Method and device for tracing beef producing area based on artificial intelligence, and electronic equipment

Abstract

The invention provides a beef origin tracing method and device based on artificial intelligence and electronic equipment. The method comprises the following steps: inputting a sample basic data extraction matrix, a sample isotope extraction matrix and a sample metabolism extraction matrix of target beef into a to-be-trained beef tracing model to predict a beef origin; calculating a loss value between the predicted beef production place and the corresponding beef production place label, adjusting model parameters of the beef traceability model to be trained based on the loss value, and if a set standard for ending the adjustment of the model parameters is not reached, continuing training until the set standard is reached; and taking the trained beef traceability model as the beef traceability model of the target beef to predict the production place of the beef to be traced, so as to judge whether the production place is the production place of the target beef or not. The multi-dimensional beef data is input into the traceability model, and the production area prediction accuracy is improved through comprehensive analysis.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and particularly to a method, device, and electronic device for tracing the origin of beef based on artificial intelligence. Background Art

[0002] In the international food market, beef, as a key protein resource, has seen a continuous increase in consumption. With the improvement of consumers' living standards and the enhancement of their health awareness, the attention to beef quality, safety, and origin information has been increasing. Beef from high-quality origins usually has unique flavors, higher nutritional value, and more stringent quality control standards, so consumers are more inclined to pay a premium.

[0003] Currently, the beef industry is also showing a booming development trend. In some guiding documents, the goal of the transformation of agriculture from yield-oriented to quality-oriented has been clearly put forward, emphasizing the green, high-quality, characteristic, and branded development of agriculture. As one of the key agricultural products, beef brands with local characteristics are being actively cultivated everywhere. The beef produced by the optimized feeding mode of beef cattle mainly based on roughage is tender, juicy, high in protein content, and rich in nutrition, and has obtained special agricultural product certifications. These characteristic beef brands not only improve the economic benefits of local agriculture but also meet consumers' demand for high-quality beef.

[0004] However, the current beef market is suffering from serious problems of origin confusion and passing off inferior goods as superior. The market is flooded with a large number of "labeled beef", and illegal merchants sell ordinary beef as beef from high-quality origins, which not only damages consumers' rights and interests but also disrupts the normal order of the market. This phenomenon seriously hinders the development of beef brands, leading consumers to doubt the authenticity of beef origin information and thus reducing their trust in beef brands.

[0005] In addition, beef origin supervision also faces major challenges. Traditional beef origin identification methods mainly rely on artificial experience and limited detection technologies, such as reviewing breeding records and detecting some chemical components. These methods have problems of low efficiency, high cost, and insufficient accuracy and reliability. Artificial experience judgment is easily affected by subjective factors, and limited detection indicators are difficult to comprehensively reflect the internal relationship between beef and its origin, and cannot effectively identify complex confusion means. Summary of the Invention

[0006] The present disclosure provides a method, device, and electronic device for tracing the origin of beef based on artificial intelligence.

[0007] The technical solutions provided by the present disclosure are as follows: A method for tracing the origin of beef based on artificial intelligence, which includes: Vectorize the isomorphic sample basic data, sample isotope data, and sample metabolome data respectively to generate a sample basic vector, a sample isotope vector, and a sample metabolite vector; Perform encoding conversion on the sample basic vector, sample isotope vector, and sample metabolite vector respectively to generate a sample basic data extraction matrix, a sample isotope extraction matrix, and a sample metabolite extraction matrix; Input the sample basic data extraction matrix, sample isotope extraction matrix, and sample metabolite extraction matrix of the target beef into the beef traceability model to be trained for predicting the origin of the beef; Calculate the loss value between the predicted beef origin and the corresponding beef origin label, and adjust the model parameters of the beef traceability model to be trained based on the loss value. If the set standard for ending the model parameter adjustment is not reached, continue training until the set standard is reached; Use the trained beef traceability model as the beef traceability model of the target beef to predict the origin of the beef to be traced, and determine whether the origin is the origin of the target beef.

[0008] An apparatus for tracing the origin of beef based on artificial intelligence, which includes: A first program unit for inputting the sample basic data extraction matrix, sample isotope extraction matrix, and sample metabolite extraction matrix of the target beef into the beef traceability model to be trained for predicting the origin of the beef; A second program unit for calculating the loss value between the predicted beef origin and the corresponding beef origin label, and adjusting the model parameters of the beef traceability model to be trained based on the loss value. If the set standard for ending the model parameter adjustment is not reached, continue training until the set standard is reached; A third program unit for using the trained beef traceability model as the beef traceability model of the target beef to predict the origin of the beef to be traced, and determining whether the origin is the origin of the target beef.

[0009] An electronic device, which includes a memory and a processor. A computer executable program is stored on the memory, and the processor is used to run the computer executable program to implement the training and prediction of the beef traceability model.

[0010] A model training method, which includes: Vectorize the isomorphic sample basic data, sample isotope data, and sample metabolome data respectively to generate a sample basic vector, a sample isotope vector, and a sample metabolite vector; The sample basic vector, the sample isotope vector, and the sample metabolic vector are respectively converted by encoding to generate a sample basic data extraction matrix, a sample isotope extraction matrix, and a sample metabolic extraction matrix; Inputting the sample basic data extraction matrix, sample isotope extraction matrix, and sample metabolism extraction matrix of the target beef into the beef traceability model to be trained to predict the origin of the beef; The loss value between the predicted beef origin and the corresponding beef origin label is calculated to adjust the model parameters of the beef traceability model to be trained based on the loss value. If the set standard for ending the model parameter adjustment is not met, the training is continued until the set standard is met.

[0011] A method for tracing the origin of agricultural and sideline products based on artificial intelligence, comprising: Input the sample basic data extraction matrix, sample isotope extraction matrix and sample metabolism extraction matrix of the target agricultural and sideline products into the agricultural and sideline products traceability model to be trained to predict the origin of the agricultural and sideline products; Calculating the loss value between the predicted agricultural and sideline product origin and the corresponding agricultural and sideline product origin label, so as to adjust the model parameters of the agricultural and sideline product traceability model to be trained based on the loss value, and if the set standard for ending the model parameter adjustment is not reached, continuing the training until the set standard is reached; The trained agricultural and sideline products traceability model is used as the agricultural and sideline products traceability model of the target agricultural and sideline products to predict the origin of the agricultural and sideline products to be traced, so as to determine whether the origin is the origin of the target agricultural and sideline products.

