Beef cattle meat yield prediction method and system based on dynamic physiological and ai three-dimensional body size fusion

Abstract

This invention relates to the field of beef cattle breeding technology, specifically providing a method and system for predicting beef cattle meat yield based on the fusion of dynamic physiology and AI three-dimensional body size. The method includes the following steps: acquiring real-time three-dimensional body size data of the target beef cattle, inputting it into a pre-trained AI weight estimation model to obtain real-time weight data and performing timestamp association; acquiring dynamic physiological data and integrating it with the time-series correlated body size and weight data into a multi-source time-series dataset; completing the spatiotemporal alignment of the multi-source data through a time-series attention fusion model, and mining the time-series causal relationships of weight, physiological, and body size changes through an attention mechanism to generate a fused correlation feature set; inputting this feature set into a meat yield prediction model, and outputting a result containing the predicted meat yield value and the causal relationship basis. This method breaks through the limitations of traditional static prediction, achieves dynamic high-precision prediction, and the causal basis can support breeding decisions, improving the level of precision breeding and breeding efficiency.

Description

Technical Field

[0001] This invention relates to the field of beef cattle breeding technology, specifically to a method and system for predicting beef cattle meat yield based on the fusion of dynamic physiology and AI three-dimensional body size. Background Technology

[0002] With the continuous improvement of the level of intelligent animal husbandry and the increasing demand for precision farming, the prediction technology of beef cattle meat yield has become increasingly important as a crucial basis for evaluating farming efficiency and making feeding management decisions. However, the current mainstream methods for predicting beef cattle meat yield still have many limitations: traditional methods mainly rely on static body size measurement data for linear regression prediction, which cannot reflect the dynamic changes in beef cattle during the growth process; existing technologies do not adequately explore the correlation between physiological indicators and body size changes, resulting in a large deviation between the prediction results and the actual meat yield; in addition, most prediction models lack a real-time feedback mechanism, making it difficult to respond promptly to dynamic farming scenarios such as feed adjustments and health abnormalities.

[0003] To address these issues, several improved prediction methods have been developed, such as weight prediction models using multi-source data fusion or trend extrapolation algorithms based on historical growth curves. However, these improvements still suffer from drawbacks such as poor data timeliness, weak feature correlation, and decision lag. Particularly for large-scale farms that require real-time adjustments to feeding strategies, existing technologies cannot simultaneously provide growth status diagnoses and optimization suggestions, severely hindering the improvement of farming efficiency.

[0004] Therefore, there is an urgent need to develop a novel method for predicting beef cattle meat yield that can achieve high-precision dynamic prediction and provide real-time breeding decision support. This method should overcome the limitations of static body size measurement by deeply integrating AI weight estimation data, dynamic physiological indicators, and three-dimensional body size change characteristics to establish an intelligent prediction model with time-series correlation analysis capabilities, thereby providing comprehensive data support for precision farming. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention provides a method and system for predicting beef yield of cattle based on the fusion of dynamic physiology and AI three-dimensional body size, in order to solve the problems in existing technologies.

[0006] One embodiment of the present invention provides a method for predicting beef yield of cattle based on the fusion of dynamic physiology and AI three-dimensional body size, comprising the following steps:

[0007] S10. Obtain the real-time three-dimensional body size data of the target beef cattle and input it into the pre-trained AI weight estimation model to obtain the real-time weight data, and associate the real-time three-dimensional body size data with the real-time weight data with timestamps.

[0008] S20. Acquire the dynamic physiological data of the target beef cattle, and integrate the dynamic physiological data, the three-dimensional body size data that has been time-stamped and associated, and the real-time weight data at preset time intervals to obtain a multi-source time-series dataset.

[0009] S30. Based on the pre-built temporal attention fusion model, feature alignment of temporal dimension and data dimension is performed on multi-source temporal datasets. The temporal attention mechanism is used to mine the temporal causal relationship between real-time weight data change trend, dynamic physiological data fluctuation state and three-dimensional body size data change state, and generate a fused correlation feature set.

[0010] S40. Input the fused correlation feature set into the pre-trained meat yield prediction model, and output the meat yield prediction result of the target beef cattle, which includes the meat yield prediction value and the causal correlation basis.

[0011] This application also relates to a beef cattle meat yield prediction system based on the fusion of dynamic physiology and AI three-dimensional body size, including:

[0012] The weight calculation module is used to acquire real-time three-dimensional body size data of the target beef cattle and input it into a pre-trained AI weight estimation model to obtain real-time weight data, and to associate the real-time three-dimensional body size data with the real-time weight data with timestamps.

[0013] The data integration module is used to acquire the dynamic physiological data of the target beef cattle and to integrate the dynamic physiological data, the three-dimensional body size data that has been time-stamped and associated, and the real-time weight data at preset time intervals to obtain a multi-source time-series dataset.

[0014] The association mining module is used to perform feature alignment of the temporal dimension and data dimension on multi-source temporal datasets based on a pre-built temporal attention fusion model, and to mine the temporal causal relationship between real-time weight data change trend, dynamic physiological data fluctuation state and three-dimensional body size data change state through the temporal attention mechanism, and generate a fused association feature set.

[0015] The meat yield prediction module is used to input the fused correlation feature set into the pre-trained meat yield prediction model and output the meat yield prediction result of the target beef cattle, which includes the meat yield prediction value and the causal correlation basis.

[0016] This application also relates to a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for predicting beef yield based on dynamic physiology and AI three-dimensional body size fusion.

[0017] This application also relates to a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method for predicting beef yield based on dynamic physiology and AI three-dimensional body size fusion.

[0018] The method and system for predicting beef yield based on dynamic physiology and AI three-dimensional body size fusion provided in the above embodiments have the following beneficial effects:

[0019] This invention acquires real-time 3D body size data of beef cattle and combines it with an AI weight estimation model to accurately obtain real-time weight. Simultaneously, it integrates dynamic physiological data to construct a multi-source time-series dataset. Relying on a time-series attention fusion model, it achieves spatiotemporal alignment and deep feature fusion of multi-source data. It can also accurately mine the temporal causal relationships between weight change trends, physiological fluctuations, and body size changes. Finally, through a meat yield prediction model, it outputs prediction results that combine accurate meat yield predictions with causal relationship evidence. This invention overcomes the limitations of traditional beef cattle meat yield prediction, which relies on static body size measurements and lacks sufficient data correlation mining. It effectively improves the dynamism and accuracy of meat yield prediction, solving the problems of large biases and lack of interpretability in existing prediction methods. It provides intuitive causal support for farmers, assists in making precise farming decisions, and ultimately improves the level of refined management and overall farming efficiency in beef cattle farming. Attached Figure Description

[0020] Figure 1 A flowchart of a method for predicting beef yield based on dynamic physiology and AI three-dimensional body size fusion provided in an embodiment of the present invention;

[0021] Figure 2 A flowchart of a method for predicting beef yield based on dynamic physiology and AI three-dimensional body size fusion provided in an embodiment of the present invention;

[0022] Figure 3 This is a schematic block diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions in the embodiments of the present invention will now be clearly and completely described in conjunction with the accompanying drawings.

[0024] Reference Figure 1 One embodiment of the present invention provides a method for predicting beef yield of cattle based on the fusion of dynamic physiology and AI three-dimensional body size, including the following steps:

[0025] S10. Obtain the real-time three-dimensional body size data of the target beef cattle and input it into the pre-trained AI weight estimation model to obtain the real-time weight data, and associate the real-time three-dimensional body size data with the real-time weight data with timestamps.

[0026] S20. Acquire the dynamic physiological data of the target beef cattle, and integrate the dynamic physiological data, the three-dimensional body size data that has been time-stamped and associated, and the real-time weight data at preset time intervals to obtain a multi-source time-series dataset.

[0027] S30. Based on the pre-built temporal attention fusion model, feature alignment of temporal dimension and data dimension is performed on multi-source temporal datasets. The temporal attention mechanism is used to mine the temporal causal relationship between real-time weight data change trend, dynamic physiological data fluctuation state and three-dimensional body size data change state, and generate a fused correlation feature set.

[0028] S40. Input the fused correlation feature set into the pre-trained meat yield prediction model, and output the meat yield prediction result of the target beef cattle, which includes the meat yield prediction value and the causal correlation basis.

[0029] In this embodiment, as described in steps S10-S40 above, its core is to achieve accurate weight estimation of beef cattle using AI three-dimensional body size, time-series integration of multi-source dynamic data, deep fusion of causal relationships through time-series attention, and accurate prediction using a dedicated model. This overcomes the limitations of traditional meat yield prediction, which relies on static body size, lacks sufficient correlation mining, and has no causal support. It achieves dynamic and high-precision prediction of beef cattle meat yield while outputting traceable causal relationship evidence, providing scientific data support for precision farming, and adapting to the dynamic monitoring and decision-making needs of large-scale beef cattle farming. Specifically, as follows:

[0030] Step S10 involves the accurate acquisition and data association of beef cattle weight. The core of this step is to achieve precise weight calculation through 3D body measurements and AI-based weight estimation, while simultaneously establishing a temporal correlation between the data to lay the foundation for subsequent time-series fusion. Specifically, firstly, real-time 3D body measurements of the target beef cattle are collected using a 3D scanning device (including 3D body measurements such as chest circumference, body length, and body width, as well as composite morphological features including body fullness). This accurately captures key morphological features of the beef cattle's body, replacing traditional manual static measurements and improving the accuracy and timeliness of the body measurements data. Secondly, the real-time 3D body measurements data are input into a pre-trained AI weight estimation model. The model outputs accurate real-time weight data based on the mapping logic between body measurements and weight, avoiding the cumbersome and unreal-time monitoring problems of traditional weighing operations. Finally, the real-time 3D body measurements data are timestamped and associated with the corresponding real-time weight data to ensure a one-to-one correspondence between body measurements and weight data collected within the same time period. This avoids data misalignment that could lead to subsequent analysis biases, providing a unified time reference for multi-source data integration and laying the data foundation for the entire prediction process.

