Farmed animal behavior tracking method and system based on artificial intelligence

Abstract

The present invention relates to the field of artificial intelligence technology, and to a method and system for tracking the behavior of farmed animals based on artificial intelligence. The present invention first adjusts the labels of several behavior tracking report texts that are to be jointly labeled, so that the description sentences originally having different behavior tracking cycle labels can be unified, and then generates a behavior tracking report adjustment text that matches the same behavior tracking cycle label. Secondly, based on the behavior semantic association variables and the description statement positioning vector, the context connection attention dimension of the description statement connection processing is cleverly determined, so that the associated behavior tracking description statements at different context connection attention dimensions can be accurately identified. Finally, by performing joint knowledge annotation on these associated statements, a behavior tracking report annotation text rich in semantics and context information is obtained, which provides solid data support for subsequent behavior pattern analysis, anomaly detection or breeding management decision-making.

Description

Technical Field

[0001] The present invention relates to the technical field of text report analysis, and more specifically, to a method and system for tracking the behavior of farmed animals based on artificial intelligence. Background Art

[0002] In the field of breeding management, tracking and analyzing the behavior of farmed animals is a crucial link. However, existing behavior tracking report texts often have inconsistent behavior tracking cycle labels, which greatly hinders subsequent data processing and analysis. At the same time, due to the lack of effective context connection processing methods, the association between different behavior tracking description statements is difficult to be accurately identified, resulting in the inability to fully mine and utilize the rich information behind these behavior data. In addition, existing annotation methods often ignore cross-text and cross-dimensional joint knowledge annotation, which makes the generated behavior tracking report text lack the richness of semantics and contextual information. Summary of the invention

[0003] In view of this, the present invention provides a method and system for tracking farmed animal behavior based on artificial intelligence.

[0004] The embodiment of the present invention provides a method for tracking behavior of farmed animals based on artificial intelligence, which is applied to an artificial intelligence processing system. The method includes: obtaining a plurality of behavior tracking report texts to be jointly annotated, and behavior semantic association variables of the behavior tracking report texts, wherein the plurality of behavior tracking report texts to be jointly annotated have behavior tracking description sentences with different behavior tracking cycle labels, and the behavior tracking report texts are report texts of selected farmed animals; performing label adjustment on the behavior tracking cycle data of the behavior tracking description sentences in each of the behavior tracking report texts to obtain a plurality of behavior tracking report adjustment texts matching the same behavior tracking cycle label; determining the contextual connection attention dimension of the description sentence connection processing according to the behavior semantic association variables and the description sentence positioning vector of the behavior tracking description sentence in the behavior tracking report adjustment text; determining the associated behavior tracking description sentence of the behavior tracking report adjustment text in the contextual connection attention dimension from each of the behavior tracking report adjustment texts; and for each contextual connection attention dimension, performing joint knowledge annotation on the associated behavior tracking description sentence corresponding to the contextual connection attention dimension to obtain the behavior tracking report annotation text corresponding to each contextual connection attention dimension.

[0005] An embodiment of the present invention further provides an artificial intelligence processing system, comprising: a memory for storing program instructions and data; and a processor for coupling with the memory and executing instructions in the memory to implement the above method.

[0006] An embodiment of the present invention further provides a computer storage medium, comprising instructions, which implement the above method when executed on a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 A schematic flowchart of the steps of a method for tracking the behavior of farmed animals based on artificial intelligence is provided in an embodiment of the present invention.

[0008] Figure 2 A structural block diagram of an artificial intelligence processing system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0009] The embodiments of the present invention effectively solve the problems of inconsistent behavior tracking cycle labels, difficult to identify contextual associations, and insufficient prior training labels through steps such as label adjustment, determination of contextual connection attention dimensions, and joint knowledge annotation, providing strong technical support for the intelligent and refined breeding management.

[0010] In conjunction with the above, please refer to Figure 1 , Figure 1 It is a flow chart of a method for tracking the behavior of farmed animals based on artificial intelligence provided by an embodiment of the present invention. The method is applied to an artificial intelligence processing system and may further include steps 110 to 150.

[0011] In modern farming, animal behavior tracking and analysis is an important means to improve farming efficiency and animal welfare. The present invention will introduce in detail how to use an artificial intelligence processing system to jointly annotate the behavior tracking report in the pig farming process from the perspective of NLP text to improve the accuracy and efficiency of behavior analysis.

[0012] Step 110: Obtain a plurality of behavior tracking report texts to be jointly annotated, and behavior semantic association variables of the behavior tracking report texts.

[0013] The plurality of behavior tracking report texts to be jointly labeled have behavior tracking description sentences with different behavior tracking period labels, and the behavior tracking report texts are report texts for selected farmed animals.

[0014] First, the artificial intelligence processing system needs to obtain several behavior tracking report texts that are to be jointly labeled. The behavior tracking report text is a behavior tracking report on the selected farmed animal - the pig, and contains behavior tracking description statements with different behavior tracking cycle labels. For example, some texts may record the pig's eating behavior in a day, while others may record the pig's activity patterns in a week. At the same time, the artificial intelligence processing system also needs to obtain behavioral semantic association variables related to these behavior tracking report texts. Behavioral semantic association variables can include the pig's age, gender, weight, health status, etc., which are of great significance for understanding the pig's behavior patterns.

[0015] Step 120: perform label adjustment on the behavior tracking period data of the behavior tracking description sentences in each of the behavior tracking report texts to obtain a plurality of behavior tracking report adjustment texts matching the same behavior tracking period label.

[0016] Next, the AI ​​processing system adjusts the labels of the behavior tracking description sentences in each behavior tracking report text. The purpose of this step is to adjust the description sentences with different behavior tracking cycle labels to description sentences with the same behavior tracking cycle labels for subsequent joint labeling. For example, if some texts record behavior in days and some in weeks, the AI ​​processing system needs to unify these data under the same cycle label, such as weeks. In this way, the data about the pigs' behavior in the same week in different texts can be integrated together.

[0017] Step 130: adjusting the description sentence location vector of the behavior tracking description sentence in the text according to the behavior semantic association variable and the behavior tracking report, and determining the contextual connection attention dimension of the description sentence connection processing.

[0018] After label adjustment, the AI ​​processing system needs to determine the contextual connection attention dimension of the description statement connection processing based on the behavior semantic association variables and the description statement location vector of the behavior tracking description statement in the behavior tracking report adjustment text. The core of this step is to identify which description statements are semantically related to each other and how these associations affect the understanding of pig behavior. For example, the pig's eating behavior and weight change may be related to each other, so these two aspects of information need to be considered together during analysis.

[0019] Step 140: Determine, from each of the behavior tracking report adjustment texts, the associated behavior tracking description sentences of the behavior tracking report adjustment text in the context connection attention dimension.

[0020] After determining the contextual connection attention dimension, the AI ​​processing system needs to find the behavior tracking description sentences that are related to each other on this dimension from the individual behavior tracking report adjustment texts. These sentences may be distributed in different texts or in different parts of the same text. For example, if the AI ​​processing system determines that "eating behavior" and "weight change" are an important contextual connection attention dimension, then it needs to find all the sentences describing the pig's eating behavior and all the sentences describing the pig's weight change, regardless of whether they appear in the same text.

[0021] Step 150: For each contextual connection attention dimension, perform joint knowledge annotation on the associated behavior tracking description sentences corresponding to the contextual connection attention dimension to obtain the behavior tracking report annotation text corresponding to each contextual connection attention dimension.

[0022] Finally, the AI ​​processing system performs joint knowledge annotation on the associated behavior tracking description sentences corresponding to each contextual connection attention dimension. The purpose of this step is to integrate the relevant information scattered in different texts to form a more comprehensive and accurate behavior analysis report. For example, for the dimensions of "eating behavior" and "weight change", the AI ​​processing system can annotate: "During the observation period, the amount of food consumed by pigs was positively correlated with their weight changes, and the weight of pigs with increased food intake also increased significantly." Such annotations not only provide specific information about pig behavior, but also reveal the intrinsic connection between behaviors.

[0023] Through the above steps 110 to 150, the artificial intelligence processing system can effectively jointly annotate the behavior tracking reports in the pig farming process, which not only improves the accuracy and efficiency of behavior analysis, but also provides more comprehensive and in-depth data support for farmers, helping them to better understand the behavior patterns of pigs, optimize farming strategies, and improve farming efficiency and animal welfare.

[0024] In the specific implementation process, the artificial intelligence processing system needs to have strong data processing capabilities and machine learning algorithm support. It needs to parse and understand a large amount of behavior tracking report text, extract key information from it, and perform complex semantic analysis and association reasoning. At the same time, the artificial intelligence processing system also needs to continuously learn and optimize its own labeling strategy to adapt to different breeding environments and changes in pig herd characteristics. In addition, in order to ensure the accuracy and reliability of the labeling results, the artificial intelligence processing system also needs to work closely with human experts. Human experts can provide professional breeding knowledge and experience support to help the artificial intelligence processing system better understand and interpret the behavior patterns of pigs. At the same time, they can also verify and correct the labeling results of the artificial intelligence processing system to ensure the accuracy and reliability of the data.

[0025] In summary, the artificial intelligence processing system can jointly label the behavior tracking reports in the pig farming process, providing farmers with more comprehensive and in-depth data support and analysis tools, which will help promote the farming industry to develop in a more intelligent and efficient direction.

[0026] Next, based on the above content, each step is introduced in detail.

[0027] In step 110, the behavior tracking report text is a detailed record of the behavior observation of a specific farmed animal (in this case, a pig). These texts are usually generated by farmers or automated monitoring systems and contain various behaviors of pigs in different time periods. For example, a behavior tracking report text may record the number of times a pig eats, the activity time, the rest pattern, etc. in a day. These texts are important bases for farm management and behavior analysis.

[0028] It is additional information related to the behavior described in the behavior tracking report text, used to provide richer context and understanding. In the pig farming scenario, these variables may include the pig's age, sex, weight, health status, farming environment (such as temperature, humidity), etc. For example, when analyzing the pig's eating behavior, considering its age and weight can help determine whether its appetite is normal, while the temperature of the farming environment may affect its activity level.

[0029] Behavior tracking period labels are used to indicate the time period corresponding to the behavior data recorded in the behavior tracking report text. These labels help organize the data into different time frames for comparison and analysis across time. For example, the period label can be "Day", "Week", "Month", etc. In the behavior analysis of pigs, the use of different period labels can reveal short-term and long-term behavior patterns and their changes.

[0030] Behavior tracking description sentences are specific sentences in the behavior tracking report text that describe the specific behavior of the pig during a specific time period. These sentences usually contain detailed information such as the behavior type, occurrence time, duration, frequency, etc. For example, "Pig A ate twice between 9 and 10 am, each time lasting about 15 minutes." Such description sentences provide specific data points for behavior analysis.

[0031] In detail, based on step 110, the artificial intelligence processing system first needs to collect several behavior tracking report texts about the selected farmed animals (pigs). These texts are the basis for subsequent analysis, and they contain various behavioral descriptions of pigs in different time periods. At the same time, the system also needs to obtain behavioral semantic associated variables related to these texts, such as the age, gender, weight, etc. of the pigs. These variables are crucial for understanding the behavior patterns of pigs because they can provide additional contextual information to help the system interpret and annotate behavioral data more accurately. Specifically, the system can extract these behavior tracking report texts and associated variables from the database of the farm. These texts may be recorded regularly by farmers or generated in real time by an automated monitoring system. When extracting texts, the system ensures that they have behavior tracking description sentences with different behavior tracking cycle labels for subsequent cross-time comparison and analysis. For example, some texts may record the behavior of pigs in a day, while others may record the behavior patterns of pigs in a week or a month. Through this step, the artificial intelligence processing system prepares the necessary data basis for subsequent joint annotation work. These data and associated variables will be used to train and optimize the system's annotation strategy to improve the accuracy and efficiency of pig behavior analysis.

