Method and device for predicting quality parameters of fermented grains of Luzhou-flavor liquor and proportioning fermented grains

Abstract

The invention relates to the technical field of wine brewing, discloses a method and a device for predicting quality parameters of fermented grains of Luzhou-flavor liquor and proportioning the fermented grains, and aims to solve the problems of low efficiency and accuracy of the conventional method for determining the quality parameters of the fermented grains, and the scheme mainly comprises the following steps: establishing a multi-modal database of the fermented grains; constructing a multi-modal mapping model based on the multi-modal database, wherein the multi-modal mapping model at least comprises an image multi-modal mapping sub-model; establishing a mapping database of the fermented grains in and out of the cellar; constructing a cellar access fermented grain mapping model based on the mapping database; and determining image data of the to-be-predicted cellar-out fermented grains, and obtaining a quality parameter prediction result of the to-be-predicted cellar-out fermented grains and a quality parameter prediction result of the corresponding cellar-in fermented grains based on the image multi-modal mapping sub-model and the cellar-in fermented grain mapping model. The determination efficiency and accuracy of the quality parameters of the fermented grains are improved, and the method is suitable for producing Luzhou-flavor liquor.

Description

Technical Field

[0001] The present invention relates to the technical field of brewing, and particularly relates to a method and device for predicting quality parameters of Luzhou-flavor liquor fermented grains and adjusting ingredients according to the fermented grains. Background Art

[0002] The fermented grains taken out of the cellar and the fermented grains put into the cellar are two important stages in the production process of Luzhou-flavor liquor. The fermented grains taken out of the cellar refer to the fermented grains taken out of the cellar pool after fermentation. The fermented grains taken out of the cellar have undergone long-term fermentation and contain rich flavor substances, which are important raw materials for brewing liquor. The fermented grains put into the cellar refer to the fermented grains that are prepared to enter the cellar pool again for fermentation after a series of treatments. The quality parameters of the fermented grains taken in and out of the cellar are crucial for the formation of the liquor body flavor, directly affecting the quality, liquor yield and production efficiency of the liquor. Enterprises can also continuously optimize the ingredient and fermentation processes according to the quality parameters of the fermented grains taken in and out of the cellar to improve the stability and controllability of liquor production.

[0003] In the prior art, the quality parameters of the fermented grains taken in and out of the cellar usually include sensory data and physical and chemical data. Among them, the sensory data is mainly obtained by winemakers observing the sensory characteristics of the fermented grains such as color, shape, and smell. This method relies on the experience of winemakers and is easily affected by subjectivity, with poor efficiency and accuracy. The physical and chemical data is usually determined by corresponding detection methods. For example, the moisture of the fermented grains is detected by the drying method, the acidity is detected by the NaOH standard solution neutralization titration method, the reducing sugar is determined by the Fehling reagent method, and the starch is detected by the hydrochloric acid hydrolysis standard glucose solution back titration method. These methods all rely on manual detection, with cumbersome operation steps, strict detection condition limitations, high requirements for the experience and skills of the detection personnel, long time consumption, low efficiency, and difficult to meet the requirements of modern production for high efficiency, precision and real-time monitoring. In addition, the quality parameters of the fermented grains in the prior art often only include sensory data and physical and chemical data, which are difficult to fully reflect the mapping relationship between vision, sense, physical and chemical properties and texture, difficult to comprehensively reflect the quality of the fermented grains, and have low accuracy.

[0004] In addition, the prior art also includes the process of adjusting ingredients according to the fermented grains. Adjusting ingredients according to the fermented grains means adjusting the ingredient ratio of the fermented grains according to the quality parameters of the fermented grains taken out of the cellar. The commonly used method is to adjust the ingredient ratio of the fermented grains according to the experience of winemakers, production processes and regulations. However, the adjustment of the ingredient ratio of the fermented grains is a complex and delicate process that requires comprehensive consideration of multiple factors. This method has high requirements for the experience and skills of winemakers and also has the problem of strong subjectivity, resulting in low efficiency and accuracy. Summary of the Invention

[0005] The present invention aims to solve the problem of low efficiency and accuracy in the existing method for determining the quality parameters of fermented grains, and proposes a method and device for predicting the quality parameters of Luzhou-flavor liquor fermented grains and adjusting ingredients according to the fermented grains.

[0006] The technical solution adopted by the present invention to solve the above technical problems is as follows:

[0007] In a first aspect, the present invention provides a method for predicting the quality parameters of Luzhou-flavor fermented grains, the method comprising:

[0008] Establish a multi-modal database of fermented grains, where the multi-modal database includes image data of multiple fermented grain samples and their corresponding quality parameters, and the quality parameters include sensory data, texture data, and physicochemical data;

[0009] Construct a multi-modal mapping model based on the multi-modal database, where the multi-modal mapping model at least includes an image multi-modal mapping sub-model, and the image multi-modal mapping sub-model is used to predict the corresponding quality parameters according to the image data of the fermented grains;

[0010] Establish a mapping database of fermented grains entering and leaving the cellar, where the mapping database includes the quality parameters of the fermented grains leaving the cellar, the quality parameters of the fermented grains entering the cellar, and the fermented grain ingredient data;

[0011] Construct an in-out cellar fermented grain mapping model based on the mapping database, and the in-out cellar fermented grain mapping model is used to predict the quality parameters of the corresponding fermented grains entering the cellar according to the quality parameters of the fermented grains leaving the cellar and the fermented grain ingredient data;

[0012] Determine the image data of the fermented grains to be predicted leaving the cellar, input the image data of the fermented grains to be predicted leaving the cellar into the image multi-modal mapping sub-model, and obtain the quality parameter prediction result of the fermented grains to be predicted leaving the cellar;

[0013] Input the quality parameter prediction result of the fermented grains to be predicted leaving the cellar and the fermented grain ingredient data into the in-out cellar fermented grain mapping model, and obtain the quality parameter prediction result of the corresponding fermented grains entering the cellar.

[0014] Further, the multi-modal mapping model further includes a sensory multi-modal mapping sub-model, and the sensory multi-modal mapping sub-model is used to predict the corresponding texture data and physicochemical data according to the sensory data of the fermented grains;

[0015] The method further comprises:

[0016] Determine the sensory data of the fermented grains to be predicted leaving the cellar, input the sensory data of the fermented grains to be predicted leaving the cellar into the sensory multi-modal mapping sub-model, and obtain the texture data prediction result and the physicochemical data prediction result of the fermented grains to be predicted leaving the cellar;

[0017] Determine the sensory data of the fermented grains to be predicted entering the cellar, input the sensory data of the fermented grains to be predicted entering the cellar into the sensory multi-modal mapping sub-model, and obtain the texture data prediction result and the physicochemical data prediction result of the fermented grains to be predicted entering the cellar.

