Feed microorganism detection data analysis method and system based on artificial intelligence

Abstract

The invention provides a feed microbiological detection data analysis method and system based on artificial intelligence, and the method comprises the steps: obtaining a basal feed microbiological detection sample, carrying out the feature shielding processing of key flora description knowledge points in a microbiological detection report, and generating a microbiological detection report after the feature shielding processing, and then knowledge point estimation is carried out on logic node information in combination with an initialized artificial intelligence network, so that the accuracy and comprehensiveness of flora description knowledge points are effectively improved, the generated extended feed microorganism detection sample enriches training data, and the performance of a microorganism feature extraction model is optimized. Finally, the trained microbial feature extraction model can output flora description knowledge points more accurately, data analysis support is provided for feed microbial detection, and the detection efficiency and accuracy are improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology. Specifically, it relates to a method and system for analyzing feed microorganism detection data based on artificial intelligence. Background Art

[0002] In the process of feed production, microorganism detection is a key link to ensure feed quality and safety. Traditional microorganism detection methods mainly rely on means such as laboratory culture and microscope observation. These methods are not only time-consuming and laborious, but also greatly affected by human factors, making it difficult to ensure the accuracy and consistency of detection results. Summary of the Invention

[0003] In view of the problems mentioned above, in combination with the first aspect of this application, embodiments of this application provide a method for analyzing feed microorganism detection data based on artificial intelligence. The method includes: Obtain basic feed microorganism detection samples. Each basic feed microorganism detection sample includes a microorganism detection report, key flora description knowledge points in the microorganism detection report, and logical node information of the key flora description knowledge points in the microorganism detection report; Perform feature masking processing on the key flora description knowledge points in the microorganism detection report to generate a microorganism detection report after feature masking processing; Use an initialized artificial intelligence network to estimate knowledge points of key logical nodes reflected by the logical node information based on the microorganism detection report after feature masking processing and the logical node information, and generate estimated flora description knowledge points; Update the key flora description knowledge points in the microorganism detection report of the basic feed microorganism detection sample to the estimated flora description knowledge points to generate an extended feed microorganism detection sample; the extended feed microorganism detection sample is used to train a microorganism feature extraction model to output flora description knowledge points based on the trained microorganism feature extraction model.

[0004] In another aspect, embodiments of this application also provide a feed production monitoring system, including a processor and a machine-readable storage medium. The machine-readable storage medium is connected to the processor. The machine-readable storage medium is used to store programs, instructions, or codes, and the processor is used to execute the programs, instructions, or codes in the machine-readable storage medium to implement the above method.

[0005] Based on the above aspects, in the embodiments of the present application, by obtaining a basic feed microbial detection sample, performing feature masking processing on the key flora description knowledge points in the microbial detection report to generate a microbial detection report after feature masking processing, and then combining with an initialized artificial intelligence network to estimate the knowledge points of logical nodes, the accuracy and comprehensiveness of the flora description knowledge points are effectively improved. The extended feed microbial detection sample generated therefrom enriches the training data, thereby optimizing the performance of the microbial feature extraction model. Finally, the trained microbial feature extraction model can more accurately output the flora description knowledge points, providing data analysis support for feed microbial detection and improving the detection efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 FIG. is a schematic execution flowchart of a method for analyzing feed microbial detection data based on artificial intelligence provided by an embodiment of the present application.

[0007] Figure 2 FIG. is a schematic hardware architecture diagram of a feed production monitoring system provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0008] The present application will be specifically described below with reference to the accompanying drawings of the specification. Figure 1 FIG. is a schematic flowchart of a method for analyzing feed microbial detection data based on artificial intelligence provided by an embodiment of the present application. The method for analyzing feed microbial detection data based on artificial intelligence will be introduced in detail below.

[0009] Step S110: Obtain a basic feed microbial detection sample. Each basic feed microbial detection sample includes a microbial detection report, key flora description knowledge points in the microbial detection report, and logical node information of the key flora description knowledge points in the microbial detection report.

[0010] Specifically, the basic feed microbial detection sample refers to a complete record containing microbial detection-related information generated by a feed production enterprise for a specific feed sample during microbial detection. Each basic feed microbial detection sample represents an independent detection instance.

[0011] The microbial detection report details the complete process from sample collection to detection results, including detection methods, use of detection instruments, detection data, etc. The key flora description knowledge points are the detection results and descriptions for specific flora (such as Escherichia coli, Salmonella, etc.) in the microbial detection report.

[0012] In this embodiment, assume that a feed production enterprise has multiple feed production workshops. To ensure the quality and safety of the feed, microbial detection of the feed is required. For example, for a certain type of pig feed produced in one of the workshops, its microbial detection report contains various aspects of content. The microbial detection report is a detailed document that records a series of information from the collection of feed samples to various detection methods, the use of detection instruments, and the final detection results.

[0013] The key flora description knowledge points therein may include specific types of harmful flora, such as the detection situation of Escherichia coli. In this report, the detected quantity, active state, etc. of Escherichia coli are detailedly recorded, which are the key flora description knowledge points about Escherichia coli.

[0014] And the logical node information of these key flora description knowledge points in the microbial detection report is also very important. For example, the report first elaborates that the sample collection location is a specific storage area in the feed production workshop, and this information has a logical connection with the subsequent flora detection. The detection result of Escherichia coli may be in a specific section of the report, such as in the "Harmful Microorganism Detection Results" section. The order, position of this section in the overall report structure, and its association with other sections (such as the sample collection section, detection method section, etc.) are all part of the logical node information. For another example, the starting and ending positions of the description knowledge points about Escherichia coli in this section are accurately marked, and these identification information constitutes the logical node information of this key flora description knowledge point about Escherichia coli in the microbial detection report.

[0015] The server will obtain such a complete basic feed microbial detection sample from the enterprise's detection laboratory information system, including the overall content of the microbial detection report, the description knowledge points of specific key flora (such as Escherichia coli, etc.), and the logical node information of these knowledge points in the report, for subsequent processing procedures.

[0016] Step S120: Perform feature masking processing on the key flora description knowledge points in the microbial detection report to generate a microbial detection report after feature masking processing.

[0017] In this embodiment, after receiving the basic feed microbial detection sample, the server starts the feature masking processing. First, the server obtains the set feature masking parameters. Assume that the enterprise determines a feature masking parameter based on past detection experience and data analysis, and this feature masking parameter is determined based on the global scale and specific requirements of the key flora description knowledge points in the microbial detection report.

[0018] Taking the previously mentioned microbial detection report of pig feed as an example, the key microbial group description knowledge points about Escherichia coli in the report cover multiple aspects and the overall global scale is relatively large. If the feature masking parameter is set to mask some key features according to a certain ratio, the server will determine the scale of the knowledge points for feature masking processing of the key microbial group description knowledge points of Escherichia coli based on this ratio.

[0019] Next, the server determines the starting point according to the structure and content distribution law of the microbial detection report. For example, in the microbial detection report of pig feed, the report is written in the order from the whole to the part, from the general detection process to the specific microbial group detection results. The server will start from the beginning of the report or the beginning of a specific section as the starting point, and extract the key microbial group description knowledge points from the microbial detection report according to the determined scale of knowledge points. For example, for the detection part of Escherichia coli, when the number of key microbial group description knowledge points related to Escherichia coli extracted reaches the required number of the scale of knowledge points, the extraction operation is stopped to form a set of key microbial group description knowledge points to be processed by feature masking.

[0020] Then, the server makes a preliminary mark for each key microbial group description knowledge point in the extracted set of key microbial group description knowledge points. For the knowledge points related to Escherichia coli, mark the relative position information in the microbial detection report, such as from which line it starts in the "Harmful Microorganism Detection Results" section and how many lines it occupies in total; at the same time, mark the type to which the knowledge point belongs, such as quantity type (the detection quantity of Escherichia coli) or status type (the active state of Escherichia coli).

[0021] According to the preliminary marking results, the server constructs an association graph of the key microbial group description knowledge points. In this graph, nodes represent each key microbial group description knowledge point, and edges represent the association relationships between knowledge points. For the two knowledge points of the detection quantity and active state of Escherichia coli, there is a logical connection between them, because the active state may affect the evaluation of the detection quantity, so there will be an edge connecting them in the association graph. The determination of the association relationship is based on the logical connection of the key microbial group description knowledge points in the microbial detection report and the relevance of the content of the key microbial group description.

[0022] In the association graph, the server determines the masking order according to the weights of nodes, the weights of edges, and the connection relationships between nodes. Suppose in the graph, the node of the detection quantity of Escherichia coli has a higher weight because it is more critical for evaluating the safety of feed quality, and compared with other auxiliary description knowledge points (such as auxiliary reagents used in the detection), its connection relationship with the node of the active state is closer. Therefore, when determining the masking order, those knowledge points with lower weights and looser connection relationships with the key nodes will be considered for masking first.

[0023] Perform feature masking processing on the key flora description knowledge point set according to the determined feature masking order and preliminary marking results. For each key flora description knowledge point, different masking strategies are adopted according to the type of the mark of the key flora description knowledge point and its positional relationship in the associated map. For example, for the knowledge point of the detection quantity of Escherichia coli of the quantity type, if its position in the map is relatively critical, a partial masking strategy may be adopted, such as only masking a part of the values or replacing the values with a specific identifier; while for some auxiliary description knowledge points, they may be directly completely masked. During the masking processing, the server updates the status of the key flora description knowledge point set in real time, generating an intermediate status knowledge point set during the feature masking processing, and this intermediate status knowledge point set records the results of each step during the masking process.