[0032] In step 120, the behavior tracking cycle data refers to the time period information associated with a specific behavior description statement in the behavior tracking report text. These data usually exist in the form of tags to indicate the time period when the behavior occurs, such as day, week, month, etc. In the application scenario of pig farming, the behavior tracking cycle data may include daily, weekly or monthly records of pigs' eating, activities, resting and other behaviors. These data are crucial for analyzing pigs' behavior patterns, health status and farming efficiency.

[0033] Label adjustment refers to modifying or converting the behavior tracking cycle data in the behavior tracking report text so that the behavior data in different texts can be matched to the same cycle label. In the application scenario of pig farming, since different texts may use different time periods to record the behavior of pigs, label adjustment is required to facilitate subsequent cross-text behavior data analysis and comparison. For example, if some texts record behavior in days and other texts in weeks, then label adjustment is to unify all data under the same cycle label, such as in weeks.

[0034] The behavior tracking report adjustment text refers to the behavior tracking report text with the same behavior tracking cycle label after label adjustment. The behavior data in these texts have been reorganized and sorted according to the unified cycle label, so that the behavior data in different texts can be effectively compared and analyzed. In the application scenario of pig farming, the behavior tracking report adjustment text may include data such as the daily food intake, activity time, and rest time of the pigs in a week. These data have been sorted according to the unified cycle label (such as week).

[0035] In detail, based on step 120, the artificial intelligence processing system will adjust the labels of the behavior tracking cycle data in the collected behavior tracking report text. The purpose of this step is to unify the behavior data in different texts under the same cycle label, so as to facilitate the subsequent cross-text behavior data analysis and comparison. Specifically, the system will first identify the behavior tracking cycle data in each text, that is, determine the time cycle label corresponding to each behavior description statement. Then, the system adjusts the cycle data in each text according to the preset unified cycle label (such as week). If the behavior data in a text is recorded in days, the system will convert these data into data in weeks to match other texts recorded in weeks. Through this step, the artificial intelligence processing system can obtain several behavior tracking report adjustment texts with the same behavior tracking cycle label. The behavior data in these adjusted texts have been reorganized and sorted according to the unified cycle label, which facilitates the subsequent behavior data analysis and comparison. At the same time, this step also helps to eliminate the data inconsistency problem caused by inconsistent cycle labels between different texts, and improves the accuracy and reliability of data analysis.

[0036] In step 130, the description sentence location vector refers to a vector used to represent the position and direction of the behavior tracking description sentence in the behavior tracking report adjustment text in a multidimensional space. These vectors are usually generated by natural language processing technology and can capture the semantic features and contextual relationships of the description sentence. In the application scenario of pig farming, the description sentence location vector may include words and phrases related to pig behavior and the grammatical and semantic relationships between them. For example, the location vector of the description sentence "Pig A eats at 9 am" may contain semantic components such as "Pig A", "9 am" and "eating".

[0037] The contextual cohesion attention dimension refers to the multiple aspects or angles used to measure and focus on the contextual cohesion relationship between different descriptive statements when processing the behavior tracking report adjustment text. These dimensions may include time sequence, spatial relationship, behavioral logic, etc., which are used to guide the system to connect different descriptive statements to form a coherent text. In the application scenario of pig farming, the contextual cohesion attention dimension may include the time sequence of pig behavior (such as eating first and then moving around), spatial relationship (such as eating in a specific area), and behavioral logic (such as there is usually a period of rest after eating).

[0038] In detail, based on step 130, the artificial intelligence processing system will use the behavior semantic association variables and the behavior tracking report to adjust the description statement positioning vector of the behavior tracking description statement in the text to determine the contextual connection attention dimension that needs to be considered when connecting the description statement. Specifically, the system will first analyze the behavior semantic association variables to understand the behavior patterns, habits, and possible environmental influencing factors of the pigs. These variables may include the age, gender, weight, health status of the pigs, and the temperature and humidity of the breeding environment. By analyzing these variables, the system can have a more comprehensive understanding of the behavior of the pigs, and infer the possible logical and temporal relationships between different behaviors. Next, the system uses the description statement positioning vector to capture the semantic features and contextual relationships of the behavior tracking description statement. These vectors can reflect the similarities and differences between the description statements, thereby helping the system to identify which statements are semantically similar and which statements are in chronological order. Finally, the system integrates the information of the behavior semantic association variables and the description statement positioning vector to determine the contextual connection attention dimension that needs to be considered when connecting the description statement. These dimensions may include time sequence (ensuring that the connected text can reflect the actual time sequence of the pig's behavior), spatial relationship (ensuring that the connected text can reflect the differences in pig behavior in different areas), and behavior logic (ensuring that the connected text can reflect the reasonable logical relationship between pig behaviors). By considering these dimensions, the system can generate more coherent and accurate behavior tracking report text.

[0039] In step 140, the associated behavior tracking description sentence refers to other description sentences in the behavior tracking report adjustment text that are identified as logically associated or sequentially associated with a specific behavior tracking description sentence based on the contextual connection attention dimension (such as time sequence, spatial relationship, behavior logic, etc.). In the application scenario of pig farming, the associated behavior tracking description sentence may include other behaviors related to the eating behavior of a specific pig, such as activities before and after eating, resting or drinking water behavior. These associated sentences together constitute a complete description of the pig's behavior pattern, which helps farmers or researchers to have a more comprehensive understanding of the pig's behavioral habits and health status.

[0040] In detail, based on step 140, the artificial intelligence processing system will identify other descriptive statements associated with a specific behavior tracking description statement in the contextual connection attention dimension from each behavior tracking report adjustment text. Specifically, the system first analyzes each behavior tracking report adjustment text to determine all behavior tracking description statements contained therein. Then, based on the previously determined contextual connection attention dimensions (such as time sequence, spatial relationship, behavior logic, etc.), the system looks for other statements associated with each description statement. For example, if the system is focusing on the time sequence dimension, it will look for other behavior descriptions that occur before or after a specific description statement. When identifying associated behavior tracking description statements, the system can consider a variety of factors, including semantic similarity, time interval, spatial location, and consistency of behavior logic between description statements. By comprehensively considering these factors, the system can more accurately identify other statements associated with a specific description statement. Finally, the system will generate a list of associated behavior tracking description statements, which contains all other description statements associated with each specific description statement in the contextual connection attention dimension. This list provides an important data basis for subsequent behavior pattern analysis, anomaly detection, or breeding management decisions.

[0041] In step 150, joint knowledge annotation refers to the process of integrating knowledge information from different sources or different levels and annotating specific content in the text (such as behavior tracking description sentences) when processing the behavior tracking report text. In the application scenario of pig farming, joint knowledge annotation may include combining various knowledge such as pig behavior patterns, environmental factors, health status, etc. with behavior tracking description sentences to add richer semantic and contextual information to these sentences. For example, for the description sentence "Pig A eats at 9 am", joint knowledge annotation can add information about the normal frequency and duration of eating behavior and its association with factors such as ambient temperature and humidity.

[0042] The behavior tracking report annotation text corresponding to each contextual connection attention dimension refers to the text containing the corresponding prior training labels generated for each specific contextual connection attention dimension (such as time sequence, spatial relationship, behavior logic, etc.) when processing the behavior tracking report text. These prior training labels reflect the contextual relationship and logical association of the description statement in this dimension. In the application scenario of pig farming, for the time sequence dimension, the annotation text can highlight the temporal relationship between different behavior description statements; for the spatial relationship dimension, the annotation text can mark the specific location or area where different behaviors occur; for the behavior logic dimension, the annotation text can explain the causal relationship or logical order between different behaviors.

[0043] In detail, based on step 150, the artificial intelligence processing system will perform in-depth joint knowledge annotation on the previously determined associated behavior tracking description statements for each contextual connection attention dimension. Specifically, the system first identifies the associated behavior tracking description statements corresponding to each contextual connection attention dimension. Then, the system uses knowledge information from different sources or different levels, such as pig behavior patterns, environmental factors, health conditions, etc. to annotate these associated statements. The annotated content may include the normal frequency, duration, association with environmental factors, logical order between behaviors, etc. of the behavior. Through joint knowledge annotation, the system can generate a behavior tracking report annotation text containing rich semantic and contextual information for each contextual connection attention dimension. These annotation texts not only reflect the contextual relationship and logical association between the description statements, but also provide a more detailed and accurate data basis for subsequent behavior pattern analysis, anomaly detection or breeding management decisions. For example, in the time sequence dimension, the annotation text can help breeders understand the behavior patterns of pigs throughout the day; in the spatial relationship dimension, the annotation text can reveal the behavior preferences of pigs in different areas; in the behavior logic dimension, the annotation text can explain the causal relationship between pig behaviors.

[0044] In summary, the embodiment of the present invention first adjusts the labels of several behavior tracking report texts that are to be jointly labeled, so that the description sentences that originally have different behavior tracking cycle labels can be unified, and then the behavior tracking report adjustment text that matches the same behavior tracking cycle label is generated. This step greatly improves the convenience and accuracy of subsequent processing. Secondly, based on the behavior semantic association variables and the description statement positioning vector, the context connection attention dimension of the description statement connection processing is cleverly determined, so that the associated behavior tracking description sentences at different context connection attention dimensions can be accurately identified. Finally, by performing joint knowledge annotation on these associated statements, the embodiment of the present invention obtains a behavior tracking report annotation text rich in semantics and contextual information, which provides solid data support for subsequent behavior pattern analysis, anomaly detection or breeding management decisions, and greatly improves the intelligence level and decision-making efficiency of breeding management.

[0045] In some examples, the behavior semantic association variable includes a first behavior semantic association variable; then based on the behavior semantic association variable and the description sentence positioning vector of the behavior tracking description sentence in the behavior tracking report adjustment text, the context connection attention dimension of the description sentence connection processing is determined, including: based on the first behavior semantic association variable, determining the first initial attention dimension of the description sentence connection processing; determining the first description sentence positioning vector corresponding to each behavior tracking description sentence in the behavior tracking report adjustment text in the first initial attention dimension; and determining the context connection attention dimension of the description sentence connection processing based on the first description sentence positioning vector.

[0046] In this embodiment, we can focus on how the behavior semantic association variables specifically affect the determination process of the contextual connection attention dimension. In particular, when the behavior semantic association variables include the first behavior semantic association variable, the following are specific steps for determining the contextual connection attention dimension describing the sentence connection process based on this variable:

[0047] 1) Determine the first initial attention dimension based on the first behavior semantic association variable

[0048] The first behavioral semantically associated variable usually refers to one or more variables that are directly related to pig behavior and have a significant impact. For example, age, sex, weight, or health status. These variables are crucial to understanding pig behavior patterns because they can reveal behavioral differences that pigs may exhibit in different physiological states.

[0049] First initial attention dimension: Based on the first behavior semantic association variable, the system will first determine an initial attention dimension. This dimension is the basis for subsequent analysis, and it reflects an aspect that the system needs to pay special attention to when processing behavior tracking description statements. For example, if the first behavior semantic association variable is the weight of the pig, then the first initial attention dimension may be "behavioral changes related to weight", which means that the system pays more attention to behavior description statements that are directly related to weight changes.

[0050] 2) Determine the first description sentence location vector corresponding to the first initial attention dimension of the behavior tracking description sentence in the behavior tracking report adjustment text

[0051] Description statement location vector: As mentioned above, it is a vector used to represent the position and direction of the behavior tracking description statement in the multidimensional space. In this step, the system generates a corresponding first description statement location vector for each behavior tracking description statement based on the first initial attention dimension. These vectors can capture the semantic features and contextual relationships of the description statement in the first initial attention dimension.