[0012] The technical solution of this application has the following technical advantages: (1) This method inputs the target beef sample basic data extraction matrix, sample isotope extraction matrix, and sample metabolic extraction matrix into the beef traceability model to be trained. These multi-dimensional data reflect the characteristics of beef from different levels. The basic data include information such as the appearance and physical properties of beef; the isotope data can reflect the geological and water source characteristics of the cattle's growth environment; and the metabolome data is related to the cattle's diet and health status. By comprehensively analyzing these multi-source data, the model can dig out a more comprehensive and in-depth relationship between beef and its origin, thereby significantly improving the accuracy of beef origin prediction and effectively combating the confusion of origin and the practice of selling inferior products as good ones.

[0013] (2) Using an artificial intelligence model for beef origin tracing can quickly process a large amount of data. Once the model training is completed, the origin prediction of the beef to be traced can be completed in a short time. Compared with the cumbersome and time-consuming processes of reviewing breeding records and detecting chemical components in traditional methods, this method greatly improves the identification efficiency, helps regulatory authorities to promptly discover and handle problematic beef in the market, and maintains the normal order of the market.

[0014] (3) This method is based on artificial intelligence technology. Although it requires certain computing resources and data collection costs during the model training stage, in subsequent actual applications, only the data of the beef to be traced needs to be input into the model to obtain the origin prediction result, without the need for complex manual operations and a large number of detection experiments. Therefore, in the long run, it can significantly reduce the cost of beef origin identification and relieve the economic burden on regulatory authorities and enterprises.

[0015] (4) This method adjusts the model parameters by calculating the loss value between the predicted beef origin and the corresponding beef origin label, enabling the model to continuously learn and optimize. As new data is continuously added, the model can better adapt to the characteristics of beef under different regions and different breeding methods, continuously improving the accuracy and reliability of origin tracing, and having strong adaptability and flexibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic flowchart of the method for beef origin tracing based on artificial intelligence in the embodiment of the present application; Figure 2 It is a schematic structural diagram of the device for beef origin tracing based on artificial intelligence in the embodiment of the present application; Figure 3 It is a schematic structural diagram of the electronic device in this embodiment; Figure 4 It is the hardware structure of the electronic device in this embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] Figure 1 It is a schematic flowchart of the method for beef origin tracing based on artificial intelligence in the embodiment of the present application. As Figure 1 shown, it includes: inputting the sample basic data extraction matrix, sample isotope extraction matrix, and sample metabolite extraction matrix of the target beef into the beef tracing model to be trained for beef origin prediction; calculating the loss value between the predicted beef origin and the corresponding beef origin label to adjust the model parameters of the beef tracing model to be trained based on the loss value. If the set standard for ending the model parameter adjustment is not reached, continue training until the set standard is reached; using the beef tracing model that has completed training as the beef tracing model of the target beef to predict the origin of the beef to be traced to determine whether the origin is the origin of the target beef.

[0018] Optionally, the method further includes: Vectorize the isomorphic sample basic data, sample isotope data, and sample metabolome data respectively to generate a sample basic vector, a sample isotope vector, and a sample metabolism vector; Perform encoding conversion on the sample basic vector, the sample isotope vector, and the sample metabolism vector respectively to generate a sample basic data extraction matrix, a sample isotope extraction matrix, and a sample metabolism extraction matrix.

[0019] Optionally, the sample basic data includes at least one of bovine species, collection date, meat classification, muscle part, meat quality characteristics, and nutritional components; the sample metabolome data includes metabolites; the sample isotope data includes the measured isotope types, and the isotope types include nitrogen, carbon, and oxygen.

[0020] Optionally, the step of vectorizing the isomorphic sample basic data, sample isotope data, and sample metabolome data respectively to generate a sample basic vector, a sample isotope vector, and a sample metabolism vector includes: Perform descriptive form annotation on the isomorphic sample basic data, sample isotope data, and sample metabolome data respectively; Based on the annotated descriptive form, match the corresponding coding components from the coding rule library to perform embedding on the isomorphic sample basic data, sample isotope data, and sample metabolome data respectively and generate a sample basic vector, a sample isotope vector, and a sample metabolism vector accordingly; Wherein, the coding rule library includes numerical data coding components, categorical data coding components, and text data coding components.

[0021] Optionally, the method further includes: Obtain a knowledge graph constructed for the target beef; Map the sample basic data, sample isotope data, and sample metabolome data to the entities and relationships in the knowledge graph to perform isomorphic processing on the feature representation dimensions of the sample basic data, sample isotope data, and sample metabolome data of the target beef.

[0022] Optionally, the entities of the knowledge graph include feeding environment, feed ingredients, metabolites, bovine species, and muscle parts, and the edges in the knowledge graph are connections between entities, representing the association relationship between the target beef and the feeding environment, the association relationship between the feed ingredients and the metabolites, the relationship between the bovine species and the growth cycle, and the relationship between the muscle parts and the metabolic characteristics.

[0023] Optionally, mapping the sample basic data, sample isotope data, and sample metabolome data to the entities and relationships in the knowledge graph to perform isomorphic processing on the feature characterization dimensions of the sample basic data, sample isotope data, and sample metabolome data of the target beef includes: Invoking a pre-constructed mapping rule library; Analyzing the heterogeneous semantic association relationships among the sample basic data, sample isotope data, and sample metabolome data; Based on the heterogeneous semantic association relationships, using the mapping rule library, mapping the sample basic data, sample isotope data, and sample metabolome data to the entities and relationships in the knowledge graph to perform isomorphic processing on the feature characterization dimensions of the sample basic data, sample isotope data, and sample metabolome data of the target beef.

[0024] Optionally, the mapping rule library includes sample basic data mapping rules, sample isotope mapping rules, sample metabolome data mapping rules, and cross-data type relationship mapping rules. The sample basic data mapping rules include sample basic data entity mapping rules and sample basic data relationship mapping rules. The sample isotope mapping rules include isotope data entity mapping rules and isotope data relationship mapping rules. The sample metabolome data mapping rules include metabolome data entity mapping rules and metabolome data relationship mapping rules.

[0025] 1. The sample basic data mapping rules can include, for example: In the entity mapping rules: Cattle breed: Directly map the cattle breed names recorded in the sample basic data, such as "Angus cattle", "Wagyu cattle", "Charolais cattle", etc., precisely to the corresponding cattle breed entities in the knowledge graph. For example, if the cattle breed in the data is "Angus cattle", it is mapped to the "Angus cattle" entity node in the knowledge graph.