[0024] Finally, the server performs a rationality check on the intermediate status knowledge point set. Specifically, by analyzing the mark information, associated map and operation records of the masking process of each key flora description knowledge point, it is judged whether the feature masking processing result is reasonable. The content of the rationality check includes: whether the processing is carried out according to the predetermined masking order, whether the necessary logical relationships between the key flora description knowledge points are retained, and whether excessive masking causes the loss of key information. For example, if during the masking process, due to the wrong masking order, the logical relationship between the detection quantity and the active state of Escherichia coli is destroyed, or excessive masking causes the complete loss of the key information about the detection of Escherichia coli, then the server will detect an unreasonable situation. If an unreasonable situation is detected during the rationality check process, then return to the previous step for adjustment, re-determine the masking order or adjust the masking strategy until no unreasonable situation is detected during the rationality check process, and replace the original knowledge points in the microbial detection report with the knowledge points in the intermediate status knowledge point set in the masked state to generate a microbial detection report after feature masking processing.

[0025] Step S130, use the initialized artificial intelligence network to estimate the knowledge points of the key logical nodes reflected by the logical node information based on the microbial detection report after the feature masking processing, and generate estimated flora description knowledge points.

[0026] In this embodiment, after obtaining the microbial detection report after the feature masking processing, start using the initialized artificial intelligence network to estimate the knowledge points.

[0027] First, use the feature extraction sub-network of the initialized artificial intelligence network to perform feature extraction based on the microbial detection report after the feature masking processing and the logical node information, and generate a feature extraction result corresponding to the microbial detection report.

[0028] For the microbial detection report of pig feed, the server parses the report structure of the report after feature masking. For example, it parses each section in the report, such as the sample collection section, the detection method section, the flora detection result section, etc., and the content structure in each section. For example, in the flora detection result section, the detection results are listed in sequence according to different flora types.

[0029] Perform logical mapping on the parsed report structure data according to the logical node information. For example, the identification information in the logical node information indicates that the detection result of Escherichia coli is in a specific sub-region in the flora detection result section. The server corresponds this identification with the corresponding part in the report structure data, marking the association possibility of each part with the key flora description knowledge points (such as knowledge related to Escherichia coli) and its position relationship in the entire report logic of the microbial detection report.

[0030] Then, through a pre-set keyword library and semantic rules related to microbial detection, perform information screening on the report structure data after logical mapping. The keyword library contains vocabulary related to microbial detection such as "Escherichia coli", "colony forming unit", "activity", etc. The semantic rules stipulate the logical relationship and semantic combination method between these vocabulary. The server screens the report structure data after logical mapping based on these, removes the content irrelevant to the key information of microbial detection, and on the basis of the screened information, performs classification pre-definition on the remaining report structure data. For example, classify the information related to the flora type into one category, and classify the information related to the detection quantity into another category, generating the target microbial detection report content after screening and pre-definition classification.

[0031] According to the pre-defined classification categories, the server performs advanced classification on the target microbial detection report content, and groups and integrates the target microbial detection report content after advanced classification. Each group corresponds to a classification category, generating independent content groups. For example, take the content related to the detection quantity of Escherichia coli as one group, and take the content related to its activity status as another group.

[0032] Next, use the feature extraction sub-network of the initialized artificial intelligence network to extract intra-group semantic features for each content group. For the group of Escherichia coli detection quantity, extract semantic features such as "quantity range", "comparison with the standard value", etc., and mark the extracted semantic features to clarify the group to which each semantic feature belongs and its specific semantic role within the group, generating a set of intra-group semantic features with marks.

[0033] The server analyzes the semantic relationships between the semantic feature sets within each group and constructs a semantic relationship network based on these relationships. For example, there may be a logical association between the "quantity range" in the Escherichia coli detection quantity grouping and the "high or low activity" in the activity status grouping. This relationship is represented in the semantic relationship network in the form of a graph, where the nodes represent the respective semantic features and the edges represent the relationships between them.

[0034] Based on the identifiers in the logical node information, the server determines the logical connections between different content groupings and integrates the cross-group logical connections into the overall semantic relationship network to generate a target logical semantic network that includes the semantic relationships of each grouping and the logical relationships between the groupings.

[0035] Starting from the nodes of the target logical semantic network, the server extracts the node features of each node. The node features include the semantic connotation of the node (such as the specific numerical range meaning of the "quantity range" node), the connection degree of the node in the target logical semantic network (the number of connections with other nodes), and the type identifier of the node (whether it is a quantity type or a status type, etc.). At the same time, the server extracts the features of each edge in the target logical semantic network. The edge features include the relationship type (whether it is a causal relationship, a parallel relationship, or other relationships) and the relationship strength (indicating the degree of closeness of the association).

[0036] Finally, the server fuses the node features of each node and the features of each edge and optimizes the fused features. For example, it removes duplicate or redundant feature information. If there are partially overlapping descriptions of the same semantic connotation for two nodes, one of them is removed; it normalizes some features, such as unifying the quantity ranges represented in different formats into a standard format. The optimized feature representation serves as the feature extraction result corresponding to the microbial detection report.

[0037] Next, use the feature sub-network that initializes the artificial intelligence network to perform feature reduction based on the connection feature vectors of the key logical nodes reflected and the feature extraction result corresponding to the microbial detection report, and generate an estimated description knowledge point of the microbial flora.

[0038] The server receives the connection feature vectors of the key logical nodes and the feature extraction result corresponding to the microbial detection report. For example, the connection feature vectors of the key logical nodes contain the feature connection information between the key logical nodes related to the Escherichia coli detection result, and this information reflects the information on the association characteristics between the detection quantity and the activity status in the logical structure of the microbial detection report.

[0039] The server analyzes the connection feature vectors of the key logical nodes to determine the meaning of the connection features represented by each element in the connection feature vectors. For example, a certain element may represent the degree of positive correlation between the detection quantity and the activity status.

[0040] Map the parsed connection feature vector to the feature extraction results corresponding to the microbial detection report to establish the mapping relationship between the connection feature vector and the features in the feature extraction results, and determine the grouping basis for feature grouping according to the established mapping relationship. For example, if the connection feature vector indicates a strong correlation between the detection quantity and the activity state, then in the feature extraction results, the grouping basis for these two related features is determined as a strong correlation relationship.

[0041] Group the feature extraction results corresponding to the microbial detection report according to the determined grouping basis, and each group contains features related to the set type of connection relationship.

[0042] Perform within-group feature adjustment for each group, and construct the association between features within the group to generate the grouping features for each group. For example, in the group of detection quantity and activity state, according to their association relationship, adjust the proportional relationship or numerical conversion relationship between the quantity feature and the state feature, construct a closer association between them, and generate the grouping features.

[0043] According to the global logical relationship in the connection feature vector and the logical structure of the microbial detection report, the server determines the integration logic for the grouping features, and integrates the grouping features of each group according to the determined integration logic to combine the grouping features in different groups according to the logical relationship, generating the global feature representation.

[0044] Using the feature reduction sub-network that initializes the artificial intelligence network, the server constructs a feature reduction mapping, which is established based on the relationship between the original logical structure of the microbial detection report, the connection feature vector, and the feature extraction results.

[0045] According to the constructed reduction mapping, the server performs a preliminary reduction on the integrated global feature representation to generate a preliminarily reduced feature representation.

[0046] According to the logical integrity requirements of the microbial detection report and the specification requirements of the flora description knowledge points, the server determines the optimization and adjustment basis for the preliminarily reduced feature representation, and based on the optimization and adjustment basis, optimizes and adjusts the preliminarily reduced feature representation, and then performs an integrity check on the reduced feature after optimization and adjustment. If the integrity check passes, the reduced feature after optimization and adjustment is determined as the estimated flora description knowledge point. For example, for the estimated flora description knowledge point of Escherichia coli, it should conform to the specification format of the flora description in the microbial detection report, such as the representation of the quantity should be accurate to a certain degree, and the description of the activity state should use specific terms, etc. If the preliminarily reduced feature representation does not meet these requirements, it is optimized and adjusted until the integrity check passes, and it is determined as the final estimated flora description knowledge point.

[0047] Another way to estimate knowledge points using an initialized artificial intelligence network is as follows: The server uses the initialized artificial intelligence network to estimate the heat map corresponding to each key logical node reflected by the logical node information based on the microbial detection report and logical node information after feature masking processing. For example, for the key logical node related to Escherichia coli in the pig feed microbial detection report, the heat map will reflect the confidence level of each detection knowledge segment in the detection knowledge segment sequence as an estimated flora description knowledge point of the key logical node (such as the logic related to Escherichia coli detection).

[0048] For each key logical node, the server extracts i reference detection knowledge segments from the detection knowledge segment sequence according to the confidence level. Assuming i = 3, a detection knowledge segment is randomly extracted from the top 3 reference detection knowledge segments with higher confidence levels as the estimated flora description knowledge point corresponding to the key logical node. For example, if a knowledge segment about the detected quantity of Escherichia coli has a high confidence level in the heat map, it may be randomly selected as the estimated flora description knowledge point corresponding to the key logical node of the detected quantity of Escherichia coli.

[0049] Step S140: Update the key flora description knowledge points in the microbial detection report of the basic feed microbial detection sample to the estimated flora description knowledge points to generate an extended feed microbial detection sample. The extended feed microbial detection sample is used to train the microbial feature extraction model so as to output flora description knowledge points based on the trained microbial feature extraction model.

[0050] In this embodiment, after obtaining the estimated flora description knowledge points, an extended feed microbial detection sample is generated. Still taking the microbial detection sample of pig feed as an example, for the microbial detection report in the previously obtained basic feed microbial detection sample, the server updates the key flora description knowledge points (such as the original detected quantity and active state of Escherichia coli, etc.) to the estimated flora description knowledge points generated in step S130.