[0052] First statement location vectors: In particular, these vectors will focus on reflecting the degree of association between the statement and the first semantically associated variable (such as weight). For example, if a statement is about the increase in the amount of food consumed by pigs, and this increase is directly related to the weight gain of pigs, then the first statement location vector of this statement will show a high weight or similarity score in this aspect.

[0053] 3) Based on the first description sentence positioning vector, determine the contextual connection attention dimension for the description sentence connection processing

[0054] In this step, the system will use the first description sentence localization vector to further refine and optimize the contextual connection attention dimension. Specifically, the system analyzes the first description sentence localization vectors between different description sentences to identify which sentences are semantically related to each other and how these relationships are reflected in the first initial attention dimension.

[0055] Through this process, the system can not only determine which description statements are closely related to the semantic association variables of the first behavior, but also reveal the contextual connection relationship between these statements. For example, the system can find that the description statements of increased food intake and the description statements of weight gain have a high similarity in the first description statement positioning vector, thereby inferring that there is a direct causal relationship between these two types of behaviors. Based on such analysis, the system can further refine the contextual connection attention dimension to make it more accurately reflect the internal connection and logical order between pig behaviors.

[0056] In summary, by combining the first behavior semantic association variable and the first description sentence location vector, the artificial intelligence processing system can more accurately determine the contextual connection attention dimension of the description sentence connection processing. This process not only improves the accuracy of behavior analysis, but also provides more in-depth and comprehensive data support for subsequent behavior pattern analysis, anomaly detection or breeding management decision-making.

[0057] In some other examples, the behavior semantic association variable also includes a second behavior semantic association variable; then based on the first description statement positioning vector, the contextual connection attention dimension of the description statement connection processing is determined, including: when the first description statement positioning vector meets the set description statement positioning requirements, based on the second behavior semantic association variable, the second initial attention dimension of the description statement connection processing is determined; the second description statement positioning vector corresponding to each behavior tracking description statement in the behavior tracking report adjustment text is determined in the target attention cycle, wherein the target attention cycle is determined based on the first initial attention dimension and the second initial attention dimension; based on the second description statement positioning vector, the contextual connection attention dimension of the description statement connection processing is determined from the target attention cycle.

[0058] Taking into account that the behavioral semantic association variables are not limited to the first behavioral semantic association variables but also include the second behavioral semantic association variables, the process of the artificial intelligence processing system determining the contextual connection attention dimension describing the sentence connection processing will become more complicated but also more precise.

[0059] 1) When the first description sentence positioning vector meets the set description sentence positioning requirements

[0060] Set description sentence location requirements: This is a predefined set of conditions or criteria used to evaluate whether the first description sentence location vector is accurate or representative enough for further analysis. These requirements may involve the vector's dimensionality, similarity score, or degree of match with other known behavior patterns.

[0061] Evaluation and Confirmation: The system first evaluates whether the first description statement location vector of each behavior tracking description statement meets these set requirements. Only when the vector meets these requirements will the system continue to analyze based on the second behavior semantic association variable to ensure the accuracy and effectiveness of subsequent steps.

[0062] 2) Based on the second behavior semantic association variable, determine the second initial attention dimension describing the sentence connection processing

[0063] Second behavioral semantically associated variables: This is another or more important variables in addition to the first behavioral semantically associated variables, which also have a significant impact on the behavior patterns of pigs. For example, in addition to body weight, the temperature, humidity or light conditions of the breeding environment may also be important second behavioral semantically associated variables.

[0064] Second initial attention dimension: Based on the second behavior semantic association variable, the system determines a new dimension different from the first initial attention dimension, namely the second initial attention dimension. This dimension reflects another aspect that the system needs to pay extra attention to when processing the behavior tracking description sentence in order to more comprehensively understand the pig's behavior pattern.

[0065] 3) Determine the second description sentence location vector corresponding to the target attention cycle of the behavior tracking description sentence in each behavior tracking report adjustment text

[0066] Target attention period: Since there are now two initial attention dimensions involved (first and second), the system needs to determine a "target attention period" during which the effects of both dimensions are considered. This period may be a specific time period that combines multiple factors (such as time, environmental conditions, and the pig's behavioral characteristics).

[0067] Second description statement location vector: For each behavior tracking description statement, the system generates the corresponding second description statement location vector within the target attention cycle. These vectors will focus on reflecting the degree of association between the description statement and the second behavior semantic association variable, while considering the impact of the target attention cycle.

[0068] 4) Based on the second description sentence location vector, determine the contextual connection attention dimension of the description sentence connection processing from the target attention cycle

[0069] In this step, the system uses the second description sentence location vector to deeply analyze the description sentence in the target attention cycle to determine the final contextual connection attention dimension. Specifically, the system:

[0070] The similarities and differences between the second description statement location vectors are analyzed to identify which description statements are related to each other in terms of semantics and behavioral logic.

[0071] Combined with the characteristics of the target attention cycle, consider the specific behavioral patterns or changing trends that pigs may exhibit during this cycle.

[0072] Based on the above information, the contextual connection attention dimension of the description sentence connection processing is determined. This dimension will not only reflect the direct connection between the description sentences, but also take into account the influence of multiple factors such as time, environmental conditions and behavioral characteristics.

[0073] Through this process, the AI ​​processing system can more comprehensively consider the impact of multiple behavioral semantic association variables on pig behavior patterns, thereby more accurately determining the contextual connection attention dimension that describes the sentence connection processing. This provides richer and deeper data support for subsequent behavioral analysis, anomaly detection, and breeding management decisions.

[0074] In the next step, based on the second description sentence positioning vector, the contextual connection attention dimension of the description sentence connection processing is determined from the target attention cycle, including: when the second description sentence positioning vector represents that any of the behavior tracking report adjustment texts has a behavior tracking description sentence to be associated within the target attention cycle, the behavior tracking cycle data of the behavior tracking description sentence to be associated is obtained; based on the behavior tracking cycle data of the behavior tracking description sentence to be associated, the contextual connection attention dimension of the description sentence connection processing is determined from the target attention cycle.

[0075] As the artificial intelligence processing system continues to determine the contextual connection attention dimension for descriptive statement connection processing, when the second descriptive statement location vector indicates that there is a behavior tracking descriptive statement to be associated within the target attention cycle, the system will take the following steps to further refine the contextual connection attention dimension.

[0076] 1) Obtain the behavior tracking cycle data of the behavior tracking description statement to be associated

[0077] Behavior tracking description sentences to be associated: These sentences are identified as behavior tracking descriptions that are potentially associated with other description sentences during the target attention period based on the analysis results of the location vector of the second description sentence. They may describe the pigs eating, moving, resting or other behaviors during a specific time period.

[0078] Behavior tracking cycle data: For each behavior tracking description statement to be associated, the system needs to obtain its corresponding behavior tracking cycle data. This data includes the specific time period (such as start time, end time) of the behavior recorded by the description statement and possible periodic information (such as daily and weekly repetition patterns).

[0079] 2) Based on the behavior tracking cycle data of the behavior tracking description sentence to be associated, determine the contextual connection attention dimension of the description sentence connection processing from the target attention cycle

[0080] In this step, the system uses the behavior tracking cycle data of the behavior tracking description sentence to be associated to further analyze and determine the contextual connection attention dimension. The specific process is as follows:

[0081] Time alignment and segmentation: First, the system divides the target attention cycle into finer time periods or "sub-periods" based on the behavior tracking cycle data. These sub-periods should correspond to the specific time periods of the behavior tracking description sentences to be associated for subsequent analysis.

[0082] Behavior pattern recognition: During each sub-cycle, the system analyzes the behavior trace description statements to be associated to identify specific behavior patterns that pigs may exhibit. These patterns may include the frequency, duration, intensity, or association with other behaviors of the behavior.

[0083] Contextual relationship construction: Next, the system constructs the contextual relationship between different description statements based on the behavior pattern recognition results. This includes determining which description statements are adjacent in time and which are logically related to each other (such as causal relationship, sequential relationship, etc.).

[0084] Determine the contextual connection attention dimension: Finally, the system combines the above information to determine the contextual connection attention dimension for the description statement connection processing. These dimensions may include the time order dimension (describing the time order of behavior), the spatial relationship dimension (describing the spatial location relationship of behavior), the behavior logic dimension (describing the causal relationship or logical order between behaviors), etc. Each dimension reflects a specific relationship or pattern between description statements, which helps to understand the behavior pattern of pigs more deeply.

[0085] Through this process, the AI ​​processing system can more accurately determine the contextual attention dimension describing the sentence connection processing from the target attention cycle. This not only improves the accuracy of behavioral analysis, but also provides more detailed and reliable data support for breeding management decisions. In practical applications, these dimensions can help breeders identify abnormal pig behavior, optimize the breeding environment, and improve breeding efficiency.

[0086] In some other possible embodiments, label adjustment is performed on the behavior tracking cycle data of the behavior tracking description sentences in each of the behavior tracking report texts to obtain a number of behavior tracking report adjustment texts matching the same behavior tracking cycle label, including: determining a key behavior tracking description sentence from the behavior tracking description sentences in each of the behavior tracking report texts, and obtaining key behavior tracking cycle data corresponding to the key behavior tracking description sentence; performing behavior tracking cycle updating on the behavior tracking cycle data of the behavior tracking description sentences in each of the behavior tracking report texts based on the key behavior tracking cycle data to obtain the period update data of the behavior tracking description sentences in each of the behavior tracking report texts; and performing semantic alignment processing on the period update data of the behavior tracking description sentences in each of the behavior tracking report texts to obtain a number of behavior tracking report adjustment texts matching the same behavior tracking cycle label.

[0087] When exploring in depth the specific application of artificial intelligence processing systems in the processing of pig farming behavior tracking reports, we need to focus on a core task: how to efficiently and accurately adjust the labels of the behavior tracking cycle data in each behavior tracking report text to ensure that these data from different sources and different recording habits can be integrated and analyzed within a unified time frame.

[0088] 1. Accurately lock the key behavior tracking description statement

[0089] The AI ​​processing system first starts the natural language processing (NLP) module to conduct in-depth analysis of the input behavior tracking report texts. These texts are usually recorded manually by farmers or generated by automated monitoring systems, and the content covers various behaviors of pigs in different time periods, such as eating, moving, resting, etc.

[0090] For example, there are two behavior tracking report texts. Text A records the detailed behavior of Pig A in one day: "Pig A eats from 9 to 10 a.m., moves from 2 to 3 p.m., and rests from 8 to 9 p.m." Text B records the behavior profile of Pig B in a week: "Pig B eats at 10 a.m. every day from Monday to Friday, moves for 1 hour at 4 p.m., and rests at irregular times at night."

[0091] The artificial intelligence processing system uses the NLP module to perform pre-processing steps such as word segmentation, part-of-speech tagging, and named entity recognition on the two texts, identifying key words and phrases related to behavior, such as "eating", "activity", "rest", and time expressions "9 am to 10 am", "Monday to Friday", etc. The system then uses semantic analysis technology to evaluate the importance of each descriptive statement. In this process, the system can refer to factors such as the number of behavior types contained in the statement, the accuracy of the time description, and the logical relationship between behaviors.

[0092] Named Entity Recognition (NER): It is a technology in natural language processing that is used to identify entities with specific meanings from text, such as names of people, places, time expressions, etc.

[0093] Semantic Analysis: It is the process of deeply understanding natural language text, aiming to reveal the implicit meaning, emotional tendency, logical structure and other information in the text.

[0094] After comprehensive evaluation, the system identified key behavior tracking description sentences in text A and text B. These sentences are directly related to the main behavior patterns of pigs and have an important impact on subsequent analysis.