[0026] Gender: Map the gender information in the sample basic data, such as "bull" and "cow", to the gender entities related to cattle in the knowledge graph. If the data shows "cow", it is mapped to the "cow" entity in the knowledge graph.

[0027] Age: Perform segmented mapping according to the numerical range of age. For example, map cattle aged 0 - 6 months to the "calf" entity in the knowledge graph; map those aged 6 months - 18 months to the "young cattle" entity; and map those over 18 months to the "adult cattle" entity.

[0028] Weight: Map according to the weight range. For example, map cattle with a weight less than 200 kg to the "lightweight cattle" entity; map those with a weight of 200 - 500 kg to the "medium-weight cattle" entity; and map those with a weight greater than 500 kg to the "heavyweight cattle" entity.

[0029] Feeding methods: Map feeding methods such as "free-range", "captive breeding", and "semi-free-range and semi-captive breeding" to the corresponding feeding method entities in the knowledge graph. If the data is "free-range", it corresponds to the "free-range" entity in the knowledge graph.

[0030] In the relationship mapping rules: Cattle breed and meat quality characteristics: Establish the relationship between the cattle breed entity and the meat quality characteristic entity. For example, the "Angus cattle" entity is connected to the meat quality characteristic entities such as "rich marbling" and "tender meat" through the "has" relationship, indicating that Angus cattle have these meat quality characteristics.

[0031] Feeding method and growth environment: The "free-range" feeding method entity is connected to the growth environment entities such as "grassland" and "mountain" through the "associated" relationship because free-range cattle usually grow in such environments; the "captive breeding" feeding method entity establishes an "associated" relationship with the "cowshed" growth environment entity.

[0032] Age and weight: In the knowledge graph, establish a "corresponding" relationship between the "calf" entity and the "lightweight cattle" entity, the "young cattle" entity and the "medium-weight cattle" entity, and the "adult cattle" entity and the "heavyweight cattle" entity, reflecting the approximate corresponding relationship between age and weight.

[0033] 2. The mapping rules for sample isotope data include the following rules: In the entity mapping rules: Carbon isotope ( ): According to the numerical range of the carbon isotope, map it to the entity related to the feed type in the knowledge graph. If the value is between 10‰ and 14‰, map it to the " plant feed" entity; if it is between 18‰ and 22‰, map it to the " plant feed" entity.

[0034] Nitrogen isotope ( ): According to the numerical situation of the nitrogen isotope, map it to the entity related to the nutrition level. For example, when the value is relatively high, map it to the "high-protein feed feeding" entity; when the value is relatively low, map it to the "ordinary feed feeding" entity.

[0035] Oxygen isotope ( ): Based on the correlation between the oxygen isotope value and the geographical region, map it to the geographical region entity in the knowledge graph. For example, a specific value range corresponds to a specific water source or breeding area entity.

[0036] Relationship mapping rules Carbon isotope and feed type: " The entity of "plant feed" is connected to the entity of the numerical range of carbon isotope ( ) through the "influence" relationship, indicating that plant feed will cause the carbon isotope in cattle to show a corresponding numerical range; the entity of " plant feed" also establishes an "influence" relationship with the corresponding entity of the numerical range of carbon isotope.

[0037] Nitrogen isotope and nutritional level: The entity of "feeding with high-protein feed" is connected to the entity of the high value range through the "cause" relationship, meaning that feeding with high-protein feed will cause the nitrogen isotope in cattle to show a higher value; the entity of "feeding with ordinary feed" establishes a "cause" relationship with the entity of the low value range.

[0038] Oxygen isotope and geographical region: The geographical region entity is connected to the corresponding entity of the numerical range of oxygen isotope through the "association" relationship, indicating that factors such as the water source in this geographical region will cause the oxygen isotope in cattle to show a specific numerical range.

[0039] 3. The sample metabolome data mapping rules include the following rules: Entity mapping rules Amino acids: Map various amino acids detected in the sample metabolome data, such as "alanine", "glycine", "glutamic acid", etc., to the corresponding amino acid entities in the knowledge graph.

[0040] Fatty acids: Map different types of fatty acids, such as "saturated fatty acids", "unsaturated fatty acids", "monounsaturated fatty acids", etc., to the corresponding fatty acid entities in the knowledge graph.

[0041] Carbohydrates: Map the detected carbohydrates, such as "glucose", "fructose", "lactose", etc., to the carbohydrate entities in the knowledge graph.

[0042] Relationship mapping rules Amino acids and nutritional functions: The entity of "alanine" is connected to the nutritional function entities such as "providing energy" and "participating in protein synthesis" through the "has" relationship, indicating that alanine has these nutritional effects; other amino acid entities also establish similar relationships with the corresponding nutritional function entities.

[0043] Fatty acids and health effects: The entity of "saturated fatty acids" is connected to the health effect entities such as "increasing cholesterol content" through the "association" relationship; the entity of "unsaturated fatty acids" establishes an "association" relationship with the health effect entities such as "reducing cholesterol content".

[0044] Carbohydrates and energy metabolism: The "glucose" entity is connected to the "main substrate of energy metabolism" entity through the "acts as" relationship, reflecting the important role of glucose in energy metabolism; other carbohydrate entities are also corresponding to entities related to energy metabolism.

[0045] 4. The comprehensive mapping rules include the following rules: Cross-data type relationship mapping rules Cattle breeds and metabolites: Certain cattle breed entities have a "tend to produce" relationship with specific metabolite entities. For example, the "Wagyu" entity is connected to the "high content of unsaturated fatty acids" entity through the "tend to produce" relationship because the meat of Wagyu usually contains a relatively high proportion of unsaturated fatty acids.

[0046] Feeding methods and isotope characteristics: The "free-range" feeding method entity is connected to specific carbon, nitrogen, and oxygen isotope value range entities through the "associated with" relationship, reflecting the relationship between the isotope characteristics of free-range cattle and the feeding method; the same applies to the "captive" feeding method.

[0047] Metabolites and meat quality characteristics: The "high content of inosinic acid" metabolite entity is connected to the "delicious meat" meat quality characteristic entity through the "affects" relationship, indicating the influence of inosinic acid content on meat deliciousness; other metabolite entities have similar relationships with corresponding meat quality characteristic entities.