[0018] Furthermore, the multimodal mapping model further includes a coupled multimodal mapping sub-model, which is used to couple the image data of the mash with the sensory data to generate image-sensory coupling features, and predict the corresponding texture data and physical and chemical data according to the image-sensory coupling features;

[0019] The method further comprises:

[0020] Determine the sensory data of the to-be-predicted lees, input the image data and sensory data of the to-be-predicted lees into the coupled multimodal mapping sub-model, and obtain the texture data prediction results and the physicochemical data prediction results of the to-be-predicted lees;

[0021] The sensory data of the mash to be predicted to be put into the cellar is determined, and the image data and sensory data of the mash to be predicted to be put into the cellar are input into the coupled multimodal mapping sub-model to obtain the texture data prediction results and the physical and chemical data prediction results of the mash to be predicted to be put into the cellar.

[0022] Further, the image multimodal mapping submodel is a combination of an image feature extraction module, a sensory feature mapping module, a texture feature mapping module, and a physicochemical feature mapping module; the sensory multimodal mapping submodel is a combination of a sensory feature extraction module, a texture feature mapping module, and a physicochemical feature mapping module; the coupled multimodal mapping submodel is a combination of an image feature extraction module, a sensory feature extraction module, a feature coupling module, a texture feature mapping module, and a physicochemical feature mapping module;

[0023] The image feature extraction module is used to extract high-dimensional features of image data and process the high-dimensional features into a one-dimensional tensor of image features through a flattening operation;

[0024] The sensory feature extraction module is used to extract high-dimensional features of sensory data and generate a one-dimensional tensor of sensory features;

[0025] The feature coupling module is used to splice the one-dimensional tensor of image features and the one-dimensional tensor of sensory features, and then input the spliced ​​features into the feature coupling module. After the feature coupling module performs calculations, it outputs the image-sensory coupling features represented by the one-dimensional tensor.

[0026] The sensory feature mapping module, texture feature mapping module and physical and chemical feature mapping module are used to output sensory data prediction results, texture data prediction results and physical and chemical data prediction results according to the one-dimensional image feature tensor, the one-dimensional sensory feature tensor or the image-sensory coupling feature.

[0027] Furthermore, the image feature extraction module, sensory feature extraction module, feature coupling module, sensory feature mapping module, texture feature mapping module and physical and chemical feature mapping module are constructed based on a neural network, and the neural network is CNN, MLP, LSTM, RNN or 1D-CNN.

[0028] Furthermore, the image data is a color photo, and the shooting parameters corresponding to each color photo are the same;

[0029] The sensory data includes: visual data, olfactory data, taste data, tactile data and overall sensory data. The visual data includes: color data, appearance data, uniformity data, and grain gelatinization degree data; the olfactory data includes: wine aroma data, sourness data, ester aroma data, lees aroma data, cellar aroma data, grain aroma data and koji aroma data; the taste data includes: sourness data, astringency data and sweetness data; the tactile data includes: structure data and tenderness data;

[0030] The texture data include: hardness data, fracture data, adhesion data, elasticity data, cohesion data, viscosity data, chewiness data and recovery data;

[0031] The physical and chemical data include: moisture data, acidity data, starch content data and sugar content data;

[0032] The fermented mash ingredient data includes: weight data of a single steamer fermented mash, data on the amount of water added to moisten the grain, data on the amount of grain added, data on the amount of bran added, data on the amount of water used to beat the mash, data on the amount of koji added and data on the duration of steaming the grain.

[0033] In a second aspect, the present invention provides a device for predicting quality parameters of Luzhou-flavor liquor lees, the device comprising:

[0034] A database establishment module is used to establish a multimodal database of fermented grains, wherein the multimodal database includes image data of multiple fermented grains samples and their corresponding quality parameters, wherein the quality parameters include sensory data, texture data, and physical and chemical data; establish a mapping database of fermented grains entering and leaving the cellar, wherein the mapping database includes quality parameters of fermented grains leaving the cellar, quality parameters of fermented grains entering the cellar, and fermented grains ingredient data;

[0035] A mapping model construction module is used to construct a multimodal mapping model based on the multimodal database, the multimodal mapping model at least comprising an image multimodal mapping sub-model, the image multimodal mapping sub-model is used to predict the corresponding quality parameters according to the image data of the fermented grains; a mapping model for fermented grains entering and leaving the cellar is constructed based on the mapping database, the mapping model for fermented grains entering and leaving the cellar is used to predict the quality parameters of the corresponding fermented grains entering the cellar according to the quality parameters of the fermented grains leaving the cellar and the fermented grains ingredient data;

[0036] A prediction module, configured to determine the image data of the cellar-out fermented grains to be predicted, input the image data of the cellar-out fermented grains to be predicted into the image multi-modal mapping sub-model to obtain the prediction result of the quality parameters of the cellar-out fermented grains to be predicted; input the prediction result of the quality parameters of the cellar-out fermented grains to be predicted and the fermented grains ingredient data into the cellar-in and cellar-out fermented grains mapping model to obtain the prediction result of the quality parameters of the corresponding cellar-in fermented grains.

[0037] Thirdly, the present invention provides a method for ingredient matching of Luzhou-flavor liquor fermented grains by observing the fermented grains, and the method includes:

[0038] Constructing an ingredient matching recommendation model based on the image multi-modal mapping sub-model and the cellar-in and cellar-out fermented grains mapping model, where the image multi-modal mapping sub-model and the cellar-in and cellar-out fermented grains mapping model are the image multi-modal mapping sub-model and the cellar-in and cellar-out fermented grains mapping model in the quality parameter prediction method of Luzhou-flavor liquor fermented grains described in the first aspect;

[0039] Obtaining the image data of the cellar-out fermented grains of the fermented grains to be ingredient-matched by observing the fermented grains, inputting the image data of the cellar-out fermented grains of the fermented grains to be ingredient-matched by observing the fermented grains and the quality parameters of the target cellar-in fermented grains into the ingredient matching recommendation model for optimization, and obtaining the recommended fermented grains ingredient data of the fermented grains to be ingredient-matched by observing the fermented grains, where the quality parameters include sensory data, texture data, and physical and chemical data.

[0040] Further, the ingredient matching recommendation model is constructed based on an evolutionary algorithm, and the evolutionary algorithm is a genetic algorithm, a simulated annealing algorithm, or an evolutionary strategy algorithm;

[0041] The fitness function of the evolutionary algorithm is as follows:

[0042]

[0043] where Fitness represents fitness, obj j represents the quality parameter of the jth target cellar-in fermented grains, and pre j represents the quality parameter of the jth cellar-in fermented grains predicted based on the image multi-modal mapping sub-model and the cellar-in and cellar-out fermented grains mapping model, and M represents the number of quality parameters of the cellar-in fermented grains.

