Early warning method, device, equipment, storage medium and program product for chicken coccidiosis

Abstract

The application relates to the technical field of cross technology of intelligent breeding, bioinformatics and artificial intelligence, and provides a chicken coccidiosis early warning method, device, equipment, storage medium and program product, which has strong anti-interference capability in chicken coccidiosis early warning. According to sound wave data and video data in a current time window, soundprint feature vectors and body state feature vectors of chickens are obtained to obtain a disease posterior probability and a group infection index; the group infection index is input into a preset chicken coccidiosis SIR-Coccidia transmission dynamics model to simulate epidemic evolution in a future time step; when the epidemic evolution simulation result exceeds a preset safety threshold, a preliminary prevention and control strategy is generated based on a multi-objective optimization algorithm, the preliminary prevention and control strategy is subjected to compliance matching processing with a preset veterinary prevention and control knowledge base, a corrected prevention and control strategy is obtained and risk prediction is carried out, and a target early warning instruction set is generated.

Description

Technical Field

[0001] This application relates to the interdisciplinary fields of smart farming, bioinformatics and artificial intelligence, and in particular to a method, device, computer equipment, storage medium and computer program product for early warning of coccidiosis in chickens. Background Technology

[0002] Coccidiosis is an acute or chronic protozoan disease caused by Eimeria coccidia parasitizing the intestinal epithelial cells of chickens. It is distributed worldwide and is one of the most economically damaging parasitic diseases in poultry farming. Traditional monitoring methods rely mainly on manual observation of bloody stools or microscopic examination, which have a significant time lag. By the time obvious bloody stools are visible to the naked eye, the chicken's intestines have already suffered severe damage, missing the optimal window for treatment.

[0003] While existing technologies offer intelligent monitoring solutions based on a single modality (sound or image), these methods suffer from weak interference resistance. The complex environment of chicken coops makes single-mode sound monitoring susceptible to fan noise, while single-mode visual monitoring is affected by changes in lighting and shading. Furthermore, current monitoring solutions lack predictive capabilities based on disease transmission mechanisms, cannot quantitatively extrapolate future epidemic trends during the subclinical phase, and typically only provide alarms, failing to automatically generate closed-loop prevention and control strategies that comply with veterinary standards. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, device, computer equipment, storage medium, and computer program product for early warning of coccidiosis in chickens, addressing the aforementioned technical problems.

[0005] This application provides a method for early warning of coccidiosis in chickens, the method comprising:

[0006] Time alignment and spatial coordinate mapping were performed on acoustic and video data in the chicken farming environment to obtain a standardized monitoring data set with spatiotemporal consistency, so as to determine the acoustic and video data of chickens in the current time window;

[0007] Based on the acoustic and video data of the chicken within the current time window, the voiceprint feature vector and body posture feature vector of the chicken are obtained.

[0008] Based on the voiceprint feature vector and body feature vector of the chicken, the posterior probability of the chicken being sick in the current time window is obtained.

[0009] The population infection index of the monitoring area in the current time window is obtained based on the posterior probability of disease of each chicken in the monitoring area in the current time window.

[0010] The population infection index is input into a pre-set SIR-Coccidia transmission dynamics model for avian coccidiosis to simulate the epidemic evolution at future time steps;

[0011] When the simulation results of the epidemic evolution exceed the preset safety threshold, a preliminary prevention and control strategy is generated based on a multi-objective optimization algorithm. The preliminary prevention and control strategy is then matched with a pre-set veterinary prevention and control knowledge base for compliance processing to obtain a revised prevention and control strategy.

[0012] Risk prediction is performed on the revised prevention and control strategy to generate a target early warning instruction set that includes environmental control instructions and drug administration recommendations.

[0013] This application provides an early warning device for chicken coccidiosis, the device comprising:

[0014] The data processing module is used to perform time alignment and spatial coordinate mapping on the acoustic and video data in the chicken farming environment to obtain a standardized monitoring data set with spatiotemporal consistency, so as to determine the acoustic and video data of chickens in the current time window.

[0015] The feature vector acquisition module is used to obtain the voiceprint feature vector and body posture feature vector of the chicken based on the sound wave data and video data of the chicken in the current time window.

[0016] The disease probability prediction module is used to obtain the posterior probability of disease of the chicken in the current time window based on the chicken's voiceprint feature vector and body feature vector.

[0017] The herd infection index determination module is used to obtain the herd infection index of the monitoring area in the current time window based on the posterior probability of disease of each chicken in the monitoring area in the current time window;

[0018] The epidemic evolution simulation module is used to input the population infection index into a preset chicken coccidiosis SIR-Coccidia transmission dynamics model to simulate the epidemic evolution at future time steps;

[0019] The strategy processing module is used to generate a preliminary prevention and control strategy based on a multi-objective optimization algorithm when the simulation results of the epidemic evolution exceed a preset safety threshold. The preliminary prevention and control strategy is then matched with a pre-set veterinary prevention and control knowledge base for compliance processing to obtain a revised prevention and control strategy.

[0020] The instruction set generation module is used to perform risk prediction on the modified prevention and control strategy and generate a target early warning instruction set that includes environmental control instructions and drug administration recommendations.

[0021] This application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0022] This application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0023] This application provides a computer program product having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0024] The solution provided in this application employs multimodal perception to obtain acoustic and video data from the chicken farming environment, thereby enabling early warning of coccidiosis in chickens. It exhibits strong anti-interference capabilities, avoiding the susceptibility of single-mode sound monitoring to fan noise interference caused by the complex chicken house environment, and the susceptibility of single-mode visual monitoring to changes in lighting and obstruction. By aligning the acoustic and video data temporally and mapping their spatial coordinates, a standardized monitoring data set with spatiotemporal consistency can be obtained. This allows for the determination of acoustic and video data of chickens within the current time window, enabling the extraction of voiceprint and body feature vectors. Subsequently, the group infection index can be calculated based on these vectors and input into a pre-set SIR-Coccidia coccidiosis transmission dynamics model for further analysis. The system simulates the evolution of the epidemic over time, enabling prediction of future epidemic trends. Then, a preliminary prevention and control strategy is generated based on a multi-objective optimization algorithm. This strategy is then matched with a pre-set veterinary prevention and control knowledge base for compliance, resulting in a revised strategy. Risk prediction is performed on this revised strategy, generating a target early warning instruction set containing environmental control instructions and medication recommendations. This provides predictive capabilities based on the disease transmission mechanism in chicken coccidiosis early warning programs, allows for quantitative extrapolation of future epidemic trends during the subclinical phase, enables early warning during the subclinical infection phase, and automatically generates closed-loop prevention and control strategies compliant with veterinary standards. This effectively solves the problems of traditional inspection delays and contact sampling stress, significantly improving the accuracy and automation level of coccidiosis prevention and control. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating an early warning method for coccidiosis in chickens in one embodiment;

[0026] Figure 2 This is another flowchart illustrating a method for early warning of coccidiosis in chickens in one embodiment;

[0027] Figure 3 This is a schematic diagram of the framework of an early warning method for coccidiosis in chickens in one embodiment;

[0028] Figure 4 This is a graph showing the trend of the number of infected people under no intervention and intervention conditions in one embodiment;

[0029] Figure 5 This is a structural block diagram of an early warning device for coccidiosis in chickens in one embodiment;

[0030] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0032] For coccidiosis in chickens, relevant technologies are usually based on intelligent monitoring schemes using a single modality (sound or image). This approach has the following technical bottlenecks:

[0033] 1. Weak anti-interference ability: The chicken house environment is complex. Single sound monitoring is easily affected by fan noise, and single visual monitoring is easily affected by changes in light and obstruction.