[0048] In a specific application scenario, the technical processing process for realizing isomorphism is as follows: 1. Parameter description Data set: Let the sample basic data set be , in the scenario of beef origin traceability, The data elements in can be information such as cattle breed, collection date, meat classification, muscle part, meat quality characteristics, nutritional components, etc. For example represents the cattle breed information of "Simmental cattle".

[0049] The sample isotope data set is , where is the data related to the measured isotope type, such as the specific measured values of isotopes such as nitrogen, carbon, and oxygen. For example can be the measured value of carbon 13 in beef.

[0050] The sample metabolome data set is , represents the data of metabolites, such as the content of a certain amino acid or fatty acid. For example is the content of inosinic acid in beef.

[0051] Knowledge graph: The entity set in the knowledge graph is , the entity can be the feeding environment, feed ingredients, metabolites, cattle breed, muscle part, etc. For example represents the "grassland feeding environment".

[0052] The relationship set is , and the relationships represent the association relationship between the target beef and the feeding environment, the association relationship between the feed ingredients and the metabolites, the relationship between the cattle breed and the growth cycle, the relationship between the muscle part and the metabolic characteristics, etc. For example represents the relationship of "the feeding environment suitable for the cattle breed".

[0053] Mapping rule library: The set of sample basic data entity mapping rules , where are the rules for mapping sample basic data to the entities in the knowledge graph. For example is the rule for mapping the information of the cattle breed "Simmental cattle" to the entity "Simmental cattle" in the knowledge graph.

[0054] The set of sample basic data relationship mapping rules , are the rules for mapping the relationships between sample basic data to the relationships in the knowledge graph. For example can map the "corresponding relationship between cattle breed and muscle part" to the corresponding relationship in the knowledge graph.

[0055] The set of sample isotope data entity mapping rules , are the rules for mapping sample isotope data to the entities in the knowledge graph. For example maps the measurement value of carbon 13 to the entity "carbon 13 content characteristic" in the knowledge graph.

[0056] The set of sample isotope data relationship mapping rules , are the rules for mapping the relationships between sample isotope data to the relationships in the knowledge graph.

[0057] The set of sample metabolome data entity mapping rules , are the rules for mapping sample metabolome data to the entities in the knowledge graph. For example maps the inosine monophosphate content to the entity "inosine monophosphate content characteristic" in the knowledge graph.

[0058] The set of sample metabolome data relationship mapping rules , are the rules for mapping the relationships between sample metabolome data to the relationships in the knowledge graph.

[0059] The set of cross-data type relationship mapping rules , are the rules for mapping the relationships between different types of data to the relationships in the knowledge graph. For example, the potential relationship between the cattle breed and the carbon-13 content can be mapped to the corresponding relationship in the knowledge graph.

[0060] 2. Analyze heterogeneous semantic association relationships To analyze the heterogeneous semantic association relationships among the sample basic data, sample isotope data, and sample metabolome data, this application uses semantic similarity calculation, employs cosine similarity based on word vectors, and introduces context weights to more accurately measure the associations.

[0061] For two data elements and , their vector representations are respectively and , and the context weights are respectively and . The context weights can be determined according to factors such as the importance and occurrence frequency of the data in the entire dataset. For example, in the traceability of beef origin, certain key metabolite data have higher context weights.

[0062] The cosine similarity formula is: where is the dot product of the vectors, and are the norms of the vectors respectively. By calculating the similarities between all pairs of data elements, this application can obtain a similarity matrix , where , and are the elements in .

[0063] 3. Perform mapping based on the mapping rule library Entity mapping: For the sample basic data , this application needs to determine the most suitable entity mapping rule. Define a scoring function to measure the quality of using the rule for mapping. The scoring function considers semantic similarity and the credibility of the rule (the credibility can be determined according to factors such as the historical usage effect of the rule): Then the entity in the knowledge graph mapped to is: , similarly, for the sample isotope data and the sample metabolome data , there are:

[0064]

[0065]

[0066]

[0067] Relationship mapping: For the relationships between the sample basic data , define a scoring function to measure the pros and cons of using the rule for relationship mapping, considering the semantic matching degree of the relationship and the credibility of the rule : Then the mapped knowledge graph relationship is: , similarly, for the relationships between sample isotope data, relationships between sample metabolome data, and relationships across data types, corresponding scoring functions and rule sets are used for mapping respectively.

[0068] The technical benefits of the above isomorphism processing means are described as follows: 1. Semantic similarity formula Calculates the semantic similarity between two data elements and . Among them is the cosine similarity calculation method, which is used to measure the similarity degree of two vectors in direction and reflects the similarity at the semantic level of data elements. And introduces context weights and . These two weights consider factors such as the importance and occurrence frequency of data elements in the entire dataset, making the similarity calculation more in line with the actual application scenario and being able to capture the semantic association between data elements more accurately.

[0069] During the isomorphism process, it provides a key basis for subsequent entity mapping and relationship mapping. Accurate semantic similarity calculation can help this application determine the most matching mapping of data elements to entities and relationships in the knowledge graph, avoiding errors caused by simple matching, thereby improving the accuracy of mapping. For example, in the beef traceability scenario, it can more accurately judge the degree of association between the cattle breed data of "Simmental cattle" and the corresponding entity in the knowledge graph.

[0070] 2. Entity mapping scoring function and mapping formula The scoring function comprehensively considers the semantic similarity between the data element and the result after mapping using the rule , as well as the credibility of the rule The credibility of a rule can be determined based on factors such as the historical use effect of the rule, which reflects the reliability of the rule in practical applications. By calculating the scores of all mapping rules and selecting the entity corresponding to the rule with the highest score as the mapping result, it ensures that the data element can be mapped to the most suitable knowledge graph entity.

[0071] This approach avoids possible errors that may occur when mapping based solely on semantic similarity. This is because some mapping rules, although they look similar in semantics, are not accurate or reliable in practical applications. A comprehensive evaluation combined with rule credibility can improve the accuracy and stability of entity mapping. In beef traceability, various beef data (such as cattle breed, isotope content, etc.) can be more accurately mapped to appropriate entities in the knowledge graph, providing a more accurate data basis for subsequent analysis.

[0072] 3. Relationship mapping score function and mapping formula Similar to entity mapping, the scoring function of relationship mapping is Considering the data element pairs Rules of Use The semantic matching degree between the mapped relations and the credibility of the rules By calculating the scores of all possible relationship mapping rules and selecting the relationship corresponding to the rule with the highest score as the mapping result, accurate mapping of the relationship between data elements is achieved.

