Daqu grading method and device based on comprehensive two-dimensional gas chromatography

Abstract

The invention mainly relates to the technical field of wine brewing. The invention provides a Daqu grading method and device based on comprehensive two-dimensional gas chromatography, and the core idea is as follows: reconstructing a comprehensive two-dimensional gas chromatogram, and judging a Daqu grading key chromatographic region in the comprehensive two-dimensional gas chromatogram based on the reconstructed comprehensive two-dimensional gas chromatogram; establishing a Daqu grading model, and writing the Daqu grading key chromatographic region into the established Daqu grading model; the method comprises the following steps: acquiring comprehensive two-dimensional gas chromatograms of different grades of yeast for making hard liquor, performing one-to-one correspondence on the comprehensive two-dimensional gas chromatograms and the grades of the yeast for making hard liquor, establishing a comprehensive two-dimensional gas chromatography data set, and training a yeast grading model based on the established comprehensive two-dimensional gas chromatography data set; and inputting the comprehensive two-dimensional gas chromatogram of the to-be-predicted yeast sample into the trained yeast grading model to obtain a grade prediction result of the to-be-predicted yeast sample, so that the automation and intelligence levels of yeast grading can be improved, and an important technical support can be provided for optimization of a yeast production process.

Description

Technical Field

[0001] The invention mainly relates to the technical field of wine making, and in particular to a Daqu classification method and device based on full two-dimensional gas chromatography. Background Art

[0002] As an important raw material in traditional Chinese winemaking technology, Daqu plays an indispensable role in the fermentation process of liquor, and its quality directly affects the flavor and quality stability of the liquor. The quality control of Daqu mostly relies on sampling inspection, and the first-level rate of Daqu is used to evaluate whether each production batch is qualified. The grading of Daqu is based on the distribution of mycelium in the cross-section of the qu block and the manual evaluation of the qu block smell, which is often inefficient and difficult to meet the high efficiency and refinement requirements of modern production for Daqu quality control. On the contrary, full two-dimensional gas chromatography generates a two-dimensional total ion flow map by extracting the flavor of Daqu. The spectral characteristics reflect the flavor characteristics and then reflect the level of Daqu. Therefore, identifying the key chromatographic area that reflects the level of Daqu has certain guiding significance for the quality control and further refined analysis of the Daqu production process. Summary of the invention

[0003] Technical Problems to be Solved by the Invention

[0004] Provided are a Daqu classification method and device based on full two-dimensional gas chromatography, aiming to improve the accuracy and automation of Daqu grade detection.

[0005] The technical solution adopted by the present invention to solve the above technical problems

[0006] In a first aspect, the present invention provides a Daqu classification method based on comprehensive two-dimensional gas chromatography, the method comprising:

[0007] Reconstructing the full two-dimensional gas chromatogram, and determining the key chromatographic region of Daqu classification in the full two-dimensional gas chromatogram based on the reconstructed full two-dimensional gas chromatogram;

[0008] Establishing a Daqu classification model, and writing the key chromatographic regions of Daqu classification into the established Daqu classification model;

[0009] Obtaining full two-dimensional gas chromatograms of different grades of Daqu, matching the full two-dimensional gas chromatograms with the grades of Daqu one by one, establishing a full two-dimensional gas chromatogram data set, and training a Daqu classification model based on the established full two-dimensional gas chromatogram data set;

[0010] The full two-dimensional gas chromatogram of the Daqu sample to be predicted is input into the trained Daqu classification model to obtain the grade prediction result of the Daqu sample to be predicted.

[0011] Furthermore, the reconstruction of the full two-dimensional gas chromatogram includes: collecting multiple full two-dimensional gas chromatograms, dividing each full two-dimensional gas chromatogram into multiple full two-dimensional gas chromatogram rectangular area blocks with a preset pixel step and a preset pixel size; reconstructing the full two-dimensional gas chromatogram rectangular area blocks at the same position in the multiple full two-dimensional gas chromatograms into an area block data set; the position of the rectangular area block is determined according to the coordinates of the diagonal points of the rectangular area block.