[0051] Suppose the original detected quantity of Escherichia coli is 100 colony forming units per gram of feed, and the detected quantity in the estimated flora description knowledge points obtained after the estimation in step S130 is 80 - 120 colony forming units per gram of feed (this is a possible estimation result), the server will replace the original detected quantity in the microbial detection report with this estimated value.

[0052] In this way, an extended feed microbial detection sample is generated. This extended feed microbial detection sample is different from the original basic feed microbial detection sample, and it contains the key flora description knowledge points after estimation and update.

[0053] Then, the server will use this extended feed microorganism detection sample to train the microorganism feature extraction model. During the training process, the microorganism feature extraction model will learn various feature relationships in the extended feed microorganism detection sample, including the logical relationships between different flora description knowledge points, the association relationships with other detection information (such as sample collection information, detection method information, etc.).

[0054] After learning and training with multiple such extended feed microorganism detection samples, the microorganism feature extraction model gradually improves and can finally output the knowledge points of flora description based on the trained microorganism feature extraction model. For example, when a new feed microorganism detection report is input, the trained microorganism feature extraction model can accurately output the knowledge points of the description of key flora (such as Escherichia coli, etc.), including relevant information such as the detection quantity and activity status. And the output information will be more accurate and in line with the actual situation because the microorganism feature extraction model has learned and optimized on a large number of extended feed microorganism detection samples.

[0055] Based on the above steps, the embodiment of the present application effectively improves the accuracy and comprehensiveness of the knowledge points of flora description by obtaining the basic feed microorganism detection sample, performing feature masking processing on the knowledge points of the description of the key flora in the microorganism detection report to generate a microorganism detection report after feature masking processing, and then combining with the initialized artificial intelligence network to estimate the knowledge points of the logical node information. The extended feed microorganism detection sample generated thereby enriches the training data and further optimizes the performance of the microorganism feature extraction model. Finally, the trained microorganism feature extraction model can more accurately output the knowledge points of flora description, providing data analysis support for feed microorganism detection and improving the detection efficiency and accuracy.

[0056] In a possible implementation manner, the logical node information includes the identification information indicating whether each detection knowledge segment in the microorganism detection report belongs to the knowledge points of the description of the key flora, and the identification information corresponding to the start logical node and the end logical node of the knowledge points of the description of the key flora in the microorganism detection report respectively. The method further includes: Step A110, embedding a logical anchor for positioning the knowledge points of the description of the key flora in the microorganism detection report after feature masking processing to generate an anchored detection report.

[0057] Step A120, embedding the anchor identification information corresponding to the logical anchor in the logical node information to generate anchored logical node information.

[0058] Step S130 includes: using the initialized artificial intelligence network to estimate the knowledge points of the key logical nodes reflected by the logical node information based on the anchored detection report and the anchored logical node information, and generating an estimated flora description knowledge point.

[0059] In this embodiment, taking the pig feed microorganism detection report as an example, the microorganism detection report contains many detection knowledge fragments. The identification information in the logical node information can clarify whether each detection knowledge fragment belongs to the key flora description knowledge point. For example, for the detection of Escherichia coli, there is a special identification indicating that a certain description about the quantity and activity of Escherichia coli belongs to the key flora description knowledge point, while other knowledge fragments unrelated to the calibration of the detection instrument have different identifications. At the same time, there are corresponding identification information for the starting logical node and the ending logical node of the key flora description knowledge point in the microorganism detection report, which helps to accurately locate the starting and ending positions of the Escherichia coli-related description in the report.

[0060] After performing feature masking processing on the key flora description knowledge points in the microorganism detection report, the server needs to embed logical anchors for locating the key flora description knowledge points into it to generate an anchored detection report. For example, for the previously mentioned Escherichia coli detection part, in the report content after feature masking processing, a logical anchor is embedded at the position where the detection quantity of Escherichia coli originally starts. This logical anchor contains a specific identification code, which is used to uniquely determine that this position is the starting point of the key flora description knowledge point related to the detection quantity of Escherichia coli. Similarly, a corresponding logical anchor is embedded at the end of the detection quantity description. Similar operations are also performed for other key flora description knowledge points such as the activity state of Escherichia coli, so as to completely generate an anchored detection report.

[0061] In terms of the logical node information, the server needs to embed the anchor identification information corresponding to the logical anchor to generate the anchored logical node information. For the logical node information related to the detection of Escherichia coli, the identification information corresponding to the previously determined starting logical node and ending logical node is updated by adding the identification information of the logical anchor. In this way, the logical node information corresponds to the logical anchor in the detection report after feature masking processing, making the logical relationship more clear and locatable.

[0062] When using the initialized artificial intelligence network for knowledge point estimation, the server performs knowledge point estimation on the key logical nodes reflected by the logical node information based on the anchored detection report and the anchored logical node information to generate an estimated flora description knowledge point. Taking the detection of Escherichia coli as an example, the initialized artificial intelligence network reads the content in the anchored detection report and accurately finds the key logical nodes related to Escherichia coli according to the identifiers in the anchored logical node information. For example, based on the logical anchor identifier, the logical node related to the detection quantity of Escherichia coli is located. The network will perform knowledge point estimation according to other surrounding detection knowledge fragments, logical relationships, and existing model parameters. For the detection quantity, it may comprehensively judge by combining the quantity relationships of other relevant microorganisms, detection environment information, etc., so as to obtain the estimated flora description knowledge point of this key logical node of the Escherichia coli detection quantity. Similarly, for key logical nodes such as the activity state of Escherichia coli, estimation is also carried out in this way based on the anchored detection report and the anchored logical node information, and finally a complete estimated flora description knowledge point is generated. This process makes the knowledge point estimation more accurate, because the logical anchor provides more accurate positioning and association information, which helps the artificial intelligence network better understand the logical relationships in the microbial detection report, so as to obtain an estimated flora description knowledge point that more conforms to the actual situation.

[0063] In a possible implementation manner, step S120 includes: Step S121, obtain a set feature masking parameter, and based on the global scale of the key flora description knowledge points in the microbial detection report and the set feature masking parameter, determine the knowledge point scale of the key flora description knowledge points for which feature masking processing is performed on the key flora description knowledge points in the microbial detection report.

[0064] First, the server obtains a set feature masking parameter. Under the enterprise's quality control standards and data analysis strategies, this set feature masking parameter is carefully determined. For example, in the microbial detection report of pig feed, the key flora description knowledge points contain information related to various microorganisms, such as the detection of Escherichia coli, Salmonella, etc. Suppose the set feature masking parameter of the enterprise is a ratio value determined based on historical data and current quality control priorities. This ratio value is to appropriately reduce the amount of some data in the key flora description knowledge points without affecting the overall data logical relationship. Based on the global scale of the key flora description knowledge points in the microbial detection report and this set feature masking parameter, the server determines the knowledge point scale of the key flora description knowledge points for which feature masking processing is performed on the key flora description knowledge points in the microbial detection report. If the total amount of description information about various key flora in the microbial detection report is large, according to the set ratio value, the server calculates the quantity scale of the key flora description knowledge points that need to be subjected to feature masking processing.

[0065] Step S122: Determine the starting point according to the structure and content distribution rule of the microbial test report. Starting from the starting point, extract the key flora description knowledge points from the microbial test report according to the scale of the knowledge points. When the number of the extracted key flora description knowledge points reaches the required number of the scale of the knowledge points, stop the extraction operation to form a set of key flora description knowledge points to be processed by feature masking.

[0066] Next, the server determines the starting point according to the structure and content distribution rule of the microbial test report. The pig feed microbial test report has a specific structure. Generally, it first elaborates on the relevant information of sample collection, including collection location, time, sample volume, etc., then details the detection method, and finally the key flora detection results and other content. The server determines the starting point based on this structural and content logical order. For example, start from the first flora description in the key flora detection result section as the starting point, and then extract the key flora description knowledge points from the microbial test report according to the previously determined scale of the knowledge points. During the extraction process, the server counts the number of key flora description knowledge points one by one. When the number of the extracted key flora description knowledge points reaches the required number of the scale of the knowledge points, stop the extraction operation, thereby forming a set of key flora description knowledge points to be processed by feature masking. Assume that under the determined scale of the knowledge points, a certain number of description knowledge points are extracted from the detection result part of Escherichia coli, and some knowledge points from the detection result part of Salmonella together constitute this set of key flora description knowledge points to be processed by feature masking.

[0067] Step S123: Conduct preliminary marking on each key flora description knowledge point in the set of the extracted key flora description knowledge points to obtain a preliminary marking result, where the preliminary marking result includes the relative position information of each key flora description knowledge point in the microbial test report and the type to which the knowledge point belongs.

[0068] After that, the server conducts preliminary marking on each key flora description knowledge point in the set of the extracted key flora description knowledge points to obtain a preliminary marking result. For each knowledge point in this set of key flora description knowledge points, the server marks its relative position information in the microbial test report. For example, for a description knowledge point of the detection quantity of Escherichia coli, mark its row position in the entire Escherichia coli detection result section, whether it is at the beginning, middle or end of the section. At the same time, mark the type to which the knowledge point belongs. For example, the detection quantity of Escherichia coli belongs to a numerical type knowledge point, while the active state of Escherichia coli belongs to a state type knowledge point. These preliminary marking results help the server better understand the role and significance of each key flora description knowledge point in the entire report.