[0095] 2. Intelligent update of behavior tracking cycle data

[0096] After locking the key behavior tracking description sentences, the artificial intelligence processing system next needs to intelligently update the behavior tracking cycle data in these sentences so that all data can be unified under the same behavior tracking cycle label.

[0097] For example, continuing with text A and text B, the system finds that the behavior tracking cycle of text A is in hours, while that of text B is in days. In order to unify the cycle labels, the system sets a standard cycle - "day", that is, all behavior data will be converted into a daily record.

[0098] For text A, the system directly regards each record as a separate daily behavior data, because text A itself covers the complete behavior cycle of pig A in one day. For text B, the system needs to split and aggregate the records. For example, the description "Pig B eats at 10 am every day from Monday to Friday" will be split into five records, each of which represents the eating behavior of pig B at 10 am on a certain day of the week; and "Active for 1 hour at 4 pm" will be aggregated into a daily record, indicating that pig B has 1 hour of activity time at 4 pm every day.

[0099] During the update process, the system also performed data verification and cleaning to ensure that the converted data not only met the requirements of the standard cycle, but also retained the core information of the original data. For data that cannot be directly converted or is ambiguous, the system uses reasonable default values ​​or context-based reasoning to process it.

[0100] 3. Semantic Alignment Processing of Periodically Updated Data

[0101] After the behavior tracking cycle data is updated, the artificial intelligence processing system further performs semantic alignment on these cycle update data to ensure that the sentences describing the same or similar behaviors in different texts remain semantically consistent and coherent.

[0102] For example, after periodic updates, the system obtains two daily behavior tracking report texts (here referred to as text A' and text B'). Text A' records in detail the specific behavior time and duration of pig A on a certain day; while text B' summarizes the daily behavior pattern of pig B in a week.

[0103] To perform semantic alignment, the system first preliminarily categorizes the descriptions in the two texts based on the behavior type. For example, all descriptions of the behavior of "eating" are classified into one category, and all descriptions of the behavior of "activity" are classified into another category. Then, the system sorts these classified sentences according to chronological order to ensure that they are logically coherent.

[0104] However, alignment based solely on behavior type and temporal order may not be sufficient. In some cases, the system also needs to consider other factors such as spatial location, behavior intensity, environmental conditions, etc. For example, if text A' mentions that pig A spent a long time in a specific area, and text B' also mentions pig B's activities in the same area, the system can consider these two descriptions as highly related and give them more attention during the alignment process.

[0105] To achieve more refined semantic alignment, the system can use deep learning models to vectorize description sentences and calculate the similarity between these vectors. Sentences with high similarity often have similar semantic content or contextual relationships, and are therefore easier to align together.

[0106] After semantic alignment, the system finally outputs several behavior tracking report adjustment texts that match the same behavior tracking cycle label (i.e., daily records). The description sentences in these texts not only have a unified time scale, but also maintain a high degree of consistency and coherence in semantics, which greatly facilitates subsequent joint annotation and in-depth analysis.

[0107] It is understandable that traditional data integration methods often rely on manual screening and comparison, which is not only time-consuming and labor-intensive but also prone to errors. The artificial intelligence processing system greatly improves the efficiency and accuracy of data integration by automatically completing tasks such as the recognition of key behavior tracking description statements, the intelligent update of behavior tracking cycle data, and the semantic alignment of cycle update data. This enables farmers to obtain comprehensive and accurate behavior tracking data more quickly, providing strong support for subsequent analysis and decision-making.

[0108] Behavior tracking reports are an important basis for evaluating the health status, behavior patterns and adaptability of pigs to the breeding environment. Through the processing of artificial intelligence processing systems, these data can be integrated and analyzed in a unified time frame, thereby revealing the deep-seated laws and trends behind pig behavior. This helps farmers better understand the needs and response patterns of pigs, and then formulate more scientific and reasonable breeding management strategies. For example, by analyzing the feeding behavior data of pigs, farmers can promptly discover and solve the problem of insufficient or excessive feed supply; by analyzing the activity behavior data, the comfort and safety of the breeding environment can be evaluated.

[0109] Animal welfare is an important aspect of modern farming that cannot be ignored. By accurately tracking and analyzing pigs' behavioral data, farmers can promptly identify and deal with issues that may affect pigs' health and welfare, such as disease and stress reactions. This helps improve the overall health and quality of life of pigs, thereby promoting the improvement of farming efficiency. At the same time, intelligent farming management based on behavioral data can also optimize resource allocation, reduce production costs, improve product quality and other benefits, laying a solid foundation for the sustainable development of the farming industry.

[0110] In summary, the application of artificial intelligence processing system in pig farming behavior tracking report processing not only improves the efficiency and accuracy of data integration, but also promotes the intelligent upgrading of farming management and improves animal welfare and farming benefits.

[0111] Under a preferred design idea, the period label of the behavior tracking period data includes a first period label and a second period label, and the key behavior tracking period data includes the first key tracking period of the key behavior tracking description statement of each behavior tracking report text under the first period label, and the second key tracking period of the key behavior tracking description statement under the second period label; based on the key behavior tracking period data, the behavior tracking period data of the behavior tracking description statement in each behavior tracking report text is updated, including: obtaining the first behavior tracking period data of the behavior tracking description statement in each behavior tracking report text under the first period label; determining the tracking period error between the first behavior tracking period data and the first key tracking period; based on the tracking period error and the second key tracking period, determining the second behavior tracking period data of the behavior tracking description statement in each behavior tracking report text under the second period label, so as to update the behavior tracking period data of the behavior tracking description statement in the behavior tracking report text.

[0112] When deeply discussing how the artificial intelligence processing system can perform refined processing on behavior tracking cycle data with different cycle labels, the implementation process will be elaborated in detail through a preferred design idea. This design idea not only takes into account the diversity and complexity of behavior tracking cycle data, but also realizes the effective conversion and update between cycle labels through intelligent algorithms, thereby ensuring the uniformity and comparability of data.

[0113] 1. Periodic label analysis of behavior tracking periodic data

[0114] In the pig farming behavior tracking report, the behavior tracking cycle data is often assigned different cycle labels to reflect the time granularity or frequency of data recording. In this embodiment, there are two main cycle labels: the first cycle label and the second cycle label. The first cycle label may represent a finer-grained time unit (such as hours, minutes), while the second cycle label represents a coarser-grained time unit (such as days, weeks). The setting of these two cycle labels is intended to meet the needs of different analysis scenarios, but also brings challenges to data integration.

[0115] 2. Identification and extraction of key behavior tracking cycle data

[0116] The artificial intelligence processing system first uses natural language processing technology (NLP) to conduct an in-depth analysis of each behavior tracking report text to identify key behavior tracking description sentences. These sentences usually contain information that has a significant impact on the pig's behavior pattern, such as behavior type, occurrence time, duration, etc. After identifying the key behavior tracking description sentences, the system further extracts the key behavior tracking cycle data of these sentences under two cycle labels: the first key tracking cycle (corresponding to the first cycle label) and the second key tracking cycle (corresponding to the second cycle label).

[0117] For example, there is a behavior tracking report text describing the eating behavior of Pig A in a day: "Pig A eats from 9:00 to 10:00 in the morning and eats again from 15:00 to 16:00 in the afternoon." In this text, "9:00 to 10:00 in the morning" and "15:00 to 16:00 in the afternoon" are time expressions in the key behavior tracking description statement, which correspond to the first key tracking period of eating behavior (in hours). At the same time, if these data need to be converted to the second period label in days, the eating behavior of the first day can be simplified to two time points or time periods (such as "morning" and "afternoon"), but this is not the second key tracking period in the strict sense, because the data under the second period label usually needs to reflect the behavior pattern in a longer time period. However, in actual operation, the behavior pattern under the second period label can be inferred or estimated based on the first key tracking period.

[0118] 3. Intelligent update of behavior tracking cycle data

[0119] After obtaining the first behavior tracking cycle data of the behavior tracking description sentences in each behavior tracking report text, the artificial intelligence processing system next needs to update these data to a unified second cycle label. To achieve this goal, the system first calculates the tracking cycle error between the first behavior tracking cycle data and the first key tracking cycle. The "tracking cycle error" here can be understood as the difference or offset between the actual recorded time period and the key tracking cycle. Then, based on this tracking cycle error and the estimated or known information of the second key tracking cycle (although the second key tracking cycle may not be given directly in actual situations, the system can indirectly obtain it through statistical analysis, pattern recognition and other methods), the system determines the second behavior tracking cycle data of each behavior tracking description statement under the second cycle label.

[0120] Among them, tracking cycle error refers to the difference between the actual recorded behavior tracking cycle and the preset or key behavior tracking cycle. This difference may be caused by the subjectivity of the recorder, the accuracy limitation of the recording equipment, or changes in environmental factors. Pattern recognition is a machine learning technology used to automatically discover and analyze regularities or patterns from data. In the processing of behavior tracking cycle data, pattern recognition can help the system identify the correlation between different behaviors, periodic patterns, etc.

[0121] For example (continuing the previous example): It is known that Pig A generally eats twice a day, and the eating time is roughly the same each time (this is an estimate of the second key tracking cycle obtained by statistical analysis of multiple behavior tracking reports). Now, the specific eating time of Pig A on a certain day (i.e., the first behavior tracking cycle data) needs to be updated to the second cycle label in days. The system first calculates the tracking cycle error between each eating time of this day and the preset eating time (for example, if the preset morning eating time is 9:00-10:00, and the actual record is 9:15-10:15, the tracking cycle error is +15 minutes). Then, based on this tracking cycle error, the system adjusts the estimated values ​​of eating time on other days (if the error is considered to be systematic, it may be necessary to adjust the estimated values ​​of all days; if the error is considered to be random, it may only be necessary to adjust the estimated value of the current day) to obtain more accurate second behavior tracking cycle data. However, in actual operation, since we usually only care about the general outline of the behavior pattern rather than the specific details, it may not be necessary to make precise adjustments to the tracking cycle error at each time point. Instead, the system can adopt a coarser-grained aggregation or smoothing method to reflect the behavior patterns under the second period label.

[0122] IV. Integration and output of periodic update data

[0123] After the intelligent update of the behavior tracking cycle data, the artificial intelligence processing system will integrate all the updated data and generate several behavior tracking report adjustment texts matching the same second cycle label. The behavior tracking description sentences in these texts have been reorganized and expressed according to the unified second cycle label, which greatly facilitates the subsequent joint annotation and in-depth analysis.

[0124] For example (summary example): The final output behavior tracking report adjustment text may no longer contain specific eating time records to the hour or minute, but a general description of Pig A's approximate daily eating pattern: "Pig A eats once in the morning and afternoon every day, and each eating time is about 1 hour." This statement not only simplifies the data presentation method, but also highlights the main characteristics of the behavior pattern, which helps farmers to grasp the behavioral patterns of pigs more quickly and formulate corresponding management measures.

[0125] Thus, it can be seen that the artificial intelligence processing system has shown strong capabilities and significant advantages in processing behavior tracking cycle data with different cycle labels. This technical solution not only realizes the effective conversion and update of data from fine-grained to coarse-grained, but also ensures that the converted data better meets the analysis requirements while maintaining the essence of the original information. This process not only improves the efficiency and accuracy of data integration, but also provides a high-quality data foundation for subsequent joint labeling, behavior pattern analysis, anomaly detection, etc.

[0126] In other possible embodiments, determining the associated behavior tracking description sentences of the behavior tracking report adjustment text in the context connection attention dimension from each of the behavior tracking report adjustment texts includes: determining the past behavior tracking description sentences of the behavior tracking report adjustment text before the context connection attention dimension from each of the behavior tracking report adjustment texts; and determining the associated behavior tracking description sentences of the behavior tracking report adjustment text in the context connection attention dimension from the past behavior tracking description sentences based on the behavior tracking cycle data of the past behavior tracking description sentences.