[0012] Furthermore, based on the reconstructed full two-dimensional gas chromatogram, the key chromatographic regions for Daqu classification in the full two-dimensional gas chromatogram are determined to specifically include:

[0013] The rectangular area block of the full two-dimensional gas chromatogram in the area block data set is input into the established Daqu classification model, and the Daqu classification model extracts the full two-dimensional gas chromatogram features and outputs the Daqu grade;

[0014] K-fold cross validation was used to evaluate the Daqu classification ability of each area block, and the area blocks with evaluation index greater than the preset evaluation index threshold were selected as key chromatographic areas;

[0015] The evaluation index may be an average accuracy value of k-fold cross validation.

[0016] Furthermore, when reconstructing the full two-dimensional gas chromatogram, if there is overlap in the divided multiple rectangular area blocks of the full two-dimensional gas chromatogram, k-fold cross validation is used to evaluate the large-curve classification ability of each area block, and all areas with evaluation indicators greater than the preset evaluation indicator threshold are selected as candidate key chromatographic areas, and the areas in the candidate key chromatographic areas with overlapping area ratios greater than the preset overlapping area ratio threshold are merged to obtain key chromatographic areas.

[0017] Furthermore, the Daqu classification model includes an image feature extraction network and a level prediction network; wherein the image feature extraction network is a convolutional neural network; the level prediction network is a convolutional neural network or a multi-layer perceptron network; the image feature extraction network and the level prediction network are connected as neural networks;

[0018] The step of writing the key chromatographic region for Daqu classification into the established Daqu classification model includes, during the training and application of the Daqu classification model, establishing a mask tensor with the same size as the full two-dimensional gas chromatogram; determining the position of the key chromatographic region according to the diagonal points of the rectangular region block of the key chromatographic region, setting the value of the mask tensor element at the corresponding position of the key chromatographic region to 1, and setting the value of the mask tensor element outside the corresponding position of the key chromatographic region to 0; performing a Hadamard product operation on the mask tensor and the full two-dimensional gas chromatogram to write the key chromatographic region for Daqu classification into the established Daqu classification model.

[0019] Furthermore, the Daqu classification model written into the key chromatographic region of Daqu classification is trained based on the established full two-dimensional gas chromatography data set, specifically including:

[0020] The full two-dimensional gas chromatography data set is divided into a training set and a validation set according to a set ratio;

[0021] The full two-dimensional gas chromatogram of Daqu in the training set is used as the input parameter of the Daqu classification model, and the corresponding Daqu level is used as the true output label of the Daqu classification model to train the Daqu classification model; during the training process, the predicted Daqu level output by the Daqu classification model is compared with the corresponding true output label, the loss function is calculated, and the parameters of the Daqu classification model are optimized based on the back propagation of the loss function;

[0022] The accuracy of the Daqu classification model is verified based on the validation set. When the accuracy of the validation set is greater than the set accuracy threshold, the training of the Daqu classification model is completed.

[0023] Furthermore, the establishing of a comprehensive two-dimensional gas chromatography data set specifically includes:

[0024] A comprehensive two-dimensional gas chromatogram of the Daqu sample is obtained using a comprehensive two-dimensional gas chromatograph according to the set detection parameters and plotting range;

[0025] The full two-dimensional gas chromatogram of the Daqu sample is labeled one-to-one with the Daqu level manually evaluated and recognized by engineers or experts in the field, forming a data set containing the full two-dimensional gas chromatogram data of Daqu and the Daqu level data. The Daqu level is represented by a one-hot vector or digital form.

[0026] Furthermore, the method further comprises preprocessing the full two-dimensional gas chromatogram in the full two-dimensional gas chromatography data set, and the preprocessing specifically comprises:

[0027] After the value area in the full two-dimensional gas chromatogram is cropped, it is adjusted to a full two-dimensional gas chromatogram of a preset resolution by scaling and resizing;

[0028] The full two-dimensional gas chromatogram adjusted to a preset resolution is standardized using a normalization method, which is embedded in the training process of the Daqu classification model.

[0029] Furthermore, the normalization method is implemented using the transforms function in the torchvision library.