[0069] Step S124: According to the preliminary labeling results, construct an association graph of key microbiota description knowledge points. In the association graph, nodes represent each key microbiota description knowledge point, and edges represent the association relationships between knowledge points. The determination basis of the association relationships is the logical connection of key microbiota description knowledge points in the microbial test report and the relevance of the content of the key microbiota description.

[0070] For example, one node represents the detected quantity of Escherichia coli, another node represents the activity state of Escherichia coli, and there are also nodes representing the detection situation of Salmonella, etc. Edges represent the association relationships between knowledge points. The determination basis of the association relationships is the logical connection of key microbiota description knowledge points in the microbial test report and the relevance of the content of the key microbiota description. From the perspective of logical connection, there is a logical association between the detected quantity of Escherichia coli and its activity state because the activity state may affect the accuracy of the detected quantity. Therefore, there will be an edge connecting these two nodes in the association graph. From the perspective of content relevance, for example, both Escherichia coli and Salmonella belong to harmful microbiota, and they have a certain association in the overall feed microbial safety assessment. Therefore, there will also be an edge connection between the description knowledge point nodes of these two related microbiota.

[0071] Step S125: In the association graph, determine the shielding order according to the weights of nodes, the weights of edges, and the connection relationships between nodes, and perform feature shielding processing on the key microbiota description knowledge point set according to the determined feature shielding order and the preliminary labeling results. Among them, when performing feature shielding processing, for each key microbiota description knowledge point, different shielding strategies are adopted according to the type of label of the key microbiota description knowledge point and its positional relationship in the association graph. During the shielding process, the state of the key microbiota description knowledge point set is updated in real time to generate an intermediate state knowledge point set during the feature shielding process, and the intermediate state knowledge point set records the results of each step during the shielding process.

[0072] After the construction of the association graph is completed, the server determines the shielding order in the association graph according to the weights of nodes, the weights of edges, and the connection relationships between nodes. The weight of a node reflects the importance of the key microbiota description knowledge point in the entire microbial test report. For example, the detected quantity of Escherichia coli is more crucial for judging whether the feed meets the microbial safety standards, so the weight of this node may be higher. The weight of an edge represents the tightness of the association relationship between knowledge points. For example, the edge weight between the detected quantity of Escherichia coli and its activity state is higher because their logical relationship is very tight. Based on these weights and the connection relationships between nodes, the server determines the shielding order. For example, for the key microbiota description knowledge point corresponding to a node with a lower weight and a relatively loose connection relationship with other key nodes, it may be given priority to be considered for shielding.

[0073] When performing feature masking processing, for each key flora description knowledge point, different masking strategies are adopted according to the type of the knowledge point marked by the key flora description knowledge point and its positional relationship in the associated graph. For the numerical knowledge point of the detection quantity of Escherichia coli, if its position in the associated graph is relatively independent and the weight is not particularly high, a partial masking strategy may be adopted, such as fuzzifying the specific value or replacing some numbers according to certain rules. For the status-type knowledge point of the activity state of Escherichia coli, if its association with other key knowledge points is close, a strategy of retaining some key descriptions and masking some auxiliary descriptions may be adopted. During the masking process, the server updates the status of the key flora description knowledge point set in real time to generate an intermediate status knowledge point set during the feature masking process, and this intermediate status knowledge point set records the results of each step during the masking process. For example, for each masked knowledge point, information such as which knowledge point is masked, what masking strategy is adopted, and the overall status of the masked knowledge point set is recorded.

[0074] Step S126, perform a rationality check on the intermediate status knowledge point set. Specifically, by analyzing the marking information of each key flora description knowledge point, the associated graph, and the operation record of the masking process, it is judged whether the feature masking processing result is reasonable. The content of the rationality check includes: whether the processing is carried out in accordance with the predetermined masking order, whether the necessary logical relationships between the key flora description knowledge points are retained, and whether excessive masking causes the loss of key information.

[0075] Then, by analyzing the marking information of each key flora description knowledge point, the associated graph, and the operation record of the masking process, it is judged whether the feature masking processing result is reasonable. The content of the rationality check includes whether the processing is carried out in accordance with the predetermined masking order, whether the necessary logical relationships between the key flora description knowledge points are retained, and whether excessive masking causes the loss of key information. For whether the processing is carried out in accordance with the predetermined masking order, the server checks whether the order in the actual masking process is consistent with the previously determined masking order. For example, if it is predetermined to mask the knowledge points corresponding to the nodes with low weight and loose connection first, but in the actual operation, the knowledge points corresponding to the nodes with high weight and tight connection are masked first, this does not meet the requirements. In terms of whether the necessary logical relationships between the key flora description knowledge points are retained, the server checks whether the logical relationship between the detection quantity of Escherichia coli and the activity state can still be traced and understood after masking. If excessive masking causes the loss of key information, for example, completely masking the detection quantity of Escherichia coli without retaining any information that can infer its approximate range, this will affect subsequent analysis and processing.

[0076] Step S127, if an unreasonable situation is detected during the rationality check, return to the above steps for adjustment, re-determine the shielding order or adjust the shielding strategy until no unreasonable situation is detected during the rationality check. Then, replace the original knowledge points in the microbial test report with the knowledge points in the intermediate state knowledge point set in the shielded state to generate a microbial test report after feature shielding processing.

[0077] If an unreasonable situation is detected during the rationality check, return to the above steps for adjustment. If it is a problem with the shielding order, the server will re-determine the shielding order; if it is a problem with the shielding strategy, the shielding strategy will be adjusted. For example, if it is found that a key flora description knowledge point is overly shielded, the server will adjust the shielding strategy for this knowledge point, changing from complete shielding to partial shielding or adjusting the degree of shielding. This adjustment process will continue until no unreasonable situation is detected during the rationality check. Then, the server will replace the original knowledge points in the microbial test report with the knowledge points in the intermediate state knowledge point set in the shielded state to generate a microbial test report after feature shielding processing. The microbial test report generated after such feature shielding processing not only reduces unnecessary data volume but also retains key logical relationships and important information, making it better for subsequent analysis, model training, and other operations.

[0078] In a possible implementation manner, step S130 includes: Step S131, use the feature extraction sub-network of the initialized artificial intelligence network to perform feature extraction based on the microbial test report after feature shielding processing and the logical node information to generate a feature extraction result corresponding to the microbial test report.

[0079] Step S132, use the feature restoration sub-network of the initialized artificial intelligence network to perform feature restoration based on the connection feature vector reflecting the key logical nodes and the feature extraction result corresponding to the microbial test report to generate an estimated flora description knowledge point.

[0080] In a possible implementation manner, step S131 includes: Step S1311, perform report structure parsing on the microbial test report after feature shielding processing to generate corresponding report structure data, where the report structure data includes each section in the microbial test report and the content structure in each section.

[0081] Step S1312: Logically map the parsed report structure data according to the logical node information to generate the logically mapped report structure data. Specifically, correspond the identification information in the logical node information to each part in the report structure data, and mark the association possibility between each part and the key flora description knowledge points as well as the positional relationship in the overall report logic of the microbial test report.

[0082] Step S1313: Screen the information of the logically mapped report structure data through a pre-set keyword library and semantic rules related to microbial detection, and on the basis of the screened information, perform pre-definition classification on the remaining report structure data to generate the target microbial test report content after screening and pre-definition classification.

[0083] Step S1314: Perform advanced classification on the target microbial test report content according to the pre-defined classification categories, and group and integrate the advanced classified target microbial test report content. Each group corresponds to a classification category to generate independent content groups.

[0084] Step S1315: Use the feature extraction sub-network of the initialized artificial intelligence network to extract the intra-group semantic features for each content group, and identify the extracted semantic features to clarify the group to which each semantic feature belongs and its specific semantic role within the group, generating an intra-group semantic feature set with identification.

[0085] Step S1316: Analyze the semantic relationships between each intra-group semantic feature set, and construct a semantic relationship network according to the semantic relationships. The semantic relationship network represents the connection relationships between semantic features in the form of a graph.

[0086] Step S1317: Determine the logical connections between different content groups according to the identification in the logical node information, and integrate the cross-group logical connections into the overall semantic relationship network to generate a target logical semantic network including the semantic relationships of each group and the logical relationships between groups.

[0087] Step S1318: Starting from the nodes of the target logical semantic network, extract the node features of each node. The node features include the semantic connotation of the node, the connection degree of the node in the target logical semantic network, and the type identification of the node.

[0088] Step S1319, extract the features of each edge in the target logical semantic network. The features of the edge include the relationship type and the relationship strength. Integrate the node features of each node and the features of each edge, and optimize the integrated features to generate an optimized feature representation as the feature extraction result corresponding to the microbial detection report. The content of the optimization includes removing duplicate or redundant feature information and normalizing some features.

[0089] In a possible implementation manner, step S132 includes: Step S1321, receive the connection feature vector of the key logical node and the feature extraction result corresponding to the microbial detection report. The connection feature vector of the key logical node contains the feature connection information related to the key logical node, and the feature connection information reflects the association characteristics between the key logical nodes in the logical structure of the microbial detection report.

[0090] Step S1322, parse the connection feature vector of the key logical node to determine the meaning of the connection features represented by each element in the connection feature vector.

[0091] Step S1323, map the parsed connection feature vector to the feature extraction result corresponding to the microbial detection report to establish a mapping relationship between the connection feature vector and the features in the feature extraction result, and determine the grouping basis for feature grouping according to the established mapping relationship.

[0092] Step S1324, group the feature extraction result corresponding to the microbial detection report according to the determined grouping basis. Each group contains features related to the connection relationship of the set type.

[0093] Step S1325, perform intra-group feature adjustment for each group and construct the association between features within the group to generate the group features of each group.