[0034] 2. Lack of mechanism prediction: Existing methods are mostly "black box" discrimination, which can only tell the current status and cannot combine the coccidia life history (such as the influence of temperature and humidity on oocyst sporulation) to predict the epidemic trend in the next few days.

[0035] 3. Lack of decision-making loop: Monitoring systems usually stop at issuing alarms and cannot automatically generate treatment strategies that comply with veterinary standards based on the severity of the epidemic.

[0036] To address this issue, this application provides an early warning method for coccidiosis in chickens. This method enhances its anti-interference capabilities through multimodal perception and, by combining the SIR-Coccidia transmission dynamics model for future time-step epidemic evolution simulation, it can predict future epidemic trends. During the generation of the target early warning instruction set, it incorporates a pre-built veterinary prevention and control knowledge base for compliance matching, automatically generating treatment strategies that conform to veterinary standards and improving safety. In SIR, S stands for Susceptible, representing susceptible flocks—chickens that are not yet infected but may be infected by coccidia; I stands for Infected, representing infected flocks—chickens that have been infected with coccidia and are infectious; and R stands for Recovered, representing recovered flocks—chickens that have recovered from infection and acquired some immunity (they may relapse into a susceptible state, depending on the model variant). Coccidia represents the chicken coccidia pathogen.

[0037] In one exemplary embodiment, such as Figure 1 As shown, a method for early warning of coccidiosis in chickens is provided. This method can be executed by a computer device and may include the following steps:

[0038] Step S101: Perform time alignment and spatial coordinate mapping on the acoustic and video data in the chicken farming environment to obtain a standardized monitoring data set with spatiotemporal consistency, so as to determine the acoustic and video data of the chickens in the current time window.

[0039] By utilizing a distributed, high-sensitivity microphone array and multi-angle high-definition infrared cameras deployed on top of the chicken farming environment, real-time environmental acoustic data (which can be called acoustic data) and video stream data (which can be called video data) of the chicken flock can be collected simultaneously. Parameters such as temperature, humidity, ammonia concentration, and light intensity can be obtained through environmental micro-stations; these data can be called environmental parameter data.

[0040] It can perform microsecond-level time synchronization of acoustic data, video data, and environmental parameter data based on Precise Time Protocol (PTP), and output time-aligned multimodal data sequences.

[0041] Furthermore, acoustic and video data from time-aligned multimodal data sequences can be mapped onto a unified three-dimensional spatial grid, resulting in a standardized multimodal monitoring dataset with spatiotemporal consistency. Using this standardized multimodal monitoring dataset, the acoustic and video data for each chicken in the monitoring area within the current time window can be determined.

[0042] Step S102: Based on the acoustic wave data and video data of the chicken within the current time window, obtain the chicken's voiceprint feature vector and body posture feature vector.

[0043] For the acoustic data of chickens within the current time window, pathological voiceprint features can be extracted using deep learning. Specifically, an adaptive denoising algorithm based on Variational Mode Decomposition (VMD) can be used to decompose the acoustic data, remove stationary mechanical noise mode components that overlap with the frequencies of the fan and feeder, and reconstruct a clean audio signal. The reconstructed clean audio signal is then converted into a multi-scale Mel-Frequency Cepstral Coefficients (MS-MFCC) spectrogram. The multi-scale refers to extracting Mel-Frequency Cepstral Coefficients using multiple time window lengths (e.g., 25ms, 50ms, 100ms). Next, the multi-scale Mel-Frequency Cepstral Coefficients (MS-MFCC) spectrogram can be input into a Residual Network with Squeeze-and-Excitation Networks (ResNet-SE) that incorporates a Coordinate Attention mechanism. During the training phase, the residual network employs a contrastive learning strategy. By constructing positive sample pairs (painful calls from the same chicken at similar times) and negative sample pairs (painful calls versus normal feeding sounds and fighting sounds), it widens the distance between coccidiosis-specific pain calls (usually short screams or low groans in the 3kHz-4kHz frequency band) and normal feeding sounds, fighting sounds, and ambient background sounds in the feature space. During the inference phase, the residual network can output a voiceprint feature vector representing the degree of pain stress.

[0044] For video data of chickens within the current time window, pathological posture recognition based on skeletal behavioral dynamics can be used. Specifically, the YOLOv8-Pose object detection network can be used to locate and track individual chickens in video data. Combined with the HRNet high-resolution network, key skeletal points of the head, neck, back, tail, wingtips, and claws are extracted to construct a spatiotemporal skeleton map sequence. The temporal dimension of the spatiotemporal skeleton map sequence consists of multiple consecutive frames, while the spatial dimension corresponds to the coordinates of the key skeletal points of the chicken in each frame. Geometric dynamics analysis can be performed on the spatiotemporal skeleton map sequence to calculate the rate of change of the vertical distance between the key wing points and the center point of the torso to generate the drooping wing index. Simultaneously, texture analysis can be performed on the video data, using texture entropy to analyze the smoothness changes of the chicken's contour edges to generate feather ruffleness. The average displacement velocity of the flock can be calculated based on optical flow to generate a movement vitality value. The drooping wing index, feather ruffleness, and movement vitality value are input into a spatiotemporal graph convolutional network (ST-GCN) to identify typical pathological postures such as standing still, clustering, drooping wings, and ruffled feathers, and output a posture feature vector.

[0045] Step S103: Based on the chicken's voiceprint feature vector and body feature vector, obtain the posterior probability of the chicken's disease in the current time window.

[0046] After obtaining the voiceprint feature vector and body feature vector of the chicken, Bayesian inference can be performed based on the voiceprint feature vector and body feature vector to obtain the posterior probability of the chicken being sick in the current time window.

[0047] Step S104: Based on the posterior probability of disease in each chicken within the monitoring area during the current time window, obtain the population infection index of the monitoring area during the current time window.

[0048] After obtaining the posterior probability of disease for each chicken within the monitoring area in the current time window, the weighted average of the posterior probabilities of disease for all chickens within the monitoring area in the current time window can be calculated, which can be expressed by the formula: .

[0049] in, This represents the total number of individuals monitored. For the first The weight of each individual is determined by video visibility. and voiceprint signal-to-noise ratio The geometric mean is determined and can be expressed by the formula: Video visibility values ​​typically range from 0 to 1; the signal-to-noise ratio (SNR) is measured in dB and can be normalized to 0 to 1. The output is a cluster infection index representing the severity of the epidemic in the monitored area within the current time window. The value ranges from 0 to 1.

[0050] Step S105: Input the group infection index into the preset chicken coccidiosis SIR-Coccidia transmission dynamics model to simulate the epidemic evolution in future time steps.

[0051] The pre-defined SIR-Coccidia transmission dynamics model includes environmental oocyst density variables. Its system of differential equations is as follows:

[0052] ;

[0053] ;

[0054] .

[0055] in Indicates time, Indicates temperature. Indicates humidity. For the number of susceptible individuals, For the number of infected individuals, The contact transmission rate is affected by humidity. For the environmental oocyst infection rate, For recovery rate, The rate of oocyst sporulation affected by temperature. The inactivation rate of oocysts affected by temperature.

[0056] Group infection index Input a pre-defined SIR-Coccidia transmission dynamics model to simulate the epidemic evolution at future time steps.

[0057] Step S106: When the epidemic evolution simulation results exceed the preset safety threshold, a preliminary prevention and control strategy is generated based on a multi-objective optimization algorithm. The preliminary prevention and control strategy is then matched with a pre-set veterinary prevention and control knowledge base for compliance processing to obtain a revised prevention and control strategy.