[0030] On the other hand, the present invention also provides a Daqu classification device based on full two-dimensional gas chromatography, which is used to implement the Daqu classification method based on full two-dimensional gas chromatography. The Daqu classification device based on full two-dimensional gas chromatography includes: a key area identification module, a full two-dimensional gas chromatography data set construction module, a Daqu classification model training module and a Daqu classification module;

[0031] The key region identification module is used to reconstruct the full two-dimensional gas chromatogram, and determine the key chromatographic region of Daqu classification in the full two-dimensional gas chromatogram based on the reconstructed full two-dimensional gas chromatogram;

[0032] The full two-dimensional gas chromatography data set construction module is used to construct a full two-dimensional gas chromatography data set;

[0033] The Daqu classification model training module is used to write the key chromatographic region of Daqu classification into the established Daqu classification model, and use the established full two-dimensional gas chromatography data set to train the Daqu classification model after the key chromatographic region of Daqu classification is written;

[0034] The Daqu grading module stores a trained Daqu grading model, and obtains a grade prediction result of the Daqu sample to be predicted based on the full two-dimensional gas chromatogram of the Daqu sample to be predicted.

[0035] Beneficial Effects of the Invention

[0036] The method and device for grading Daqu based on full two-dimensional gas chromatography described in the present invention obtain full two-dimensional gas chromatograms of Daqu of different grades, identify key chromatographic regions for judging Daqu grade by combining the division and evaluation of regional blocks, write the identified key chromatographic regions for Daqu grading into the established Daqu grading model, and combine neural network technology with full two-dimensional gas chromatograms to accurately predict the grade of Daqu samples, thereby improving the automation and intelligence level of Daqu grading, and providing important technical support for the optimization of Daqu production technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 The flowchart of the Daqu classification method based on comprehensive two-dimensional gas chromatography is as follows;

[0038] Figure 2 A schematic diagram of the technical composition of the Daqu fractionation method based on comprehensive two-dimensional gas chromatography;

[0039] Figure 3 A schematic diagram of the key chromatographic regions for Daqu classification in the embodiment;

[0040] Figure 4 This is a schematic diagram of the Daqu classification device based on comprehensive two-dimensional gas chromatography. DETAILED DESCRIPTION

[0041] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present application more clear, the present application is further described in detail below in combination with the implementation methods. Unless otherwise defined, all scientific and technological terms used herein have the same meanings as understood by ordinary technicians in the field.

[0042] Obviously, the described embodiments are only some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] In a first aspect, the present invention provides a Daqu classification method based on comprehensive two-dimensional gas chromatography, such as Figure 1 and Figure 2 As shown, including:

[0044] The full two-dimensional gas chromatogram is reconstructed, and the key chromatographic region for Daqu classification in the full two-dimensional gas chromatogram is determined based on the reconstructed full two-dimensional gas chromatogram, and the key chromatographic region for Daqu classification is written into the established Daqu classification model.

[0045] The identification of key chromatographic regions for Daqu classification is achieved through ergodic region block reconstruction and evaluation;

[0046] Divide the full two-dimensional gas chromatogram data into a plurality of rectangular area blocks, each area block is determined by the coordinates of two diagonal points;

[0047] The traversal region block construction refers to dividing the region blocks in the full two-dimensional gas chromatogram data with a preset pixel step size and a preset region block pixel size;

[0048] Wherein, when the preset pixel step size is smaller than the preset region block pixel size, there is overlap between the region blocks;

[0049] The region blocks corresponding to the same region block coordinates in all chromatogram data are combined to form a region block data set; each chromatogram is divided into a plurality of region blocks, the number of which is consistent with the number of generated region block data sets;

[0050] The Daqu classification model is trained with each regional block data set. During the training process, the accuracy of the test set is evaluated after the training set is trained for a preset number of rounds.