[0094] Step S1326, determine the integration logic of the group features according to the global logical relationship in the connection feature vector and the logical structure of the microbial detection report, and integrate the group features of each group according to the determined integration logic to combine the group features in different groups according to the logical relationship to generate a global feature representation.

[0095] Step S1327, use the feature reduction sub-network of the initialized artificial intelligence network to construct a feature reduction mapping. The feature reduction mapping is established based on the relationship between the original logical structure of the microbial detection report, the connection feature vector, and the feature extraction result.

[0096] Step S1328: Based on the constructed reduction mapping, perform a preliminary reduction on the integrated global feature representation to generate a preliminarily reduced feature representation.

[0097] Step S1329: According to the logical integrity requirements of the microbial test report and the specification requirements of the flora description knowledge points, determine the basis for optimizing and adjusting the preliminarily reduced feature representation. After optimizing and adjusting the preliminarily reduced feature representation based on the optimization and adjustment basis, perform an integrity check on the reduced feature after optimization and adjustment. If the integrity check passes, determine the reduced feature after optimization and adjustment as the estimated flora description knowledge points.

[0098] In this embodiment, first, use the feature extraction sub-network of the initialized artificial intelligence network to perform feature extraction based on the microbial test report and logical node information after feature masking processing to generate the feature extraction result corresponding to the microbial test report. This process starts from parsing the report structure of the microbial test report after feature masking processing. Taking the microbial test report of pig feed as an example, the report structure parsing aims to clarify each section in the report and the content structure in each section. The report may include a sample collection section, which details the location, time, collection method, etc. of the pig feed sample collection; the detection method section elaborates on the microbial detection methods used, such as the culture techniques, detection instruments, etc.; the key flora detection result section lists the detection data of flora such as Escherichia coli and Salmonella. Through such parsing, the corresponding report structure data is generated.

[0099] Next, perform a logical mapping on the parsed report structure data according to the logical node information to generate the logically mapped report structure data. The identification information in the logical node information plays an important role here. For example, there may be an identification in the logical node information indicating that a certain specific content is related to the Escherichia coli detection result. When performing the logical mapping, the server will correspond such identification information to each part of the report structure data. For each part of the report, mark the possibility of its association with the key flora description knowledge points and its position relationship in the entire report logic of the microbial test report. For example, some content in the sample collection section may have a potential association with the subsequent Escherichia coli detection result. The server will mark this association possibility and at the same time determine the position of this part in the entire report logic sequence, whether it is pre-information or auxiliary information parallel to the detection result, etc.

[0100] Then, through a pre-set keyword library and semantic rules related to microorganism detection, information screening is performed on the logically mapped report structure data. The keyword library contains words closely related to microorganism detection such as "Escherichia coli", "colony forming unit", "activity detection", "microorganism species", etc. The semantic rules stipulate the logical combination and semantic relationships between these words. The server screens the logically mapped report structure data based on these, removing content that is irrelevant to the key information of microorganism detection. For example, some parts of the environmental description during sample collection that are unrelated to microorganisms, such as irrelevant building information around the collection location, will be screened out. Based on the screened information, the remaining report structure data is pre-defined for classification, generating the target microorganism detection report content that has been screened and pre-defined for classification. For example, information related to the types of bacterial flora is grouped together, such as content related to the names of bacterial flora like Escherichia coli, Salmonella, etc.; information related to the detection quantity is grouped into another category, such as the number of colony forming units; content related to activity detection is grouped into another category, including descriptions of the level of activity, etc.

[0101] According to the pre-defined classification categories, the server further classifies the target microorganism detection report content and groups and integrates the further classified target microorganism detection report content. Each group corresponds to a classification category, thus generating independent content groups. Taking the classification mentioned earlier as an example, for the bacterial flora type group, it only contains information related to various bacterial flora names; the detection quantity group is purely about the detection quantities of different bacterial flora; the activity detection group is specifically for information on the activity detection results of the bacterial flora.

[0102] Using the feature extraction sub-network of the initialized artificial intelligence network, intra-group semantic feature extraction is performed for each content group. For the bacterial flora type group, semantic features such as the name features of specific bacterial flora and whether it is a harmful bacterial flora may be extracted; for the detection quantity group, semantic features such as the range of the quantity and the comparison of the quantity with the standard value are extracted; for the activity detection group, semantic features such as the description of the degree of activity and the relationship between activity and environmental factors are extracted. And the extracted semantic features are marked to clarify the group to which each semantic feature belongs and its specific semantic role within the group, generating a set of intra-group semantic features with markings. For example, for the semantic feature of "comparison of quantity with the standard value" in the detection quantity group, it is marked as belonging to the detection quantity group, and its role within the group is to evaluate whether the detection result meets the standard.

[0103] The server then analyzes the semantic relationships between the semantic feature sets within each group and constructs a semantic relationship network based on these relationships. In this semantic relationship network, the connection relationships between semantic features are represented in the form of a graph. For example, there is a logical association between the semantic feature "Escherichia coli" in the flora species grouping and the semantic feature "Escherichia coli detection quantity" in the detection quantity grouping. In the semantic relationship network, there will be an edge connecting these two nodes. Because only when it is known that it is Escherichia coli does the detection quantity have practical significance, and this relationship is reflected by the edge in the network.

[0104] Based on the identifiers in the logical node information, the server determines the logical connections between different content groupings and integrates the cross-group logical connections into the overall semantic relationship network to generate a target logical semantic network that includes the semantic relationships of each grouping and the logical relationships between groupings. For example, the logical node information may indicate that there is a logical connection based on a specific detection process between the flora species grouping and the detection quantity grouping, and the server will integrate this logical connection into the semantic relationship network to make the entire network more completely reflect the logical structure in the microbial detection report.

[0105] Starting from the nodes of the target logical semantic network, the server extracts the node features of each node. The node features include the semantic connotation of the node. For example, for the node "Escherichia coli detection quantity", its semantic connotation is the information about the detected quantity of Escherichia coli; the connection degree of the node in the target logical semantic network, that is, the number of connections with other nodes. The higher the connection degree, the more important and relevant the node is in the entire logical relationship; and the type identifier of the node, such as whether the node belongs to a quantity type node or a name type node, etc. At the same time, the server extracts the features of each edge in the target logical semantic network. The features of the edge include the relationship type, such as causal relationship, parallel relationship, or subordinate relationship, etc., and the relationship strength, which represents the tightness of the association. For example, a strongly associated edge indicates that the logical relationship between the two nodes is very tight.

[0106] Finally, the server fuses the node features of each node and the features of each edge, and optimizes the fused features to generate an optimized feature representation as the feature extraction result corresponding to the microbial detection report. The optimization content includes removing duplicate or redundant feature information. For example, if there are two node features that are basically the same in semantic connotation but only slightly different in expression, one of them is removed. Normalizing some features, such as unifying the quantities represented in different formats into a standard format. For example, some detection quantities may be represented in scientific notation, and some in ordinary numbers, and they are unified into a format that is convenient for subsequent processing.

[0107] After completing feature extraction to obtain the feature extraction results corresponding to the microbial detection report, use the feature reduction sub-network of the initialized artificial intelligence network to perform feature reduction based on the connection feature vectors of the key logical nodes reflected and the feature extraction results corresponding to the microbial detection report, and generate an estimated flora description knowledge point.

[0108] The server first receives the connection feature vectors of the key logical nodes and the feature extraction results corresponding to the microbial detection report. The connection feature vectors of the key logical nodes contain the feature connection information related to the key logical nodes. For example, for the key logical nodes related to Escherichia coli detection, the connection feature vectors contain the correlation characteristic information between the detection quantity, activity status, etc. of the key logical nodes.

[0109] Then, analyze the connection feature vectors of the key logical nodes to determine the meaning of the connection features represented by each element in the connection feature vectors. For example, a certain element may represent the positive correlation degree between the detection quantity and the activity status, or represent the influence weight of different detection methods on the accuracy of the detection results, etc.

[0110] Map the analyzed connection feature vectors to the feature extraction results corresponding to the microbial detection report to establish a mapping relationship between the connection feature vectors and the features in the feature extraction results, and determine the grouping basis for feature grouping according to the established mapping relationship. For example, if the connection feature vectors indicate a strong correlation between the detection quantity and the activity status, then in the feature extraction results, the grouping basis for these two related features is determined as a strong correlation relationship.

[0111] Group the feature extraction results corresponding to the microbial detection report according to the determined grouping basis, and each group contains the features related to the set type of connection relationship. For example, group the features related to the detection quantity and the activity status into one group, and group the features related to the detection results of other flora into another group, etc.

[0112] Perform intra-group feature adjustment for each group and construct the association between the features within the group to generate the grouping features of each group. For the group of detection quantity and activity status, according to their association relationship, adjust the proportional relationship or numerical conversion relationship between the quantity feature and the status feature, construct a closer association between them, and generate the grouping features.

[0113] According to the global logical relationship in the connection feature vectors and the logical structure of the microbial detection report, the server determines the integration logic of the grouping features, and integrates the grouping features of each group according to the determined integration logic, so as to combine the grouping features in different groups according to the logical relationship and generate a global feature representation.

[0114] Restore the sub-network using the features of the initialized artificial intelligence network, and construct a feature restoration mapping, which is established based on the relationship between the original logical structure of the microbial detection report, the connection feature vector, and the feature extraction result.

[0115] According to the constructed restoration mapping, the server performs a preliminary restoration on the integrated global feature representation to generate a preliminarily restored feature representation.