[0127] In the in-depth exploration of the diverse applications of artificial intelligence processing systems in pig farming behavior tracking report processing, we will focus on a specific embodiment, that is, how to accurately identify the associated behavior tracking description sentences from the adjusted behavior tracking report text under the specific contextual connection attention dimension. This process requires not only the system to have a deep understanding of the text content, but also to be able to flexibly use the behavior tracking cycle data to build logical connections between behaviors.

[0128] 1. In pig farming management, behavior tracking reports are an important way to understand the health status, behavior patterns and adaptability of pigs to the farming environment. However, due to the diversity and complexity of report texts, extracting valuable information directly from them often faces many challenges. In particular, when it is necessary to analyze the association between specific behaviors, it is necessary to consider the contextual connection attention dimension, that is, those dimensions that can guide the understanding of the logical relationship between behaviors, such as time sequence, causality, spatial location, etc.

[0129] The artificial intelligence processing system uses advanced natural language processing (NLP) and machine learning algorithms to automatically process and analyze behavior tracking report texts and identify related behavior tracking description sentences in different contexts and attention dimensions. This process not only improves the efficiency and accuracy of data analysis, but also provides strong support for breeding management decisions.

[0130] 2. Determine the description statement for tracking past behaviors

[0131] Before determining the associated behavior tracking description sentences from the behavior tracking report adjustment text, the artificial intelligence processing system first needs to identify the past behavior tracking description sentences that are located before the specific contextual connection attention dimension. These sentences constitute the background information for analyzing the current behavior pattern and are crucial for understanding the logical relationship between behaviors.

[0132] For example, there is an adjusted behavior tracking report text that describes the behavior pattern of Pig A in a day: "At 7 o'clock in the morning, Pig A started eating; at 9 o'clock in the morning, Pig A rolled in the mud; at 2 o'clock in the afternoon, Pig A returned to the pigsty to rest." In this text, if the contextual connection attention dimension is "time order", then "At 7 o'clock in the morning, Pig A started eating" and "At 9 o'clock in the morning, Pig A rolled in the mud" constitute the past behavior tracking description sentence before the behavior of "At 2 o'clock in the afternoon, Pig A returned to the pigsty to rest."

[0133] The artificial intelligence processing system parses these sentences through NLP technology, extracts time expressions and behavior description words, and then sorts them according to chronological order. The sorted sentence list is the past behavior tracking description sentence set required for subsequent analysis.

[0134] 3. Determine the associated behavior tracking description statement based on the behavior tracking cycle data

[0135] After determining the past behavior tracking description sentences, the artificial intelligence processing system needs to further analyze the behavior tracking cycle data in these sentences to determine other sentences associated with the specific behavior tracking description sentences under the current contextual connection attention dimension.

[0136] For example (continuing the previous example): the behavior tracking description sentence of interest is "at 2 pm, Pig A returned to the pig house to rest." In order to determine the past behaviors associated with this behavior, the system needs to review the previous past behavior tracking description sentences and analyze the behavior tracking cycle data between them.

[0137] First, the system noticed that the two behaviors of "Pig A started eating at 7 a.m." and "Pig A rolled in the mud at 9 a.m." occurred immediately before "Pig A returned to the pigsty to rest at 2 p.m.." However, the time sequence alone is not enough to determine the direct connection between them. Therefore, the system needs to further consider the causal relationship or logical connection between the behaviors.

[0138] In this example, although eating and rolling are temporally adjacent to resting, it is unclear whether they are the direct cause of resting. However, if the system finds through historical data analysis that pigs often have a resting behavior pattern after eating and rolling, it can infer that there is some degree of correlation between these two past behaviors and the current resting behavior.

[0139] To verify this association, the system can use machine learning algorithms to analyze a large number of historical behavior tracking reports and build a correlation model between behaviors. In practical applications, this model may be implemented based on technologies such as time series analysis, association rule mining, or graph neural networks. By analyzing the behavior tracking cycle data in past behavior tracking description statements and their characteristics in terms of time, space, causality, etc., the system can identify other statements associated with specific behaviors.

[0140] 4. Integration and output of associated behavior tracking description statements

[0141] After determining the past behavior tracking description sentences associated with a specific behavior tracking description sentence, the artificial intelligence processing system integrates these sentences together and outputs a behavior tracking report adjustment text segment or summary containing the associated information. This segment or summary not only shows the logical relationship between behaviors, but also provides intuitive and comprehensive information support for breeding management decisions.

[0142] For example (summary example): After the above analysis process, the artificial intelligence processing system may output the following text fragment containing associated behavior tracking description statements: "Pig A started eating at 7 o'clock in the morning and rolled in the mud at 9 o'clock in the morning. After these activities, Pig A returned to the pig house to rest at 2 o'clock in the afternoon." Such a text fragment clearly shows the behavioral sequence of Pig A in a day and the possible correlations between them, which helps farmers better understand the behavioral patterns and health status of pigs.

[0143] Thus, it can be seen that the artificial intelligence processing system has demonstrated strong capabilities and significant advantages in processing pig farming behavior tracking reports. This technical solution achieves accurate identification and integration of the correlation between behaviors by introducing the analysis method of contextual connection attention dimension and behavior tracking cycle data. This process not only improves the depth and breadth of data analysis, but also provides more comprehensive and accurate information support for farming management decisions.

[0144] In the next step, based on the behavior tracking cycle data of the past behavior tracking description sentences, the associated behavior tracking description sentences of the behavior tracking report adjustment text in the context connection attention dimension are determined from the past behavior tracking description sentences, including: determining a set of behavior tracking description sentences corresponding to each of the behavior tracking report adjustment texts, wherein the behavior tracking description sentence set is used to cache past behavior tracking description sentences of the behavior tracking report adjustment text before the context connection attention dimension; optimizing the behavior tracking description sentence set based on the context connection attention dimension and the behavior tracking cycle data of the past behavior tracking description sentences; and determining the associated behavior tracking description sentences of the behavior tracking report adjustment text in the context connection attention dimension from the optimized behavior tracking description sentence set.

[0145] In the in-depth stage of pig farming behavior tracking report processing, the artificial intelligence processing system demonstrated its outstanding ability in complex data analysis and association identification. Next, we will elaborate on how the artificial intelligence processing system accurately determines the associated behavior tracking description statements from the cached behavior tracking description statement set under the specific context connection attention dimension based on the behavior tracking cycle data of past behavior tracking description statements. This process not only reflects the system's sophisticated processing capabilities for time series data, but also demonstrates its intelligent advantages in building behavior logic chains.

[0146] 1. Construction and caching of behavior tracking description statement set

[0147] Before determining the associated behavior tracking description sentences, the AI ​​processing system first adjusts the text for each behavior tracking report and builds a special collection of behavior tracking description sentences. The purpose of this collection is to cache past behavior tracking description sentences that occurred before the contextual connection attention dimension. These sentences are the basis for subsequent analysis, and they provide the necessary background information and contextual references for understanding the current behavior.

[0148] For example, there is a behavior tracking report adjustment text that describes the behavior sequence of Pig A over a period of time: "At 8 a.m., Pig A started eating; at 10 a.m., Pig A rolled in the mud; at 12 noon, Pig A returned to the pigsty to rest." In this example, if the context connection attention dimension is "time order", then the two description sentences "At 8 a.m., Pig A started eating" and "At 10 a.m., Pig A rolled in the mud" will be cached in the behavior tracking description sentence set corresponding to this report adjustment text.

[0149] The artificial intelligence processing system automatically parses the text content through NLP technology, extracts time expressions and behavior description words, and organizes them into a set of behavior tracking description sentences in chronological order. This process is highly automated and can process large amounts of text data and quickly generate corresponding sentence sets.

[0150] 2. Optimization of attention dimension and behavior tracking cycle data based on contextual connection

[0151] After constructing a set of behavior tracking description statements, the AI ​​processing system does not directly determine the associated behavior tracking description statements from them. Instead, the system further optimizes this set based on the contextual connection attention dimension and the behavior tracking cycle data of past behavior tracking description statements. The purpose of the optimization is to eliminate those statements that are irrelevant or have low relevance to the current analysis target, thereby highlighting those truly important associated behaviors.

[0152] For example (continuing the previous example): After determining "time order" as the contextual attention dimension, the AI ​​processing system optimizes the cached set of behavior tracking description statements. This optimization process may involve the following aspects:

[0153] Time window screening: The system may set a time window (such as a few hours before and after), and only retain the past behavior tracking description statements that fall within this time window. This helps reduce the impact of noise data and improve the accuracy of analysis.

[0154] Behavior type matching: The system may identify past behaviors that match or are logically related to the current behavior (e.g., "Pig A returns to the barn to rest at 12 noon") based on common behavior patterns of pigs or historical data analysis results. For example, if historical data shows that eating and rolling are often followed by resting behaviors, the system will pay more attention to these two types of behaviors.

[0155] Cycle data analysis: The system conducts in-depth analysis of the behavior tracking cycle data in the past behavior tracking description statements. This data may include the frequency, duration, and periodic pattern of the behavior. By analyzing this data, the system can identify the periodic association or time dependency between behaviors.

[0156] Semantic relationship reasoning: In addition to analysis based on statistical data, the system may also use semantic analysis technology to infer the causal relationship or logical sequence between behaviors. This involves understanding the deep meaning of the sentence and comprehensive consideration of the context.

[0157] Through the above optimization process, the artificial intelligence processing system can significantly reduce the redundant information in the set of behavior tracking description statements and improve the efficiency and accuracy of associated behavior identification.

[0158] 3. Determination and Output of Related Behavior Tracking Description Statements

[0159] After completing the optimization of the behavior tracking description sentence set, the artificial intelligence processing system will determine the related behavior tracking description sentences in the optimized set that are connected to the current behavior tracking report adjustment text in a specific contextual connection attention dimension. These sentences are not only adjacent or close to the current behavior in time, but also have a close logical relationship.

[0160] For example (summary example): After optimization, the set of behavior tracking description sentences may only contain a small number of past description sentences that are highly relevant to the behavior "At 12 noon, Pig A returned to the pig house to rest", such as "At 8 a.m., Pig A began to eat" and "At 10 a.m., Pig A rolled in the mud." The artificial intelligence processing system comprehensively analyzes the time information, behavior type, cycle data, and semantic relationships in these sentences, and ultimately determines that they are related behavior tracking description sentences, and outputs this result to the user.

[0161] The output results may be presented in a variety of forms, such as text summaries, chart reports, or visual interfaces. Regardless of the form adopted, the purpose is to enable users to intuitively understand the behavior patterns of pigs and the correlations between them, thereby providing strong support for breeding management decisions.

[0162] Thus, it can be seen that the artificial intelligence processing system has demonstrated powerful capabilities and significant advantages in processing pig farming behavior tracking reports. This technical solution achieves in-depth analysis and associated identification of pig behavior patterns through a series of steps, such as constructing a set of behavior tracking description statements, optimizing the processing of attention dimensions and behavior tracking cycle data based on contextual connection, and accurately determining associated behavior tracking description statements. This process not only improves the accuracy and depth of data analysis, but also provides more comprehensive and accurate information support for farming management decisions.

[0163] In some other possible embodiments, joint knowledge annotation is performed on the associated behavior tracking description statements corresponding to the contextual connection attention dimension, including: determining a text paragraph to be annotated of the associated behavior tracking description statement corresponding to the contextual connection attention dimension; decomposing the text paragraph to be annotated into at least one text report unit; configuring a corresponding NLP knowledge annotation algorithm for each of the text report units, and performing joint knowledge annotation on the associated behavior tracking description statements in the text report unit through the NLP knowledge annotation algorithm.