[0051] Each regional block data set is evaluated through k-fold cross validation, where the k value is usually 5 or 10. The k-fold test set accuracy is summarized and the average value is calculated as the evaluation indicator of the regional block classification ability;

[0052] Selecting the region blocks whose evaluation index is greater than a preset evaluation index threshold as the key chromatographic region to be selected;

[0053] Merging the regions in the candidate key chromatographic regions where the overlap area ratio is greater than a preset overlap area ratio threshold value to obtain a key chromatographic region;

[0054] When merging overlapping areas, priority is given to retaining the area block with the highest classification performance index; if the indicators of multiple overlapping areas are similar, the new area boundary is calculated by weighted average.

[0055] In this embodiment, if Figure 3 As shown in the figure, the preset pixel step is 10, and the preset regional block pixel size is 64, so the 256×256 full two-dimensional gas chromatogram data can be divided into 400 parts, named block_000 to block_399 respectively, so 400 regional block data sets are obtained;

[0056] Each regional block data set is trained with a Daqu classification model. In this embodiment, a 5-fold cross validation is adopted for evaluation. After the training set is trained for a preset number of rounds, the accuracy of the test set is evaluated. The preset number of rounds is 50 times.

[0057] Taking block_395 as an example, we can get:

[0058] block_395:

[0059] Fold 1: Test Accuracy: 100.00%

[0060] Fold 2: Test Accuracy: 100.00%

[0061] Fold 3: Test Accuracy: 100.00%

[0062] Fold 4: Test Accuracy: 100.00%

[0063] Fold 5: Test Accuracy: 100.00%

[0064] AverageAccuracy:100.00%

[0065] Therefore, the average accuracy of 5 trainings (i.e., 5 folds) is 100%, which is the evaluation index of the regional block classification ability, and the classification ability evaluation index of block_395 is 100%; then the regional block whose evaluation index is greater than the preset evaluation index threshold is selected as the key chromatographic region to be selected. In this embodiment, the threshold is 100%, that is, the regional block whose classification ability evaluation index is 100% is selected as the key chromatographic region to be selected; then the regions whose overlapping area ratio in the key chromatographic region to be selected is greater than the preset overlapping area ratio threshold are merged to obtain the key chromatographic region. In this embodiment, 50% is selected as the area threshold; when merging overlapping regions, the regional block with the highest classification performance index is preferentially retained; if the indicators of multiple overlapping regions are similar, the new regional boundary is calculated by weighted average; since the indicators of the blocks to be selected in this embodiment are all 100%, the new regional boundary should be calculated by weighted average; this embodiment is only displayed as a method, therefore, in this embodiment, one of the similar regional blocks is directly retained as the key chromatographic region.

[0066] In the process of constructing a full two-dimensional gas chromatography data set, a full two-dimensional gas chromatograph is used to obtain a full two-dimensional gas chromatogram of a Daqu sample; the full two-dimensional gas chromatogram is a two-dimensional color pixel image, encoded in RGB color, and includes chromatographic peak signals of one-dimensional retention time and two-dimensional retention time; in the Daqu-level full two-dimensional gas chromatogram data set, the full two-dimensional gas chromatograms of all Daqu samples maintain consistent detection parameters and drawing ranges during acquisition; the full two-dimensional gas chromatogram data of the Daqu samples are labeled one-to-one with the Daqu levels manually assessed and identified by engineers or experts in the field to form a full two-dimensional gas chromatography data set; the Daqu level data includes levels in the styles of primary, secondary, tertiary, etc., which can be represented in the form of one-hot vectors or numbers such as 0, 1, 2, etc.

[0067] In this embodiment, Baobaoqu of Luzhou-flavor liquor is taken as a sample, and two levels, primary and secondary, are considered, with 0 representing primary qu and 1 representing secondary qu.

[0068] When constructing a Daqu grading model, the Daqu grading model includes an image feature extraction network and a level prediction network; wherein the image feature extraction network is a convolutional neural network; the level prediction network is a convolutional neural network or a multi-layer perceptron network; the image feature extraction network and the level prediction network are connected as neural networks; during the training and application of the Daqu grading model, a mask tensor with the same size as the full two-dimensional gas chromatogram is established; the position of the key chromatographic area is determined according to the diagonal points of the rectangular area block of the key chromatographic area, and the mask tensor element value at the corresponding position of the key chromatographic area is set to 1, and the mask tensor element value outside the corresponding position of the key chromatographic area is set to 0; the mask tensor and the full two-dimensional gas chromatogram are subjected to a Hadamard product operation to realize writing the Daqu grading key chromatographic area into the established Daqu grading model.