[0073] In the isomorphism process, accurate relationship mapping can capture the complex associations between different data elements. In the beef traceability scenario, there are multiple relationships between different types of data (such as cattle breeds and breeding environments, feed ingredients and metabolites, etc.). Through this relationship mapping method that combines semantic matching and rule credibility, these relationships can be more comprehensively and accurately mapped to the knowledge graph, which helps to explore the deeper intrinsic connections between data.

[0074] Optionally, the encoding conversion of the sample basic vector, the sample isotope vector, and the sample metabolic vector is performed respectively to generate a sample basic data extraction matrix, a sample isotope extraction matrix, and a sample metabolic extraction matrix, including: Based on the input layer, the normalized sample basis vector, sample isotope vector, and sample metabolic vector are input into the encoder layer, wherein the encoder layer includes a plurality of hidden layers, and the number of neurons decreases as the number of layers increases; Based on the encoder layer, the sample basis vector, the sample isotope vector, and the sample metabolism vector are converted into a latent space with a dimension lower than that of each basis vector to generate a sample basis hidden vector, a sample isotope hidden vector, and a sample metabolism hidden vector; Based on the decoder layer, reconstruct the sample base hidden vector, sample isotope hidden vector, and sample metabolite hidden vector to reconstruct the sample base data extraction vector, sample isotope extraction vector, and sample metabolite extraction vector, where the decoder layer includes multiple hidden layers, and the number of neurons increases with the increase in the number of layers; Based on the output layer, generate a sample base data extraction matrix, a sample isotope extraction matrix, and a sample metabolite extraction matrix based on the sample base data extraction vector, the sample isotope extraction vector, and the sample metabolite extraction vector.

[0075] Optionally, based on the encoder layer, compress the sample base vector, sample isotope vector, and sample metabolite vector respectively to convert the sample base vector, sample isotope vector, and sample metabolite vector into a latent space with a dimension lower than theirs and thereby generate a sample base hidden vector, a sample isotope hidden vector, and a sample metabolite hidden vector.

[0076] Optionally, combine the sample base data extraction vector, the sample isotope extraction vector, and the sample metabolite extraction vector row by row or column by column to generate a sample base data extraction matrix, a sample isotope extraction matrix, and a sample metabolite extraction matrix.

[0077] Specifically, in an application scenario, the technical processing process of generating each extraction matrix is as follows: 1. Parameter Definition The normalized sample base vector is , where represents the dimension of the sample base vector. In the scenario of beef origin traceability, contains the results of vectorization and normalization of information such as cattle breed, collection date, meat classification, muscle part, meat quality characteristics, and nutritional components. For example, the cattle breed can be represented by one-hot encoding, and the nutritional components can be represented by specific numerical values. The normalized sample isotope vector is , is the dimension of the sample isotope vector, and its elements are the processed values of the measured values of isotopes such as nitrogen, carbon, and oxygen. The normalized sample metabolite vector is , is the dimension of the sample metabolite vector, which contains the results of vectorization and normalization of metabolite-related data, such as the content of a certain amino acid or fatty acid. The sample base hidden vector is , ; the sample isotope hidden vector is , ; the sample metabolite hidden vector is , These hidden vectors are obtained after being compressed by the encoder layer and are in a lower-dimensional latent space. The sample basic data extraction vector is , the sample isotope data extraction vector is , the sample metabolite data extraction vector is , and they are obtained after being reconstructed by the decoder layer, with the same dimension as the original input vector. The sample basic data extraction matrix is or (depending on combination by rows or columns), the sample isotope data extraction matrix is or , the sample metabolite data extraction matrix is or .

[0078] Network layer parameters: The weight matrix of the rd layer of the encoder layer is , and the bias vector is . For encoding the sample basic vector, assume the encoder has layers, then . The weight matrix of the rd layer of the decoder layer is , and the bias vector is . For decoding the sample basic vector, assume the decoder has layers, then .

[0079] Activation function: Assume the activation function is , and common activation functions such as the ReLU function .

[0080] 2. Working process of the encoder layer The encoder layer compresses the input vector through multiple hidden layers and maps it to a lower-dimensional latent space. For the sample basic vector , the output of the rd layer of the encoder layer is calculated as follows: , where . After being encoded through layers, the sample basic hidden vector is obtained.

[0081] Similarly, for the sample isotope vector and the sample metabolite vector , there are: , , where , . Finally, the sample isotope hidden vector and the sample metabolite hidden vector 。

[0082] The encoder layer gradually reduces the dimension of the vector through a series of linear transformations (weight matrix multiplication and bias addition) and non-linear transformations (activation functions), extracts and compresses the important features in the original data into a low-dimensional latent space. The number of neurons decreases as the number of layers increases, enabling the model to automatically learn the core features of the data and remove redundant information.

[0083] In the scenario of beef origin traceability, the original sample vector contains a large amount of information, and some of this information is redundant or not directly helpful for origin traceability. Through the compression of the encoder layer, the most critical features can be extracted, reducing the complexity of the data and the workload of subsequent calculations. Moreover, the low-dimensional latent space can better represent the internal structure of the data, facilitating the learning and analysis of subsequent models.

[0084] 3. Technical processing process of the decoder layer The decoder layer reconstructs the hidden vector to recover a vector similar to the original input vector. For the sample-based hidden vector , the output of the th layer of the decoder layer is calculated as follows: , where . After layers of decoding, the sample-based data extraction vector is obtained. Similarly, for the sample isotope hidden vector and the sample metabolite hidden vector , there are: , , where , . Finally, the sample isotope data extraction vector and the sample metabolite data extraction vector are obtained.

[0085] The decoder layer is the inverse process of the encoder layer. Through a series of linear and non-linear transformations, it restores the low-dimensional hidden vector to the dimension of the original vector. Its goal is to make the reconstructed vector as close as possible to the original input vector, thereby verifying whether the features extracted by the encoder layer contain sufficient information to restore the original data.

[0086] In beef origin traceability, the reconstruction process of the decoder layer can help this application evaluate the quality of the features extracted by the encoder layer. If the reconstructed vector is very close to the original vector, it indicates that the encoder layer has successfully retained the key information of the data. At the same time, the reconstructed vector can remove the noise and interference in the original data, obtaining cleaner and more representative data, which is beneficial for subsequent analysis and modeling.