[0044] Fourthly, the present invention provides an ingredient matching device for Luzhou-flavor liquor fermented grains by observing the fermented grains, and the device includes:

[0045] A construction module, configured to construct an ingredient matching recommendation model based on the image multi-modal mapping sub-model and the cellar-in and cellar-out fermented grains mapping model, where the image multi-modal mapping sub-model and the cellar-in and cellar-out fermented grains mapping model are the image multi-modal mapping sub-model and the cellar-in and cellar-out fermented grains mapping model in the quality parameter prediction method of Luzhou-flavor liquor fermented grains described in the first aspect;

[0046] A recommendation module, which is used to obtain the image data of the cellar-exited fermented grains of the fermented grains to be visually inspected for ingredient preparation, input the image data of the cellar-exited fermented grains of the fermented grains to be visually inspected for ingredient preparation and the quality parameters of the target cellar-entered fermented grains into the visual inspection and ingredient preparation recommendation model for optimization, and obtain the recommended fermented grain ingredient data of the fermented grains to be visually inspected for ingredient preparation. The quality parameters include sensory data, texture data, and physical and chemical data.

[0047] The beneficial effects of the present invention are as follows: The method and device for predicting the quality parameters of the fermented grains of Luzhou-flavor liquor provided by the present invention can predict and obtain the quality parameters of the cellar-exited fermented grains and the corresponding cellar-entered fermented grains according to the image data of the cellar-exited fermented grains by constructing a multi-modal mapping model and an in-out cellar fermented grains mapping model, thereby improving the efficiency and accuracy of determining the quality parameters of the fermented grains. At the same time, the present invention also considers using texture data as the quality parameters of the fermented grains, and the predicted quality parameters can comprehensively evaluate the quality of the fermented grains, thereby improving the accuracy of the quality evaluation of the fermented grains. In addition, for the method and device for visually inspecting and preparing ingredients for the fermented grains of Luzhou-flavor liquor provided by the present invention, by constructing a visual inspection and ingredient preparation recommendation model, after setting the quality parameters of the target cellar-entered fermented grains, the recommended fermented grain ingredient data that meet the quality parameters of the target cellar-entered fermented grains can be sought, so as to adjust and optimize the fermented grain ingredients, thereby improving the efficiency and accuracy of the adjustment of the fermented grain ingredients. Through the present invention, the stability of the fermented grains fermentation and the quality consistency of the products can be ensured, thereby enhancing the flavor and market competitiveness of Luzhou-flavor liquor. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic flowchart of a method for predicting the quality parameters of the fermented grains of Luzhou-flavor liquor provided by an embodiment;

[0049] Figure 2 It is a schematic flowchart of another method for predicting the quality parameters of the fermented grains of Luzhou-flavor liquor provided by an embodiment;

[0050] Figure 3 It is a schematic diagram of the image data of the fermented grain sample provided by an embodiment;

[0051] Figure 4 It is a schematic diagram of the acidity prediction result based on the image multi-modal mapping sub-model provided by an embodiment;

[0052] Figure 5 It is a schematic structural diagram of a device for predicting the quality parameters of the fermented grains of Luzhou-flavor liquor provided by an embodiment;

[0053] Figure 6 It is a schematic flowchart of a method for visually inspecting and preparing ingredients for the fermented grains of Luzhou-flavor liquor provided by an embodiment;

[0054] Figure 7 It is a schematic flowchart of another method for visually inspecting and preparing ingredients for the fermented grains of Luzhou-flavor liquor provided by an embodiment;

[0055] Figure 8 Structural schematic diagram of the koji-based ingredient device for Luzhou-flavor liquor fermented grains provided for the embodiment. Detailed implementation manners

[0056] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the present embodiment will be clearly and completely described below in conjunction with the accompanying drawings in the present embodiment.

[0057] In some processes described in the specification of the present invention and the above-mentioned accompanying drawings, a plurality of operations appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order in which they appear in this article or may be executed in parallel. The serial numbers of the operations are only used to distinguish the different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel.

[0058] At present, the method for determining the quality parameters of fermented grains is mainly that the winemaker obtains sensory data through observation, and the tester obtains physical and chemical data by using corresponding detection methods. At present, the method of koji-based ingredient is mainly that the winemaker flexibly adjusts the ingredient ratio of fermented grains according to experience. The inventor has found through research that these methods have problems of low efficiency and accuracy.

[0059] Based on this, the technical solution of the present invention is proposed. In the present invention, a multi-modal database of fermented grains is established. The multi-modal database includes image data of multiple fermented grain samples and their corresponding quality parameters. The quality parameters include sensory data, texture data, and physical and chemical data. A multi-modal mapping model is constructed based on the multi-modal database. The multi-modal mapping model at least includes an image multi-modal mapping sub-model, and the image multi-modal mapping sub-model is used to predict the corresponding quality parameters according to the image data of the fermented grains. A mapping database of in-cellar and out-cellar fermented grains is established. The mapping database includes the quality parameters of out-cellar fermented grains, the quality parameters of in-cellar fermented grains, and fermented grain ingredient data. An in-cellar and out-cellar fermented grain mapping model is constructed based on the mapping database. The in-cellar and out-cellar fermented grain mapping model is used to predict the quality parameters of the corresponding in-cellar fermented grains according to the quality parameters of out-cellar fermented grains and fermented grain ingredient data. Determine the image data of the to-be-predicted out-cellar fermented grains, input the image data of the to-be-predicted out-cellar fermented grains into the image multi-modal mapping sub-model, and obtain the quality parameter prediction result of the to-be-predicted out-cellar fermented grains. Input the quality parameter prediction result of the to-be-predicted out-cellar fermented grains and the fermented grain ingredient data into the in-cellar and out-cellar fermented grain mapping model, and obtain the quality parameter prediction result of the corresponding in-cellar fermented grains.

[0060] In addition, based on the image multi-modal mapping sub-model and the in-out cellar fermented grains mapping model constructed by the above technical solution, the present invention further includes: constructing a fermented grains ingredient recommendation model based on the image multi-modal mapping sub-model and the in-out cellar fermented grains mapping model; obtaining the image data of the out-cellar fermented grains of the fermented grains to be visually inspected for ingredient determination, and inputting the image data of the out-cellar fermented grains of the fermented grains to be visually inspected for ingredient determination and the quality parameters of the target in-cellar fermented grains into the fermented grains ingredient recommendation model for optimization to obtain the recommended fermented grains ingredient data of the fermented grains to be visually inspected for ingredient determination, where the quality parameters include sensory data, texture data, and physicochemical data.