[0164] In the in-depth exploration of pig farming behavior tracking report processing, the artificial intelligence processing system demonstrated its powerful capabilities in complex data analysis and knowledge annotation. Next, we will elaborate on how the artificial intelligence processing system performs joint knowledge annotation on the associated behavior tracking description sentences corresponding to the contextual connection attention dimension in some possible embodiments. This process not only involves the fine processing of text content, but also integrates advanced knowledge annotation algorithms in the field of natural language processing (NLP) to achieve in-depth understanding and expression of behavior patterns and their associations.

[0165] 1. Determination of the text paragraphs to be annotated

[0166] Before joint knowledge annotation, the AI ​​processing system first needs to determine the text paragraphs to be annotated where the associated behavior tracking description sentences corresponding to the contextual connection attention dimension are located. These paragraphs usually contain complete sentences or sentence combinations that describe specific behavior patterns and their associated behaviors before and after, and are the basis for subsequent annotation work.

[0167] For example, after the previous analysis steps, the AI ​​processing system has identified the past behavior tracking description sentences associated with the behavior of "Pig A started eating at 3 pm" under the contextual connection attention dimension of "time order", such as "Pig A had a full activity at 10 am" and "At 12 noon, Pig A returned to the pigsty to rest for a while". These sentences together constitute the text paragraph to be annotated that describes the before-after association pattern of Pig A's eating behavior.

[0168] The system automatically identifies and extracts these sentences through NLP technology, and combines them into a logically coherent text paragraph for subsequent joint knowledge annotation.

[0169] 2. Disassembly of the Text Report Unit

[0170] After determining the text paragraph to be annotated, the AI ​​processing system will then break it down into at least one text report unit. These text report units are the basic units of annotation work, and they may be a complete sentence, a clause, or a phrase containing a description of a specific behavior. The purpose of breaking down is to simplify complex text content into small units that are easier to process and analyze, thereby improving the accuracy and efficiency of annotation.

[0171] For example (continuing the previous example):

[0172] The AI ​​processing system may break down the text paragraph to be annotated into the following text report units:

[0173] “Pig A was fully active at 10 a.m.”;

[0174] "At 12 noon, Pig A returned to the pigsty and rested for a while";

[0175] "Pig A starts eating at 3pm".

[0176] Each text report unit contains a clear behavior description and time information, providing a clear analysis object for subsequent NLP knowledge annotation algorithms.

[0177] 3. Configuration and application of NLP knowledge annotation algorithm

[0178] After disassembling the text report units, the artificial intelligence processing system needs to configure the corresponding NLP knowledge annotation algorithms for them. These algorithms are key tools for realizing joint knowledge annotation, and they can conduct in-depth analysis and annotation of text content based on natural language processing technology and domain knowledge base.

[0179] For example (algorithm configuration and application):

[0180] Part-of-speech tagging: First, the system may apply a part-of-speech tagging algorithm to identify nouns, verbs, adjectives, and other parts of speech in the text reporting unit. This helps the system understand the structure and meaning of the sentence, laying the foundation for subsequent behavior recognition and association analysis.

[0181] Named Entity Recognition: Next, the system may use named entity recognition algorithms to extract key entities in text report units, such as pig names, behavior types, time expressions, etc. These entities are important nodes for building a behavioral pattern association network.

[0182] Relation extraction: Based on part-of-speech tagging and named entity recognition, the system further applies relation extraction algorithms to identify the associations between behaviors in text report units. These relationships may include temporal order relationships, causal relationships, spatial location relationships, etc., which together constitute the logical framework of the behavior pattern.

[0183] Domain knowledge integration: Finally, the system will integrate the above analysis results with the domain knowledge base. The domain knowledge base contains professional terms, common behavior patterns, abnormal behavior characteristics and other information in the field of pig farming. By integrating domain knowledge, the system can verify and optimize the annotation results and improve the accuracy and reliability of the annotation.

[0184] Through the configuration and application of the above-mentioned NLP knowledge annotation algorithm, the artificial intelligence processing system can realize the joint knowledge annotation of the associated behavior tracking description statements in the text report unit. The annotation results not only contain the descriptive information of the behavior itself, but also reveal the association and logical order between the behaviors, providing more comprehensive and in-depth data support for breeding management decisions.

[0185] 4. Output and Utilization of Joint Knowledge Annotation Results

[0186] After completing the joint knowledge annotation, the artificial intelligence processing system will output the annotation results to the user in an appropriate form. These results may be presented in various forms such as text reports, charts, and visual interfaces, so that users can intuitively understand the behavior patterns of pigs and their associations.

[0187] For example (result output and utilization):

[0188] The output annotation results may include the following information:

[0189] Behavior Description: Lists in detail the specific contents of each associated behavior tracking description statement.

[0190] Temporal relationship: clearly mark the temporal sequence between behaviors.

[0191] Causation: Infer and label possible cause-and-effect relationships between behaviors.

[0192] Abnormal behavior prompts: Identify and annotate abnormal behaviors based on the domain knowledge base.

[0193] Users can use these annotation results to evaluate the health status, behavioral adaptability, and suitability of the breeding environment of pigs, and formulate corresponding management measures and improvement plans accordingly. For example, if it is found that pigs frequently exhibit abnormal behaviors within a specific period of time, users may need to adjust feed formulas, improve breeding environments, or increase activity levels to improve pig welfare.

[0194] Thus, it can be seen that the joint knowledge annotation technology of the artificial intelligence processing system in the processing of pig farming behavior tracking reports has shown significant advantages. This technical solution not only improves the depth and breadth of behavior analysis, but also provides richer and more accurate information support for farming management decisions.

[0195] In another possible embodiment, obtaining a plurality of behavior tracking report texts to be jointly annotated includes: obtaining a plurality of initial behavior tracking report texts and tracking environment characteristics of the initial behavior tracking report texts; based on the tracking environment characteristics, semantically annotating the initial behavior tracking report texts to obtain a plurality of behavior tracking report texts to be jointly annotated.

[0196] When discussing the diverse applications of artificial intelligence processing systems in pig farming behavior tracking report processing, we have to mention a key step: how to effectively prepare and preprocess data for subsequent joint labeling and in-depth analysis. In another possible embodiment, the artificial intelligence processing system obtains several initial behavior tracking report texts and their tracking environment features, and semantically labels the texts based on these features, thereby generating several behavior tracking report texts to be jointly labeled. This process not only improves the availability and accuracy of the data, but also lays a solid foundation for subsequent analysis. The following is a detailed example explanation of this embodiment.

[0197] 1. Obtaining the text of the initial behavior tracking report

[0198] As the starting point of data processing, the artificial intelligence processing system first needs to obtain several initial behavior tracking report texts. These texts are usually recorded manually by farmers or generated by automated monitoring systems, and the content covers various behavioral performances of pigs in different time periods and related environmental information. These texts are the basic data source for subsequent analysis, and their quality and integrity directly affect the accuracy and reliability of the analysis results.

[0199] For example, a pig farm has an automated monitoring system that can monitor and record the behavior data of pigs in real time. Every day, the system generates an initial behavior tracking report text, which describes in detail the pigs' eating, activity, rest and other behavior patterns on that day. In addition, the system also records some environmental characteristic data related to behavior, such as temperature, humidity, light intensity, etc. The artificial intelligence processing system is connected to the monitoring system through an interface to regularly obtain these initial behavior tracking report texts and their tracking environmental characteristic data.

[0200] 2. Identification and extraction of tracking environment features

[0201] After obtaining the initial behavior tracking report text, the AI ​​processing system needs to identify and extract the tracking environment features. These features are important for understanding the environmental factors behind pig behavior because they can reveal the potential relationship between behavior and environment.

[0202] Tracking environment characteristics: refers to environmental parameters or conditions related to pig behavior tracking, such as temperature, humidity, light intensity, air quality, etc. These characteristics are one of the important factors affecting pig behavior.

[0203] For example, in the initial behavior tracking report text, in addition to the specific sentences describing the pigs’ behavior, it may also contain some data points or descriptive text related to environmental characteristics. For example, “Today, the temperature in the pig house is maintained at around 25°C, the humidity is moderate, and the light is sufficient.” This sentence includes three tracking environmental characteristics: temperature, humidity, and light intensity. The artificial intelligence processing system parses these texts through NLP technology, extracts the environmental characteristic data, and associates it with the corresponding behavior description sentences.

[0204] 3. Semantic Annotation Based on Tracking Environment Features

[0205] After extracting the tracking environment features, the AI ​​processing system will semantically annotate the initial behavior tracking report text. Semantic annotation is a process of converting entities, relationships, events, etc. in natural language text into structured data, which helps the system understand the text content more deeply and identify the key information and logical structure.

[0206] For example: Based on the extracted tracking environment feature data, the artificial intelligence processing system will semantically annotate each behavior description sentence in the initial behavior tracking report text. For example, for the behavior description sentence "Pig A started eating at 9 am", the system may annotate the pig's name (Pig A), behavior type (eating), occurrence time (9 am) and related environmental characteristics (such as temperature, humidity, etc.). If environmental characteristics have a significant impact on behavior (such as high temperature causing decreased appetite), the system will also try to establish a correlation between behavior and environmental characteristics and reflect it in the annotation.

[0207] In the process of semantic annotation, the artificial intelligence processing system may use a variety of NLP technologies and domain knowledge bases to assist in the annotation work. For example, the system may use part-of-speech tagging algorithms to identify nouns, verbs and other parts of speech in the text; use named entity recognition algorithms to extract key entities such as pig names and time expressions; use relationship extraction algorithms to identify the association between behavior and environmental features, etc. By combining these technologies and methods, the system can achieve comprehensive and in-depth semantic annotation of the initial behavior tracking report text.

[0208] 4. Generation of Behavior Tracking Report Text for Joint Annotation

[0209] After semantic annotation, the initial behavior tracking report text is converted into a form containing rich semantic information and structured data. These annotated texts are the behavior tracking report texts to be jointly annotated. They not only retain the behavior description information in the original text, but also add detailed descriptions of the tracking environment characteristics and the revelation of associations, providing a more comprehensive and accurate data foundation for subsequent joint annotation and in-depth analysis.

[0210] For example, after the above steps, the artificial intelligence processing system generated the following behavior tracking report text fragment to be jointly annotated: "Pig A started eating at 9 am in an environment with a temperature of 25°C and moderate humidity. During the eating process, Pig A showed a high appetite and activity. Subsequently, in a well-lit pig house, Pig A was active for about half an hour." This text not only describes Pig A's eating and activity behaviors and the time of their occurrence, but also clearly points out the tracking environment characteristics related to these behaviors (temperature, humidity, light intensity) and their possible effects on behavior (such as changes in appetite and activity). Such annotation results enable subsequent analysis to more accurately consider the impact of environmental factors on pig behavior, thereby improving the accuracy and reliability of the analysis.

[0211] Thus, it can be seen that the artificial intelligence processing system has significant advantages in acquiring and preprocessing behavior tracking report text. This technical solution not only improves the availability and accuracy of data by introducing the recognition and extraction of tracking environment features and the semantic annotation process based on these features, but also lays a solid foundation for subsequent joint annotation and in-depth analysis.

[0212] In some other examples, the tracking environment features include a target text semantic fine-grained vector of the initial behavior tracking report text; based on the tracking environment features, the initial behavior tracking report text is semantically annotated to obtain a number of behavior tracking report texts to be jointly annotated, including: obtaining the initial text semantic fine-grained vector of the initial behavior tracking report text; pairing the initial text semantic fine-grained vector with the target text semantic fine-grained vector; and based on the pairing result, performing semantic fine-grained annotation on the initial behavior tracking report text to obtain a number of behavior tracking report texts to be jointly annotated.