[0087] 4. Output layer generates matrix Extract the sample basic data extraction vector , the sample isotope data extraction vector and the sample metabolite data extraction vector are combined row by row or column by column respectively to obtain the corresponding data extraction matrix. If combined by row, for the sample basic data extraction matrix , assuming there are sample basic data extraction vectors , then . Similarly, the sample isotope data extraction matrix and the sample metabolite data extraction matrix can be obtained.

[0088] The process of combining vectors into a matrix is to facilitate subsequent data processing and analysis. The matrix form can organize the characteristic data of multiple samples together, facilitating batch calculation and operation.

[0089] In beef origin traceability, the matrix-form data can be directly input into the subsequent beef traceability model for training and prediction. For example, when using a deep learning model for origin prediction, the matrix-form data can be processed in batches as input to improve calculation efficiency. Moreover, the matrix structure can clearly show the characteristic relationships between different samples, helping the model better learn and discover the patterns in the data.

[0090] Optionally, inputting the sample basic data extraction matrix, the sample isotope extraction matrix, and the sample metabolite extraction matrix into the beef traceability model to be trained for beef origin prediction includes: Based on the vector extraction layer, the sample basic data extraction vector, the sample isotope extraction vector, and the sample metabolite extraction vector are extracted from the sample basic data extraction matrix, the sample isotope extraction matrix, and the sample metabolite extraction matrix; based on the positional encoding layer, sequence position information perception is performed on the extracted sample basic data extraction vector, sample isotope extraction vector, and sample metabolite extraction vector, so as to add positional encoding to the sample basic data extraction vector, sample isotope extraction vector, and sample metabolite extraction vector and generate a sample basic position perception vector, a sample isotope position perception vector, and a sample metabolite position perception vector accordingly; based on the multi-head self-attention mechanism layer, multi-dimensional information fusion is performed on the sample basic position perception vector, the sample isotope position perception vector, and the sample metabolite position perception vector to generate a three-source data collaborative feature vector; based on the feed-forward neural network layer, non-linear abstraction is performed on the comprehensive feature vector to generate a multi-modal deep abstraction feature vector; based on the classifier layer, classification decision is performed on the multi-modal deep abstraction feature vector to map the multi-modal deep abstraction feature vector into the constructed beef origin category space, obtain scores for different beef origins, and convert the scores into probability distributions.

[0091] Specifically, in an application scenario, the innovative implementation process of the above model training is as follows: 1. Parameter Definition Sample basic data extraction matrix , where represents the number of samples, is the dimension of the sample basic data extraction vector. In the beef origin traceability scenario, each row of the matrix represents the sample basic data extraction vector of a sample, and these vectors contain the features after encoding and conversion of information such as cattle breed, collection date, and meat classification. Sample isotope data extraction matrix , is the dimension of the sample isotope data extraction vector, and the elements are the features related to the measurement values of isotopes such as nitrogen, carbon, and oxygen. Sample metabolite data extraction matrix , is the dimension of the sample metabolite data extraction vector, which contains metabolite-related features. Sample basic data extraction vector ( ), sample isotope data extraction vector , sample metabolite data extraction vector . Sample basic position perception vector , sample isotope position perception vector , sample metabolite position perception vector . Three-source data collaborative feature vector , multi-modal deep abstraction feature vector .

[0092] Position encoding matrix , , , used to add position information to the vector.

[0093] Weight matrix in the multi-head self-attention mechanism: query matrix , key matrix , value matrix ([[]] , is the number of heads, is the dimension of the input vector, where are respectively taken , , ), the concatenated weight matrix .

[0094] Weight matrix of the feed-forward neural network layer , , bias vector , .

[0095] Weight matrix of the classifier layer , bias vector , where is the number of categories of beef origin.

[0096] For the vector extraction layer: Extract vectors from the sample basic data extraction matrix , sample isotope data extraction matrix , sample metabolite data extraction matrix , expressed as: , this step is to convert the data in matrix form into vector form, which is convenient for subsequent layers to process the feature vectors of each sample. In the beef traceability scenario, different samples come from different beef individuals. Separating their feature vectors can enable more detailed analysis of each sample.

[0097] For the position encoding layer: Add position encoding to each vector to perceive the sequence position information. For the sample basic data extraction vector , its position-encoded vector is: , similarly, for the sample isotope data extraction vector and the sample metabolite data extraction vector, there are: , , position encoding adopts a trigonometric function form. For example, for the encoding value at the pos-th position and the -th dimension: , .

[0098] Positional encoding enables the model to perceive the relative position of vectors in a sequence. In multi-source data fusion, there is an order relationship among different types of data, and adding position information helps the model better understand the structure and association between data. In beef traceability, the sampling order of samples, the arrangement order of different data, etc. contain useful information, and positional encoding can incorporate this information into the feature vectors.

[0099] For the multi-head self-attention mechanism layer, the single-head attention calculation: For each head , calculate the query , key and value : , where is the concatenated position-aware vector (e.g., etc.).

[0100] Attention score calculation: , and the multi-head attention concatenation combines the attention results of heads: , and finally obtains the collaborative feature vector of the three-source data . The multi-head self-attention mechanism can capture the dependencies between data from multiple different representation subspaces, realizing multi-dimensional information fusion. In beef traceability, the basic data of samples, isotope data, and metabolic data each contain information in different aspects. The multi-head self-attention mechanism can automatically learn the internal associations between these data and fuse them into a collaborative feature vector to represent the characteristics of beef more comprehensively.

[0101] For the feed-forward neural network layer, perform a non-linear transformation on the collaborative feature vector of the three-source data : , , and the feed-forward neural network further abstracts and transforms the collaborative feature vector through a non-linear activation function (such as ReLU) to mine the deep features in the data. In beef traceability, there are complex non-linear relationships in the characteristics of beef from different origins, and the feed-forward neural network can learn these complex relationships to generate more discriminative multi-modal deep abstract feature vectors.

[0102] For the classifier layer, map the multi-modal deep abstract feature vector to the beef origin category space: , where is the score vector for different beef origins.

[0103] Convert the scores to a probability distribution through the softmax function: 。

[0104] The classifier layer makes a classification decision on the beef origin based on the deep abstract feature vector. The softmax function converts the scores into a probability distribution, making the results interpretable. In beef traceability, this probability distribution can intuitively represent the possibility of the beef sample coming from each origin, helping to determine the true origin of the beef.