[0061] Specifically, the present invention constructs a multi-modal database of image data, sensory data, texture data, and physicochemical data, and constructs an image multi-modal mapping sub-model based on this, so as to realize the prediction from image data to quality parameters. By constructing a mapping database of the quality parameters of out-cellar fermented grains, the quality parameters of in-cellar fermented grains, and fermented grains ingredient data, and constructing an in-out cellar fermented grains mapping model based on this, the prediction between the quality parameters of fermented grains at different stages is realized. The present invention avoids the problems of cumbersome operation, high detection requirements, and subjectivity existing in manual observation and detection methods, and improves the efficiency and accuracy of determining the quality data of fermented grains. In addition, by constructing a fermented grains ingredient recommendation model, after setting the quality parameters of the target in-cellar fermented grains, the recommended fermented grains ingredient data that meet the quality parameters of the target in-cellar fermented grains can be sought, so as to adjust and optimize the fermented grains ingredients, thereby improving the efficiency and accuracy of fermented grains ingredient adjustment.

[0062] Next, the technical solutions in this embodiment will be clearly and completely described in conjunction with the accompanying drawings in this embodiment. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments.

[0063] Figure 1 and Figure 2 respectively show the schematic flowcharts of a method for predicting the quality parameters of fermented grains of Luzhou-flavor liquor. Please refer to Figure 1 and Figure 2 , and this method includes the following steps:

[0064] S101. Establish a multi-modal database of fermented grains, where the multi-modal database includes the image data of multiple fermented grains samples and their corresponding quality parameters, and the quality parameters include sensory data, texture data, and physicochemical data.

[0065] Please refer to Figure 3, in this embodiment, the image data is a color photo, and the shooting parameters corresponding to each color photo are the same. The same shooting parameters include the same spatial resolution, the same light source background, the same white balance, the same focal length, and aperture value. In practical applications, an image can be obtained by shooting with a camera, then compressed and cropped into a square, and then processed into a 256×256 RGB three-channel color image using the Python Imaging Library; of course, it can also be processed into other resolutions, such as 128×128 or 512×512, or even 64×64. These are all common image resolutions in convolutional neural network processing. However, it should be noted that the resolution of the image should be able to clearly distinguish the structure, color, and other characteristics of the fermented grains, and should not blindly reduce the resolution.

[0066] The sensory data includes: visual data, olfactory data, gustatory data, tactile data, and overall sensory data. The visual data includes: color data, appearance data, uniformity data, and degree of grain gelatinization data; the olfactory data includes: wine aroma data, sour gas data, ester aroma data, fermented grains aroma data, cellar aroma data, grain aroma data, and koji aroma data; the gustatory data includes: sourness data, astringency data, and sweetness data; the tactile data includes: structure data and softness and ripeness data. In practical applications, the sensory data can be obtained through questionnaires by multiple experienced brewing engineers in the field.

[0067] The texture data includes: hardness data, fracturability data, adhesiveness data, elasticity data, cohesiveness data, gumminess data, chewiness data, and resilience data. In practical applications, the texture data can be obtained by measuring with a texture analyzer.

[0068] The physical and chemical data includes: moisture data, acidity data, starch content data, and sugar content data. The physical and chemical data can be obtained through physical and chemical tests or a near-infrared physical and chemical analyzer.

[0069] This embodiment considers using sensory data, texture data, and physical and chemical data as quality parameters of the fermented grains, which can comprehensively evaluate the quality of the fermented grains and thus improve the accuracy of the quality evaluation of the fermented grains.

[0070] S102. Construct a multi-modal mapping model based on the multi-modal database. The multi-modal mapping model at least includes an image multi-modal mapping sub-model, and the image multi-modal mapping sub-model is used to predict the corresponding quality parameters according to the image data of the fermented grains.

[0071] In this embodiment, the image multimodal mapping submodel is a combination of an image feature extraction module, a sensory feature mapping module, a texture feature mapping module, and a physical and chemical feature mapping module. The image feature extraction module, the sensory feature mapping module, the texture feature mapping module, and the physical and chemical feature mapping module are constructed based on a neural network, and the neural network can be, but is not limited to, CNN, MLP, LSTM, RNN, or 1D-CNN.

[0072] Among them, the image feature extraction module is used to extract high-dimensional features of image data, and process the high-dimensional features into one-dimensional tensors of image features through flattening operations. Specifically, the image feature extraction module extracts high-dimensional features of image data based on convolutional neural networks. For example, the convolutional neural network processes a 256×256 three-channel image into a 6×6×1 three-dimensional tensor, which is the high-dimensional feature of the image data. After that, the 6×6×1 three-dimensional tensor can be directly processed into a one-dimensional tensor of length 36 through a flattening operation, which is the one-dimensional tensor of image features.

[0073] In the image multimodal mapping sub-model, the sensory feature mapping module is used to output the sensory data prediction results based on the one-dimensional tensor of image features, the texture feature mapping module is used to output the texture data prediction results based on the one-dimensional tensor of image features, and the physicochemical feature mapping module is used to output the physicochemical data prediction results based on the one-dimensional tensor of image features.

[0074] In practical applications, the image data in the multimodal database is used as input, and the corresponding quality parameters are used as true labels to train the neural network model. After the training is completed, the image multimodal mapping submodel can be obtained, and the image multimodal mapping submodel can predict the corresponding quality parameters according to the image data of the mash. The model training method belongs to the prior art and will not be described in detail in this embodiment.

[0075] S103, establishing a mapping database of the fermented grains entering and leaving the cellar, wherein the mapping database includes the quality parameters of the fermented grains leaving the cellar, the quality parameters of the fermented grains entering the cellar, and fermented grains ingredient data.

[0076] Similarly, the quality parameters of the uncellared and in-cellared mash in this embodiment are of the same type as the quality parameters of the mash sample, and also include sensory data, texture data and physical and chemical data, which will not be described in detail in this embodiment.

[0077] The data of fermented mash ingredients include: the weight data of a single steamer fermented mash, the amount of water added to moisten the grain, the amount of grain added, the amount of bran added, the amount of water measured, the amount of koji added and the time of steaming the grain.

[0078] S104, constructing a mapping model for the fermented grains entering and leaving the cellar based on the mapping database, wherein the mapping model for the fermented grains entering and leaving the cellar is used to predict the quality parameters of the fermented grains entering the cellar according to the quality parameters of the fermented grains leaving the cellar and the fermented grains ingredient data.

[0079] In practical applications, the quality parameters of the fermented grains taken out of the cellar and the data of the fermented grains formulation are used as inputs, and the quality parameters of the fermented grains put into the cellar are used as the true labels to train the neural network model. After the training is completed, the mapping model of the fermented grains taken out of and put into the cellar can be obtained. The mapping model of the fermented grains taken out of and put into the cellar can predict the quality parameters of the corresponding fermented grains put into the cellar according to the quality parameters of the fermented grains taken out of the cellar and the data of the fermented grains formulation. The model training method belongs to the prior art and will not be elaborated in this embodiment.