[0058] When the simulated epidemic evolution results exceed a preset safety threshold, a prevention and control strategy generation mechanism can be triggered. This mechanism generates a preliminary prevention and control strategy based on a multi-objective optimization algorithm. This preliminary strategy may include values ​​for fan speed, drug type, and drug concentration. A pre-built veterinary prevention and control knowledge base is then used to perform compliance matching on the preliminary strategy, resulting in a revised prevention and control strategy.

[0059] Step S107: Perform risk prediction on the revised prevention and control strategy and generate a target early warning instruction set that includes environmental control instructions and drug administration recommendations.

[0060] The prevention and control strategy generation mechanism can also include risk prediction of the revised prevention and control strategy to determine whether it can control the epidemic within a safe range. If so, the revised prevention and control strategy can be encoded into a target early warning instruction set containing environmental control instructions (such as fan speed adjustment and humidity control) and drug administration recommendations (such as drug type, dosage, and administration area). The target early warning instruction set belongs to industrial control instructions. Among them, environmental control instructions can correspond to fan speed values, and drug administration recommendations can correspond to drug type and drug concentration values. The target early warning instruction set can be sent to the farm's environmental control system via the MQTT (Message Queuing Telemetry Transport) protocol to achieve unmanned closed-loop management. It can also trigger graded alarms based on risk levels (blue, yellow, red).

[0061] The aforementioned early warning method for coccidiosis in chickens employs multimodal sensing to obtain acoustic and video data from the chicken farming environment. This method offers high anti-interference capabilities, avoiding the susceptibility of single-mode sound monitoring to fan noise and single-mode visual monitoring to light variations and obstructions caused by complex chicken house environments. By aligning the acoustic and video data temporally and mapping their spatial coordinates, a standardized monitoring data set with spatiotemporal consistency can be obtained. This allows for the determination of acoustic and video data of chickens within the current time window, enabling the extraction of voiceprint and body feature vectors. Subsequently, the group infection index can be calculated based on these vectors and input into a pre-defined SIR-Coccidia transmission dynamics model. By simulating the evolution of the epidemic over future time steps, it is possible to predict the future trend of the epidemic. Then, based on a multi-objective optimization algorithm, a preliminary prevention and control strategy is generated. The preliminary prevention and control strategy is then matched with a pre-set veterinary prevention and control knowledge base for compliance processing to obtain a revised prevention and control strategy. Risk prediction is performed on the revised prevention and control strategy to generate a target early warning instruction set containing environmental control instructions and drug administration recommendations. This provides predictive capabilities based on the disease transmission mechanism in the chicken coccidiosis early warning program, enables quantitative extrapolation of the future epidemic trend during the subclinical stage, achieves early warning in the subclinical infection stage, and automatically generates a closed-loop prevention and control strategy that complies with veterinary standards. It can effectively solve the problems of traditional inspection lag and contact sampling stress, and significantly improve the accuracy and automation level of coccidiosis prevention and control.

[0062] In one exemplary embodiment, acoustic and video data from a chicken farming environment are time-aligned and spatially mapped to obtain a standardized monitoring data set with spatiotemporal consistency, including:

[0063] After time alignment of acoustic and video data from the chicken farming environment, noise reduction is performed on the acoustic data, and keyframe extraction is performed on the video data. Based on the transformation matrix between the spatial positioning coordinate system of the microphone array and the physical world coordinate system, and the transformation matrix between the pixel coordinate system of the camera and the physical world coordinate system, the noise-reduced acoustic data and the video data after keyframe extraction are mapped onto a unified three-dimensional spatial grid to obtain a standardized monitoring data set with spatiotemporal consistency. Among them, the acoustic data is collected by the microphone array, and the video data is collected by the camera.

[0064] The camera can be a multi-angle high-definition infrared camera, and the corresponding video data collected is multi-angle video data.

[0065] Precision Time Protocol (PTP) can be used to align acoustic data, video data, and environmental parameter data along the timeline. Then, adaptive Wiener filtering can be applied to the time-aligned acoustic data for noise reduction, and keyframes can be extracted from the time-aligned video data. By combining the camera's intrinsic parameter matrix with the spatial geometric topology of the microphone array, transformation matrices can be established between the camera's pixel coordinate system and the physical world coordinate system, as well as between the microphone array's spatial positioning coordinate system and the physical world coordinate system. Based on these transformation matrices, the noise-reduced acoustic data and the video data with extracted keyframes can be mapped onto a unified three-dimensional spatial grid, generating a standardized monitoring data set with spatiotemporal consistency.

[0066] In an exemplary embodiment, the posterior probability of a chicken's disease in the current time window is obtained based on its voiceprint feature vector and body posture feature vector, including:

[0067] The abnormal voiceprint index and the abnormal body posture index are obtained from the voiceprint feature vector of the chickens. A multimodal joint feature representation is obtained from the voiceprint feature vector and the abnormal body posture index. The multimodal joint feature representation, the abnormal voiceprint index and the abnormal body posture index are used as observational evidence, and the environmental parameter data within the current time window are used as prior probability conditions. Bayesian inference is performed based on the observational evidence and the prior probability conditions to obtain the posterior probability of the chickens being sick in the current time window.

[0068] Classification can be performed based on deep learning. Specifically, the voiceprint feature vector can be input into a classifier or regression model to calculate the voiceprint anomaly index. The voiceprint anomaly index can be used to quantify the degree of anomaly in voiceprints within the current time window. Its value typically ranges from 0 to 1. The voiceprint anomaly index can be used for subsequent multimodal feature fusion. Alternatively, a body posture feature vector can be input into a classifier or regression model to calculate the body posture anomaly index. The body posture abnormality index can be used to quantify the degree of abnormality of body posture within the current time window. The value range is usually from 0 to 1. The body posture abnormality index is used for subsequent multimodal feature fusion.

[0069] A sliding window with a time window of 30 seconds can be defined, and the current time window can correspond to a time length of 30 seconds.

[0070] The voiceprint feature vector, body feature vector, voiceprint abnormality index, and body abnormality index of chickens can be input into a multimodal fusion diagnostic model based on Bayesian inference to calculate the infection risk and obtain the population infection index of the monitored area in the current time window.

[0071] The specific process of calculating infection risk using a multimodal fusion diagnostic model based on Bayesian inference may include:

[0072] The voiceprint feature vector within the current time window is calculated using a cross-modal interactive attention module. With body feature vector The cross-covariance matrix between Singular value decomposition of the cross-covariance matrix yields the principal singular values. and Modal confidence weights are calculated using softmax normalization: , The confidence weights of auditory and visual modalities for coccidiosis discrimination are quantified within the current time window.

[0073] Tensor fusion technology is used to weight the voiceprint feature vector and body feature vector using modal confidence weights: , Perform an outer product operation on the weighted feature vectors: After flattening, a multimodal joint feature representation is formed. .

[0074] By combining Bayesian confidence networks, multimodal joint feature representation and voiceprint anomaly index are integrated. and postural abnormality index As observational evidence, environmental parameter data within the current time window (Especially bedding humidity and temperature) are used as prior probability conditions to calculate the posterior probability of disease in individual chickens within the current time window. .in,

[0075] .

[0076] and As a modal reliability indicator, when both are greater than 0.6, the confidence weight of the posterior probability of disease increases.