[0031] Step S20 is the multi-source dynamic data time series integration step. The core is to collect dynamic physiological dimension core data (focusing on the collection of exercise and metabolism-related indicators) to achieve time series normalization of multi-source data and provide a complete dataset for feature fusion. Specifically, dynamic physiological data of the target beef cattle is first obtained through wearable monitoring devices (such as smart ear collars for beef cattle) or livestock monitoring terminals. This dynamic physiological data includes, but is not limited to, step count data (a key indicator), heart rate, body temperature, and rumen motility frequency. Step count data directly reflects the activity level of the beef cattle: tethered beef cattle, due to limited space, typically have a lower daily step count (e.g., 500-1000 steps), resulting in low energy consumption and easy accumulation of body fat (high percentage of fat), thus affecting meat yield and lean meat percentage. Free-range or non-tethered beef cattle, with ample space, have a higher daily step count (e.g., 1500-3000 steps), balanced energy consumption, lower body fat percentage, and higher lean meat percentage, resulting in better meat yield and lean meat percentage. Rumen motility frequency directly reflects the digestive and metabolic efficiency of the beef cattle and is a core indicator of nutrient absorption. Normal motility ensures the full decomposition and absorption of feed nutrients, providing sufficient nutrients for body size development, weight gain, and lean meat synthesis. The system comprehensively reflects the growth, health, exercise, and digestive metabolism of beef cattle using four data categories: steps, heart rate, body temperature, and rumen motility frequency. This provides triple support from physiological, exercise, and metabolic perspectives for subsequent correlation analysis. Then, at preset time intervals (e.g., every 12 hours or 24 hours), the dynamic physiological data is integrated with the already timestamped 3D body size data and real-time weight data. Simultaneously, corresponding data at the same timestamp within a preset time period (e.g., the previous three days) are supplemented. The changes in multi-source data are traced along a timeline to eliminate temporal heterogeneity caused by differences in data collection frequency. Finally, a multi-source time-series dataset containing three core dimensions—body size, weight, and physiology—is formed. This ensures that the data covers key indicators of beef cattle growth and possesses temporal continuity, avoiding biases caused by the randomness of single-time-point data. This provides complete data support for subsequent exploration of temporal causal relationships and avoids prediction biases caused by single-dimensional data.

[0032] Step S30 involves deep fusion of multi-source data and causal correlation mining. The core is to achieve feature alignment and causal mining through a temporal attention fusion model, enabling predictions to be both relevant and interpretable. Specifically, the multi-source time-series dataset is first input into a pre-built temporal attention fusion model. The model first completes dual feature alignment in both the temporal and data dimensions, ensuring consistency in the spatiotemporal dimensions of data from different dimensions and collection frequencies, eliminating differences in units and temporal deviations. Then, through the built-in temporal attention mechanism, it focuses on key time nodes and core characteristic indicators in the growth process of beef cattle, mining the temporal causal correlation between the changing trends of real-time weight data, the fluctuations of dynamic physiological data, and the changes in three-dimensional body size data. This clarifies both the strength of the correlation between data and defines the causal logic. The causal logic includes positive and negative causality. Positive causality includes "daily step count stabilizes within a reasonable range (tethered 8...)". (00-1000 steps / 2000-2500 steps for free-range) → Balanced energy consumption → Controllable body fat percentage → Synergistic improvement in body fullness and lean meat ratio → Steady weight gain → Simultaneous optimization of meat yield; Reverse causal relationships include "too low step count (<500 steps / day) → body fat accumulation → weight gain but decreased lean meat ratio → potential decrease in meat yield", or "abnormal physiological indicators (such as elevated body temperature) → sudden decrease in step count → slowed weight gain → delayed body size development"; Finally, the aligned features, attention weights, and causal relationships are integrated to generate a fused correlation feature set. This feature set contains both quantitative data and feature importance and causal basis, providing high-value and interpretable core feature support for subsequent meat yield prediction.

[0033] Step S40 is the precise prediction and causal output stage for beef cattle meat yield. The core is to achieve high-precision prediction of meat yield while outputting causal correlation evidence, making the prediction results traceable and reliable. Specifically, the fused correlation feature set is input into the pre-trained meat yield prediction model. This model is constructed based on the growth patterns, exercise and metabolic characteristics of beef cattle, and the correlation logic of meat yield. It can accurately identify core indicators strongly correlated with meat yield in the fused feature set (such as step count, chest circumference, and body fullness). The model first calculates the predicted meat yield value of the target beef cattle based on the fused features, avoiding the limitations of traditional linear regression prediction and improving prediction accuracy. Then, it correlates the temporal causal relationships in the fused correlation feature set to extract the causal logic directly related to the predicted meat yield value (such as "daily step count of 90 in tethered state"). "0 steps, 5cm increase in body length, 8cm increase in chest circumference and normal metabolic indicators → 15kg increase in weight, body fat percentage controlled at 18% → 2% increase in meat yield," thus establishing a causal relationship. The final output includes a complete prediction result containing the predicted meat yield and the corresponding causal relationship, allowing farmers to understand the target beef cattle's meat yield and identify the core influencing factors (such as insufficient steps leading to a low meat yield). This provides precise decision-making basis for subsequent feeding adjustments (such as appropriately increasing activity space and adjusting feeding programs) and the selection of slaughter timing, truly achieving data-driven precision farming.

[0034] Reference Figure 2 In one embodiment, after step S40, the following step is further included:

[0035] S50. Monitor the dynamic physiological data of the target beef cattle in real time, and compare the dynamic physiological data with the preset physiological index threshold range in real time.

[0036] S60. When dynamic physiological data is detected to exceed the threshold range of physiological indicators, it is determined to be an abnormal physiological state, and dynamic adjustment of the meat yield prediction result is performed. The dynamic adjustment includes:

[0037] Trigger the adaptive correction of the parameters of the AI ​​weight estimation model to adapt to the weight estimation logic under abnormal physiological conditions;

[0038] Based on the real-time weight data output by the corrected AI weight estimation model, the temporal causal association logic of the fused associated feature set is corrected.

[0039] The corrected fusion-related feature set is input into the meat yield prediction model, and the meat yield prediction result, which includes the predicted meat yield value and the causal relationship basis, is dynamically adjusted.

[0040] S70. Based on the dynamically adjusted meat yield prediction results, retrieve the preset physiological abnormality-breeding strategy association rule library, and output a dynamic breeding optimization strategy that matches the physiological state of the target beef cattle and the dynamically adjusted meat yield prediction value.

[0041] In this embodiment, as described in steps S50-S70 above, the core is to construct a closed-loop precision farming system of "prediction-monitoring-adjustment-optimization" through dynamic physiological monitoring, abnormal linkage adjustment, and intelligent strategy output. This system overcomes the limitations of traditional static meat yield prediction, which cannot adapt to the physiological fluctuations of beef cattle, making meat yield prediction more closely aligned with the real-time growth status of beef cattle. Simultaneously, it enables the precise implementation of prediction results into farming strategies, further improving the adaptability and practicality of precision farming. Specifically, as follows:

[0042] Step S50 involves real-time monitoring and threshold comparison of the physiological state of beef cattle. Its core function is to capture dynamic physiological changes in beef cattle in real time, promptly identify abnormal states, and provide early warnings for subsequent prediction, adjustment, and strategy optimization. Specifically, based on the same wearable monitoring device (such as a smart ear collar for beef cattle) or livestock monitoring terminal as in Step S20, the dynamic physiological data of the target beef cattle is continuously monitored in real time. This eliminates the need for repeated data collection equipment, ensuring the continuity and consistency of monitoring. The dynamic physiological data remains consistent with that of Step S20, including step count data (a key indicator), heart rate, body temperature, and rumen motility frequency. Simultaneously, a preset threshold range for physiological indicators matching the growth stage of the beef cattle is established. This threshold range is based on the characteristics of the beef cattle breed, the nutritional needs of the growth stage, and industry healthy farming standards. It is necessary to differentiate between growth stages and farming methods to avoid misjudgments caused by a single threshold. The real-time monitored dynamic physiological data is compared with the corresponding preset threshold range in real time to determine whether the physiological state is stable, ensuring that abnormal states (such as metabolic abnormalities and health risks) are captured immediately, reserving a response window for subsequent adjustments.