[0116] According to the logical integrity requirements of the microbial detection report and the specification requirements of the flora description knowledge points, the server determines the basis for optimizing and adjusting the preliminarily restored feature representation, and based on the basis for optimizing and adjusting, optimizes and adjusts the preliminarily restored feature representation. After that, an integrity check is performed on the restored feature after optimization and adjustment. If the integrity check passes, the restored feature after optimization and adjustment is determined as the estimated flora description knowledge point. For example, for the estimated flora description knowledge point of Escherichia coli, it should conform to the specification format of the flora description in the microbial detection report. For example, the representation of the quantity should be accurate to a certain degree, and the description of the active state should use specific terms, etc. If the preliminarily restored feature representation does not meet these requirements, it is optimized and adjusted until the integrity check passes and is determined as the final estimated flora description knowledge point.

[0117] In a possible implementation manner, step S130 may further include: Using the initialized artificial intelligence network, estimate the heat map corresponding to each key logical node reflected by the logical node information based on the microbial detection report after the feature masking process. The heat map reflects the confidence that each detection knowledge segment in the detection knowledge segment sequence is the estimated flora description knowledge point of the key logical node.

[0118] For each key logical node, extract i reference detection knowledge segments from the detection knowledge segment sequence according to the confidence, and randomly extract one detection knowledge segment from the i reference detection knowledge segments as the estimated flora description knowledge point corresponding to the key logical node. Wherein, i is a positive integer.

[0119] In this embodiment, taking the microbial detection report of pig feed as an example, after this report has undergone feature masking processing, it contains numerous detection knowledge fragments. For each key logical node, such as the key logical node related to the detection of Escherichia coli, this logical node covers multiple logical associations, such as the detection quantity of Escherichia coli, the active state, and the accuracy of the detection method. The server analyzes these logical nodes through initializing an artificial intelligence network and generates a corresponding heat map. This heat map reflects the confidence level of each detection knowledge fragment in the detection knowledge fragment sequence being an estimated flora description knowledge point of the key logical node. For example, in the detection knowledge fragment sequence, a description fragment about a specific value range of the detection quantity of Escherichia coli may show a relatively high confidence level in the heat map, indicating that this fragment has a greater possibility of being an accurate estimated flora description knowledge point related to the key logical node of Escherichia coli detection. For some detection knowledge fragments with relatively weak relevance to Escherichia coli detection or doubtful data accuracy, they will show a relatively low confidence level in the heat map.

[0120] Next, assuming that i is set to 3, the server will select 3 reference detection knowledge fragments from the detection knowledge fragment sequence in the order of decreasing confidence level. For example, for the key logical node related to the detection of Escherichia coli, a detection knowledge fragment about the detection quantity of Escherichia coli being in a certain specific interval and having a relatively high confidence level may be selected, a detection knowledge fragment about the performance of the active state of Escherichia coli in a specific environment and having the second-highest confidence level, and a detection knowledge fragment about the accuracy of the instrument used to detect this flora and having a relatively high confidence level. These reference detection knowledge fragments are selected based on the confidence levels in the previously generated heat map, and they represent to a certain extent the relatively important and possibly accurate information related to this key logical node.

[0121] Then, still taking the key logical node related to the detection of Escherichia coli as an example, among the 3 previously selected reference detection knowledge fragments, the server randomly selects a detection knowledge fragment about the detection quantity of Escherichia coli being in a certain specific interval and determines it as the estimated flora description knowledge point corresponding to the key logical node related to the detection of Escherichia coli. This random selection method takes into account to a certain extent the possibility of making an even selection among multiple detection knowledge fragments with relatively high confidence levels, avoiding the one-sidedness that may be caused by relying solely on a certain specific high-confidence fragment. In this way, for each key logical node, the server can generate the corresponding estimated flora description knowledge point, and these estimated flora description knowledge points will play an important role in subsequent data analysis, model training, or quality assessment, etc. For example, they are used to evaluate whether the microbial situation in the feed meets the quality standards, or to train a microbial feature extraction model to improve the model's ability to accurately output flora description knowledge points.

[0122] In a possible implementation manner, the method further includes: Step B110: Using the initialized microbial feature extraction model to extract microbial features from the extended feed microbial detection sample, generating a microbial feature extraction result, and determining the logical node information corresponding to the extended feed microbial detection sample based on the microbial feature extraction result.

[0123] Step B120: Extracting, from the extended feed microbial detection sample, the extended feed microbial detection samples whose determined logical node information matches the logical node information corresponding to the key flora description knowledge points in the microbial detection report of the corresponding basic feed microbial detection sample, and training the microbial feature extraction model based on the extracted extended feed microbial detection samples.

[0124] In this embodiment, first, taking pig feed as an example, the extended feed microbial detection sample contains the content of the microbial detection report after being processed by the previous steps, and the key flora description knowledge points may have been updated to estimated flora description knowledge points. The initialized microbial feature extraction model starts to analyze these extended feed microbial detection samples. During this process, the model will identify various microbial-related features in the report. For example, for the Escherichia coli detection part in the report, the model will extract the features related to Escherichia coli, such as the detected colony number range, relevant numerical values or descriptions of the flora activity status, etc. These features together constitute the microbial feature extraction result. Based on this microbial feature extraction result, the server determines the logical node information corresponding to the extended feed microbial detection sample. This process involves the analysis of the logical relationship between the extracted features and the report structure and content. For example, for the detected Escherichia coli colony number feature, information such as its position in the report and its relationship with other microbial detection results will be determined as part of the logical node information.

[0125] Next, assume that the logical node information corresponding to the key flora description knowledge points in a pig feed detection report in the basic feed microbial detection sample contains specific structural and content logical relationships. The server will compare and match the logical node information in the extended feed microbial detection sample with the logical node information in these basic feed microbial detection samples. For example, if the logical node information regarding the Escherichia coli detection part in the basic feed microbial detection sample includes specific detection sequences, result presentation methods, etc. logical relationships, the server will find samples with similar logical relationships in the extended feed microbial detection sample. Based on the extracted matching extended feed microbial detection samples, the server trains the microbial feature extraction model. During the training process, the model will learn according to the microbial features and logical relationships in these matching samples, and adjust its own parameters to better adapt to the processing of feed microbial detection samples.

[0126] In a possible implementation manner, the method further includes: respectively using the extended feed microorganism detection sample and the basic feed microorganism detection sample to train a microorganism feature extraction model, and generating a trained microorganism feature extraction model.

[0127] In this embodiment, for the basic feed microorganism detection sample, it contains the key flora description knowledge points and logical node information in the untreated original microorganism detection report. These original information provide the basic microorganism detection data pattern for the model. The extended feed microorganism detection sample contains the information processed in the previous steps and has more variable and estimated information. By using these two types of samples for training simultaneously, the microorganism feature extraction model can more comprehensively learn the features and logical relationships related to feed microorganism detection. For example, while the model learns the original Escherichia coli detection data pattern in the basic feed microorganism detection sample, it can also learn different processing methods and logical relationships from the estimated Escherichia coli detection data in the extended feed microorganism detection sample. Through learning and training with a large number of such samples, the model continuously optimizes its own parameters and finally generates a trained microorganism feature extraction model. This trained microorganism feature extraction model can more accurately extract microorganism features from new feed microorganism detection samples, and can efficiently and accurately output relevant microorganism feature and logical relationship information for both the original basic feed microorganism detection sample and the processed extended feed microorganism detection sample, thus providing strong support for the microorganism detection and evaluation of feed quality.

[0128] In a possible implementation manner, the training step of initializing the artificial intelligence network includes: Step C110, obtaining a to-be-learned feed microorganism detection sample, where each to-be-learned feed microorganism detection sample includes a microorganism detection report, the key flora description knowledge points in the microorganism detection report, and the logical node information of the key flora description knowledge points in the microorganism detection report.

[0129] Step C120, performing feature masking processing on the key flora description knowledge points in the microorganism detection report to generate a microorganism detection report after feature masking processing.

[0130] Step C130, using a candidate artificial intelligence network to estimate the knowledge points of the key logical nodes reflected by the logical node information based on the microorganism detection report after feature masking processing and the logical node information, and generating estimated flora description knowledge points.

[0131] Step C140: Based on the error between the estimated flora description knowledge points and the key flora description knowledge points in the microbial detection report of the to-be-learned feed microbial detection sample, construct a training cost parameter, and optimize the candidate artificial intelligence network based on the training cost parameter to generate the initialized artificial intelligence network. The initialized artificial intelligence network is used to estimate knowledge points for the microbial detection report after feature masking processing.

[0132] In this embodiment, taking pig feed as an example, these to-be-learned feed microbial detection samples are a large amount of detection data accumulated by enterprises during the daily feed production quality monitoring process. Each to-be-learned feed microbial detection sample contains rich information, and the microbial detection report details the whole process from feed sample collection, transportation, preservation to microbial detection. In the microbial detection report, there are many key flora description knowledge points. For example, for the detection of Escherichia coli, the report will clearly record information such as the detected quantity, activity status, and whether there are mutations. Moreover, the logical node information of these key flora description knowledge points in the microbial detection report is also accurately recorded. The logical node information covers the position of each key flora description knowledge point in the report, the logical association with other knowledge points, etc. For example, the detected quantity of Escherichia coli may be located at a specific position under the "Harmful Flora Detection Results" section in the report, and there is a logical sequence or causal relationship with other knowledge points such as the detection method and sample source.