[0213] In-depth discussion of the advanced application of artificial intelligence processing system in pig farming behavior tracking report processing involves the embodiment of tracking environment feature refinement and semantic fine-grained annotation. This embodiment further improves the preprocessing accuracy of behavior tracking report text and the depth of subsequent analysis by introducing the concept of target text semantic fine-grained vector.

[0214] 1. Deepening of tracking environment features: target text semantic fine-grained vector

[0215] In traditional behavior tracking report processing, tracking environment features are often simplified into some macro environmental parameters (such as temperature, humidity, etc.). However, in practical applications, these macro parameters may not fully reflect the subtle impact of the environment on pig behavior. Therefore, in this embodiment, the concept of target text semantic fine-grained vector is introduced as a deeper expression of tracking environment features.

[0216] Target text semantic fine-grained vector: refers to the feature vector that can reflect the semantic details of the text after in-depth analysis of the text content through natural language processing technology. Compared with traditional vocabulary vectors or sentence vectors, the target text semantic fine-grained vector has higher semantic resolution and richer information content, and can more accurately capture the semantic nuances in the text.

[0217] In the context of pig farming behavior tracking reports, the target text semantic fine-grained vector may contain multi-dimensional information about pig behavior descriptions, environmental conditions descriptions, and potential relationships between the two. This information is crucial for a deep understanding of the environmental factors behind pig behavior and their influencing mechanisms.

[0218] 2. Acquisition of Initial Text Semantic Fine-Grained Vector

[0219] In order to semantically annotate the initial behavior tracking report text based on the target text semantic fine-grained vector, the artificial intelligence processing system first needs to obtain the initial text semantic fine-grained vector of these texts. This process usually involves the following steps:

[0220] Text preprocessing: Perform preprocessing operations such as word segmentation, part-of-speech tagging, and stop word removal on the initial behavior tracking report text to facilitate subsequent in-depth semantic analysis.

[0221] Semantic feature extraction: Semantic analysis techniques in natural language processing (such as word embedding, pre-trained language models, etc.) are used to extract semantic features from text and represent them in the form of initial text semantic fine-grained vectors. These vectors not only contain information at the vocabulary level, but also incorporate semantic information in the context.

[0222] Feature optimization: Optimize the extracted initial text semantic fine-grained vector to improve its expressiveness and discrimination. Optimization methods may include dimensionality reduction, feature selection, or feature fusion.

[0223] 3. Pairing of Initial Text Semantic Fine-Grained Vector and Target Text Semantic Fine-Grained Vector

[0224] After obtaining the initial text semantic fine-grained vectors of the initial behavior tracking report text, the artificial intelligence processing system needs to pair these vectors with the target text semantic fine-grained vectors. The purpose of pairing is to find the semantic fragments in the initial text that best match the target environment features, so as to facilitate subsequent semantic fine-grained annotation.

[0225] The pairing process may involve the following key steps:

[0226] Similarity calculation: Use vector space models or deep learning methods to calculate the similarity or distance between the initial text semantic fine-grained vector and the target text semantic fine-grained vector. Vector pairs with higher similarity are closer in semantics and are therefore more likely to represent the same or similar environmental features.

[0227] Threshold setting: Set a similarity threshold based on actual needs and data characteristics. Only when the similarity between vector pairs exceeds the threshold, it is considered that there is a valid pairing relationship between them.

[0228] Pairing result screening: Filter out valid pairing results based on similarity calculation results and set thresholds. These results will serve as the basis for subsequent semantic annotation.

[0229] 4. Semantic fine-grained annotation

[0230] After the initial text semantic fine-grained vector is paired with the target text semantic fine-grained vector, the artificial intelligence processing system can perform semantic fine-grained annotation on the initial behavior tracking report text according to the pairing results. This process aims to enhance the comprehensibility and analyzability of the text through refined semantic information.

[0231] The annotation process may include the following steps:

[0232] Annotation element identification: Based on the pairing results, semantic elements (such as words, phrases, or sentence fragments) in the initial text that are associated with the target environment features are identified. These elements are the core objects of annotation work.

[0233] Prior training label generation: Generate corresponding prior training labels for the identified semantic elements. Prior training labels may include detailed descriptions of environmental features, predictions of possible impacts on behavior, and associations with other elements. This information will be attached to the original text in a structured form.

[0234] Integration of annotation results: Integrate the generated prior training labels into the initial behavior tracking report text to form several behavior tracking report texts to be jointly annotated. These texts not only retain the original behavior description information, but also add rich environmental feature descriptions and semantic association information.

[0235] For example, one initial behavior tracking report text describes the pigs’ eating behavior at different times and the conditions of their environment: “In the cool air of the early morning, Pig A began to eat breakfast leisurely; as the sun gradually rose, the temperature in the pig house also rose, but Pig A’s enthusiasm for eating did not diminish.”

[0236] The AI ​​processing system first obtains the initial text semantic fine-grained vector of the text and pairs it with the target text semantic fine-grained vector (representing features such as cool air in the morning and high temperature environment). Through similarity calculation and threshold screening, the system identifies the semantic fragments in the text that describe cool air in the morning and high temperature environment.

[0237] Next, the system generates corresponding prior training labels for these semantic fragments, such as "Cool air in the morning: It is conducive to improving the appetite of pigs" and "High temperature environment: Although it exists, it has little effect on the eating enthusiasm of pig A". Then, the system integrates these prior training labels into the original text to form the behavior tracking report text to be jointly annotated.

[0238] The annotated text not only clearly describes the pigs’ eating behavior and its temporal changes, but also deeply reveals the possible impact of environmental factors on behavior. Such text provides richer and more accurate data support for subsequent joint annotation and in-depth analysis.

[0239] Thus, it can be seen that the introduction of semantic fine-grained vectors of target text and the annotation method based on semantic fine-grainedness have significant advantages in the processing of pig farming behavior tracking reports. This technical solution not only deepens the understanding and expression of tracking environment characteristics, but also improves the preprocessing accuracy of behavior tracking report text and the depth of subsequent analysis.

[0240] In some independent embodiments, after obtaining the behavior tracking report annotation text corresponding to each context connection attention dimension, the method also includes: performing structured disaster recovery processing on the behavior tracking report annotation text corresponding to each context connection attention dimension.

[0241] Specifically, the behavior tracking report annotation text corresponding to each contextual connection attention dimension is subjected to structured disaster recovery processing, including:

[0242] The target deep structured semantic network is used to extract structured disaster recovery vectors from the behavior tracking report annotation text corresponding to each context connection attention dimension, and structured disaster recovery processing is performed based on the extracted structured disaster recovery vectors;

[0243] The debugging step of the target deep structured semantic network includes: obtaining at least one sample behavior tracking report annotation text, each sample behavior tracking report annotation text includes a sample structured disaster recovery vector and a debugging learning label, and the debugging learning label includes a prior distribution feature of at least one vector member in the sample structured disaster recovery vector in a preset structured disaster recovery feature map; performing multiple cycles of debugging on the deep structured semantic network to be debugged according to the at least one sample behavior tracking report annotation text until the loss assessment requirement is met, thereby obtaining the deep structured semantic network;

[0244] Among them, for each sample behavior tracking report annotation text, each loop debugging includes: inputting the sample structured disaster recovery vector of the sample behavior tracking report annotation text into the deep structured semantic network to be debugged in this loop, and obtaining the chain embedded semantic vector of each vector member in the preset structured disaster recovery feature graph, and the chain embedded semantic vector is used to characterize the confidence weight feature of the distribution feature of the corresponding vector member on the sample structured disaster recovery vector in the preset structured disaster recovery feature graph; for each vector member, according to the chain embedded semantic vector of the vector member, obtaining the attention embedded semantic vector of the vector member in X intersecting feature subgraphs in the preset structured disaster recovery feature graph, The attention embedding semantic vector is used to characterize the confidence weight characteristics of the local relationship characteristics of the corresponding vector member in the corresponding feature subgraph; X is a positive integer; for each vector member, according to the local prior relationship characteristics of the vector member in the corresponding feature subgraph and the set discrete evaluation index, the local weight matrix of the local relationship characteristics of the vector member in the corresponding feature subgraph is determined, and according to the difference between the attention embedding semantic vector and the local weight matrix of the vector member in each feature subgraph, the training error of the vector member in each feature subgraph is determined; based on the training errors of each vector member in each feature subgraph, the network neuron weights of the deep structured semantic network to be debugged are adjusted.

[0245] In the in-depth exploration of pig farming behavior tracking report processing, the artificial intelligence processing system is not only committed to improving the accuracy and efficiency of data annotation, but also focuses on how to ensure that these annotation texts can maintain their integrity and availability in the face of potential data loss, corruption or system failure. To this end, in some independent embodiments, the system introduces a structured disaster recovery processing mechanism to further enhance the reliability and durability of the behavior tracking report annotation text. The following is a detailed example explanation of this embodiment.

[0246] 1. Background of the Introduction of Structured Disaster Recovery

[0247] In the processing flow of pig farming behavior tracking reports, the annotated text of the behavior tracking report after joint knowledge annotation is an important basis for subsequent analysis and decision-making. However, these text data may face various risks during storage, transmission or use, such as hardware failure, software errors, human operation errors, etc., which may lead to data loss or damage. In order to reduce the impact of these risks on data integrity and availability, the artificial intelligence processing system introduces a structured disaster recovery processing mechanism.

[0248] 2. Specific implementation steps of structured disaster recovery

[0249] 1) Using the target deep structured semantic network to extract structured disaster recovery vectors

[0250] The first step of structured disaster recovery processing is to use the target deep structured semantic network (hereinafter referred to as "deep semantic network") to extract structured disaster recovery vectors from the behavior tracking report annotation text corresponding to each context connection attention dimension. This step aims to express the key information in the text in a structured form to form a disaster recovery vector for subsequent disaster recovery processing.

[0251] The deep semantic network is a semantic analysis model based on deep learning technology. It can deeply understand the semantic structure of text content and convert it into a computable numerical vector. In the process of extracting structured disaster recovery vectors, the deep semantic network will perform a comprehensive semantic analysis on the annotated text, identify key entities, relationships, events and other elements, and represent them in the form of structured disaster recovery vectors. These vectors not only contain the core information of the text, but also retain its semantic structure and contextual relationship.

[0252] 2) Processing of structured disaster recovery vectors

[0253] After extracting the structured disaster recovery vectors, the AI ​​processing system will further perform disaster recovery on these vectors. The specific operations of this step may include data backup, redundant storage, error detection and recovery, etc. Through these measures, the system can ensure that the complete annotated text information can still be restored in the event of data loss or damage.

[0254] 3. Debugging and Optimizing the Target Deep Structured Semantic Network

[0255] In order to ensure the accuracy and reliability of the deep semantic network in the extraction of structured disaster recovery vectors, the system needs to be fully debugged and optimized. The following are the debugging steps for the target deep structured semantic network:

[0256] 1) Obtain sample behavior tracking report annotation text and its debugging learning label

[0257] First, the system needs to obtain at least one sample behavior tracking report annotated text as training data. Each sample text contains a sample structured disaster recovery vector and a debugging learning label. The debugging learning label is manually annotated by experts based on the semantic content and structural features of the sample text. It reflects the prior distribution characteristics of each vector member in the sample structured disaster recovery vector in the preset structured disaster recovery feature map. This prior knowledge is crucial for guiding the learning and debugging of the network.