[0043] Step S60 is the dynamic adjustment step for physiological abnormality judgment and meat yield prediction results. Its core is to achieve adaptive correction of the prediction model and results under abnormal conditions, ensuring the real-time performance and accuracy of meat yield prediction and avoiding prediction deviations caused by abnormal physiological states. Specifically, when real-time monitoring shows that dynamic physiological data exceeds the physiological indicator threshold range, it is directly judged as an abnormal physiological state of the target beef cattle, without the need for additional judgment logic, thus improving response efficiency. Subsequently, the entire chain of dynamic adjustment of meat yield prediction results is initiated. The adjustment process links the preceding AI weight estimation model, integrates the associated feature set, and the meat yield prediction model to achieve full-process adaptation and correction. The specific adjustments include three aspects:

[0044] 1) Trigger the adaptive correction of parameters of the AI ​​weight estimation model to adapt to the weight estimation logic under abnormal physiological conditions. The core is to adjust the mapping weight between body size features and weight according to the abnormal physiological type (such as abnormal digestion and metabolism, abnormal exercise state) to avoid the interference of abnormal body size data (such as high body fullness caused by puffiness) on weight estimation and ensure the accuracy of weight data.

[0045] 2) Based on the real-time weight data output by the corrected AI weight estimation model, the temporal causal relationship logic of the fused correlation feature set is corrected. That is, the correlation relationship of "weight-physiological data-body size data" under abnormal conditions is updated synchronously, replacing the original causal logic under normal conditions, so as to ensure that the correlation feature set can truly reflect the current growth status of beef cattle and avoid using normal logic to analyze abnormal data.

[0046] 3) Input the corrected fusion correlation feature set into the meat yield prediction model, dynamically adjust the meat yield prediction result containing the predicted meat yield value and the causal correlation basis, so that the predicted value matches the abnormal physiological state, and at the same time the causal correlation basis is updated to the correlation logic under the abnormal state, ensuring that the interpretability and accuracy of the prediction result are consistent.

[0047] Step S70 is the dynamic breeding optimization strategy output stage. Its core is to establish a seamless implementation link between "anomaly prediction" and "breeding strategy," ensuring that the meat yield prediction results truly guide precision breeding and realize the core value of "prediction serving production." Specifically, a pre-built physiological anomaly-breeding strategy association rule library is constructed. This rule library stores exclusive breeding optimization strategies corresponding to different growth stages, different types of physiological anomalies, and different meat yield prediction adjustment ranges. The strategies cover four core dimensions: feeding adjustment, activity adjustment, health intervention, and prediction tracking, achieving precise matching of "anomaly type - anomaly severity - meat yield adjustment range - breeding strategy." It should be noted that those skilled in the art can flexibly construct this association rule library based on large-scale beef cattle breeding big data, industry standards, and breeding scenario requirements; no limitations are imposed here. The construction of the association rule library involves various aspects of the breeding process. By processing various data such as abnormal thresholds and strategy adaptation parameters, those skilled in the art can make appropriate adjustments based on the breed, farming model, and growth stage of beef cattle to adapt to different application scenarios. This is clear to those skilled in the art. After obtaining the dynamically adjusted meat yield prediction result, the system automatically retrieves the rule base, accurately matches the current physiological state (abnormality type and degree) of the target beef cattle with the dynamically adjusted meat yield prediction value, and outputs a targeted dynamic farming optimization strategy. The core objective of the strategy is to correct abnormal physiological states, ensure the meat yield potential of beef cattle, and avoid stress reactions caused by excessive adjustments, thereby achieving a balance between farming efficiency and beef cattle health.

[0048] In one embodiment, step S10, training the pre-trained AI weight estimation model, includes the following steps:

[0049] S101. Collect historical three-dimensional body size data and corresponding measured weight data of multiple batches of sample beef cattle, and associate the historical three-dimensional body size data with the corresponding measured weight data by timestamp to obtain the basic training dataset.

[0050] S102. Perform data preprocessing on the basic training dataset to obtain a standardized training dataset;

[0051] S103. Construct an initial AI weight estimation network model, wherein the initial AI weight estimation network model has a built-in mapping logic between three-dimensional body size features and body weight;

[0052] S104. Input the standardized training dataset into the initial AI weight estimation network model, perform iterative training of the model according to the preset training batch and learning rate, and output the weight prediction error in real time during the training process.

[0053] S105. When the weight prediction error is lower than the preset error threshold, stop the model training and obtain the trained AI weight estimation model; if the preset error threshold is not reached, adjust the model network parameters and repeat the model iterative training until the training termination condition is met.

[0054] In this embodiment, as described in steps S101-S105 above, the core is to train a high-precision AI weight estimation model suitable for beef cattle weight estimation scenarios through a standardized process of "data acquisition-preprocessing-model building-iterative training-termination upon reaching the target". This ensures that the model has stable three-dimensional body size features and weight mapping capabilities, providing core support for the real-time accurate weight estimation in the subsequent step S10. This overcomes the limitations of traditional weight estimation models, such as poor generalization ability and high prediction error, and improves the basic data accuracy of the overall meat yield prediction system, as detailed below:

[0055] Step S101 is the basic training dataset collection and association step, the core of which is to obtain high-quality, strongly correlated training data to provide a reliable data foundation for model training. Specifically, it is necessary to cover sample beef cattle of multiple breeds, growth stages, and farming models (to ensure the model's generalization ability), and collect historical three-dimensional body size data and corresponding measured weight data for each sample beef cattle. Among them, the collection dimensions of historical three-dimensional body size data are consistent with those in step S10, including core features such as chest circumference, body length, body width, and body fullness, and are collected at fixed time intervals (e.g., once a day) using a three-dimensional scanning device. The measured weight data for the corresponding time period is collected synchronously using high-precision livestock weighing equipment (such as weighbridges and dynamic weighing bars) to ensure the authenticity and accuracy of the weight data. Subsequently, the historical three-dimensional body size data of the same beef cattle and the measured weight data for the corresponding time period are timestamped to ensure the temporal correspondence between "body size" and "weight" and avoid data misalignment. Finally, all the associated data are integrated to form a basic training dataset, ensuring the diversity and representativeness of the dataset and laying the foundation for the model to learn the generalized body size-weight mapping law.

[0056] Step S102 is the training data preprocessing stage, the core of which is to optimize data quality, eliminate data noise and anomalies, and provide standardized input for model training. Specifically, data preprocessing includes three core operations: First, missing value handling, for a small number of missing data in the basic training dataset (such as missing body size data in certain time periods), the mean of data of the same species and growth stage is used to fill in the missing data or linear interpolation to avoid the missing data affecting the training effect; Second, outlier removal, outlier data (such as abnormally large / small body size data due to equipment failure, or sudden changes in measured weight data) is identified and removed through the 3σ principle or box plot method to ensure the purity of the dataset; Third, data standardization, the three-dimensional body size data and measured weight data are uniformly mapped to the [0,1] interval (such as using Min-Max normalization) to eliminate the training bias caused by the difference in different indicator units (cm vs kg), so that the model focuses on the correlation between features rather than the absolute size of indicator values, thereby improving training efficiency and model accuracy.

[0057] Step S103 is the initial AI weight estimation network model construction stage. The core is to build a network structure that adapts to the "three-dimensional body size-weight" mapping relationship, ensuring that the model has feature extraction and nonlinear fitting capabilities. Specifically, the initial AI weight estimation network model adopts a deep learning architecture, with built-in mapping logic between three-dimensional body size features and weight. The model structure includes an input layer, a feature extraction layer, and a mapping output layer: the input layer dimension is consistent with the three-dimensional body size feature dimension (e.g., if there are four features such as chest circumference, body length, body width, and body fullness, the input layer is 4-dimensional); the feature extraction layer uses a convolutional neural network (CNN) or a fully connected layer to automatically extract high-order correlation features of body size data (e.g., the synergistic effect of chest circumference and body width); the mapping output layer is a single neuron that outputs the predicted weight value; the design of the model structure needs to adapt to the dimension and correlation complexity of the body size features of beef cattle to ensure that the nonlinear mapping relationship between body size parameters and weight can be fully explored.

[0058] Step S104 is the model iterative training stage, the core of which is to optimize model parameters through data iteration and minimize weight prediction error. Specifically, the standardized training dataset is divided into a training set and a validation set in a 7:3 ratio (the training set is used for parameter updates, and the validation set is used for real-time evaluation of generalization ability); the training set is input into the initial AI weight estimation network model for iterative training according to a preset training batch size (e.g., batch size 32) and learning rate (e.g., initial learning rate 0.001); during training, mean squared error (MSE) is used as the loss function to calculate the error between the model's output weight prediction value and the measured weight data (i.e., weight prediction error) in real time, and the model network parameters (e.g., hidden layer weights, bias terms) are updated through the backpropagation algorithm; at the same time, the validation set is used to monitor whether the model is overfitting. If the validation set error continues to rise, the regularization parameters are adjusted to ensure that the model has both accuracy and generalization ability.