[0077] After calculating the infection risk as described above and obtaining the posterior probability of disease for individual chickens within the current time window, spatial clustering and statistical aggregation can be performed on the posterior probability of disease for individual chickens within the monitoring area within the current time window. Specifically, this includes using the DBSCAN algorithm (neighborhood radius...). , the minimum number of points ) for high posterior probability of disease ( Individuals are spatially clustered, and clusters containing more than 5 individuals are marked as high-risk areas for the epidemic. The spatial center coordinates and boundary range of the high-risk areas are recorded. DBSCAN stands for Density-Based Spatial Clustering of Applications with Noise.

[0078] In an exemplary embodiment, the population infection index is input into a pre-set SIR-Coccidia transmission dynamics model to simulate the epidemic evolution at future time steps, including:

[0079] Based on the temperature and humidity data of the current time window, the parameters affected by temperature and humidity in the pre-set SIR-Coccidia transmission dynamics model of chicken coccidiosis are corrected to obtain the corrected SIR-Coccidia transmission dynamics model of chicken coccidiosis; the epidemic evolution of future time steps is simulated based on the population infection index and the corrected SIR-Coccidia transmission dynamics model of chicken coccidiosis.

[0080] A pre-defined SIR-Coccidia transmission dynamics model for chicken coccidiosis can be initialized, and an environmental oocyst density variable can be introduced into the model. It can display the cluster infection index of the monitored area within the current time window. Mapped to the initial number of infected individuals ,in Total population size (unit: individuals). This represents the number of infections at the initial time (t=0) (unit: infected).

[0081] Initial number of susceptible groups (Unit: individual).

[0082] Environmental cyst density (Unit: units / m²) This setting can be based on historical monitoring data. If there is no historical data, it can be initialized to 0.

[0083] Based on real-time collected environmental temperature and humidity data, the temperature and humidity data corresponding to the current time window can be obtained. Then, the relevant parameters of the SIR-Coccidia transmission dynamics model for chicken coccidiosis can be dynamically corrected through the following process:

[0084] First, based on the current humidity data (Unit: %) Calculate the contact transmission rate ,in Baseline contact transmission rate (unit: individuals) -1 ·sky -1 ), Humidity influence coefficient (unit: %) -1 (dimensionless); secondly, based on the current temperature data. (Unit: K, absolute temperature) Calculation of oocyst sporulation rate using the Arrhenius equation and inactivation rate ,in Sporulation frequency factor (unit: individuals) -1 ·sky -1 ), Inactivation frequency factor (unit: days) -1 ), and Activation energy (unit: J·mol) -1 ), This is a gas constant; in practical applications, a simplified exponential model based on Celsius temperature can be used to approximate the calculation instead of the Arrhenius equation. Environmental infection rate. (Unit: individual) -1 ·indivual -1 ·m²·day -1 () represents the infectivity of a unit oocyst density to a susceptible individual; recovery rate (Unit: days) -1 () indicates the daily recovery rate of infected individuals.

[0085] The current state can be , and Corrected parameters , and ,as well as and The differential equations of the SIR-Coccidia transmission dynamics model for chicken coccidiosis are input into the model, and then the fourth-order Runge-Kutta method is used to numerically solve for the disease evolution within a preset future time step (e.g., 24 to 72 hours) to predict susceptible populations. Infected groups and environmental follicle density The curve showing the change.

[0086] The time-dimensional change curve obtained from the numerical solution is mapped back to the spatial grid map of the farm. Specifically: First, the farm space can be divided into uniform grids of 2m × 2m (for a standard chicken house of 100m × 20m, it is divided into 50 × 10 = 500 grid units), and each grid unit records its center coordinates and index number; Second, combined with the spatial location coordinates of the high-risk areas of the epidemic obtained above (cluster centers and boundaries obtained by DBSCAN clustering), each high-risk area is mapped to the corresponding grid unit index set to establish a spatial mapping table from region to grid; Then, based on the spatial mapping table, the global infection count of the entire monitoring area is calculated. The distribution of local infections in each high-risk area is determined by decomposing the individual density proportions within each high-risk area. This local infection distribution is then further allocated to the corresponding grid cells based on the individual density distribution within each grid cell, and finally determined by the transmission probability of adjacent grid cells (derived from the contact transmission rate). and environmental infection rate (Decision) Predict the spatial spread path of the epidemic in the next 24 to 72 hours and generate an epidemic spread risk distribution map that includes spatial dimensions (which grid cells) and temporal dimensions (the infection status at each moment in the next 24 to 72 hours).

[0087] Analyzing the distribution map of the epidemic spread risk can extract the peak time of future outbreaks, the coordinates of high-risk areas, and the key environmental inducing factors that lead to rapid sporulation of oocysts. Among them, the peak time is used to determine the optimal intervention time, the coordinates of high-risk areas are used for targeted drug administration and environmental regulation, and the key environmental inducing factors (which may include temperature T and humidity H) are used as constraints for multi-objective optimization.

[0088] In one exemplary embodiment, a preliminary prevention and control strategy is generated based on a multi-objective optimization algorithm, including:

[0089] A multi-objective optimization function is obtained, aiming to minimize the spread rate of the epidemic, minimize drug costs, and minimize the risk of drug residues. The decision variables in the multi-objective optimization function include fan speed, drug type, and drug concentration. Based on the multi-objective optimization function, the Pareto optimal frontier is searched in the solution space using the NSGA-II algorithm. The fan speed, drug type, and drug concentration values ​​corresponding to the optimal solution are selected to obtain the preliminary prevention and control strategy.

[0090] Based on the influence mechanism of key environmental inducing factors on the evolution of the epidemic, a multi-objective optimization function can be constructed with the objectives of minimizing the epidemic spread rate, minimizing drug costs, and minimizing drug residue risks. The epidemic spread rate is related to economic losses, and the product of the epidemic spread rate and the economic loss per chicken reflects the economic loss. Therefore, minimizing the epidemic spread rate can be transformed into minimizing economic losses. The resulting multi-objective optimization function is as follows:

[0091] ;

[0092] ;

[0093] .

[0094] in, To predict the number of infections after a time step (unit: animals), this number is indirectly determined by environmental control parameters. The specific transmission mechanism is as follows: based on the empirical model of the farm's environmental control system, the decision variable is the fan speed. Adjustments will cause corresponding changes in ambient temperature and humidity, resulting in the adjusted temperature. and humidity Where ΔT and ΔH are determined by the fan speed. With baseline speed The difference between them is determined through an empirical model of the environmental control system, and then the adjusted temperature is... and humidity Substitute the SIR model parameters , , Calculations are performed, which indirectly affect the number of infections. Meanwhile, decision variables (Drug concentration) indirectly affects the number of infections by increasing the recovery rate γ. .

[0095] Total population size (unit: individuals); Economic loss per chicken (unit: yuan); Decision variables (Drug type) Corresponding unit cost of the drug (unit: yuan·kg) -1 ); The single dose (unit: kg·unit) corresponding to the drug concentration is the decision variable. -1 ); Energy costs for environmental regulation (unit: yuan·kWh) -1 ); Decision variables Corresponding wind turbine power increment (unit: kW); For the predicted time step (in hours); The drug residue coefficient (dimensionless) corresponding to the decision variable drug (drug type); The empirical withdrawal period (in days) corresponding to the selected drug type. This represents the remaining time (in days) until the planned slaughter date for the current flock. During the withdrawal period... Greater than the remaining time to slaughter When the withdrawal period cannot be completed before the planned slaughter, there is a risk of drug residue; when the withdrawal period... Less than or equal to the remaining slaughter time hour, A value of zero indicates that there is no risk of drug residue.

[0096] Based on the multi-objective optimization function, the NSGA-II (Non-dominated Sorting Genetic Algorithm II) algorithm can be used to solve for the optimal control strategy, search for the Pareto optimal front, select the fan speed, drug type, and drug concentration values ​​corresponding to the optimal solution, and obtain the preliminary prevention and control strategy.

