromatograph, and / or a three-dimensional fluorescence spectrometer, the liquor samples including Daqu sauce-flavored liquor, non-Daqu sauce-flavored liquor, and / or imitation sauce-flavored liquor, the collected data including chromatographic data and / or spectral data corresponding to each liquor sample; Preprocessing the sample data based on a chemometric modeling analysis module to obtain preprocessed sample data; The preprocessed sample data is input into an initial identification model constructed based on a graph neural network or a partial least squares discriminant method, and the initial identification model is trained to obtain a target identification model. The target identification model is used to identify the process category and authenticity of the input wine sample to be identified, and output an identification result. The identification result indicates the process category and authenticity of the wine sample to be identified. The target identification model includes a gas chromatography identification model, an ion chromatography identification model and / or a three-dimensional fluorescence spectrum identification model. The process category includes Daqu sauce-flavored liquor, non-Daqu sauce-flavored liquor and / or imitation sauce-flavored liquor. The imitation sauce-flavored liquor indicates that the wine sample to be identified is a counterfeit sauce-flavored liquor.

2. The method for constructing a model for identifying the process classification and authenticity of Maotai-flavor liquor according to claim 1, characterized in that: The collected data includes chromatographic data, and the chemometric modeling analysis module preprocesses the sample data to obtain the preprocessed sample data including: Performing chromatographic peak integration, background subtraction and baseline correction on the chromatographic data; The retention time alignment processing is performed on the chromatographic peaks of the samples whose compound retention time drifts in the sample data after chromatographic peak integration, background subtraction and baseline correction.

3. The method for constructing a model for identifying the process classification and authenticity of Maotai-flavor liquor according to claim 1, characterized in that: The collected data includes three-dimensional fluorescence spectrum data, and the chemometric modeling analysis module preprocesses the sample data to obtain the preprocessed sample data including: Correcting the three-dimensional fluorescence spectrum data using a standard sample; The spline interpolation algorithm was used to perform scattering subtraction on the corrected three-dimensional fluorescence spectrum data.

4. The method for constructing a model for identifying the process classification and authenticity of Maotai-flavor liquor according to claim 1, characterized in that: Inputting the preprocessed sample data into an initial identification model constructed based on a graph neural network, training the initial identification model, and obtaining a target identification model includes: Input the preprocessed sample data into the initial feature processing module and output the initial feature representation of each data node. The initial feature processing module includes a message function, an aggregation function and an update function; The parameters of each function in the initial feature processing module are updated according to the initial feature representation, and the target feature representation is output. The feature processing module after the parameter update is used as the target identification model.

5. The method for constructing a model for identifying the process classification and authenticity of Maotai-flavor liquor according to claim 4, characterized in that: The step of inputting the pre-processed sample data into the initial feature processing module and outputting the initial feature representation of each data node includes: Extracting feature data from preprocessed sample data; Constructing an adjacency matrix of each data node based on the feature data to obtain a set of neighbor nodes of each data node, wherein the adjacency matrix represents the connection relationship between the data nodes; Based on the functions in the initial feature processing module and the connection relationship, an initial feature representation of each data node is determined.

6. The method for constructing a model for identifying the process classification and authenticity of Maotai-flavor liquor according to claim 5, characterized in that: The determining of the initial feature representation of each data node based on each feature processing function in the initial feature processing module and the connection relationship includes: Determining neighbor node information of each data node based on the message passing function; Determining aggregated information of each data node based on the neighbor node information and the aggregation function; Based on the update function and the aggregate message, the feature data is updated to obtain an initial feature representation corresponding to each data node.

7. The method for constructing a model for identifying the process classification and authenticity of Maotai-flavor liquor according to claim 1, characterized in that: When the collected data includes three-dimensional fluorescence spectrum data, before inputting the pre-processed sample data into the initial identification model constructed based on the partial least squares discriminant method, the method further includes: The three-dimensional fluorescence spectrum data is factored.

8. The method for constructing a model for identifying the process classification and authenticity of Maotai-flavor liquor according to claim 7, characterized in that: Inputting the pre-processed sample data into an initial identification model constructed based on the partial least squares discriminant method, training the initial identification model, and obtaining a target identification model includes: Extract the characteristic matrix of the sample data after preprocessing or factor decomposition as the independent variable matrix; Performing principal component analysis on the independent variable matrix and the category matrix to obtain a corresponding score matrix; The score matrix is ​​input into an initial linear regression function, multiple linear regression is performed, and a target regression coefficient matrix is ​​determined. The linear regression function corresponding to the target regression coefficient matrix is ​​used as the target identification model.

9. A method for distinguishing the process type and authenticity of Maotai-flavor liquor, characterized in that: The identification method is performed by using the target identification model constructed in the method for constructing a process classification and authenticity identification model of Maotai-flavor liquor according to any one of claims 1 to 8, and the identification method comprises: Acquiring collected data of a wine sample to be identified, wherein the collected data is collected from the wine sample to be identified by a gas chromatograph, an ion chromatograph, and / or a three-dimensional fluorescence spectrometer, wherein the wine sample to be identified includes Daqu sauce-flavored liquor, non-Daqu sauce-flavored liquor, and / or imitation sauce-flavored liquor, and the collected data includes chromatographic data and / or spectral data corresponding to the wine sample to be identified; Preprocessing the collected data based on a chemometric modeling analysis module to obtain preprocessed collected data; The preprocessed collected data is input into a pre-built target prediction model, and an identification result is output, wherein the identification result indicates the process category and authenticity of the wine sample to be identified, wherein the process category includes Daqu sauce-flavored liquor, non-Daqu sauce-flavored liquor and / or imitation sauce-flavored liquor, and the imitation sauce-flavored liquor indicates that the wine sample to be identified is a counterfeit sauce-flavored liquor, and the target model includes a gas chromatography identification model, an ion chromatography identification model and / or a three-dimensional fluorescence spectrum identification model.

10. A computer device, characterized in that: The computer device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the computer program to implement the method for constructing a model for identifying the process category and authenticity of a sauce-flavored liquor as described in any one of claims 1 to 8 or the method for identifying the process category and authenticity of a sauce-flavored liquor as described in claim 9.

Images