[0105] Specifically, in an application scenario, the calculation process of the above three-source data collaborative feature vector is as follows: Let the sample basic position perception vector set be ,where ; the sample isotope position perception vector set is ,where ; the sample metabolic position perception vector set is ,where 。Here is the number of samples, 、 、 are the dimensions of the three position perception vectors respectively.

[0106] To facilitate subsequent calculations, the three position perception vectors of each sample are concatenated to obtain the concatenated vector ,where 。

[0107] 2. Multi-head self-attention mechanism calculation 2.1 Single-head attention calculation The multi-head self-attention mechanism consists of multiple single-head attentions. For the th head ( , is the number of heads), it is necessary to calculate the query matrix 、the key matrix and the value matrix 。

[0108] First, define the weight matrix: the query matrix ,the key matrix ,the value matrix ,where and are the dimensions of the query / key space and the value space respectively.

[0109] For the concatenated vector of all samples, calculate: , , 。Then, calculate the attention scores: ,where, The function normalizes the attention scores so that their values are between [0, 1] and their sum is 1. The similarity between the query and the key is calculated, which is to prevent the dot product result from being too large.

[0110] 2.2 Multi-Head Attention Concatenation Concatenate the attention results of

[0111] .

[0112] Finally, the concatenated result is mapped to the desired dimension through a linear transformation to obtain the collaborative feature vector of the three-source data : , where: is the weight matrix after concatenation.

[0113] For a single sample , its collaborative feature vector of the three-source data is the th row of the matrix .

[0114] Through the above steps, the multi-head self-attention mechanism can capture the dependencies between the sample base position perception vector, the sample isotope position perception vector, and the sample metabolism position perception vector from multiple different representation subspaces, fuse the multi-source data in multiple dimensions, and finally generate the collaborative feature vector of the three-source data. This fusion method can fully explore the internal correlations between different data sources and provide more comprehensive and valuable feature information for subsequent beef origin prediction.

[0115] Figure 2 This is a schematic structural diagram of the device for beef origin tracing based on artificial intelligence in the embodiment of the present application. As Figure 2 shown, it includes: A first program unit for inputting the sample base data extraction matrix, the sample isotope extraction matrix, and the sample metabolism extraction matrix of the target beef into the beef origin tracing model to be trained for beef origin prediction; A second program unit for calculating the loss value between the predicted beef origin and the corresponding beef origin label, adjusting the model parameters of the beef origin tracing model to be trained based on the loss value, and continuing to train until the set standard for ending the model parameter adjustment is reached if the set standard is not met; A third program unit is used to use the beef traceability model that has completed training as the beef traceability model of the target beef to predict the origin of the beef to be traced, so as to determine whether the origin is the origin of the target beef.

[0116] Figure 3 This is a schematic structural diagram of the electronic device according to this embodiment. As Figure 3 shown, it includes a memory 301 and a processor 302. The memory stores computer-executable instructions, and the processor is used to run the computer-executable instructions to execute the method according to any one of the embodiments of the present application.

[0117] Figure 4 This is the hardware structure of the electronic device according to this embodiment. As Figure 4 shown, the hardware structure of the electronic device may include: a processor 401, a communication interface 402, a computer-readable medium 403, and a communication bus 404; wherein the processor 401, the communication interface 402, and the computer-readable medium 403 complete communication with each other through the communication bus 404; optionally, the communication interface 402 may be an interface of a communication module, such as an interface of a GSM module; wherein, the processor 401 may be specifically configured to execute the method of any of the above embodiments. The processor 401 may be a general-purpose processor, including a central processing unit (CPU for short), a network processor (NP for short), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The computer-readable medium 403 may be, but is not limited to, a random access storage medium (Random Access Memory, RAM), a read-only storage medium (Read Only Memory, ROM), a programmable read-only storage medium (Programmable Read-Only Memory, PROM), an erasable programmable read-only storage medium (Erasable Programmable Read-Only Memory, EPROM), an electrically erasable programmable read-only storage medium (Electric Erasable Programmable Read-Only Memory, EEPROM), etc.

[0118] The embodiment of the present application also provides a model training method, which includes: Input the sample basic data extraction matrix, sample isotope extraction matrix, and sample metabolite extraction matrix of the target beef into the beef traceability model to be trained for predicting the origin of the beef; Calculate the loss value between the predicted beef origin and the corresponding beef origin label, and adjust the model parameters of the beef traceability model to be trained based on the loss value. If the set standard for ending the model parameter adjustment is not reached, continue training until the set standard is achieved.

[0119] The embodiment of the present application also provides a method for tracing the origin of agricultural and sideline products based on artificial intelligence, which includes: Input the sample basic data extraction matrix, sample isotope extraction matrix, and sample metabolite extraction matrix of the target agricultural and sideline products of meat into the agricultural and sideline products traceability model to be trained for predicting the origin of the agricultural and sideline products; Calculate the loss value between the predicted origin of the agricultural and sideline products and the corresponding agricultural and sideline products origin label, and adjust the model parameters of the agricultural and sideline products traceability model to be trained based on the loss value. If the set standard for ending the model parameter adjustment is not reached, continue training until the set standard is achieved; Use the trained agricultural and sideline products traceability model as the agricultural and sideline products traceability model of the target agricultural and sideline products to predict the origin of the agricultural and sideline products to be traced, so as to determine whether the origin is the origin of the target agricultural and sideline products.

[0120] The above agricultural and sideline products can be, for example, pork, mutton, etc. The difference is that for different meats, the sample basic data, sample isotope data, and sample metabolite data can be constructed specifically. At the same time, when training the model, try different model architectures, such as convolutional neural network (CNN), recurrent neural network (RNN), etc., to determine the model structure most suitable for the meat data.

[0121] In addition, in terms of the selection of the loss function, the origin distribution and data characteristics of different meats affect the selection of the loss function. For example, if the origin distribution of a certain meat is unbalanced, the number of samples in some origins is much more than that in other origins. Using the traditional cross-entropy loss function will cause the model to be biased towards the origins with more samples. At this time, a weighted cross-entropy loss function can be adopted to assign higher weights to the origins with fewer samples to balance the influence of different origins.

[0122] Furthermore, in terms of training parameter adjustment, different meat data characteristics and model structures require different training parameters. For example, the choice of learning rate affects the convergence speed and performance of the model. For meats with large data fluctuations, a smaller learning rate needs to be selected to avoid oscillations during model training; while for meats with relatively stable data, the learning rate can be appropriately increased to accelerate the model's convergence speed. In addition, parameters such as batch size and number of training epochs also need to be adjusted, and the optimal combination of training parameters is determined through multiple experiments and validations.