[0069] The full two-dimensional gas chromatogram data of Daqu is used as the input feature of the Daqu classification model, and the corresponding Daqu level data is used as the label feature of the model; the loss function is calculated by the output features and label features of the Daqu classification model, and the loss function adopts a cross entropy loss function or a focal loss function, and the model parameters are optimized by back propagation of the loss function;

[0070] The Daqu-level full two-dimensional gas chromatogram dataset needs to be divided into a training set and a validation set in proportion to train and validate the model. When the classification accuracy of the validation set reaches the preset threshold, the trained Daqu grading model is obtained.

[0071] In this embodiment, the image feature extraction network is a convolutional neural network, and the level prediction network is an MLP network, and the two are connected as a neural network. In this embodiment, it is implemented through the PyTorch library, and the schematic architecture is as follows:

[0072] class SimpleCNN(nn.Module):

[0073] def __init__(self):

[0074] super(SimpleCNN,self).__init__()

[0075] self.conv1=nn.Conv2d(3,3,kernel_size=3,padding=1)

[0076] self.conv2=nn.Conv2d(3,3,kernel_size=3,padding=1)

[0077] self.conv3=nn.Conv2d(3,1,kernel_size=3,padding=1)

[0078] self.pool = nn.MaxPool2d(2,2)

[0079] self.fc1=nn.Linear(1*32*32,128)

[0080] self.fc2 = nn.Linear(128,2)

[0081] defforward(self,x):

[0082] x=self.pool(torch.relu(self.conv1(x)))

[0083] x=self.pool(torch.relu(self.conv2(x)))

[0084] x=self.pool(torch.relu(self.conv3(x)))

[0085] x=x.view(-1,1*32*32)

[0086] x = torch.relu(self.fc1(x))

[0087] x = self.fc2(x)

[0088] return x

[0089] In this embodiment, the Daqu-level full two-dimensional gas chromatogram data set needs to be divided into a training set and a validation set in proportion, with the training set accounting for 80% and the validation set accounting for 20%.

[0090] At the same time, in order to ensure the consistency of the input parameters of the Daqu classification model, the full two-dimensional gas chromatogram data needs to be preprocessed in both the Daqu classification model training stage and the application stage. The preprocessing steps include:

[0091] The value area in the Daqu full two-dimensional gas chromatogram is cropped and adjusted to an image of a preset resolution by scaling and resizing; the value area in the Daqu full two-dimensional gas chromatogram refers to the chromatogram area after removing the image areas such as the border and the coordinate axis that are not related to the chromatogram itself; the preset resolution is set by engineers or experts according to data characteristics and model architecture, and typical values ​​include square images of 512×512, 256×256, or rectangular images of 256×512, 128×256, and 512×1024, and other suitable sizes can also be selected;

[0092] The Daqu full two-dimensional gas chromatogram data also needs to be standardized when input into the Daqu classification model. The standardization method includes normalizing the data of each channel of the full two-dimensional gas chromatogram data to a distribution with a mean of 0 and a standard deviation of 1. The standardization method also includes minimum and maximum value normalization and other normalization methods set by engineers or experts. The normalization method is embedded in the training process of the Daqu classification model.

[0093] In this embodiment, a full two-dimensional gas chromatograph is used to obtain a full two-dimensional gas chromatogram of the Daqu sample with a resolution of 1347×912, which is represented by an RGB color image. The area from 200 to 1000 pixels in the x-direction and the area from 250 to 575 pixels in the y-direction are intercepted as the valuable area in the full two-dimensional gas chromatogram of Daqu. Then, Python programming is used to adjust the image to the preset 256×256 resolution size through the resize command in the PIL library.

[0094] In this embodiment, the normalization method is implemented using the transforms function in the torchvision library, and the data of each channel of the full two-dimensional gas chromatogram data is normalized to a distribution with a mean of 0 and a standard deviation of 1.