[0059] Step S105 is the model training termination and confirmation stage, the core of which is to ensure that the trained AI weight estimation model meets the preset accuracy requirements and has practical application value. Specifically, a preset weight prediction error threshold (e.g., MSE < 0.5) is set. The model is trained to a weight prediction absolute error of <0.7kg. The weight prediction error during training is compared with the preset threshold in real time. When the weight prediction error is lower than the preset error threshold, it means that the model accuracy meets the standard. The model training is stopped and the currently trained AI weight estimation model is output. If the preset error threshold is not reached after training to the preset maximum number of iterations (e.g., 100 rounds), the model network parameters are adjusted (e.g., increasing the number of hidden layer neurons, reducing the learning rate to 0.0001, and increasing the regularization strength). The model iteration training in step S104 is repeated until the weight prediction error meets the standard or the training termination condition is met (e.g., maximum number of iterations, error decrease rate <0.0001). The trained AI weight estimation model is obtained. This model can accurately adapt to the weight estimation needs of beef cattle. The real-time weight data can be directly called in the subsequent step S10.

[0060] It should be noted that the model network structure, training batches, learning rate, error threshold, and other parameters involved in this embodiment can be flexibly adjusted by those skilled in the art according to the sample size, data accuracy, and the needs of the aquaculture scenario. No limitation is made here, and all of them can achieve the training effect of a high-precision AI weight estimation model.

[0061] In one embodiment, step S10, which involves acquiring real-time three-dimensional body size data of the target beef cattle and inputting it into a pre-trained AI weight estimation model to obtain real-time weight data, specifically includes the following steps:

[0062] S11. Perform a three-dimensional scan on the target beef cattle, collect the original three-dimensional body size point cloud data, and remove environmental noise points in the point cloud data to obtain the original three-dimensional body size data after noise removal.

[0063] S12. Extract features from the denoised original three-dimensional body size data to obtain three-dimensional body size features including chest circumference, body length and body width, and composite morphological features including body fullness. Integrate the three-dimensional body size features and composite morphological features to form real-time three-dimensional body size data.

[0064] S13. Collect real-time physiological data of the target beef cattle, determine the physiological state based on the preset physiological index threshold range, and generate corresponding real-time physiological state labels.

[0065] S14. Input the real-time three-dimensional body size data and real-time physiological state labels into the pre-trained AI weight estimation model. The AI ​​weight estimation model performs the following logical calculations and outputs real-time weight data:

[0066] 1) If the real-time physiological state label is normal physiological state, the first mapping logic based on the physiological morphology rules of beef cattle is called, and the three-dimensional body size features and composite morphological features are substituted into the preset quantitative correlation formula to calculate the initial weight value of the target beef cattle under normal physiological state.

[0067] 2) If the real-time physiological state label is an abnormal physiological state, the second mapping logic based on the abnormal physiological-weight association rule is called, and the corresponding weight growth rate decay coefficient is matched according to the abnormal physiological state. The weight growth rate decay coefficient is substituted into the preset quantitative correction formula to adapt and correct the initial weight value, and the weight correction value of the target beef cattle under the abnormal physiological state is calculated.

[0068] 3) Output the real-time weight data of the target beef cattle that matches the current time period. The real-time weight data is the initial weight value under normal physiological conditions or the weight correction value under abnormal physiological conditions.

[0069] In this embodiment, as described in steps S11-S14 above, the core is to achieve high-precision output of real-time weight data of the target beef cattle through a refined process of "3D scanning denoising - precise feature extraction - physiological state adaptation - dual logic weight calculation". This solves the problems of traditional weight estimation ignoring differences in physiological state and data noise affecting accuracy, and further consolidates the basic data quality for subsequent multi-source data integration and meat yield prediction, as detailed below:

[0070] Step S11 is the 3D body size point cloud data acquisition and denoising stage. The core is to obtain pure raw body size data and avoid feature extraction deviations caused by environmental interference. Specifically, a 3D scanning device (such as a smart 3D scanning channel for livestock farms) consistent with step S101 is used to perform a full-range 3D scan of the target beef cattle, acquiring raw 3D body size point cloud data containing the complete morphology of the beef cattle's body. Since there may be interference from fences, feed, ground debris, etc. in the scanning environment, environmental noise points need to be removed using point cloud denoising algorithms. Denoising methods include statistical filtering and radius filtering. Statistical filtering removes isolated noise points exceeding 3σ by calculating the mean and standard deviation of the distance between each point and its neighboring points. Radius filtering removes noise points with insufficient neighboring points by setting the neighborhood radius and minimum number of points, ensuring that the denoised data only contains the morphological information of the beef cattle's body, providing a high-purity data foundation for subsequent feature extraction.

[0071] Step S12 is the three-dimensional body size feature extraction and integration stage. The core is to extract key body size features from the denoised data to form standardized real-time three-dimensional body size data. Specifically, based on computer vision and morphological analysis methods, feature extraction is performed on the denoised three-dimensional body size raw data: a three-dimensional model of the beef cattle body is constructed using a point cloud fitting algorithm, and basic three-dimensional body size features such as chest circumference (circumference at the widest part of the body), body length (straight-line distance from shoulder to rump), and body width (horizontal distance between the two acromions) are measured; body fullness is calculated by the ratio of body volume to standard body size volume, forming composite morphological features. The formula for calculating body fullness is: Body Fullness = Actual Body Volume of the Target Beef Cattle ÷ Standard Body Size Volume of the Same Breed, Growth Stage, and Weight (Note: Standard body size volume is based on scale). The statistical settings for healthy beef cattle samples are as follows: the actual body volume is calculated from point cloud data using a 3D convex hull algorithm or voxelization method, with a ratio range of 0.7~1.1. A ratio of 0.85~0.95 corresponds to a compact body with a high lean meat percentage; 0.7~0.85 corresponds to a lean body with a low muscle percentage and insufficient total lean meat; and 0.95~1.1 corresponds to a plump body with a high fat percentage and a soft body. The extracted basic 3D body size features and composite morphological features are integrated according to a preset format to form real-time 3D body size data, ensuring that the feature dimensions are consistent with the requirements of steps S10 and S30, achieving seamless data link connection.

[0072] Step S13 involves real-time physiological data acquisition and physiological state tag generation. The core of this step is to correlate the physiological state of the beef cattle, providing a basis for subsequent weight calculation. Specifically, through wearable monitoring devices (such as smart ear collars for beef cattle) used in step S20, real-time physiological data of the target beef cattle is collected synchronously. Data types include step count, heart rate, body temperature, and rumen motility frequency. The physiological indicator threshold ranges preset in step S50 (adapted to breed, growth stage, and farming mode) are then called. The real-time physiological data is compared with these threshold ranges to determine the current physiological state (normal / abnormal), and corresponding real-time physiological state tags (such as "normal physiological state," "abnormal digestion and metabolism," "abnormal body temperature," etc.) are generated. This achieves the linkage between physiological state and body size data, providing a crucial basis for dual-logic weight calculation.

[0073] Step S14 is the dual-logic weight calculation and real-time output stage. Its core is adapting different mapping logics based on physiological state labels to ensure the accuracy of weight data under abnormal physiological states. Specifically, the real-time 3D body size data integrated in step S12 and the real-time physiological state labels generated in step S13 are input into the pre-trained AI weight estimation model. The model calculates real-time weight data by calling the corresponding mapping logic based on the physiological state labels. The specific logic is as follows:

[0074] 1) If the real-time physiological state label is normal, then the first mapping logic based on the physiological morphology rules of beef cattle is invoked. This logic is trained based on the body size-weight correlation data of multiple batches of beef cattle in normal physiological states, and has a built-in preset quantitative correlation formula: initial weight = chest circumference × +body oblique length × +body width × +Body fullness × (in , , , The weight coefficients are optimized for model training (adapted and adjusted based on the correlation between body size and weight of beef cattle of different breeds and growth stages). The real-time three-dimensional body size features are substituted into the formula to calculate the initial weight value under normal physiological conditions.

[0075] 2) If the real-time physiological status label is an abnormal physiological status, the second mapping logic based on the abnormal physiological-weight association rule is invoked. This logic presets the weight growth rate decay coefficient corresponding to different abnormal types (e.g., the decay coefficient for digestive and metabolic abnormalities is 0.997~0.992, and the decay coefficient for body temperature abnormalities is 0.995~0.990, which is derived from large-scale abnormal data statistics). The corresponding decay coefficient is matched according to the current abnormal type (e.g., digestive and metabolic abnormalities) and substituted into the quantitative correction formula: weight correction value = initial weight value × weight growth rate decay coefficient × abnormal duration correction factor (the factor is 1.0 for abnormal duration ≤24h, and 0.998 for 24~48h). The interference of abnormal physiological status on weight estimation is eliminated through dual-factor correction, ensuring that the weight data matches the actual growth status of beef cattle.

[0076] 3) Output real-time weight data that matches the current time period: output the initial weight value under normal physiological conditions and output the weight correction value under abnormal physiological conditions to ensure that the weight data is adapted to the real-time growth status of beef cattle, and provide accurate basic data for subsequent data integration of S20 and causal mining of S30.

[0077] In one embodiment, step S30 specifically includes the following steps:

[0078] S31. Input the multi-source time series dataset into the pre-built time series attention fusion model, and perform spatiotemporal alignment on the multi-source time series dataset based on periodic calibration logic and livestock data standardization logic to generate a spatiotemporal alignment feature set.