[0123] Origin Traceability Prediction Example Test Sample 1: Angus cattle grazed on the grasslands of Xilingol League, Inner Mongolia (36 months old) Input Data: δ¹³C = -21.3‰ (Typical C3 plant feeding characteristics) Metabolome shows high linoleic acid content (0.82 mg / g) Model Output: Probability distribution: [Inner Mongolia: 0.89, Shandong: 0.07, Shanxi: 0.04] Conclusion: Correctly traced back to the origin in Inner Mongolia Test Sample 2: Luxi Yellow Cattle in Shandong (fed with grains, 18 months old) Input Data: δ¹ 5 N = 8.2‰ (High-protein feed marker) Metabolome shows low inosinic acid content (0.45 mg / g) Model Output: Probability distribution: [Shandong: 0.84, Inner Mongolia: 0.12, Shanxi: 0.04] Conclusion: Correctly traced back to the origin in Shandong Comparative Test (Traditional Method vs Method of the Present Application):

[0124] The above description is only a preferred embodiment of the present application and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solutions formed by the specific combination of the above technical features.

Claims

1. A method for tracing the origin of beef based on artificial intelligence, characterized in that, Including: Vectorize the isomorphic sample basic data, sample isotope data, and sample metabolome data respectively to generate a sample basic vector, a sample isotope vector, and a sample metabolism vector; Perform encoding conversion on the sample basic vector, sample isotope vector, and sample metabolism vector respectively to generate a sample basic data extraction matrix, a sample isotope extraction matrix, and a sample metabolism extraction matrix; Input the sample basic data extraction matrix, sample isotope extraction matrix, and sample metabolism extraction matrix of the target beef into the beef traceability model to be trained for predicting the origin of the beef; Calculate the loss value between the predicted beef origin and the corresponding beef origin label, and adjust the model parameters of the beef traceability model to be trained based on the loss value. If the set standard for ending the model parameter adjustment is not reached, continue training until the set standard is reached; Use the trained beef traceability model as the beef traceability model of the target beef to predict the origin of the beef to be traced, and determine whether the origin is the origin of the target beef.

2. The method according to claim 1, wherein The step of vectorizing the isomorphic sample basic data, sample isotope data, and sample metabolome data respectively to generate a sample basic vector, a sample isotope vector, and a sample metabolism vector includes: Perform description form annotation on the isomorphic sample basic data, sample isotope data, and sample metabolome data respectively; Based on the annotated description form, match the corresponding coding components from the coding rule library to perform embedding on the isomorphic sample basic data, sample isotope data, and sample metabolome data respectively, and generate a sample basic vector, a sample isotope vector, and a sample metabolism vector accordingly; Among them, the coding rule library includes numerical data coding components, categorical data coding components, and text data coding components.

3. The method according to claim 2, wherein The method further includes: Obtain the knowledge graph constructed for the target beef; Map the sample basic data, sample isotope data, and sample metabolome data to the entities and relationships in the knowledge graph to perform isomorphic processing on the feature representation dimensions of the sample basic data, sample isotope data, and sample metabolome data of the target beef.

4. The method according to claim 3, wherein The entities of the knowledge graph include feeding environment, feed ingredients, metabolites, cattle breeds, and muscle parts. The edges in the knowledge graph are connections between entities, representing the association relationship between the target beef and the feeding environment, the association relationship between the feed ingredients and metabolites, the relationship between the cattle breed and the growth cycle, and the relationship between the muscle part and the metabolic characteristics.

5. The method according to claim 4, wherein The step of mapping the sample basic data, sample isotope data, and sample metabolome data to the entities and relationships in the knowledge graph to perform isomorphic processing on the feature representation dimensions of the sample basic data, sample isotope data, and sample metabolome data of the target beef includes: Call the pre-constructed mapping rule library; Analyze the heterogeneous semantic association relationships among the sample basic data, sample isotope data, and sample metabolome data; Based on the heterogeneous semantic association relationship, using the mapping rule library, map the sample basic data, sample isotope data, and sample metabolome data to the entities and relationships in the knowledge graph, so as to perform isomorphic processing on the sample basic data, sample isotope data, and sample metabolome data of the target beef in terms of feature characterization dimensions.

6. The method according to claim 5, wherein The mapping rule library includes sample basic data mapping rules, sample isotope mapping rules, sample metabolome data mapping rules, and cross-data type relationship mapping rules. The sample basic data mapping rules include sample basic data entity mapping rules and sample basic data relationship mapping rules. The sample isotope mapping rules include isotope data entity mapping rules and isotope data relationship mapping rules. The sample metabolome data mapping rules include metabolome data entity mapping rules and metabolome data relationship mapping rules.

7. An apparatus for tracing the origin of beef based on artificial intelligence, characterized in that, It includes: A first program unit for inputting the sample basic data extraction matrix, sample isotope extraction matrix, and sample metabolism extraction matrix of the target beef into the beef traceability model to be trained for predicting the origin of the beef. A second program unit for calculating the loss value between the predicted beef origin and the corresponding beef origin label, so as to adjust the model parameters of the beef traceability model to be trained based on the loss value. If the set standard for ending the adjustment of the model parameters is not reached, continue training until the set standard is reached. A third program unit for using the beef traceability model that has completed training as the beef traceability model of the target beef to predict the origin of the beef to be traced, so as to determine whether the origin is the origin of the target beef.

8. An electronic device, characterized in that, It includes a memory and a processor. A computer-executable program is stored on the memory, and the processor is used to run the computer-executable program to realize the training and prediction of the beef traceability model.

9. A model training method, characterized in that, It includes: Vectorize the isomorphic sample basic data, sample isotope data, and sample metabolome data respectively to generate a sample basic vector, a sample isotope vector, and a sample metabolism vector. Perform encoding conversion on the sample basic vector, sample isotope vector, and sample metabolism vector respectively to generate a sample basic data extraction matrix, a sample isotope extraction matrix, and a sample metabolism extraction matrix. Input the sample basic data extraction matrix, sample isotope extraction matrix, and sample metabolism extraction matrix of the target beef into the beef traceability model to be trained for predicting the origin of the beef. Calculate the loss value between the predicted beef origin and the corresponding beef origin label, so as to adjust the model parameters of the beef traceability model to be trained based on the loss value. If the set standard for ending the adjustment of the model parameters is not reached, continue training until the set standard is reached.

Images