[0097] In an exemplary embodiment, risk prediction is performed on the modified prevention and control strategy to generate a target early warning instruction set containing environmental control instructions and drug administration recommendations, including:

[0098] A second simulation was conducted based on the revised prevention and control strategy and the SIR-Coccidia transmission dynamics model for chicken coccidiosis. When the results of the second simulation met the preset conditions, a target early warning instruction set containing environmental control instructions and drug administration recommendations was obtained based on the revised prevention and control strategy.

[0099] The revised prevention and control strategy was then substituted into the SIR-Coccidia transmission dynamics model for secondary simulation. The specific process included the following steps:

[0100] The first step is to convert the 80% fan speed in the revised control strategy into an environmental parameter change. Based on the empirical model of the farm's environmental control system, increasing the fan speed from 40% to 80% will result in a 5% decrease in humidity (from...). Down to ), the temperature rises by 2°C (from Rise to );

[0101] The second step is to update the parameters affected by humidity and temperature in the SIR-Coccidia transmission dynamics model of chicken coccidiosis, and calculate the updated contact transmission rate and the updated oocyst inactivation rate.

[0102] The updated contact transmission rate is:

[0103] Only -1 ·sky -1 ,

[0104] The updated follicle inactivation rate is:

[0105] sky -1 ;

[0106] The third step is to model the therapeutic effect of diclazuril (a drug in the revised prevention and control strategy) as an increase in the recovery rate: Among them, the basic rehabilitation rate sky -1 Drug efficacy coefficient ppm -1 ·sky -1 diclazuril concentration =15ppm, therefore the updated recovery rate sky -1 (That is, the average infection period is shortened from 10 days to 2.5 days);

[0107] Fourth step, use the updated parameters , , The differential equations of the SIR-Coccidia transmission dynamics model in chickens were resolved using the fourth-order Runge-Kutta method with a time step of [missing information]. Hours, predicting the next 72 hours , , The curve showing the change.

[0108] The results of the second simulation show that:

[0109] Initial number of infections However, after 72 hours, the number of infections dropped. Only, infection rate It can be confirmed that, under the premise of compliant medication use, the epidemic can be controlled within a safe range, and the number of infections is showing a downward trend. Therefore, when the results of the second simulation meet the preset conditions, the revised prevention and control strategy can be encoded into a target early warning instruction set that includes environmental control instructions and medication recommendations, and the instructions can be issued.

[0110] To better understand the above method, an application example of the early warning method for coccidiosis in chickens according to this application is described in detail below. This embodiment provides an early warning scheme for coccidiosis in chickens based on voiceprint recognition and body posture analysis, which utilizes deep learning technology for non-contact monitoring of respiratory and digestive tract diseases in poultry. It belongs to the field of smart farming and non-contact monitoring technology for animal diseases. This scheme constructs a technical chain of "multimodal perception - feature decoupling and fusion - dynamics inference - closed-loop decision-making". First, it collects sound wave, video, and environmental parameter data of the farming environment to construct a multi-source spatiotemporal simulation dataset. Second, it uses an improved variational mode decomposition (VMD) and attention mechanism to extract pain voiceprint features and uses a spatiotemporal graph convolutional network (ST-GCN) to analyze pathological body posture features. Next, it constructs a multimodal fusion model based on Bayesian inference to calculate the infection risk and inputs a chicken coccidiosis SIR-Coccidia transmission dynamics model that incorporates environmental factors to simulate the epidemic evolution. Finally, it generates environmental regulation and drug administration strategies based on a multi-objective optimization algorithm. This enables early warning during the subclinical infection phase, effectively addressing the issues of delayed traditional inspections and stress associated with contact sampling. It significantly improves the accuracy and automation of coccidiosis control, resolving problems such as monitoring lag, high reliance on manual intervention, low accuracy of single-modality methods, and lack of predictive decision-making in related technologies. For example... Figure 2 As shown, the scheme may include steps S110 to S150, wherein, Figure 3 This is a schematic diagram of the framework of the scheme.

[0111] Step S110: Collect multimodal data from the breeding environment. The multimodal data includes real-time acoustic data of the chicken flock's environment, multi-angle video data, and environmental parameter data. Perform microsecond-level time synchronization on the multimodal data, and perform spatial coordinate mapping on the acoustic data and video data to generate a standardized monitoring data set with spatiotemporal consistency. The multimodal data is the input dataset of the multi-source spatiotemporal breeding environment simulation data, and the standardized monitoring data set is the output dataset of the multi-source spatiotemporal breeding environment simulation data.

[0112] Step S120: Decouple and extract multi-source features from the standardized monitoring data set to obtain the voiceprint feature vector set representing pain stress and the body feature vector set representing pathological behavior, and input the feature vector set into the multimodal fusion diagnostic model to calculate the herd infection index;

[0113] After multi-source feature decoupling and feature extraction, this step allows for Bayesian inference of infection risk. Specifically, multi-source feature decoupling and feature extraction can include:

[0114] For acoustic data, variational mode decomposition (VMD) can be used to decompose the signal into several intrinsic mode functions (IMFs). The number of modes K in the variational mode decomposition can be set to 3 to 7, and the penalty factor α can be set to 1000 to 3000. Low-frequency mechanical noise components are removed, and multi-scale Mel-frequency cepstral coefficients (MS-MFCCs) are extracted as the voiceprint feature vector. The voiceprint anomaly index is then calculated using a classifier or regression model. Here, "multi-scale" refers to using multiple time window lengths to extract the Mel-frequency cepstral coefficients.

[0115] For video data, the YOLOv8-Pose network can be used to extract key points of the chicken skeleton, construct a spatiotemporal skeleton sequence, calculate the drooping wing index and feather texture entropy as body posture feature vectors, and calculate the body posture abnormality index through a classifier or regression model; where the drooping wing index is the angle between the line connecting the wing key point and the center point of the trunk and the vertical direction, or the ratio of the vertical distance between the wing key point and the center point of the trunk to the trunk height.

[0116] Voiceprint feature vector, body posture feature vector, voiceprint abnormality index, and body posture abnormality index are input into the multimodal fusion diagnostic model. Among them, the voiceprint abnormality index and body posture abnormality index are used as observational evidence for Bayesian inference to participate in the calculation of the herd infection index.

[0117] Step S130: Input the population infection index into the preset chicken coccidiosis SIR-Coccidia transmission dynamics model to simulate the epidemic evolution in future time steps. When the simulation results exceed the preset safety threshold, trigger the prevention and control strategy generation mechanism.

[0118] Step S140: Generate a preliminary prevention and control strategy based on a multi-objective optimization algorithm, and perform compliance matching processing between the preliminary prevention and control strategy and the pre-set veterinary prevention and control knowledge base to generate a revised prevention and control strategy;

[0119] The compliance matching process may include: parsing the medication guidelines in the pre-set veterinary prevention and control knowledge base to generate a drug contraindication list and a set of withdrawal period thresholds; comparing the medication recommendations in the initial prevention and control strategy with the drug contraindication list, and if there are any violations, triggering an alternative drug recommendation algorithm for correction.

[0120] Step S150: Perform risk prediction and dynamic adjustment on the revised prevention and control strategy, and generate a target early warning instruction set that includes environmental control parameters and drug administration recommendations.

[0121] The solution provided in this embodiment can be applied to large-scale standardized white-feathered broiler farms (such as single-building farms with 50,000 birds, floor-raising mode). This scenario is used to introduce the specific implementation process of the above steps.