[0258] 2) Loop debugging of deep structured semantic networks to be debugged

[0259] Next, the system will debug the deep structured semantic network to be debugged multiple times based on the sample behavior tracking report annotation text. In each cycle, the system will input the sample structured disaster recovery vector of the sample text into the network and observe whether the output of the network is consistent with the debugging learning label. If not, the training error is calculated based on the set loss assessment index, and the network's neuron weights are adjusted to reduce the error. This process will be repeated until the loss assessment requirements are met.

[0260] 3) Generation of chained semantic embedding vectors and attention semantic embedding vectors

[0261] During the loop debugging process, the system generates a chain embedding semantic vector and an attention embedding semantic vector for each vector member. The chain embedding semantic vector is used to characterize the confidence weight features of the overall distribution features of the vector members in the preset structured disaster recovery feature graph; while the attention embedding semantic vector is used to characterize the confidence weight features of the local relationship features of the vector members in a specific feature subgraph. The generation of these vectors helps the system to more accurately understand the position and functional relationship of the vector members in the structured disaster recovery feature graph.

[0262] 4) Determination of local weight matrix of local relationship features and calculation of training error

[0263] For each vector member, the system determines the local weight matrix based on its local prior relationship features in the corresponding feature subgraph and the set discrete evaluation index. Then the difference between the attention embedding semantic vector and the local weight matrix is ​​compared to calculate the training error. This error reflects the accuracy of the network's understanding of the vector member features at the current learning stage.

[0264] 5) Adjustment of network neuron weights

[0265] Finally, the neuron weights of the deep structured semantic network to be debugged are adjusted according to the training errors of each vector member in each feature subgraph. This process aims to gradually optimize the network parameter settings to improve its performance until the preset debugging requirements are met.

[0266] 4. For example, a behavior tracking report on pigs’ eating behavior has been annotated with joint knowledge to obtain detailed descriptions of eating time, eating amount, and related environmental characteristics. In order to ensure that this annotated text can maintain its integrity and availability in the face of potential risks, the artificial intelligence processing system first uses the debugged target deep structured semantic network to extract structured disaster recovery vectors from the text. During the extraction process, the network deeply analyzes key entities (such as pig names, eating time, etc.), relationships (such as the relationship between eating amount and time), and events (such as the occurrence of eating behavior) in the text, and represents them in the form of structured disaster recovery vectors for storage and backup.

[0267] Later, during the disaster recovery processing phase, the system will perform redundant storage and error detection on these disaster recovery vectors to ensure data reliability. At the same time, when the system detects data loss or damage, it can quickly use the backup data to restore the complete annotation text information for subsequent analysis and decision-making.

[0268] In this way, by introducing a structured disaster recovery processing mechanism and combining the application of the target deep structured semantic network, the artificial intelligence processing system realizes the organic combination of data annotation and disaster recovery protection in the processing of pig farming behavior tracking reports. This technical solution not only improves the accuracy and reliability of the annotated text, but also enhances its durability and availability in the face of potential risks. In detail, firstly, the technical solution realizes an in-depth understanding and structured representation of the annotated text content through the application of deep semantic networks, thereby improving the accuracy and efficiency of data annotation; secondly, the introduction of a structured disaster recovery processing mechanism effectively reduces the risk of data loss or damage, ensuring that the annotated text can maintain its integrity and availability in the face of potential risks; finally, the technical solution also provides more stable and reliable data support for subsequent joint annotation and in-depth analysis, which helps to promote the intelligent and refined development of farming management.

[0269] Further, Figure 2 The structural block diagram of the artificial intelligence processing system 300 is shown, including: a memory 310 for storing program instructions and data; a processor 320 for coupling with the memory 310 and executing the instructions in the memory 310 to implement the above method.

[0270] Furthermore, a computer storage medium is provided, comprising instructions, which implement the above method when executed on a processor.

[0271] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for tracking the behavior of farmed animals based on artificial intelligence, characterized in that: Applied to an artificial intelligence processing system, the method comprises: Acquire a plurality of behavior tracking report texts to be jointly labeled, and behavior semantic associated variables of the behavior tracking report texts, wherein the plurality of behavior tracking report texts to be jointly labeled have behavior tracking description sentences with different behavior tracking cycle labels, and the behavior tracking report texts are report texts for selected farmed animals; Performing label adjustment on the behavior tracking period data of the behavior tracking description sentences in each of the behavior tracking report texts to obtain a plurality of behavior tracking report adjustment texts matching the same behavior tracking period label; Determine the contextual connection attention dimension of the description sentence connection processing according to the behavior semantic association variable and the description sentence positioning vector of the behavior tracking description sentence in the behavior tracking report adjustment text; Determine, from each of the behavior tracking report adjustment texts, an associated behavior tracking description sentence of the behavior tracking report adjustment text in the contextual connection attention dimension; For each contextual connection attention dimension, joint knowledge annotation is performed on the associated behavior tracking description sentences corresponding to the contextual connection attention dimension to obtain the behavior tracking report annotation text corresponding to each contextual connection attention dimension.

2. The method for tracking animal behavior based on artificial intelligence according to claim 1, characterized in that: The behavior semantics associated variables include a first behavior semantics associated variable; Determining the contextual connection attention dimension of the description sentence connection processing according to the behavior semantic association variable and the description sentence positioning vector of the behavior tracking description sentence in the behavior tracking report adjustment text includes: Determining a first initial attention dimension describing sentence connection processing based on the first behavior semantic association variable; Determine a first description sentence positioning vector corresponding to the behavior tracking description sentence in each behavior tracking report adjustment text in the first initial attention dimension; Based on the first description sentence positioning vector, determine the context connection attention dimension of the description sentence connection processing.

3. The method for tracking animal behavior based on artificial intelligence according to claim 2, characterized in that: The behavior semantics associated variables also include a second behavior semantics associated variable; Determining the contextual connection attention dimension of the description sentence connection processing according to the first description sentence location vector includes: When the first description sentence location vector meets the set description sentence location requirement, determining a second initial attention dimension for description sentence connection processing based on the second behavior semantic association variable; Determine a second description sentence positioning vector corresponding to a target attention cycle of a behavior tracking description sentence in each behavior tracking report adjustment text, wherein the target attention cycle is determined according to the first initial attention dimension and the second initial attention dimension; Based on the second description sentence positioning vector, the contextual connection attention dimension of the description sentence connection processing is determined from the target attention cycle.

4. The method for tracking animal behavior based on artificial intelligence according to claim 3, characterized in that: Determining a contextual connection attention dimension for description sentence connection processing from the target attention cycle according to the second description sentence location vector includes: When the second description sentence location vector represents that any of the behavior tracking report adjustment texts has a behavior tracking description sentence to be associated within the target attention cycle, acquiring behavior tracking cycle data of the behavior tracking description sentence to be associated; Based on the behavior tracking cycle data of the behavior tracking description sentence to be associated, the contextual connection attention dimension of the description sentence connection processing is determined from the target attention cycle.

5. The method for tracking animal behavior based on artificial intelligence according to claim 1, characterized in that: The behavior tracking period data of the behavior tracking description sentences in each of the behavior tracking report texts are label adjusted to obtain a plurality of behavior tracking report adjustment texts matching the same behavior tracking period label, including: Determine a key behavior tracking description sentence from the behavior tracking description sentences in each of the behavior tracking report texts, and obtain key behavior tracking cycle data corresponding to the key behavior tracking description sentence; Based on the key behavior tracking cycle data, the behavior tracking cycle data of the behavior tracking description sentences in each of the behavior tracking report texts are updated to obtain the cycle update data of the behavior tracking description sentences in each of the behavior tracking report texts; Performing semantic alignment processing on the periodic update data of the behavior tracking description sentences in each of the behavior tracking report texts to obtain a plurality of behavior tracking report adjustment texts matching the same behavior tracking period label; The period label of the behavior tracking period data includes a first period label and a second period label, and the key behavior tracking period data includes a first key tracking period of the key behavior tracking description sentence of each behavior tracking report text under the first period label, and a second key tracking period of the key behavior tracking description sentence under the second period label; Based on the key behavior tracking cycle data, the behavior tracking cycle data of the behavior tracking description sentences in each behavior tracking report text is updated, including: Acquire the first behavior tracking cycle data of the behavior tracking description sentence under the first cycle label in each of the behavior tracking report texts; determining a tracking period error between the first behavior tracking period data and the first key tracking period; Based on the tracking cycle error and the second key tracking cycle, the second behavior tracking cycle data of the behavior tracking description sentences in each of the behavior tracking report texts under the second cycle label is determined to perform a behavior tracking cycle update on the behavior tracking cycle data of the behavior tracking description sentences in the behavior tracking report texts.

6. The method for tracking animal behavior based on artificial intelligence according to claim 1, characterized in that: Determining, from each of the behavior tracking report adjustment texts, an associated behavior tracking description sentence of the behavior tracking report adjustment text in the contextual connection attention dimension, including: Determining, from each of the behavior tracking report adjustment texts, a past behavior tracking description sentence of the behavior tracking report adjustment text before the contextual connection attention dimension; Based on the behavior tracking cycle data of the past behavior tracking description sentences, determining from the past behavior tracking description sentences the associated behavior tracking description sentences of the behavior tracking report adjustment text in the contextual connection attention dimension; Based on the behavior tracking cycle data of the past behavior tracking description sentences, determining the associated behavior tracking description sentences of the behavior tracking report adjustment text in the context connection attention dimension from the past behavior tracking description sentences, including: Determine a set of behavior tracking description sentences corresponding to each of the behavior tracking report adjustment texts, wherein the set of behavior tracking description sentences is used to cache past behavior tracking description sentences of the behavior tracking report adjustment text before the context connection attention dimension; Optimizing the set of behavior tracking description statements based on the contextual connection attention dimension and the behavior tracking cycle data of the past behavior tracking description statements; From the optimized behavior tracking description sentence set, determine the associated behavior tracking description sentence of the behavior tracking report adjustment text in the context connection attention dimension.

7. The method for tracking animal behavior based on artificial intelligence according to claim 1, characterized in that: Performing joint knowledge annotation on the associated behavior tracking description sentences corresponding to the contextual connection attention dimension, including: Determine a text paragraph to be annotated of an associated behavior tracking description sentence corresponding to the contextual connection attention dimension; Decomposing the text paragraph to be annotated into at least one text report unit; A corresponding NLP knowledge annotation algorithm is configured for each of the text reporting units, and the associated behavior tracking description statements in the text reporting units are jointly annotated with knowledge through the NLP knowledge annotation algorithm.

8. The method for tracking farmed animal behavior based on artificial intelligence according to claim 1, characterized in that: Obtain several behavior tracking report texts to be jointly annotated, including: Acquire a plurality of initial behavior tracking report texts and tracking environment features of the initial behavior tracking report texts; Based on the tracking environment characteristics, semantically annotating the initial behavior tracking report text to obtain a plurality of behavior tracking report texts to be jointly annotated; The tracking environment feature includes a target text semantic fine-grained vector of the initial behavior tracking report text; Based on the tracking environment characteristics, the initial behavior tracking report text is semantically annotated to obtain a plurality of behavior tracking report texts to be jointly annotated, including: Obtaining an initial text semantic fine-grained vector of the initial behavior tracking report text; Pairing the initial text semantic fine-grained vector with the target text semantic fine-grained vector; According to the pairing result, the initial behavior tracking report text is annotated based on semantic fine-grainedness to obtain a plurality of behavior tracking report texts to be jointly annotated.

9. An artificial intelligence processing system, characterized in that: include: Memory, used to store program instructions and data; A processor, coupled to a memory, and configured to execute instructions in the memory to implement the method according to any one of claims 1 to 8.

10. A computer storage medium, characterized in that: Contains instructions, which, when executed on a processor, implement the method according to any one of claims 1 to 8.

Images