[0079] S32. Based on the temporal attention mechanism, the attention weights of the temporal alignment feature set are initialized according to the feature priority of different growth stages of the target beef cattle; and the feature contribution of key nodes in the temporal data is calculated. Based on the feature contribution, the initial weights are dynamically corrected using the gradient descent algorithm to obtain the stage-adapted feature attention weight distribution.

[0080] S33. Based on the feature association mining logic, and combined with the feature attention weight distribution, the feature association relationship between the real-time weight data change trend, the dynamic physiological data fluctuation state, and the three-dimensional body size data change state is analyzed through the Pearson correlation coefficient matrix.

[0081] S34. Map the feature associations to a preset causal association graph, match the causal logic corresponding to the animal husbandry field, and obtain a time-series causal association set;

[0082] S35. Integrate the temporal alignment feature set, feature attention weight distribution, and temporal causal relationship set to generate a fused correlation feature set containing data quantification values, feature importance weights, and causal relationship basis.

[0083] In this embodiment, as described in steps S31-S35 above, the core is to construct a temporal causal association system adapted to the livestock scenario through a refined fusion process of "spatiotemporal alignment - weight adaptation - association mining - causal matching - feature integration". This solves the problems of traditional multi-source data fusion ignoring differences in growth stages and lacking clear causal logic, and provides a high-value fusion feature set of "data-weight-causality" in three parts for subsequent meat yield prediction, as detailed below:

[0084] Step S31 is the spatiotemporal alignment of multi-source data, the core of which is to eliminate the spatiotemporal heterogeneity of multi-source data and ensure the consistency of feature fusion. Specifically, the multi-source time-series dataset (including timestamps, 3D body size, weight, and physiological data) formed in step S20 is input into a pre-constructed temporal attention fusion model. The model has two built-in core logics: first, a periodic calibration logic, which calibrates the time granularity of data collected at different frequencies according to the characteristics of the beef cattle growth cycle (such as the physiological data fluctuation cycle in the early / middle / late fattening period) (e.g., unifying the body size data collected in 12 hours and the weight data collected in 24 hours to a daily cycle); second, a livestock data standardization logic, which uses a targeted standardization strategy to map data of different dimensions such as 3D body size (cm), weight (kg), and physiological indicators (steps / day, times / minute) to the same distribution interval (mean 0, standard deviation 1). Through the dual logic, the time synchronization and spatial dimension unification of multi-source data are achieved, generating a spatiotemporally aligned feature set, ensuring that data with different collection frequencies and different dimensions are comparable, and laying a unified data foundation for subsequent feature fusion and association mining.

[0085] Step S32 is the stage-adaptive attention weight optimization step, the core of which is to enable the temporal attention fusion model to focus on the key features of different growth stages and improve the fusion targeting. Specifically, the temporal attention fusion model, based on its built-in temporal attention mechanism, first presets the feature priorities of different growth stages of target beef cattle according to knowledge in the livestock field (e.g., mid-fat stage: body fullness > rumen motility frequency > daily steps > chest circumference; early fattening stage: body length > daily steps > heart rate). It then initializes the attention weights of the spatiotemporally aligned feature set generated in step S31 (e.g., initial weight of body fullness at 0.3 and rumen motility frequency at 0.25 in mid-fat stage). Next, it calculates the feature contribution of key nodes in the temporal data (e.g., sudden changes in weight gain and abnormal fluctuations in physiological indicators) using a sliding window method, and dynamically corrects the initial weights using a gradient descent algorithm—if the contribution of body fullness to weight gain reaches 0.85, its weight is increased to 0.35; if the contribution of heart rate is only 0.1, it is decreased to 0.08. Finally, the temporal attention fusion model outputs a stage-adapted feature attention weight distribution, enabling the model to automatically focus on the core influencing factors of each growth stage, thus improving the targeting and effectiveness of feature fusion.

[0086] Step S33 is the cross-dimensional feature association mining stage, the core of which is to reveal the intrinsic relationships among three types of data—weight, physiological data, and body size—using a temporal attention fusion model. Specifically, the temporal attention fusion model, based on its built-in feature association mining logic and combined with the feature attention weight distribution output in step S32, performs in-depth processing on the spatiotemporal aligned feature set generated in step S31 to construct a Pearson correlation coefficient matrix. It focuses on analyzing the relationships among the three core data types: first, the correlation between real-time weight data trends and dynamic physiological data fluctuations; second, the correlation between dynamic physiological data and three-dimensional body size data changes; and third, the correlation between three-dimensional body size data and weight trends. The correlation strength is quantified through matrix calculation (correlation coefficient ranges from -1 to 1, with closer to 1 / -1 indicating stronger positive / negative correlation), thereby quantifying the degree of association between data from different dimensions and providing data support for subsequent causal logic matching.

[0087] Step S34 is the temporal causal association matching step, the core of which is that the temporal attention fusion model transforms data associations into causal logic that can be explained in the livestock field. Specifically, the temporal attention fusion model has a built-in causal association graph based on livestock expert knowledge and large-scale data pre-training (the graph contains full-link causal nodes of "physiological state - body size change - weight gain - meat yield influence", such as "normal rumen motility → efficient nutrient absorption → stable weight gain", "insufficient daily steps → body fat accumulation → decreased lean meat ratio", the graph parameters have been solidified in the model pre-training stage); the model calls this graph, maps the feature associations output by it in step S33 (such as a strong positive correlation between rumen motility frequency and weight gain) to the graph nodes, matches the corresponding causal logic in the livestock field, automatically eliminates false associations (such as the weak correlation between body width and heart rate with no actual causal relationship), and outputs a set of temporal causal associations, realizing the transformation from "data association" to "causal logic", and giving the feature fusion results interpretability.

[0088] Step S35 is the generation of the fused and correlated feature set. Its core is the integration of multi-dimensional information by the temporal attention fusion model to form a standardized final output. Specifically, the temporal attention fusion model integrates three types of core information from its various steps: the spatiotemporal alignment feature set (data quantification values) from step S31, the feature attention weight distribution (feature importance weights) from step S32, and the temporal causal correlation set (causal basis) from step S34, according to a preset livestock feature fusion format (such as the quadruple "feature name-quantification value-weight-causal basis"). This generates a fused and correlated feature set—which includes standardized data, clearly defines the importance weights of each feature, and includes interpretable causal logic. This solves the problem of traditional feature sets "having only data but no basis," providing core feature support that is both accurate and interpretable for subsequent meat yield prediction.

[0089] The construction of the pre-built temporal attention fusion model is specifically as follows: Based on deep learning frameworks (such as TensorFlow and PyTorch), the model architecture comprises three main modules: an input layer, a core processing layer, and an output layer. Periodic calibration logic, livestock data standardization logic, temporal attention mechanism, feature association mining logic, causal association matching logic, and multi-dimensional feature integration logic are embedded throughout the model architecture and solidified during pre-training. The input layer dimensions are consistent with the feature dimensions of the multi-source temporal dataset (e.g., 4-dimensional body size + 1-dimensional weight + 4-dimensional physiological characteristics = 9 dimensions), supporting variable-length temporal data input and reserving a timestamp feature input channel to provide a foundation for periodic calibration. The core processing layer is the core of the model, divided into six functional modules corresponding to the entire process logic, achieving one-stop coherent processing from S31 to S35.

[0090] The cycle calibration module has built-in cycle calibration logic. It uses gated cyclic units (GRUs) to capture the temporal dependence of the beef cattle growth cycle. It combines prior knowledge of the livestock growth cycle to set time windows (such as a 7-day window in the mid-fat stage). Through attention masking mechanism and dynamic time warping algorithm, it masks asynchronous timestamp data, calibrates the deviation of different cycle segments, and automatically calibrates the time granularity of data with different collection frequencies to achieve time dimension unification.

[0091] Livestock data standardization module: Built-in livestock data standardization logic, embedding the Z-score standardization algorithm into the network layer, configuring standardization parameters separately for different dimensional features such as body size, weight, and physiology, completing data normalization in real time, and achieving spatial dimension unification;

[0092] Temporal attention weight optimization module: It has a built-in temporal attention mechanism, supports preset feature priorities according to growth stage, calculates feature contribution by sliding window, dynamically corrects weights through gradient descent algorithm, and outputs stage-adaptive weight distribution;

[0093] Cross-dimensional feature association mining module: Built-in feature association mining logic, adapted to Pearson correlation coefficient matrix calculation, accurately mining the inherent associations of three types of data: weight, physiological and body size, and outputting feature association relationships;

[0094] Temporal causal association matching module: It has built-in causal association matching logic, integrates pre-trained causal association graphs in the animal husbandry field, supports feature association mapping and false association removal, and outputs a set of temporal causal associations;

[0095] Multi-dimensional fusion and correlation feature generation module: It has built-in multi-dimensional feature integration logic, integrates the output results of each step according to the preset livestock format, and generates a fusion and correlation feature set containing "data-weight-causality";

[0096] The output layer is the full-process result output layer, which can output spatiotemporal aligned feature sets, feature attention weight distribution, feature correlation, temporal causal correlation set and fused correlation feature set as needed, realizing full-link coverage from data input to final feature output. After the model architecture is built, it is pre-trained using large-scale livestock multi-source time-series data (including body size-weight-physiological data of different breeds and growth stages). During the training process, the comprehensive optimization goals are "spatiotemporal alignment accuracy, weight adaptation accuracy, correlation mining accuracy, causal matching accuracy, and feature integration completeness". The parameters of the six major modules are optimized by backpropagation, and the core logic of the whole process is solidified to ensure that the model can be directly and continuously executed when called, and finally a temporal attention fusion model that can be directly called.