[0080] S105. Determine the image data of the fermented grains taken out of the cellar to be predicted, and input the image data of the fermented grains taken out of the cellar to be predicted into the image multi-modal mapping sub-model to obtain the prediction result of the quality parameters of the fermented grains taken out of the cellar to be predicted.

[0081] When it is necessary to determine the quality parameters of the fermented grains taken out of the cellar, the prediction of the quality parameters of the fermented grains taken out of the cellar can be realized by using the image data of the fermented grains taken out of the cellar to be predicted. The image data can be quickly collected and processed. Compared with the traditional chemical analysis method, using the image data for analysis can significantly shorten the detection time, thereby improving the detection efficiency and reducing the cost. And the image analysis technology has the characteristics of non-contact and non-destructiveness. During the detection process, there is no need to sample or damage the fermented grains, thereby reducing the detection cost and reducing the waste of raw materials. In addition, the image multi-modal mapping sub-model is constructed based on the neural network model, which has a powerful non-linear modeling ability, can capture the complex non-linear relationship between the image data and the quality parameters, and has the ability to automatically learn to extract useful features and representations from the data, thereby improving the accuracy of the prediction of the quality parameters of the fermented grains.

[0082] Figure 4 Fig. shows a schematic diagram of the acidity prediction result based on the image multi-modal mapping sub-model, that is, the quality parameters (acidity) of the fermented grains taken out of the cellar are predicted by using the image multi-modal mapping sub-model. It can be seen that the prediction effect of the image multi-modal mapping sub-model is good, and the R 2 value of the comparison between the predicted value and the true value is 0.98.

[0083] S106. Input the prediction result of the quality parameters of the fermented grains taken out of the cellar to be predicted and the data of the fermented grains formulation into the mapping model of the fermented grains taken out of and put into the cellar to obtain the prediction result of the quality parameters of the corresponding fermented grains put into the cellar.

[0084] When it is necessary to determine the quality parameters of the fermented grains entering the cellar, the quality parameter prediction results of the fermented grains leaving the cellar to be predicted obtained in step S5 and the fermented grains batching data can be used to realize the prediction of the quality parameters of the corresponding fermented grains entering the cellar. That is, the quality parameters of the corresponding fermented grains entering the cellar can be predicted by using the image data of the fermented grains leaving the cellar and the fermented grains batching data. Similarly, compared with the traditional chemical analysis method, using image data for analysis can significantly shorten the detection time, thereby improving the detection efficiency and reducing the cost. Secondly, the image analysis technology has the characteristics of non-contact and non-destructiveness. During the detection process, there is no need to sample or damage the fermented grains, thereby reducing the detection cost and reducing the waste of raw materials. Moreover, the mapping model of the fermented grains entering and leaving the cellar is also constructed based on the neural network model, which can capture the complex non-linear relationship between the image data, the fermented grains batching data and the quality parameters, and has the ability to automatically learn and extract useful features and representations from the data, thereby improving the accuracy of the quality parameter prediction of the fermented grains. In addition, in this embodiment, before the processing of the fermented grains leaving the cellar is completed, the quality parameters of the fermented grains entering the cellar can be predicted according to the quality parameters of the fermented grains leaving the cellar, further improving the efficiency, so as to meet the requirements of modern production for high efficiency, accuracy and real-time monitoring.

[0085] In this embodiment, the multi-modal mapping model further includes a sensory multi-modal mapping sub-model, and the sensory multi-modal mapping sub-model is used to predict the corresponding texture data and physical and chemical data according to the sensory data of the fermented grains.

[0086] The sensory multi-modal mapping sub-model is a combination of a sensory feature extraction module, a texture feature mapping module and a physical and chemical feature mapping module. The sensory feature extraction module, the texture feature mapping module and the physical and chemical feature mapping module are constructed based on the neural network, and the neural network can be, but is not limited to, CNN, MLP, LSTM, RNN or 1D-CNN.

[0087] Among them, the sensory feature extraction module is used to extract the high-dimensional features of the sensory data and generate a one-dimensional tensor of sensory features. In the sensory multi-modal mapping sub-model, the texture feature mapping module is used to output the texture data prediction result according to the one-dimensional tensor of sensory features, and the physical and chemical feature mapping module is used to output the physical and chemical data prediction result according to the one-dimensional tensor of sensory features.

[0088] In practical applications, the sensory data in the multi-modal database is used as the input, and the corresponding texture data and physical and chemical data are used as the true labels to train the neural network model. After the training is completed, the sensory multi-modal mapping sub-model can be obtained, and the sensory multi-modal mapping sub-model can predict the corresponding texture data and physical and chemical data according to the sensory data of the fermented grains. The model training method belongs to the prior art and will not be elaborated in this embodiment.

[0089] Based on the constructed sensory multi-modal mapping sub-model, the method in this embodiment further includes:

[0090] Determine the sensory data of the fermented grains to be predicted when leaving the pit, and input the sensory data of the fermented grains to be predicted when leaving the pit into the sensory multi-modal mapping sub-model to obtain the predicted results of the texture data and physical and chemical data of the fermented grains to be predicted when leaving the pit.

[0091] Determine the sensory data of the fermented grains to be predicted when entering the pit, and input the sensory data of the fermented grains to be predicted when entering the pit into the sensory multi-modal mapping sub-model to obtain the predicted results of the texture data and physical and chemical data of the fermented grains to be predicted when entering the pit.

[0092] According to the above solution, the texture data and physical and chemical data of the corresponding fermented grains can be predicted by using the sensory data of the fermented grains when leaving the pit or entering the pit. When it is inconvenient to obtain the image data of the fermented grains when leaving the pit or entering the pit, the sensory data can be obtained and input into the sensory multi-modal mapping sub-model to obtain the corresponding texture data and physical and chemical data. In practical applications, the texture data and physical and chemical data can be flexibly selected to be predicted by using the image data or sensory data, so as to improve the flexibility of predicting the quality parameters of the fermented grains when entering and leaving the pit. Moreover, the sensory multi-modal mapping sub-model is also constructed based on the neural network model, which can capture the complex non-linear relationship between the sensory data and the texture data and physical and chemical data, and automatically learn the ability to extract useful features and representations from the data, thereby improving the accuracy of predicting the texture data and physical and chemical data of the fermented grains.

[0093] In this embodiment, the multi-modal mapping model further includes a coupled multi-modal mapping sub-model, which is used to couple the image data and sensory data of the fermented grains to generate image-sensory coupled features, and predict the corresponding texture data and physical and chemical data according to the image-sensory coupled features.

[0094] The coupled multi-modal mapping sub-model is a combination of an image feature extraction module, a sensory feature extraction module, a feature coupling module, a texture feature mapping module, and a physical and chemical feature mapping module. The image feature extraction module, the sensory feature extraction module, the feature coupling module, the texture feature mapping module, and the physical and chemical feature mapping module are constructed based on the neural network, and the neural network can be, but is not limited to, CNN, MLP, LSTM, RNN, or 1D-CNN.