[0095] In a second aspect, the present invention provides a device for Daqu classification and key chromatographic region identification based on full two-dimensional gas chromatography, such as Figure 4 As shown, it includes: a key area identification module, a full two-dimensional gas chromatography data set construction module, a Daqu classification model training module and a Daqu classification module;

[0096] The key region identification module is used to reconstruct the full two-dimensional gas chromatogram, and determine the key chromatographic region of Daqu classification in the full two-dimensional gas chromatogram based on the reconstructed full two-dimensional gas chromatogram;

[0097] The full two-dimensional gas chromatography data set construction module is used to construct a full two-dimensional gas chromatography data set;

[0098] The Daqu classification model training module is used to write the key chromatographic region of Daqu classification into the established Daqu classification model, and use the established full two-dimensional gas chromatography data set to train the Daqu classification model after the key chromatographic region of Daqu classification is written;

[0099] The Daqu grading module stores a trained Daqu grading model, and obtains a grade prediction result of the Daqu sample to be predicted based on the full two-dimensional gas chromatogram of the Daqu sample to be predicted.

Claims

1. A Daqu classification method based on comprehensive two-dimensional gas chromatography, characterized in that: include: Reconstructing the full two-dimensional gas chromatogram, and determining the key chromatographic region of Daqu classification in the full two-dimensional gas chromatogram based on the reconstructed full two-dimensional gas chromatogram; Establishing a Daqu classification model, and writing the key chromatographic regions of Daqu classification into the established Daqu classification model; Obtaining full two-dimensional gas chromatograms of different grades of Daqu, matching the full two-dimensional gas chromatograms with the grades of Daqu one by one, establishing a full two-dimensional gas chromatogram data set, and training a Daqu classification model based on the established full two-dimensional gas chromatogram data set; The full two-dimensional gas chromatogram of the Daqu sample to be predicted is input into the trained Daqu classification model to obtain the grade prediction result of the Daqu sample to be predicted.

2. The Daqu classification method based on comprehensive two-dimensional gas chromatography according to claim 1, characterized in that: The reconstruction of the full two-dimensional gas chromatogram includes: collecting multiple full two-dimensional gas chromatograms, dividing each full two-dimensional gas chromatogram into multiple full two-dimensional gas chromatogram rectangular area blocks with a preset pixel step and a preset pixel size; reconstructing the full two-dimensional gas chromatogram rectangular area blocks at the same position in the multiple full two-dimensional gas chromatograms into an area block data set; the position of the rectangular area block is determined according to the coordinates of the diagonal points of the rectangular area block.

3. The Daqu classification method based on comprehensive two-dimensional gas chromatography according to claim 2, characterized in that: Based on the reconstructed full two-dimensional gas chromatogram, the key chromatographic regions for Daqu classification in the full two-dimensional gas chromatogram are determined to include: The rectangular area block of the full two-dimensional gas chromatogram in the area block data set is input into the established Daqu classification model, and the Daqu classification model extracts the full two-dimensional gas chromatogram features and outputs the Daqu grade; K-fold cross validation was used to evaluate the Daqu classification ability of each regional block, and the regional blocks with evaluation indicators greater than the preset evaluation indicator threshold were selected as key chromatographic regions.

4. The Daqu classification method based on comprehensive two-dimensional gas chromatography according to claim 3, characterized in that: When reconstructing the full two-dimensional gas chromatogram, if there is overlap in the divided rectangular area blocks of the full two-dimensional gas chromatogram, k-fold cross validation is used to evaluate the Daqu classification ability of each area block, and all areas with evaluation indicators greater than the preset evaluation indicator threshold are selected as candidate key chromatographic areas. The areas in the candidate key chromatographic areas with overlapping area ratios greater than the preset overlapping area ratio threshold are merged to obtain the key chromatographic areas.