[0097] It should be noted that the selection of deep learning frameworks, network layer construction, module parameter initialization, pre-training optimization, and other related construction processes for the temporal attention fusion model are all conventional technical means that are well-known and achievable by those skilled in the art based on existing deep learning technology and the characteristics of livestock breeding data. This embodiment is only specifically optimized for the scenario of predicting the meat yield of beef cattle and does not involve any improvement to the basic deep learning model construction method. Those skilled in the art can flexibly adjust it according to the actual application scenario, and no limitation is made here.

[0098] In one embodiment, step S31 specifically includes the following steps:

[0099] S311. The multi-source time series dataset is divided into growth stage periodic segments, physiological periodic segments and data acquisition periodic segments by periodic calibration logic, and the time series calibration of different periodic segments is completed by dynamic time warping algorithm to obtain time series aligned dataset.

[0100] S312. Through the livestock data standardization logic, the dynamic physiological data, three-dimensional body size data and real-time weight data are standardized to obtain a standardized dataset.

[0101] S313. Integrate the temporally aligned dataset with the standardized dataset to eliminate the temporal heterogeneity and dimensional distribution differences of multi-source data and generate a spatiotemporally aligned feature set.

[0102] In this embodiment, as described in steps S311-S313 above, the core is to achieve dual unification of time series and dimensions of multi-source data through a refined decomposition process of "period division calibration - categorized standardization - dual-dimensional integration," thereby solving the core pain points of period fragmentation and dimension distribution differences in S31, and providing high-quality basic data for subsequent feature weight optimization, as detailed below:

[0103] Step S311 is the multi-cycle segmentation and temporal calibration step, the core of which is to eliminate the temporal fragmentation differences of multi-source data and achieve timeline unification. Specifically, the cycle calibration module built into the temporal attention fusion model initiates the cycle calibration logic, dividing the multi-source time-series dataset from step S20 into three layers according to three dimensions: first, growth stage cycle segments (divided into 7-14 day units based on the growth characteristics of beef cattle in the early / middle / late fattening stages, such as dividing the middle fattening stage into 7-day units); second, physiological cycle segments (divided into 24-hour units based on the natural fluctuation cycle of core physiological indicators, such as the daily cycle segments of rumen peristalsis and heart rate); and third, data acquisition cycle segments (divided into 12-hour / 24-hour units based on the actual acquisition frequency). The data is divided into segments, such as segments of body size collected over 12 hours and corresponding segments of weight collected over 24 hours. After segmentation, the Dynamic Time Warping (DTW) algorithm is used to stretch or compress physiological cycle segments and data collection cycle segments with growth stage cycle segments as the reference axis. This calibrates the temporal granularity deviation of different cycle segments (e.g., calibrating three body size data points collected over 12 hours to the corresponding time nodes within a 24-hour cycle). Finally, a time-aligned dataset is obtained, ensuring that data with different collection frequencies and different cycle attributes are completely matched in the time dimension, providing a unified time reference for subsequent cross-dimensional data fusion.

[0104] Step S312 is the categorized livestock data standardization step. The core is to adapt the standardization strategy according to the data characteristics, avoiding feature distortion caused by a single standardization approach. Specifically, the livestock data standardization module built into the temporal attention fusion model initiates the livestock data standardization logic, designing adaptation schemes for three types of core data: First, dynamic physiological data (steps / day, rumen motility times / minute, etc.), which, due to its large fluctuation range and significant influence from growth stages, adopts Z-score standardization (standardization formula: ...). ,in This represents the average of data from the same growth stage. The data distribution is designed to conform to the characteristics of livestock farming scenarios. Secondly, for three-dimensional body size data (chest circumference, body length, etc. in cm), due to their strong stability and relatively concentrated distribution, an improved Z-score standardization is used (fixing the standard deviation range of body size data in livestock farming to 10-15) to avoid extreme values ​​affecting the standardization effect. Thirdly, for real-time weight data (kg-level data), which shows a linear growth trend with different growth stages, Min-Max normalization is used (mapped to the [0,1] interval, formula: ...). ,in, (Extreme weight values ​​for the same breed and growth stage) to preserve growth trend characteristics; standardized datasets are obtained through categorized processing to normalize the distribution of livestock data with different dimensions and characteristics, avoiding interference from dimensional differences in subsequent feature weight calculations.

[0105] Step S313 is the two-dimensional data integration stage, the core of which is to integrate the temporal alignment and standardization results to completely eliminate heterogeneity. Specifically, the temporal attention fusion model integrates the temporal alignment dataset (unified daily time axis) obtained in step S311 and the standardized dataset (unified distribution range) obtained in step S312 on a time-node-by-time basis, according to the principle of "unique timestamp matching". Using the daily timestamp as an index, it associates the standardized physiological data, standardized body size data, and standardized weight data of the corresponding time nodes to form a four-dimensional data structure of "timestamp-standardized physiology-standardized body size-standardized weight", which completely eliminates the temporal heterogeneity of multi-source data (unified time granularity) and the difference in dimensional distribution (unified data distribution). Finally, it generates a spatiotemporal alignment feature set, which provides high-quality input data with a unified structure and pure data for the attention weight initialization and dynamic optimization in step S32.

[0106] In one embodiment, step S40 specifically includes the following steps:

[0107] S41. Input the fused and associated feature set into the pre-trained meat yield prediction model, wherein the meat yield prediction model has built-in stage meat yield weighting logic and causal basis binding logic.

[0108] S42. Based on the stage-based weighted logic of meat yield, match the corresponding meat yield feature weight according to the current growth stage of the target beef cattle, and weight and fuse the meat yield feature weight with the data quantification value and feature importance weight in the fusion association feature set to generate a meat yield prediction feature set.

[0109] S43. Based on the causal basis binding logic, retrieve the time-series causal relationship set from the fusion and correlation feature set, match the preset causal relationship graph, and generate a causal explanation path that is strongly correlated with the meat yield prediction feature set.

[0110] S44. Input the meat yield prediction feature set into the prediction module of the meat yield prediction model, calculate the initial predicted value of meat yield, and bind the initial predicted value of meat yield with the causal correlation basis based on the causal explanation path. After calibration by the model's deviation correction module, output the target beef cattle meat yield prediction result containing the predicted value of meat yield and the corresponding causal correlation basis.

[0111] In this embodiment, as described in steps S41-S44 above, the core is to refine the prediction process of meat yield during the mid-term tethering of Simmental cattle fattening. Through the step-by-step execution of "input activation - stage weighting - causal path generation - prediction calibration", it achieves complete alignment with the core logic of meat yield prediction in terms of numerical values, logic, and scenarios, solving the problem of the crudeness of traditional prediction processes and ensuring that the output of "predicted value + causal basis" is both accurate and unambiguous, as detailed below:

[0112] Step S41 is the model input and dual-logic activation stage, the core of which is to inherit the previous fusion features and activate the model-specific adaptation logic. Specifically, the fusion-related feature set generated in step S35 (including the quantified values ​​of core indicators, feature importance weights, and a set of temporal causal relationships, such as "stable step count → normal metabolism → steady increase in body size and weight → lean meat percentage reaching the standard") is input into the pre-trained meat yield prediction model (specifically optimized for the growth patterns and metabolic characteristics of different breeds, growth stages, and farming modes of beef cattle). This model has two built-in core logics: one is the stage-specific meat yield weighting logic (pre-stores a meat yield feature weight library for different growth stages, with weight ratios set based on research on factors affecting meat yield in the livestock field), and the other is the causal basis binding logic (which is of the same origin as the causal relationship graph of the previous temporal attention fusion model, ensuring consistency of causal logic throughout the entire chain). After receiving the feature set, the model automatically identifies the breed, growth stage, and farming mode of the target beef cattle, and activates the corresponding scenario-specific weighting logic and causal binding logic, providing an adaptation basis for subsequent step-by-step calculations.

[0113] Step S42 is the stage-weighted fusion step to generate the predicted feature set. The core is to accurately weight features according to growth stage to strengthen the correlation between features and meat yield. Specifically, the model, based on the stage-weighted meat yield logic, retrieves the meat yield feature weights corresponding to the current growth stage (e.g., early / mid / late fattening) of the target beef cattle from the built-in weight library (the weight allocation is tailored to the core influencing factors of meat yield at that stage, such as focusing on body fullness and weight gain in the mid-fattening stage, and focusing on body size growth and physiological metabolism in the early fattening stage); then, it calculates each indicator according to the preset weighting formula (meat yield prediction feature value = meat yield feature weight × data quantification value × feature importance weight). Through triple weight fusion, the contribution of core features to meat yield is highlighted. Finally, the weighted results of all indicators are integrated to generate the meat yield prediction feature set, ensuring that the feature set accurately matches the meat yield prediction needs of the current growth stage and improving the targeting of subsequent prediction calculations.