[0095] Among them, the image feature extraction module is used to extract the high-dimensional features of the image data and process the high-dimensional features into a one-dimensional tensor of image features through a flattening operation. The sensory feature extraction module is used to extract the high-dimensional features of the sensory data and generate a one-dimensional tensor of sensory features. In the coupled multi-modal mapping sub-model, the texture feature mapping module is used to output the predicted result of the texture data according to the image-sensory coupled features, and the physical and chemical feature mapping module is used to output the predicted result of the physical and chemical data according to the image-sensory coupled features.

[0096] In practical applications, image data and sensory data in a multimodal database are used as inputs, and corresponding texture data and physical and chemical data are used as true labels to train a neural network model. After the training is completed, a coupled multimodal mapping sub-model can be obtained. The coupled multimodal mapping sub-model can predict the corresponding texture data and physical and chemical data based on the image data and sensory data of fermented grains. The model training method belongs to the prior art and will not be elaborated in this embodiment.

[0097] Based on the constructed coupled multimodal mapping sub-model, the method described in this embodiment further includes:

[0098] Determine the sensory data of the fermented grains to be predicted for out-of-pit, and input the image data and sensory data of the fermented grains to be predicted for out-of-pit into the coupled multimodal mapping sub-model to obtain the prediction results of the texture data and physical and chemical data of the fermented grains to be predicted for out-of-pit.

[0099] Determine the sensory data of the fermented grains to be predicted for in-pit, and input the image data and sensory data of the fermented grains to be predicted for in-pit into the coupled multimodal mapping sub-model to obtain the prediction results of the texture data and physical and chemical data of the fermented grains to be predicted for in-pit.

[0100] According to the above solution, accurate prediction of the texture data and physical and chemical data of the corresponding fermented grains can be achieved by using the image data and sensory data of out-of-pit or in-pit fermented grains. In practical applications, it is possible to flexibly select to use image data and / or sensory data for predicting texture data and physical and chemical data, thereby improving the flexibility of predicting the quality parameters of out-of-pit and in-pit fermented grains. Moreover, the coupled multimodal mapping sub-model is also constructed based on a neural network model. By coupling data from different modalities, it can capture the associations and dependencies between different modalities, thereby enhancing the model's understanding ability and reasoning ability, improving the model's performance, and further improving the accuracy. And it can capture the complex non-linear relationships between image-sensory coupling features and texture data and physical and chemical data, and has the ability to automatically learn to extract useful features and representations from the data, thereby improving the accuracy of predicting the texture data and physical and chemical data of fermented grains.

[0101] In summary, the method for predicting the quality parameters of fermented grains in Luzhou-flavor liquor provided in this embodiment, by constructing a multimodal mapping model and an out-of-pit and in-pit fermented grains mapping model, can predict the quality parameters of out-of-pit fermented grains and the corresponding in-pit fermented grains based on the image data of out-of-pit fermented grains, and can predict the texture data and physical and chemical data of the corresponding fermented grains based on the image data and / or sensory data of out-of-pit and in-pit fermented grains, improving the efficiency and accuracy of determining the quality parameters of fermented grains. At the same time, this embodiment also considers using texture data as the quality parameters of fermented grains, and the predicted quality parameters can comprehensively evaluate the quality of fermented grains, thereby improving the accuracy of fermented grains quality evaluation.

[0102] Figure 5The structural schematic diagram of a device for predicting the quality parameters of the fermented grains of Luzhou-flavor liquor is shown. Please refer to Figure 5 , the device includes:

[0103] A database establishment module, used to establish a multi-modal database of fermented grains. The multi-modal database includes image data of multiple fermented grain samples and their corresponding quality parameters. The quality parameters include sensory data, texture data, and physical and chemical data; establish a mapping database of the fermented grains entering and leaving the cellar. The mapping database includes the quality parameters of the fermented grains leaving the cellar, the quality parameters of the fermented grains entering the cellar, and the fermented grain ingredient data;

[0104] A mapping model construction module, used to construct a multi-modal mapping model based on the multi-modal database. The multi-modal mapping model at least includes an image multi-modal mapping sub-model, and the image multi-modal mapping sub-model is used to predict the corresponding quality parameters according to the image data of the fermented grains; construct an in-out cellar fermented grain mapping model based on the mapping database, and the in-out cellar fermented grain mapping model is used to predict the quality parameters of the corresponding fermented grains entering the cellar according to the quality parameters of the fermented grains leaving the cellar and the fermented grain ingredient data;

[0105] A prediction module, used to determine the image data of the fermented grains leaving the cellar to be predicted, input the image data of the fermented grains leaving the cellar to be predicted into the image multi-modal mapping sub-model, and obtain the prediction result of the quality parameters of the fermented grains leaving the cellar to be predicted; input the prediction result of the quality parameters of the fermented grains leaving the cellar to be predicted and the fermented grain ingredient data into the in-out cellar fermented grain mapping model, and obtain the prediction result of the quality parameters of the corresponding fermented grains entering the cellar.

[0106] It can be understood that since the device for predicting the quality parameters of the fermented grains of Luzhou-flavor liquor described in this embodiment is a device for implementing the method for predicting the quality parameters of the fermented grains of Luzhou-flavor liquor described in the embodiment, for the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple. For the relevant parts, please refer to the partial description of the method and will not be elaborated here.

[0107] Figure 6 and Figure 7 respectively show a method for judging and preparing ingredients according to the fermented grains of Luzhou-flavor liquor. Please refer to Figure 6 and Figure 7 , the method includes:

[0108] S201. Construct a judging and ingredient-recommending model based on the image multi-modal mapping sub-model and the in-out cellar fermented grain mapping model. The image multi-modal mapping sub-model and the in-out cellar fermented grain mapping model are the image multi-modal mapping sub-model and the in-out cellar fermented grain mapping model in the method for predicting the quality parameters of the fermented grains of Luzhou-flavor liquor described in this embodiment.

[0109] In this embodiment, the koji ingredient recommendation model is constructed based on an evolutionary algorithm, and the evolutionary algorithm is a genetic algorithm, a simulated annealing algorithm, or an evolutionary strategy algorithm.

[0110] S202. Obtain the image data of the pit-out fermented grains of the fermented grains to be viewed for koji ingredients, and input the image data of the pit-out fermented grains of the fermented grains to be viewed for koji ingredients and the quality parameters of the target pit-in fermented grains into the koji ingredient recommendation model for optimization to obtain the recommended koji ingredient data of the fermented grains to be viewed for koji ingredients. The quality parameters include sensory data, texture data, and physicochemical data.