[0133] After obtaining the feed microorganism detection sample to be learned, the server performs feature masking processing on the key flora description knowledge points in the microorganism detection report to generate a microorganism detection report after feature masking processing. In this process, the server operates according to specific rules and standards set by the enterprise. For the key flora description knowledge points in the pig feed microorganism detection report, the server determines the parts that need to be masked according to the overall strategy. For example, if there are too many description knowledge points about Escherichia coli in the report and some information is redundant, the server will determine the knowledge point scale of the key flora description knowledge points to be subjected to feature masking processing according to the global scale of the key flora description knowledge points, such as how many description points related to Escherichia coli there are in total, in combination with the set masking standard. Then, determine the starting point according to the structure and content distribution law of the microorganism detection report. For example, start from the beginning of the part about the detection results of Escherichia coli in the report and extract the key flora description knowledge points according to the determined knowledge point scale. During the extraction process, these key flora description knowledge points are preliminarily marked with their relative position information and the type they belong to in the report. For example, the detection quantity belongs to the numerical type, and the activity status belongs to the qualitative type, etc. Then construct an association graph of the key flora description knowledge points. In the graph, each key flora description knowledge point is represented by a node, and the edge represents the association relationship between the knowledge points. This association relationship is determined according to the logical connection and content relevance of the key flora description knowledge points in the microorganism detection report. For example, there is a logical connection between the detection quantity and the activity status of Escherichia coli, and there will be an edge connecting these two nodes in the association graph. After that, determine the masking order according to the weight of the nodes, the weight of the edges, and the connection relationship between the nodes. For the key flora description knowledge points corresponding to the nodes with lower weights and looser connection relationships, they may be masked first. During the masking process, for each key flora description knowledge point, different masking strategies are adopted according to its marked type and position relationship in the association graph. For example, partial fuzzification processing may be performed on the numerical detection quantity. During the masking process, the status of the key flora description knowledge point set is updated in real time to generate an intermediate status knowledge point set. Finally, perform a rationality check on the intermediate status knowledge point set. By analyzing the marked information, association graph, and operation records of the masking process of each key flora description knowledge point, judge whether it is processed according to the predetermined masking order, whether the necessary logical relationships are retained, and whether excessive masking causes key information loss. If there are unreasonable situations, return for adjustment until the rationality check passes. Replace the original knowledge points in the microorganism detection report with the knowledge points in the intermediate status knowledge point set in the masked state, so as to generate a microorganism detection report after feature masking processing.

[0134] Subsequently, the server uses the candidate artificial intelligence network to perform knowledge point estimation on the key logical nodes reflected by the logical node information based on the microorganism detection report after feature masking processing and the logical node information, and generates an estimated flora description knowledge point. The server inputs the microorganism detection report after feature masking processing and the logical node information into the candidate artificial intelligence network. For the microorganism detection report of pig feed, the network will first analyze the content in the report. Taking the logical node related to Escherichia coli detection as an example, the network will locate the key logical nodes related to Escherichia coli according to the logical node information, such as the logical association part between the detection quantity and the activity state. Then, through the algorithm mechanism inside the network, combined with other relevant information in the content of the report after feature masking processing, such as the possible impact of the sample collection environment on Escherichia coli, etc., knowledge point estimation is performed on these key logical nodes. In this process, it may involve comprehensive analysis and reasoning of different detection knowledge fragments, so as to obtain the estimated flora description knowledge points of key logical nodes such as the detection quantity and activity state of Escherichia coli, for example, it is estimated that the detection quantity of Escherichia coli may be within a certain range, and the activity state is at a certain specific level, etc.

[0135] Finally, based on the error between the estimated flora description knowledge point and the key flora description knowledge point in the microorganism detection report of the feed microorganism detection sample to be learned, the server constructs a training cost parameter, and optimizes the candidate artificial intelligence network based on the training cost parameter to generate an initial artificial intelligence network. The server will compare the generated estimated flora description knowledge point with the key flora description knowledge point in the original microorganism detection report of the feed microorganism detection sample to be learned. For example, for the detection quantity of Escherichia coli, there is a certain difference between the estimated value and the actual detection value in the original report, and this difference is part of the error. The server comprehensively considers the error situations of all key flora description knowledge points and constructs a training cost parameter. This training cost parameter reflects the deviation degree between the current estimation result of the candidate artificial intelligence network and the real result. Then, the server optimizes the candidate artificial intelligence network according to this training cost parameter. During the optimization process, the network will adjust various internal parameters, such as adjusting the weight coefficients related to different key flora description knowledge points, etc. By continuously adjusting the parameters to reduce the value of the training cost parameter, the estimation result of the network is made closer to the real value. After such an optimization process, an initial artificial intelligence network is finally generated. This initial artificial intelligence network can be used to perform knowledge point estimation on the microorganism detection report after feature masking processing, provide strong support for subsequent feed microorganism detection data processing and analysis, improve the estimation accuracy of the key flora description knowledge points in the microorganism detection report, and thus help enterprises better monitor and manage the microorganism in feed quality.

[0136] Figure 2The following shows the hardware structure diagram of the feed production monitoring system 100 provided by the embodiments of the present application for implementing the above-mentioned artificial intelligence-based feed microorganism detection data analysis method, as Figure 2 shown, the feed production monitoring system 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

[0137] In a possible design, the feed production monitoring system 100 may be a single server or a server group. The server group may be centralized or distributed (for example, the feed production monitoring system 100 may be a distributed system). In some embodiments, the feed production monitoring system 100 may be local or remote. For example, the feed production monitoring system 100 may access information and / or data stored in the machine-readable storage medium 120 via a network. Again, for example, the feed production monitoring system 100 may be directly connected to the machine-readable storage medium 120 to access the stored information and / or data. In some embodiments, the feed production monitoring system 100 may be implemented on a feed production monitoring system. Only by way of example, the feed production monitoring system may include a private cloud, a semantic-related cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc. or any aggregation thereof.

[0138] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data obtained from an external terminal. In some embodiments, the machine-readable storage medium 120 may store the data and / or instructions used by the feed production monitoring system 100 to execute or use to complete the exemplary methods described in the present application.

[0139] In a specific implementation process, one or more processors 110 execute the computer-executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the artificial intelligence-based feed microorganism detection data analysis method of the above method embodiment. The processor 110, the machine-readable storage medium 120, and the communication unit 140 are connected through the bus 130, and the processor 110 may be used to control the transceiver actions of the communication unit 140.

[0140] For the specific implementation process of the processor 110, reference may be made to the various method embodiments executed by the above feed production monitoring system 100. The implementation principles and technical effects are similar, and will not be elaborated here in this embodiment.

[0141] In addition, the embodiments of the present application further provide a readable storage medium, in which computer-executable instructions are set. When the processor executes the computer-executable instructions, the above-mentioned artificial intelligence-based feed microorganism detection data analysis method is implemented.

[0142] It should be noted that, in order to simplify the description of the disclosure of the present application and thus assist in the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, sometimes multiple features are merged into one embodiment, drawing, or description thereof. Similarly, it should be noted that, in order to simplify the description of the disclosure of the present application and thus assist in the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, sometimes multiple features are merged into one embodiment, drawing, or description thereof.

Claims

1. A feed microorganism detection data analysis method based on artificial intelligence, characterized in that: The method comprises: Obtain basic feed microbial detection samples, each basic feed microbial detection sample includes a microbial detection report, key flora description knowledge points in the microbial detection report, and logical node information of the key flora description knowledge points in the microbial detection report; Performing feature shielding processing on key flora description knowledge points in the microbial detection report to generate a microbial detection report after feature shielding processing; Using an initialized artificial intelligence network, based on the microbial detection report after the feature shielding process and the logical node information, knowledge points of key logical nodes reflected by the logical node information are estimated to generate estimated flora description knowledge points; The key flora description knowledge points in the microbial detection report of the basic feed microbial detection sample are updated to the estimated flora description knowledge points to generate an extended feed microbial detection sample; the extended feed microbial detection sample is used to train the microbial feature extraction model to output the flora description knowledge points based on the trained microbial feature extraction model.

2. The artificial intelligence-based feed microbial detection data analysis method according to claim 1, characterized in that: The logical node information includes identification information of whether each detection knowledge segment in the microbial detection report belongs to a key flora description knowledge point, and identification information of the starting logical node and the end logical node of the key flora description knowledge point in the microbial detection report respectively corresponding to each other; the method further includes: Embed a logical anchor point for locating key flora description knowledge points in the microbial detection report after the feature shielding process to generate an anchored detection report; Embedding the anchor point identification information corresponding to the logical anchor point in the logical node information to generate anchored logical node information; The utilizing initialized artificial intelligence network, based on the microbial detection report after the feature shielding processing and the logical node information, performs knowledge point estimation on the key logical nodes reflected by the logical node information, and generates estimated flora description knowledge points, including: By utilizing the initialized artificial intelligence network, knowledge points of key logical nodes reflected by the logical node information are estimated based on the anchored detection report and the anchored logical node information to generate estimated flora description knowledge points.