[0114] Step S43 is the causal explanation path generation stage, the core of which is to uncover the inherent logic of meat yield prediction and give the prediction results traceability. Specifically, the model, based on the causal basis binding logic, first retrieves the time-series causal relationship set from the fused correlation feature set; then, through the causal graph matching module, it accurately matches these relationships with the model's built-in livestock field causal relationship graph (including the entire link node of "feature-growth state-meat yield"), filters out weak correlation logics without actual livestock significance, and extracts a structured and interpretable causal explanation path: the path must clearly present the transmission relationship of "core feature → growth state → meat yield impact", such as "body fullness meets the standard → lean meat ratio is controllable → meat yield is stable", "normal physiological metabolism → efficient nutrient absorption → appropriate weight gain → meat yield meets the standard", etc.; finally, it is integrated into a core causal explanation path that is strongly correlated with the meat yield prediction feature set, realizing the transformation from "data features" to "meat yield impact logic", solving the pain point of traditional prediction "only giving results, without evidence".

[0115] Step S44 involves prediction calculation, binding calibration, and result output. Its core is accurate prediction and logical closed-loop processing to ensure the reliability of the output results. Specifically, it is executed in three steps:

[0116] ① Prediction Calculation: Input the meat yield prediction feature set into the prediction module of the model (built based on the improved gradient boosting tree algorithm, which optimizes the feature sensitivity parameters for the beef cattle meat yield prediction scenario and avoids the limitations of traditional linear regression). Through feature mapping and probability calculation, the initial predicted value of meat yield is output.

[0117] ② Causal binding: Bind the initial predicted value to the causal explanation path generated in step S43 one by one to form a preliminary result of "predicted value + step-by-step causal basis", and clarify the core supporting logic of each prediction result;

[0118] ③ Calibration Output: The model's deviation correction module calls the historical calibration database (which stores past prediction deviation records and adaptation correction rules) of the same species, growth stage, and breeding mode. According to the adaptation correction rules of key indicators such as body fat percentage and growth rate, the initial prediction value is calibrated to further improve the prediction accuracy.

[0119] The final output includes a complete prediction result containing "predicted meat yield and corresponding causal relationship basis". In some scenarios, a net meat yield reference value can be output as needed (calculated based on the predicted meat yield and the lean meat ratio coefficient of the corresponding growth stage), providing accurate data support and logical basis for subsequent feeding program optimization and slaughter timing selection.

[0120] The construction of the pre-trained meat yield prediction model is specifically as follows: Based on deep learning frameworks (such as TensorFlow and PyTorch), the model architecture comprises three main modules: an input layer, a core processing layer, and an output layer. The stage-based meat yield weighting logic, causal basis binding logic, prediction calculation logic, and bias correction logic are all embedded within the model architecture and solidified through large-scale data pre-training. The input layer dimension is consistent with the feature dimensions of the fused and associated feature set (including three categories of features: data quantification values, feature importance weights, and time-series causal relationship sets), supporting the adaptation of beef cattle feature inputs from different breeds, growth stages, and farming models, and reserving interfaces for breed, growth stage, and farming model recognition. The core processing layer is the core of the model, divided into four functional modules to achieve closed-loop processing throughout the prediction process.

[0121] Stage-based meat yield weighting module: Built-in stage-based meat yield weighting logic, pre-stored exclusive meat yield feature weight library for different growth stages such as early / middle / late fattening, supports automatic matching of corresponding weights according to the target beef cattle growth stage, and completes weighted fusion calculation;

[0122] Causal basis binding module: It has built-in causal basis binding logic, integrates the causal relationship graph of the animal husbandry field which is the same as the previous temporal attention fusion model, supports the retrieval, matching and structured path generation of temporal causal relationship set, and ensures the consistency and interpretability of causal logic;

[0123] Prediction Calculation Module: Based on the improved gradient boosting tree algorithm, it optimizes the feature sensitivity and fitting accuracy parameters for the beef yield prediction scenario. It outputs the initial prediction value of meat yield through feature mapping and probability calculation, avoiding the limitations of traditional linear regression in adapting to nonlinear features.

[0124] Deviation correction module: Built-in historical calibration database (stores prediction deviation records and targeted correction rules for beef cattle of different breeds, growth stages and breeding modes), supports deviation calibration according to key indicators such as body fat percentage, growth rate, and physiological state, and improves the scene adaptation accuracy of prediction results.

[0125] The output layer is a multi-dimensional result output layer, which can output the predicted meat yield, reference value of net meat yield, and step-by-step causal explanation path as needed, realizing the integrated output of "prediction results + causal basis". After the model architecture is built, it is pre-trained using large-scale training data (including the fusion and correlation feature data of beef cattle of different breeds, growth stages and breeding modes, and corresponding measured meat yield data). During the training process, the core optimization objectives are "prediction accuracy, causal binding accuracy, and stage and breed adaptability". The parameters of the four modules are iteratively optimized through gradient descent algorithm, and the prediction logic and calibration rules under each scenario are solidified to ensure that the model can be directly and coherently executed when called, and finally a pre-trained model that is accurately adapted to the beef cattle meat yield prediction scenario is formed.

[0126] It should be noted that the selection of deep learning framework, network layer construction, module parameter initialization, pre-training optimization, and other related construction processes for the meat yield prediction model are all conventional technical means that are well-known and achievable by those skilled in the art based on existing deep learning technology and the characteristics of beef cattle breeding data. This embodiment only makes targeted optimizations for the beef cattle meat yield prediction scenario and does not involve improvements to the basic deep learning model construction method. Those skilled in the art can flexibly adjust it according to the characteristics of beef cattle of different breeds, growth stages, and breeding models, and no limitations are made here.

[0127] In one embodiment, a beef cattle meat yield prediction system based on the fusion of dynamic physiology and AI three-dimensional body size is provided. This beef cattle meat yield prediction system based on the fusion of dynamic physiology and AI three-dimensional body size corresponds to the beef cattle meat yield prediction method based on the fusion of dynamic physiology and AI three-dimensional body size in the above embodiment. This beef cattle meat yield prediction system based on the fusion of dynamic physiology and AI three-dimensional body size includes:

[0128] The weight calculation module is used to acquire real-time three-dimensional body size data of the target beef cattle and input it into a pre-trained AI weight estimation model to obtain real-time weight data, and to associate the real-time three-dimensional body size data with the real-time weight data with timestamps.

[0129] The data integration module is used to acquire the dynamic physiological data of the target beef cattle and to integrate the dynamic physiological data, the three-dimensional body size data that has been time-stamped and associated, and the real-time weight data at preset time intervals to obtain a multi-source time-series dataset.

[0130] The association mining module is used to perform feature alignment of the temporal dimension and data dimension on multi-source temporal datasets based on a pre-built temporal attention fusion model, and to mine the temporal causal relationship between real-time weight data change trend, dynamic physiological data fluctuation state and three-dimensional body size data change state through the temporal attention mechanism, and generate a fused association feature set.

[0131] The meat yield prediction module is used to input the fused correlation feature set into the pre-trained meat yield prediction model and output the meat yield prediction result of the target beef cattle, which includes the meat yield prediction value and the causal correlation basis.

[0132] Optional, also includes:

[0133] The real-time monitoring module is used to monitor the dynamic physiological data of the target beef cattle in real time and compare the dynamic physiological data with the preset physiological index threshold range in real time.

[0134] The anomaly detection module is used to determine an abnormal physiological state when dynamic physiological data exceeds the threshold range of physiological indicators, and to dynamically adjust the meat yield prediction results accordingly.

[0135] The strategy output module is used to retrieve a preset physiological abnormality-breeding strategy association rule library based on the dynamically adjusted meat yield prediction results, and output a dynamic breeding optimization strategy that matches the physiological state of the target beef cattle and the dynamically adjusted meat yield prediction value.

[0136] For specific limitations regarding the beef cattle yield prediction system based on the fusion of dynamic physiology and AI three-dimensional body measurements, please refer to the limitations of the beef cattle yield prediction method based on the fusion of dynamic physiology and AI three-dimensional body measurements mentioned above, which will not be repeated here. Each module in the aforementioned beef cattle yield prediction system based on the fusion of dynamic physiology and AI three-dimensional body measurements can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0137] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows. Figure 3 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The database is used for data storage, data processing, and data analysis. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for predicting beef cattle meat yield based on dynamic physiological data and AI three-dimensional body scale fusion.

[0138] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a method for predicting beef yield based on dynamic physiology and AI three-dimensional body size fusion.

[0139] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements a method for predicting beef yield based on the fusion of dynamic physiology and AI three-dimensional body size.