[0111] Specifically, the method for obtaining the image data of the pit-out fermented grains of the fermented grains to be viewed for koji ingredients is the same as the method for obtaining the image data of the fermented grain samples in the quality parameter prediction method of the fermented grains of Luzhou-flavor liquor described in this embodiment. Similarly, the quality parameters of the target pit-in fermented grains in this embodiment are of the same type as those of the fermented grain samples, and also include sensory data, texture data, and physicochemical data, which will not be elaborated here. The quality parameters of the target pit-in fermented grains are the set quality parameters expected to be achieved.

[0112] In practical applications, the optimization process of the koji ingredient recommendation model is specifically as follows: Obtain the image data of the pit-out fermented grains of the fermented grains to be viewed for koji ingredients, input it into the image multi-modal mapping sub-model to obtain the quality parameter prediction result of the corresponding pit-out fermented grains, input it and all possible koji ingredient data into the pit-in and pit-out fermented grain mapping model to predict the quality parameters of the corresponding pit-in fermented grains, then compare the quality parameters of the predicted pit-in fermented grains with the quality parameters of the target pit-in fermented grains, calculate the fitness according to the fitness function, determine the optimal quality parameters of the predicted pit-in fermented grains according to the fitness, the koji ingredient data corresponding to the optimal quality parameters is the optimal koji ingredient data, and finally use it as the recommended koji ingredient data of the fermented grains to be viewed for koji ingredients.

[0113] In this embodiment, the fitness function of the evolutionary algorithm is as follows:

[0114]

[0115] Among them, Fitness represents the fitness, obj j represents the quality parameter of the j-th target pit-in fermented grains, pre j represents the quality parameter of the j-th pit-in fermented grains predicted based on the image multi-modal mapping sub-model and the pit-in and pit-out fermented grain mapping model, and M represents the number of quality parameters of the pit-in fermented grains.

[0116] In practical applications, the fitness of each quality parameter can be calculated respectively by the above method, and the average fitness is used as the final fitness.

[0117] The method and device for observing and proportioning fermented grains of Luzhou-flavor liquor provided in this embodiment can, by constructing a recommendation model for observing and proportioning fermented grains, seek recommended fermented grain proportioning data that meet the quality parameters of the target fermented grains in the cellar after setting the quality parameters of the target fermented grains in the cellar, so as to adjust and optimize the proportioning of fermented grains, thereby improving the efficiency and accuracy of the adjustment of fermented grain proportioning.

[0118] Figure 8 The structural schematic diagram of a device for observing and proportioning fermented grains of Luzhou-flavor liquor is shown. Please refer to Figure 8 , the device includes:

[0119] A construction module, configured to construct a recommendation model for observing and proportioning fermented grains based on an image multi-modal mapping sub-model and an in-out cellar fermented grain mapping model, where the image multi-modal mapping sub-model and the in-out cellar fermented grain mapping model are the image multi-modal mapping sub-model and the in-out cellar fermented grain mapping model in the method for predicting the quality parameters of Luzhou-flavor liquor fermented grains described in the embodiment;

[0120] A recommendation module, configured to obtain the image data of the fermented grains out of the cellar of the fermented grains to be observed and proportioned, input the image data of the fermented grains out of the cellar of the fermented grains to be observed and proportioned and the quality parameters of the target fermented grains in the cellar into the recommendation model for observing and proportioning fermented grains for optimization, and obtain the recommended fermented grain proportioning data of the fermented grains to be observed and proportioned, where the quality parameters include sensory data, texture data, and physical and chemical data.

[0121] It can be understood that since the device for observing and proportioning fermented grains of Luzhou-flavor liquor described in this embodiment is a device for implementing the method for observing and proportioning fermented grains of Luzhou-flavor liquor described in the embodiment, for the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple. For the relevant parts, please refer to the partial description of the method and will not be elaborated here.

Claims

1. A method for predicting quality parameters of Luzhou-flavor liquor lees, characterized in that: The method comprises: Establishing a multimodal database of fermented grains, wherein the multimodal database includes image data of a plurality of fermented grains samples and corresponding quality parameters thereof, wherein the quality parameters include sensory data, texture data, and physical and chemical data; Constructing a multimodal mapping model based on the multimodal database, wherein the multimodal mapping model at least includes an image multimodal mapping sub-model, and the image multimodal mapping sub-model is used to predict corresponding quality parameters according to image data of the mash; Establishing a mapping database of fermented grains entering and leaving the cellar, wherein the mapping database includes quality parameters of fermented grains leaving the cellar, quality parameters of fermented grains entering the cellar, and fermented grains ingredient data; Constructing an in-cellar and out-cellar lees fermented grains mapping model based on the mapping database, wherein the in-cellar and out-cellar lees fermented grains mapping model is used to predict the quality parameters of the corresponding in-cellar lees fermented grains according to the quality parameters of the out-cellar lees fermented grains and the lees fermented grains ingredient data; Determine the image data of the lees to be predicted to be out of the cellar, input the image data of the lees to be predicted to be out of the cellar into the image multimodal mapping sub-model, and obtain the prediction result of the quality parameter of the lees to be predicted to be out of the cellar; The quality parameter prediction results of the mash to be predicted out of the cellar and the mash ingredient data are input into the mapping model of mash entering and leaving the cellar to obtain the quality parameter prediction results of the corresponding mash entering the cellar.

2. The method for predicting quality parameters of Luzhou-flavor liquor lees according to claim 1, characterized in that: The multimodal mapping model further includes a sensory multimodal mapping sub-model, and the sensory multimodal mapping sub-model is used to predict corresponding texture data and physical and chemical data according to the sensory data of the fermented grains; The method further comprises: Determine the sensory data of the predicted lees to be taken out of the cellar, input the sensory data of the predicted lees to be taken out of the cellar into the sensory multimodal mapping sub-model, and obtain the texture data prediction results and the physical and chemical data prediction results of the predicted lees to be taken out of the cellar; The sensory data of the mash to be predicted to be put into the cellar are determined, and the sensory data of the mash to be predicted to be put into the cellar are input into the sensory multimodal mapping sub-model to obtain the texture data prediction results and the physical and chemical data prediction results of the mash to be predicted to be put into the cellar.

3. The method for predicting quality parameters of Luzhou-flavor liquor lees according to claim 2, characterized in that: The multimodal mapping model further includes a coupled multimodal mapping sub-model, wherein the coupled multimodal mapping sub-model is used to couple the image data of the mash with the sensory data to generate an image-sensory coupling feature, and predict the corresponding texture data and physical and chemical data according to the image-sensory coupling feature; The method further comprises: Determine the sensory data of the to-be-predicted lees, input the image data and sensory data of the to-be-predicted lees into the coupled multimodal mapping sub-model, and obtain the texture data prediction results and the physicochemical data prediction results of the to-be-predicted lees; The sensory data of the mash to be predicted to be put into the cellar is determined, and the image data and sensory data of the mash to be predicted to be put into the cellar are input into the coupled multimodal mapping sub-model to obtain the texture data prediction results and the physical and chemical data prediction results of the mash to be predicted to be put into the cellar.