[0122] In step S110, multimodal data from the aquaculture environment can be collected and spatiotemporally aligned to generate a standardized monitoring dataset. To comprehensively capture the subtle characteristics of the subclinical infection phase, a multidimensional sensing network can be constructed. Step S110 specifically includes steps S111 to S113.

[0123] Step S111: Data acquisition.

[0124] For acoustic data, high-sensitivity pickup nodes (frequency response of 10Hz-20kHz and sampling rate of 44.1kHz) are deployed every 15 meters along the central axis on the top of the chicken house to ensure that the high-frequency harmonic components of the chickens' pain-induced calls can be fully captured.

[0125] For video data, a 4K infrared camera can be deployed for acquisition, with a resolution of 2560×1440 and a frame rate of 30fps.

[0126] For environmental parameter data, IoT micro-stations are deployed to collect temperature, humidity, and ammonia concentration.

[0127] Step S112: Spatiotemporal alignment.

[0128] The IEEE 1588v2 Precision Time Protocol (PTP) was used to provide microsecond-level timing for all sensors, ensuring audio-visual synchronization with an error controlled within 1ms. Secondly, the Zhang Zhengyou calibration method was used to establish the transformation matrix between the camera's pixel coordinate system and the physical world coordinate system: a calibration board (chessboard pattern, 10cm×10cm grid) was placed on the farm floor, multi-angle calibration images were acquired, and the camera's intrinsic parameter matrix and distortion coefficients were calculated to establish the pixel coordinate system. To the physical world coordinate system The mapping relationship is established. Simultaneously, a spatial coordinate system for sound source localization is established based on the physical installation location of the microphone array.

[0129] Step S113: Standardization process.

[0130] Adaptive Wiener filtering is used to remove background noise from the fan after time alignment of the acoustic data; the filter parameters may include: window length of 256 sampling points, overlap rate of 50%, and noise estimation using the first 100ms of the silent segment.

[0131] Keyframes are extracted from time-aligned video data using the inter-frame difference method. When the pixel difference between consecutive frames exceeds 5%, they are extracted as keyframes.

[0132] The denoised acoustic data and the video data after keyframe extraction are mapped to a unified three-dimensional spatial grid (grid resolution 2m×2m×2m) using the previously established coordinate mapping relationship, generating a standardized monitoring data set in JSONL (JSON Lines) format. .

[0133] In step S120, multi-source features can be decoupled and extracted from the standardized monitoring data set to calculate the herd infection index. Step S120 may specifically include steps S121 to S123.

[0134] Step S121: Voiceprint feature extraction.

[0135] Input the acoustic data from the standardized monitoring dataset into the Variational Mode Decomposition (VMD) module and set the number of modes. Punishment factor Iterative convergence tolerance The signal was decomposed into five intrinsic mode functions (IMF1 to IMF5). IMF1 and IMF2 have center frequencies of 50Hz and 150Hz, respectively, corresponding to the mechanical noise of the fan and feeder. IMF3 to IMF5 have center frequencies of 800Hz, 2500Hz, and 3800Hz, respectively, containing the main frequency components of the chicken's calls. Low-frequency components (IMF1 and IMF2) were removed, and IMF3 to IMF5 were retained for signal reconstruction, resulting in a clean audio signal. The signal-to-noise ratio was improved from the original 12dB to 25dB.

[0136] The reconstructed signal was converted into a multi-scale Mel-Frequency Cepstral Coefficients (MS-MFCC) spectrogram. MFCC features were extracted using three time window lengths: 25ms, 50ms, and 100ms (corresponding to frame shifts of 10ms, 20ms, and 40ms, respectively). Each window extracted 13-dimensional MFCC coefficients and their first and second-order differences, resulting in a total of 39 features. The three scales were concatenated to form a 117-dimensional multi-scale feature vector. MFCC stands for Mel-Frequency Cepstral Coefficients.

[0137] A ResNet-SE network with coordinate attention mechanism (containing 4 residual blocks, each with 64-128-256-512 convolutional kernels) is used to extract pain-related vocalization feature vectors. The ResNet-SE network is trained using a contrastive learning strategy: positive sample pairs (pain-related calls from the same chicken at similar times, with a time interval of <5 seconds) and negative sample pairs (pain-related calls, normal feeding sounds, and fighting sounds) are constructed, and the InfoNCE loss function is used. Increase the distance between pathological and normal voiceprints in the feature space. Among these, The loss value of the InfoNCE loss function; This is the feature representation vector of the current anchor point sample (i.e., the target voiceprint sample to be learned) in the feature space; The feature representation vector in the feature space of the sample that forms a positive sample pair with the anchor sample (i.e., another pain-induced call of the same chicken within a similar time period); This is the feature representation vector of the k-th sample in the feature space during the training batch, including both positive and negative samples. This represents the number of samples in the training batch. This represents the total number of samples after data augmentation. For indicator functions, when When the value is 1, The value is 0, which is used to exclude the anchor sample itself in the summation of the denominator; This is the cosine similarity function, used to calculate the directional similarity between two feature vectors; This is a temperature parameter used to adjust the smoothness of the similarity distribution, with a value of 0.07. The loss function is the natural exponential function. This loss function increases the distance between pathological voiceprints and normal voiceprints in the feature space by maximizing the similarity between the anchor sample and positive samples, while minimizing the similarity between the anchor sample and all negative samples.

[0138] The ResNet-SE network focuses on the energy distribution in the 3kHz-4kHz frequency band (a characteristic band of chickens' pain-sensing calls), outputting a 128-dimensional voiceprint feature vector, and calculating the voiceprint anomaly index through a fully connected layer (128→64→1, with the sigmoid activation function). In this embodiment, the typical value of the voiceprint anomaly index for healthy chickens is 0.05-0.15; for infected chickens, the typical value is 0.65-0.92. InfoNCE stands for Information Noise-Contrastive Estimation.

[0139] Step S122: Extraction of body features.

[0140] The YOLOv8-Pose algorithm (input resolution 640×640, confidence threshold 0.5, NMS threshold 0.45) was used to perform instance segmentation and keypoint detection on chickens in video frames. Twenty-one skeletal keypoints were extracted, including the head, neck, back center, tail, left wingtip, right wingtip, left claw, and right claw. Each keypoint contains three values: (x, y, confidence), where x and y represent the coordinates of the keypoint, and confidence represents the confidence level. A spatiotemporal skeleton graph sequence was constructed, with a temporal dimension of 30 consecutive frames (corresponding to 1 second of video, frame rate 30fps) and a spatial dimension consisting of the two-dimensional coordinates of the chicken skeletal keypoints in each frame.

[0141] Calculate the drooping wing index The drooping wing index can be defined using either of the two aforementioned definitions; this embodiment uses the included angle definition because it is more robust to changes in lighting and viewing angle. First, the average position of the left and right wingtips is calculated. Then calculate the wingtip and the center point of the back. The angle between the line connecting the two directions and the vertical direction (Unit: degree). In this embodiment, normal chickens The mean is Sick chickens The mean is ,by As a discrimination threshold.

[0142] Calculate feather texture entropy: Perform gray-level co-occurrence matrix (GLCM) analysis on the chicken outline region (extracted through instance segmentation mask) to calculate texture entropy. ,in These are elements of a normalized gray-level co-occurrence matrix. Healthy chickens have neatly arranged feathers. The typical value is 4.2-5.1; diseased chickens have ruffled feathers. The typical value is 6.8-8.5. The average displacement velocity of the chicken flock was calculated based on the optical flow method (Farneback algorithm, pyramid level 3, window size 15×15). (Unit: pixels / frame), where The number of individuals detected. For the first Inter-frame displacement of individual chickens. Healthy flock of chickens. Typical values ​​are 8.5-15.2 pixels / frame, in diseased chicken flocks. Typical values ​​are 1.2-4.8 pixels per frame.