5. The Daqu classification method based on comprehensive two-dimensional gas chromatography according to claim 1, characterized in that: The Daqu classification model includes an image feature extraction network and a level prediction network; wherein the image feature extraction network is a convolutional neural network; the level prediction network is a convolutional neural network or a multi-layer perceptron network; the image feature extraction network and the level prediction network are connected as neural networks; The step of writing the key chromatographic region for Daqu classification into the established Daqu classification model includes, during the training and application of the Daqu classification model, establishing a mask tensor with the same size as the full two-dimensional gas chromatogram; determining the position of the key chromatographic region according to the diagonal points of the rectangular region block of the key chromatographic region, setting the value of the mask tensor element at the corresponding position of the key chromatographic region to 1, and setting the value of the mask tensor element outside the corresponding position of the key chromatographic region to 0; performing a Hadamard product operation on the mask tensor and the full two-dimensional gas chromatogram to write the key chromatographic region for Daqu classification into the established Daqu classification model.

6. The Daqu classification method based on comprehensive two-dimensional gas chromatography according to any one of claims 1 to 5, characterized in that: Based on the established full two-dimensional gas chromatography data set, the Daqu classification model after writing the key chromatographic region of Daqu classification is trained, which specifically includes: The full two-dimensional gas chromatography data set is divided into a training set and a validation set according to a set ratio; The full two-dimensional gas chromatogram of Daqu in the training set is used as the input parameter of the Daqu classification model, and the corresponding Daqu level is used as the true output label of the Daqu classification model to train the Daqu classification model; during the training process, the predicted Daqu level output by the Daqu classification model is compared with the corresponding true output label, the loss function is calculated, and the parameters of the Daqu classification model are optimized based on the back propagation of the loss function; The accuracy of the Daqu classification model is verified based on the validation set. When the accuracy of the validation set is greater than the set accuracy threshold, the training of the Daqu classification model is completed.

7. The Daqu classification method based on comprehensive two-dimensional gas chromatography according to claim 1, characterized in that: The establishment of a comprehensive two-dimensional gas chromatography data set specifically includes: A comprehensive two-dimensional gas chromatogram of the Daqu sample is obtained using a comprehensive two-dimensional gas chromatograph according to the set detection parameters and plotting range; The full two-dimensional gas chromatogram of the Daqu sample is labeled one-to-one with the Daqu level manually evaluated and recognized by engineers or experts in the field, forming a data set containing the full two-dimensional gas chromatogram data of Daqu and the Daqu level data.

8. The Daqu classification method based on comprehensive two-dimensional gas chromatography according to claim 1, characterized in that: The method further includes preprocessing the full two-dimensional gas chromatogram in the full two-dimensional gas chromatogram data set, wherein the preprocessing specifically includes: after cutting out the value region in the full two-dimensional gas chromatogram, adjusting it to a full two-dimensional gas chromatogram of a preset resolution by scaling and resizing; The full two-dimensional gas chromatogram adjusted to a preset resolution is standardized using a normalization method, which is embedded in the training process of the Daqu classification model.

9. The Daqu classification method based on comprehensive two-dimensional gas chromatography according to claim 8, characterized in that: The normalization method is implemented using the transforms function in the torchvision library.

10. A Daqu classification device based on comprehensive two-dimensional gas chromatography, used to implement the Daqu classification method based on comprehensive two-dimensional gas chromatography according to any one of claims 1 to 9, characterized in that: include: Key area identification module, full two-dimensional gas chromatography data set construction module, Daqu classification model training module and Daqu classification module; The key region identification module is used to reconstruct the full two-dimensional gas chromatogram, and determine the key chromatographic region of Daqu classification in the full two-dimensional gas chromatogram based on the reconstructed full two-dimensional gas chromatogram; The full two-dimensional gas chromatography data set construction module is used to construct a full two-dimensional gas chromatography data set; The Daqu classification model training module is used to write the key chromatographic region of Daqu classification into the established Daqu classification model, and use the established full two-dimensional gas chromatography data set to train the Daqu classification model after the key chromatographic region of Daqu classification is written; The Daqu grading module stores a trained Daqu grading model, and obtains the grade prediction result of the Daqu sample to be predicted based on the full two-dimensional gas chromatogram of the Daqu sample to be predicted.

Images