4. The method for predicting quality parameters of Luzhou-flavor liquor lees according to claim 3, characterized in that: The image multimodal mapping submodel is a combination of an image feature extraction module, a sensory feature mapping module, a texture feature mapping module, and a physicochemical feature mapping module; the sensory multimodal mapping submodel is a combination of a sensory feature extraction module, a texture feature mapping module, and a physicochemical feature mapping module; the coupled multimodal mapping submodel is a combination of an image feature extraction module, a sensory feature extraction module, a feature coupling module, a texture feature mapping module, and a physicochemical feature mapping module; The image feature extraction module is used to extract high-dimensional features of image data, and process the high-dimensional features into a one-dimensional tensor of image features through a flattening operation; The sensory feature extraction module is used to extract high-dimensional features of sensory data and generate a one-dimensional tensor of sensory features; The feature coupling module is used to splice the one-dimensional tensor of image features and the one-dimensional tensor of sensory features, and then input the spliced ​​features into the feature coupling module. After the feature coupling module performs calculations, it outputs the image-sensory coupling features represented by the one-dimensional tensor. The sensory feature mapping module, texture feature mapping module and physical and chemical feature mapping module are used to output sensory data prediction results, texture data prediction results and physical and chemical data prediction results according to the one-dimensional image feature tensor, the one-dimensional sensory feature tensor or the image-sensory coupling feature.

5. The method for predicting quality parameters of Luzhou-flavor liquor lees according to claim 4, characterized in that: The image feature extraction module, sensory feature extraction module, feature coupling module, sensory feature mapping module, texture feature mapping module and physical and chemical feature mapping module are constructed based on a neural network, and the neural network is CNN, MLP, LSTM, RNN or 1D-CNN.

6. The method for predicting quality parameters of Luzhou-flavor liquor lees according to claim 1, characterized in that: The image data are color photos, and the shooting parameters corresponding to each color photo are the same; The sensory data includes: visual data, olfactory data, taste data, tactile data and overall sensory data. The visual data includes: color data, appearance data, uniformity data, and grain gelatinization degree data; the olfactory data includes: wine aroma data, sourness data, ester aroma data, lees aroma data, cellar aroma data, grain aroma data and koji aroma data; the taste data includes: sourness data, astringency data and sweetness data; the tactile data includes: structure data and tenderness data; The texture data include: hardness data, fracture data, adhesion data, elasticity data, cohesion data, viscosity data, chewiness data and recovery data; The physical and chemical data include: moisture data, acidity data, starch content data and sugar content data; The fermented mash ingredient data includes: weight data of a single steamer fermented mash, data on the amount of water added to moisten the grain, data on the amount of grain added, data on the amount of bran added, data on the amount of water used to beat the mash, data on the amount of koji added and data on the duration of steaming the grain.

7. A device for predicting quality parameters of Luzhou-flavor liquor lees, characterized in that: The device comprises: A database establishment module is used to establish a multimodal database of fermented grains, wherein the multimodal database includes image data of multiple fermented grains samples and their corresponding quality parameters, wherein the quality parameters include sensory data, texture data, and physical and chemical data; establish a mapping database of fermented grains entering and leaving the cellar, wherein the mapping database includes quality parameters of fermented grains leaving the cellar, quality parameters of fermented grains entering the cellar, and fermented grains ingredient data; A mapping model construction module is used to construct a multimodal mapping model based on the multimodal database, the multimodal mapping model at least comprising an image multimodal mapping sub-model, the image multimodal mapping sub-model is used to predict the corresponding quality parameters according to the image data of the fermented grains; a mapping model for fermented grains entering and leaving the cellar is constructed based on the mapping database, the mapping model for fermented grains entering and leaving the cellar is used to predict the quality parameters of the corresponding fermented grains entering the cellar according to the quality parameters of the fermented grains leaving the cellar and the fermented grains ingredient data; The prediction module is used to determine the image data of the predicted dregs to be taken out of the cellar, input the image data of the predicted dregs to be taken out of the cellar into the image multimodal mapping sub-model, and obtain the quality parameter prediction results of the predicted dregs to be taken out of the cellar; input the quality parameter prediction results of the predicted dregs to be taken out of the cellar and the dregs ingredient data into the mapping model of the dregs entering and leaving the cellar, and obtain the quality parameter prediction results of the corresponding dregs entering the cellar.

8. A method for preparing ingredients for Luzhou-flavor liquor lees, characterized in that: The method comprises: A recommendation model for viewing lees ingredients is constructed based on an image multimodal mapping sub-model and an in-and-out lees mapping model, wherein the image multimodal mapping sub-model and the in-and-out lees mapping model are the image multimodal mapping sub-model and the in-and-out lees mapping model in the quality parameter prediction method for Luzhou-flavor liquor lees according to any one of claims 1 to 6; The image data of the uncellared mash of the to-be-checked mash with ingredients to be checked is obtained, and the image data of the uncellared mash of the to-be-checked mash and the quality parameters of the target mash entering the cellar are input into a mash ingredient recommendation model for optimization, so as to obtain the recommended mash ingredient data of the to-be-checked mash, wherein the quality parameters include sensory data, texture data and physical and chemical data.

9. The method for preparing ingredients for Luzhou-flavor liquor lees according to claim 8, characterized in that: The bad ingredients recommendation model is constructed based on an evolutionary algorithm, and the evolutionary algorithm is a genetic algorithm, a simulated annealing algorithm or an evolutionary strategy algorithm; The fitness function of the evolutionary algorithm is as follows: Among them, Fitness represents fitness, obj j represents the quality parameter of the jth target mash entering the cellar, pre j represents the quality parameter of the jth mash entering the cellar obtained by prediction based on the image multimodal mapping sub-model and the mash entering and leaving the cellar mapping model, and M represents the number of quality parameters of the mash entering the cellar.

10. A device for batching lees of Luzhou-flavor liquor, characterized in that: The device comprises: A construction module for constructing a recommendation model for viewing lees ingredients based on an image multimodal mapping submodel and an in-and-out lees mapping model, wherein the image multimodal mapping submodel and the in-and-out lees mapping model are the image multimodal mapping submodel and the in-and-out lees mapping model in the quality parameter prediction method for Luzhou-flavor liquor lees according to any one of claims 1 to 6; The recommendation module is used to obtain image data of the uncellared mash of the mash with the ingredients to be checked, input the image data of the uncellared mash of the ingredients to be checked and the quality parameters of the target mash entering the cellar into the recommendation model for checking mash ingredients for optimization, and obtain the recommended mash ingredient data of the mash with the ingredients to be checked, wherein the quality parameters include sensory data, texture data and physical and chemical data.

Images