[0143] The drooping wing index, feather texture entropy, and kinetic activity value are concatenated into a geometric feature vector (3×30=90 dimensions, containing 30 frames of temporal information). This vector is then input into a spatiotemporal graph convolutional network (ST-GCN, containing 9 graph convolutional layers, with hidden layer dimensions of 64-128-256) for temporal analysis. This identifies typical pathological postures such as standing still, clustering, drooping wings, and ruffled feathers, outputting a 256-dimensional posture feature vector. The posture abnormality index is then calculated through a fully connected layer (256→128→1, with the sigmoid activation function). In this embodiment, for healthy chickens, Typical values ​​are 0.08-0.18; for infected chickens, Typical values ​​are 0.72-0.95.

[0144] Step S123: Feature fusion and population infection index calculation.

[0145] First, through a cross-modal interactive attention module, the cross-covariance matrix between the voiceprint feature vector and the body posture feature vector within the current 30-second time window is calculated. Singular value decomposition (SVD) is then performed on the cross-covariance matrix to obtain the principal singular values. , (Typical values ​​in this embodiment). Modal confidence weights are calculated using softmax normalization: , Quantify the confidence weights of auditory and visual modalities in the diagnosis of coccidiosis.

[0146] Second, tensor fusion technology is used to weight the voiceprint feature vector and body feature vector using modal confidence weights: , Perform an outer product operation on the weighted eigenvectors: After flattening, a multimodal joint feature representation is formed. .

[0147] Third, the multimodal joint feature representation, voiceprint anomaly index, and body posture anomaly index are input into the Bayesian confidence network, combined with the current environmental humidity. and temperature As a prior probability condition, the posterior probability of disease for each individual within the monitoring area is calculated. In this embodiment, for the first... Individual, if and (If all values ​​are greater than the 0.6 threshold), then in the calculation of the individual's posterior probability, the confidence weight of the observed evidence is set to 1.2 (a 20% improvement compared to the single-modality approach); if and If all values ​​are less than 0.3, then the confidence weight of the observed evidence is set to 0.5, indicating that the individual may be in a healthy state, thus reducing the risk of false alarms.

[0148] Fourth, perform spatial clustering and statistical aggregation on all individuals within the monitoring area: using the DBSCAN algorithm (neighborhood radius). , the minimum number of points ) for high posterior probability of disease ( Individuals were spatially clustered, and clusters containing more than 5 individuals were marked as high-risk areas for the epidemic. The spatial center coordinates and boundary ranges of these clusters were recorded. In this embodiment, three high-risk areas were identified, located in the northeast corner (center coordinates: X=15m, Y=8m), the middle (center coordinates: X=50m, Y=10m), and the southwest corner (center coordinates: X=85m, Y=6m) of the chicken house.

[0149] Fifth, calculate the weighted average of the posterior probability of disease for all individuals within the monitoring area: ,in This represents the total number of individuals monitored. For the first The weight of each individual is determined by the geometric mean of video visibility and audio signal-to-noise ratio. The herd infection index is calculated in this embodiment. This indicates that approximately 12% of individuals in the current area are at risk of infection.

[0150] In step S130, the population infection index can be input into the SIR-Coccidia transmission dynamics model for chicken coccidiosis to simulate the epidemic evolution. Step S130 can specifically include steps S131 and S132.

[0151] Step S131: Model initialization.

[0152] Map the herd infection index to the initial number of infections. ,in Total population size (unit: individuals). This represents the number of infected individuals at the initial time (t=0). Initial size of the susceptible population. (Unit: individual oocysts). Environmental oocyst density. (Unit: oocysts / m²) Based on historical monitoring data, the initial value in this embodiment is 0 (assuming a new batch of chickens). Construct the environmental oocyst density. The system of differential equations:

[0153] ;

[0154] ;

[0155] .

[0156] Among them, the contact transmission rate , Only -1 ·sky -1 The baseline contact transmission rate is given, where H represents relative humidity (%); the humidity influence coefficient of 0.015 is expressed in % (%). -1 Environmental infection rate Only -1 ·indivual -1 ·m²·day -1 This indicates the infectivity of a unit oocyst density to a susceptible individual; recovery rate. sky -1 This indicates the daily recovery rate of infected individuals (i.e., the average infection period is 10 days); oocyst excretion coefficient. One -1 ·sky -1 (Number of oocysts released per infected chicken per day). In this example, for the sake of simplifying the calculation, the oocyst sporulation rate is calculated under the current ambient temperature of 25°C. Take constant value One -1 ·sky -1 ; Oocyte inactivation rate sky -1 Where T is the ambient temperature (unit: °C), this embodiment uses a simplified exponential model (temperature in degrees Celsius), the formula reflects the effect of temperature on oocyst survival (high temperature accelerates inactivation), and the baseline inactivation rate is 0.05 days. -1 The inactivation rate of oocytes at 20℃ has a temperature coefficient of 0.08℃. -1 This indicates that for every 1°C increase in temperature, the inactivation rate increases by approximately 8%.

[0157] Step S132: Threshold triggering.

[0158] Numerical simulations were performed to predict the development of the epidemic over the next 72 hours. The fourth-order Runge-Kutta method was used to solve the system of differential equations, with a time step of [missing information]. Hours. When simulation results show the future infection rate If an outbreak risk is detected, step S140 is automatically triggered. The total population is represented by 10%, which is the safety threshold. This means that when the infection rate exceeds 10% (i.e., 5,000 chickens), it is considered a risk of an outbreak.

[0159] In step S140, a preliminary strategy can be generated based on multi-objective optimization, and compliance matching verification can be performed. Step S140 may specifically include steps S141 to S143.

[0160] Step S141: Initial strategy generation.

[0161] Multi-objective optimization is performed using the NSGA-II algorithm, constructing three objective functions:

[0162] ;

[0163] ;

[0164] .

[0165] in, The number of infections after 72 hours (unit: chickens), and 15 represents the economic loss per chicken. For the unit cost of the drugs, in this example, sulfachlorpyrifos sodium is 0.08 yuan / g, diclazuril is 0.15 yuan / g, and toltrazuril is 0.22 yuan / g; This is a single dose (unit: g / animal). Calculated based on a body weight of 2kg, 10ppm corresponds to 0.02g / animal, and 30ppm corresponds to 0.06g / animal. The electricity cost is 0.6 yuan / kWh; The formula for calculating the increase in wind turbine power (unit: kW) is as follows: ,in This refers to the rated power of a single fan. Baseline rotational speed, The optimized rotational speed; The drug residue coefficients (dimensionless) for the selected drug types are: 1.0 for sulfachlorpyridazine sodium, 0.6 for diclazuril, and 0.4 for toltrazuril. The empirical withdrawal period (in days) for the selected drug type is 10 days for sulfachlorpyrifos sodium, 5 days for diclazuril, and 1 day for toltrazuril. This represents the remaining time (in days) between the current flock and the planned slaughter date; in this example, it is 7 days.

[0166] Decision variables include: wind turbine speed Drug types Dosage concentration .

[0167] Algorithm settings: population size 100, number of generations 50, crossover probability 0.9, mutation probability 0.1, Pareto front solution is selected by crowding distance sorting.

[0168] After the algorithm runs, it can obtain 15 Pareto optimal solutions. The system automatically selects the solution with the highest overall score as the initial strategy based on cost-benefit trade-offs.