3. The artificial intelligence-based feed microbial detection data analysis method according to claim 1, characterized in that: The feature shielding processing is performed on the key flora description knowledge points in the microbial detection report to generate the microbial detection report after the feature shielding processing, including: Acquire a set feature shielding parameter, and determine the knowledge point scale of the key flora description knowledge points for feature shielding the key flora description knowledge points in the microbial detection report based on the global scale of the key flora description knowledge points in the microbial detection report and the set feature shielding parameter; Determine a starting point according to the structure and content distribution law of the microbial detection report, extract key flora description knowledge points from the microbial detection report starting from the starting point according to the knowledge point scale, and stop the extraction operation when the number of extracted key flora description knowledge points reaches the number required by the knowledge point scale, so as to form a set of key flora description knowledge points to be processed by feature shielding; Preliminarily marking each key flora description knowledge point in the extracted key flora description knowledge point set to obtain a preliminary marking result, wherein the preliminary marking result includes relative position information of each key flora description knowledge point in the microbial detection report and the type of the knowledge point; According to the preliminary labeling results, a correlation map of key flora description knowledge points is constructed, in which a node represents each key flora description knowledge point, and an edge represents the correlation between knowledge points. The correlation is determined based on the logical connection between the key flora description knowledge points in the microbial detection report and the relevance of the content of the key flora description; In the association graph, a shielding order is determined according to the weight of the node, the weight of the edge and the connection relationship between the nodes, and a feature shielding process is performed on the key flora description knowledge point set according to the determined feature shielding order and the preliminary marking result, wherein, when performing the feature shielding process, for each key flora description knowledge point, different shielding strategies are adopted according to the type of the key flora description knowledge point mark and the positional relationship in the association graph, and the state of the key flora description knowledge point set is updated in real time during the shielding process, and an intermediate state knowledge point set in the feature shielding process is generated, and the intermediate state knowledge point set records the result of each step in the shielding process; Performing a rationality check on the intermediate state knowledge point set, specifically by analyzing the marking information of each key flora description knowledge point, the association map, and the operation record of the shielding process, to determine whether the feature shielding process result is reasonable. The rationality check includes: whether the process is performed in accordance with the predetermined shielding order, whether the necessary logical relationship between the key flora description knowledge points is retained, and whether excessive shielding leads to loss of key information; If unreasonable situations are detected during the rationality check, return to the above steps for adjustment, redetermine the shielding order or adjust the shielding strategy until no unreasonable situations are detected during the rationality check. Then, the knowledge points in the intermediate state knowledge point set replace the original knowledge points in the microbial detection report according to the shielded state to generate a microbial detection report after feature shielding processing.

4. The artificial intelligence-based feed microbial detection data analysis method according to claim 1, characterized in that: The utilizing initialized artificial intelligence network, based on the microbial detection report after the feature shielding processing and the logical node information, performs knowledge point estimation on the key logical nodes reflected by the logical node information, and generates estimated flora description knowledge points, including: Using the feature extraction sub-network of the initialized artificial intelligence network, feature extraction is performed based on the microbial detection report after the feature masking process and the logical node information to generate a feature extraction result corresponding to the microbial detection report; The feature restoration subnetwork of the initialized artificial intelligence network is utilized to perform feature restoration based on the connection feature vectors of the reflected key logical nodes and the feature extraction results corresponding to the microbial detection report to generate estimated flora description knowledge points.

5. The artificial intelligence-based feed microbial detection data analysis method according to claim 4, characterized in that: The step of utilizing the feature extraction subnetwork of the initialized artificial intelligence network to extract features based on the microbial detection report after feature masking and the logical node information to generate a feature extraction result corresponding to the microbial detection report includes: Performing report structure analysis on the microbial detection report after feature shielding processing to generate corresponding report structure data, wherein the report structure data includes each section in the microbial detection report and the content structure in each section; Logically mapping the parsed report structure data according to the logical node information to generate the report structure data after logical mapping, specifically corresponding the identification information in the logical node information to each part in the report structure data, marking the possibility of association between each part and the key flora description knowledge point and the position relationship in the entire report logic of the microbial detection report; The report structure data after the logic mapping is screened by using a pre-set keyword library and semantic rules related to microbial detection, and the remaining report structure data is classified and predefined based on the screened information to generate target microbial detection report content after screening and pre-defined classification; According to the predefined classification categories, the target microorganism detection report content is advanced classified, and the advanced classified target microorganism detection report content is grouped and integrated, each group corresponds to a classification category, and each independent content group is generated; Using the feature extraction subnetwork of the initialized artificial intelligence network, the semantic features within each content group are extracted, and the extracted semantic features are identified to clarify the group to which each semantic feature belongs and its specific semantic role within the group, and to generate a semantic feature set within the group with the identification; Analyzing the semantic relationship between the semantic feature sets in each group, and constructing a semantic relationship network based on the semantic relationship, wherein the semantic relationship network represents the connection relationship between the semantic features in the form of a graph; Determine the logical connections between different content groups according to the identifiers in the logical node information, and integrate the logical connections across groups into the overall semantic relationship network to generate a target logical semantic network including the semantic relationships between each group and the logical relationships between groups; Starting from the nodes of the target logical semantic network, extracting node features of each node, the node features including the semantic connotation of the node, the connectivity of the node in the target logical semantic network, and the type identification of the node; Extracting features of each edge in the target logical semantic network, wherein the features of the edge include a relationship type and a relationship strength; The node feature of each node and the feature of each edge are merged, and the merged features are optimized to generate an optimized feature representation as a feature extraction result corresponding to the microbial detection report, wherein the optimization includes removing repeated or redundant feature information and normalizing some features; And, the step of using the feature restoration subnetwork of the initialized artificial intelligence network to perform feature restoration based on the connection feature vector of the reflected key logical nodes and the feature extraction result corresponding to the microbial detection report to generate estimated flora description knowledge points includes: Receiving a connection feature vector of a key logic node and a feature extraction result corresponding to a microbial detection report, wherein the connection feature vector of the key logic node includes feature connection information related to the key logic node, and the feature connection information reflects the association characteristics between the key logic nodes in the logical structure of the microbial detection report; Analyze the connection feature vector of the key logic node to determine the connection feature meaning represented by each element in the connection feature vector; Mapping the parsed connection feature vector to the feature extraction result corresponding to the microbial detection report to establish a mapping relationship between the connection feature vector and the features in the feature extraction result, and determining a grouping basis for feature grouping according to the established mapping relationship; Grouping the feature extraction results corresponding to the microbial detection report according to the determined grouping basis, each grouping includes features related to the set type connection relationship; Adjust the intra-group features for each group, build associations between features within the group, and generate group features for each group; Determine the integration logic of the grouping features according to the global logical relationship in the connection feature vector and the logical structure of the microbial detection report, and integrate the grouping features of each group according to the determined integration logic, so as to combine the grouping features in different groups according to the logical relationship to generate a global feature representation; Using the feature restoration subnetwork of the initialized artificial intelligence network, constructing a feature restoration map, wherein the feature restoration map is established based on the original logical structure of the microbial detection report, the connection feature vector, and the relationship between the feature extraction results; According to the constructed restoration map, the integrated global feature representation is initially restored to generate a initially restored feature representation; According to the logical integrity requirements of the microbial detection report and the specification requirements of the flora description knowledge points, the optimization and adjustment basis of the feature representation after the preliminary restoration is determined, and after the feature representation after the preliminary restoration is optimized and adjusted based on the optimization and adjustment basis, an integrity check is performed on the optimized and adjusted restored features. If the integrity check passes, the optimized and adjusted restored features are determined as estimated flora description knowledge points.

6. The artificial intelligence-based feed microbial detection data analysis method according to claim 1, characterized in that: The utilizing initialized artificial intelligence network, based on the microbial detection report after the feature shielding processing and the logical node information, performs knowledge point estimation on the key logical nodes reflected by the logical node information, and generates estimated flora description knowledge points, including: Using an initialized artificial intelligence network, based on the microbial detection report after feature shielding and the logical node information, a heat map corresponding to each key logical node reflected by the logical node information is estimated, wherein the heat map reflects the confidence that each detection knowledge fragment in the detection knowledge fragment sequence is an estimated flora description knowledge point of the key logical node; For each key logic node, i reference detection knowledge fragments are extracted from the detection knowledge fragment sequence according to the confidence level, and a detection knowledge fragment is randomly extracted from the i reference detection knowledge fragments as the estimated flora description knowledge point corresponding to the key logic node; wherein i is a positive integer.

7. The artificial intelligence-based feed microbial detection data analysis method according to claim 1, characterized in that: The method further comprises: Extracting microbial features from the extended feed microbial detection sample using an initialized microbial feature extraction model to generate a microbial feature extraction result, and determining logical node information corresponding to the extended feed microbial detection sample based on the microbial feature extraction result; From the extended feed microbial detection samples, the determined logical node information is extracted and matched with the logical node information corresponding to the key flora description knowledge points in the microbial detection report of the corresponding basic feed microbial detection sample, and the microbial feature extraction model is trained based on the extracted extended feed microbial detection samples.

8. The artificial intelligence-based feed microbial detection data analysis method according to claim 1, characterized in that: The method further comprises: The microbial feature extraction model is trained using the extended feed microbial detection samples and the basic feed microbial detection samples respectively to generate a trained microbial feature extraction model.

9. The artificial intelligence-based feed microbial detection data analysis method according to any one of claims 1 to 8, characterized in that: The training steps of initializing the artificial intelligence network include: Obtain feed microbial detection samples to be studied, each feed microbial detection sample to be studied includes a microbial detection report, key flora description knowledge points in the microbial detection report, and logical node information of the key flora description knowledge points in the microbial detection report; Performing feature shielding processing on key flora description knowledge points in the microbial detection report to generate a microbial detection report after feature shielding processing; Using the candidate artificial intelligence network, based on the microbial detection report after the feature shielding process and the logical node information, the key logical nodes reflected by the logical node information are estimated, and the estimated flora description knowledge points are generated; Based on the error between the estimated flora description knowledge point and the key flora description knowledge point in the microbial detection report of the feed microbial detection sample to be learned, a training cost parameter is constructed, and the candidate artificial intelligence network is optimized based on the training cost parameter to generate the initialized artificial intelligence network; the initialized artificial intelligence network is used to estimate the knowledge points of the microbial detection report after feature masking processing.

10. A feed production monitoring system, characterized in that: The feed production monitoring system includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the artificial intelligence-based feed microorganism detection data analysis method described in any one of claims 1 to 9.

Images