[0140] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0141] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0142] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for predicting beef yield in cattle based on the fusion of dynamic physiology and AI three-dimensional body size, characterized in that, Includes the following steps: S10. Obtain real-time three-dimensional body size data of the target beef cattle and input it into the pre-trained AI weight estimation model to obtain real-time weight data, and associate the real-time three-dimensional body size data with the real-time weight data using timestamps; wherein, the training of the pre-trained AI weight estimation model includes the following steps: S101. Collect historical three-dimensional body size data and corresponding measured weight data of multiple batches of sample beef cattle, and associate the historical three-dimensional body size data with the corresponding measured weight data by timestamp to obtain the basic training dataset. S102. Perform data preprocessing on the basic training dataset to obtain a standardized training dataset; S103. Construct an initial AI weight estimation network model, wherein the initial AI weight estimation network model has a built-in mapping logic between three-dimensional body size features and body weight; S104. Input the standardized training dataset into the initial AI weight estimation network model, perform iterative training of the model according to the preset training batch and learning rate, and output the weight prediction error in real time during the training process. S105. When the weight prediction error is lower than the preset error threshold, stop the model training and obtain the trained AI weight estimation model; if the preset error threshold is not reached, adjust the model network parameters and repeat the model iterative training until the training termination condition is met. S20. Acquire the dynamic physiological data of the target beef cattle, and integrate the dynamic physiological data, the three-dimensional body size data that has been time-stamped and associated, and the real-time weight data at preset time intervals to obtain a multi-source time-series dataset. S30. Based on a pre-built temporal attention fusion model, feature alignment of the temporal dimension and data dimension is performed on multi-source temporal datasets. The temporal attention mechanism is then used to mine the temporal causal relationships among real-time weight data trends, dynamic physiological data fluctuations, and three-dimensional body size data changes, generating a fused correlation feature set. Specifically, this includes the following steps: S31. Input the multi-source time series dataset into the pre-built time series attention fusion model, and perform spatiotemporal alignment on the multi-source time series dataset based on periodic calibration logic and livestock data standardization logic to generate a spatiotemporal alignment feature set. S32. Based on the temporal attention mechanism, the attention weights of the temporal alignment feature set are initialized according to the feature priority of different growth stages of the target beef cattle; and the feature contribution of key nodes in the temporal data is calculated. Based on the feature contribution, the initial weights are dynamically corrected using the gradient descent algorithm to obtain the stage-adapted feature attention weight distribution. S33. Based on the feature association mining logic, and combined with the feature attention weight distribution, the feature association relationship between the real-time weight data change trend, the dynamic physiological data fluctuation state, and the three-dimensional body size data change state is analyzed through the Pearson correlation coefficient matrix. S34. Map the feature associations to a preset causal association graph, match the causal logic corresponding to the animal husbandry field, and obtain a time-series causal association set; S35. Integrate the temporal alignment feature set, feature attention weight distribution, and temporal causal relationship set to generate a fused correlation feature set containing data quantification values, feature importance weights, and causal relationship basis; S40. Input the fused correlation feature set into the pre-trained meat yield prediction model, and output the meat yield prediction result of the target beef cattle, which includes the meat yield prediction value and the causal correlation basis.

2. The method for predicting beef yield of cattle based on the fusion of dynamic physiology and AI three-dimensional body size as described in claim 1, characterized in that, Following step S40, the following steps are also included: S50. Monitor the dynamic physiological data of the target beef cattle in real time, and compare the dynamic physiological data with the preset physiological index threshold range in real time. S60. When dynamic physiological data is detected to exceed the threshold range of physiological indicators, it is determined to be an abnormal physiological state, and dynamic adjustment of the meat yield prediction result is performed. The dynamic adjustment includes: Trigger the adaptive correction of the parameters of the AI ​​weight estimation model to adapt to the weight estimation logic under abnormal physiological conditions; Based on the real-time weight data output by the corrected AI weight estimation model, the temporal causal association logic of the fused associated feature set is corrected. The corrected fusion-related feature set is input into the meat yield prediction model, and the meat yield prediction result, which includes the predicted meat yield value and the causal relationship basis, is dynamically adjusted. S70. Based on the dynamically adjusted meat yield prediction results, retrieve the preset physiological abnormality-breeding strategy association rule library, and output a dynamic breeding optimization strategy that matches the physiological state of the target beef cattle and the dynamically adjusted meat yield prediction value.

3. The method for predicting beef yield of cattle based on the fusion of dynamic physiology and AI three-dimensional body size as described in claim 1, characterized in that, In step S10, acquiring real-time three-dimensional body size data of the target beef cattle and inputting it into a pre-trained AI weight estimation model to obtain real-time weight data specifically includes the following steps: S11. Perform a three-dimensional scan on the target beef cattle, collect the original three-dimensional body size point cloud data, and remove environmental noise points in the point cloud data to obtain the original three-dimensional body size data after noise removal. S12. Extract features from the denoised original three-dimensional body size data to obtain three-dimensional body size features including chest circumference, body length and body width, and composite morphological features including body fullness. Integrate the three-dimensional body size features and composite morphological features to form real-time three-dimensional body size data. S13. Collect real-time physiological data of the target beef cattle, determine the physiological state based on the preset physiological index threshold range, and generate corresponding real-time physiological state labels. S14. Input the real-time three-dimensional body size data and real-time physiological state labels into the pre-trained AI weight estimation model. The AI ​​weight estimation model performs the following logical calculations and outputs real-time weight data: 1) If the real-time physiological state label is normal physiological state, the first mapping logic based on the physiological morphology rules of beef cattle is called, and the three-dimensional body size features and composite morphological features are substituted into the preset quantitative correlation formula to calculate the initial weight value of the target beef cattle under normal physiological state. 2) If the real-time physiological state label is an abnormal physiological state, the second mapping logic based on the abnormal physiological-weight association rule is called, and the corresponding weight growth rate decay coefficient is matched according to the abnormal physiological state. The weight growth rate decay coefficient is substituted into the preset quantitative correction formula to adapt and correct the initial weight value, and the weight correction value of the target beef cattle under the abnormal physiological state is calculated. 3) Output the real-time weight data of the target beef cattle that matches the current time period. The real-time weight data is the initial weight value under normal physiological conditions or the weight correction value under abnormal physiological conditions.

4. The method for predicting beef yield of cattle based on the fusion of dynamic physiology and AI three-dimensional body size as described in claim 1, characterized in that, Step S31 specifically includes the following steps: S311. The multi-source time series dataset is divided into growth stage periodic segments, physiological periodic segments and data acquisition periodic segments by periodic calibration logic, and the time series calibration of different periodic segments is completed by dynamic time warping algorithm to obtain time series aligned dataset. S312. Through the livestock data standardization logic, the dynamic physiological data, three-dimensional body size data and real-time weight data are standardized to obtain a standardized dataset. S313. Integrate the temporally aligned dataset with the standardized dataset to eliminate the temporal heterogeneity and dimensional distribution differences of multi-source data and generate a spatiotemporally aligned feature set.

5. The method for predicting beef yield of cattle based on the fusion of dynamic physiology and AI three-dimensional body size as described in claim 1, characterized in that, Step S40 specifically includes the following steps: S41. Input the fused and associated feature set into the pre-trained meat yield prediction model, wherein the meat yield prediction model has built-in stage meat yield weighting logic and causal basis binding logic. S42. Based on the stage-based weighted logic of meat yield, match the corresponding meat yield feature weight according to the current growth stage of the target beef cattle, and weight and fuse the meat yield feature weight with the data quantification value and feature importance weight in the fusion association feature set to generate a meat yield prediction feature set. S43. Based on the causal basis binding logic, retrieve the time-series causal relationship set from the fusion association feature set, match the preset causal relationship graph, and generate a causal explanation path that is strongly associated with the meat yield prediction feature set. S44. Input the meat yield prediction feature set into the prediction module of the meat yield prediction model, calculate the initial predicted value of meat yield, and bind the initial predicted value of meat yield with the causal correlation basis based on the causal explanation path. After calibration by the model's deviation correction module, output the target beef cattle meat yield prediction result containing the predicted value of meat yield and the corresponding causal correlation basis.

6. A beef cattle meat yield prediction system based on the fusion of dynamic physiology and AI three-dimensional body size, used to implement the steps of the beef cattle meat yield prediction method based on the fusion of dynamic physiology and AI three-dimensional body size as described in any one of claims 1-5, characterized in that, include: The weight calculation module is used to acquire real-time three-dimensional body size data of the target beef cattle and input it into a pre-trained AI weight estimation model to obtain real-time weight data, and to associate the real-time three-dimensional body size data with the real-time weight data with timestamps. The data integration module is used to acquire the dynamic physiological data of the target beef cattle and to integrate the dynamic physiological data, the three-dimensional body size data that has been time-stamped and associated, and the real-time weight data at preset time intervals to obtain a multi-source time-series dataset. The association mining module is used to perform feature alignment of the temporal dimension and data dimension on multi-source temporal datasets based on a pre-built temporal attention fusion model, and to mine the temporal causal relationship between real-time weight data change trend, dynamic physiological data fluctuation state and three-dimensional body size data change state through the temporal attention mechanism, and generate a fused association feature set. The meat yield prediction module is used to input the fused correlation feature set into the pre-trained meat yield prediction model and output the meat yield prediction result of the target beef cattle, which includes the meat yield prediction value and the causal correlation basis.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for predicting beef yield of cattle based on the fusion of dynamic physiology and AI three-dimensional body size as described in any one of claims 1-5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for predicting beef yield of cattle based on the fusion of dynamic physiology and AI three-dimensional body size as described in any one of claims 1-5.

Images