[0169] Fan speed 80% ( ) + 20ppm sulfachlorpyrifos sodium (corresponding to );

[0170] Total cost Yuan.

[0171] Step S142: Compliance verification.

[0172] The system can automatically access a pre-set veterinary disease prevention and control knowledge base. It was detected that the current flock is 35 days old, with only 7 days remaining until the planned slaughter date, while the empirical withdrawal period for "sulfachlorpyrifos sodium" is 10 days.

[0173] Step S143: Adjust the strategy.

[0174] The system determined that the above strategy posed a risk of drug residue violations and triggered an alternative drug recommendation algorithm. The knowledge base recommended the alternative drug "diclazuril" (with a 5-day withdrawal period), generating a revised prevention and control strategy.

[0175] In step S150, risk prediction and dynamic adjustment can be performed on the modified strategy to generate the final instruction set. Step S150 may specifically include steps S151 and S152.

[0176] Step S151: Closed-loop prediction.

[0177] The revised prevention and control strategy was then substituted into the SIR-Coccidia transmission dynamics model for secondary simulation. The specific process included the following steps:

[0178] The first step is to convert the 80% fan speed in the revised control strategy into an environmental parameter change. Based on the empirical model of the farm's environmental control system, increasing the fan speed from 40% to 80% will result in a 5% decrease in humidity (from...). Down to ), the temperature rises by 2°C (from Rise to );

[0179] The second step is to update the parameters affected by humidity and temperature in the SIR-Coccidia transmission dynamics model of chicken coccidiosis, and calculate the updated contact transmission rate and the updated oocyst inactivation rate.

[0180] The updated contact transmission rate is:

[0181] Only -1 ·sky -1 ,

[0182] The updated follicle inactivation rate is:

[0183] sky -1 ;

[0184] The third step is to model the therapeutic effect of diclazuril (a drug in the revised prevention and control strategy) as an increase in the recovery rate: Among them, the basic rehabilitation rate Drug efficacy coefficient diclazuril concentration Therefore, the updated recovery rate (That is, the average infection period is shortened from 10 days to 2.5 days);

[0185] Fourth step, use the updated parameters , , The differential equations of the SIR-Coccidia transmission dynamics model in chickens were resolved using the fourth-order Runge-Kutta method with a time step of [missing information]. Hours, predicting the next 72 hours , , The curve showing the change.

[0186] The second simulation results show: the initial number of infections However, after 72 hours, the number of infections dropped. infection rate It can be confirmed that, under the premise of compliant drug use, the epidemic can be controlled within a safe range, and the number of infections is showing a downward trend. Figure 4 The study shows the trend of the number of infected individuals obtained through epidemic evolution simulation without intervention, as well as the trend of the number of infected individuals obtained through the modified prevention and control strategy.

[0187] Step S152: Instruction issued.

[0188] Therefore, when the results of the second simulation meet the preset conditions, the revised prevention and control strategy can be coded into a target early warning instruction set that includes environmental control instructions and drug administration recommendations, for example: {"Device":"Fan_Zone3","Action":"SetSpeed","Value":80},{"Device":"DosingPump","Action":"AddDrug","Type":"Diclazuril","Concentration":15}. The two instructions have the following meanings: The first instruction sets the fan speed of Fan Zone 3 to 80%, where the "Device" field specifies the target device as the fan in Fan Zone 3, the "Action" field specifies the operation to be set (SetSpeed), and the "Value" field specifies the speed value as 80 (unit: percentage). The second instruction adds the drug diclazuril to the dosing pump at a concentration of 15 ppm. Here, the "Device" field specifies the target device as the dosing pump, the "Action" field specifies the operation to be added (AddDrug), the "Type" field specifies the drug type as diclazuril, and the "Concentration" field specifies the concentration value as 15 (unit: ppm). This set of warning instructions can be sent to the farm's environmental control system via the MQTT protocol, enabling unmanned closed-loop management. Simultaneously, a yellow warning report is generated and pushed to the farm management personnel's mobile devices, including an epidemic situation map, control strategy descriptions, and expected effect assessments.

[0189] The effects of this embodiment include: 1. Multimodal verification, reducing false alarms: By fusing voiceprint and body features through a deep correlation matrix, physiological calls and pathological pain-induced calls can be effectively distinguished, solving the robustness problem of single-modal monitoring in complex environments. 2. Integration of mechanism and data-driven approaches: Not only is AI used for current status identification, but the SIR-Coccidia dynamic model is also introduced for future situation prediction, achieving a qualitative leap from alarm to early warning. 3. Automated closed-loop decision-making: The system can directly output control commands based on multi-objective optimization, realizing intelligent and unmanned aquaculture management, which can reduce the economic losses caused by coccidiosis.

[0190] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations fall within the scope of protection of this application.

[0191] In one exemplary embodiment, such as Figure 5 As shown, an early warning device for coccidiosis in chickens is provided, comprising:

[0192] The data processing module 501 is used to perform time alignment and spatial coordinate mapping on the acoustic and video data in the chicken farming environment to obtain a standardized monitoring data set with spatiotemporal consistency, so as to determine the acoustic and video data of the chickens in the current time window.

[0193] The feature vector acquisition module 502 is used to obtain the voiceprint feature vector and body posture feature vector of the chicken based on the sound wave data and video data of the chicken in the current time window.

[0194] Disease probability prediction module 503 is used to obtain the posterior probability of disease of the chicken in the current time window based on the voiceprint feature vector and body feature vector of the chicken.

[0195] The herd infection index determination module 504 is used to obtain the herd infection index of the monitoring area in the current time window based on the posterior probability of disease of each chicken in the monitoring area in the current time window.

[0196] The epidemic evolution simulation module 505 is used to input the population infection index into a preset chicken coccidiosis SIR-Coccidia transmission dynamics model to simulate the epidemic evolution at future time steps;

[0197] The strategy processing module 506 is used to generate a preliminary prevention and control strategy based on a multi-objective optimization algorithm when the epidemic evolution simulation result exceeds a preset safety threshold, and to perform compliance matching processing between the preliminary prevention and control strategy and a preset veterinary prevention and control knowledge base to obtain a revised prevention and control strategy.

[0198] The instruction set generation module 507 is used to perform risk prediction on the modified prevention and control strategy and generate a target early warning instruction set that includes environmental control instructions and drug administration recommendations.

[0199] In one exemplary embodiment, the data processing module 501 is configured to:

[0200] After time alignment of acoustic and video data from the chicken farming environment, noise reduction is performed on the acoustic data, and keyframe extraction is performed on the video data. Based on the transformation matrix between the spatial positioning coordinate system of the microphone array and the physical world coordinate system, and the transformation matrix between the pixel coordinate system of the camera and the physical world coordinate system, the noise-reduced acoustic data and the video data after keyframe extraction are mapped onto a unified three-dimensional spatial grid to obtain a standardized monitoring data set with spatiotemporal consistency. The acoustic data is collected by the microphone array, and the video data is collected by the camera.

[0201] In one exemplary embodiment, the disease probability prediction module 503 is used for:

[0202] A voiceprint abnormality index is obtained based on the voiceprint feature vector of the chicken, and a body posture abnormality index is obtained based on the body posture feature vector. A multimodal joint feature representation is obtained based on the voiceprint feature vector and the body posture feature vector. The multimodal joint feature representation, the voiceprint abnormality index, and the body posture abnormality index are used as observational evidence, and the environmental parameter data within the current time window are used as prior probability conditions. Bayesian inference is performed based on the observational evidence and the prior probability conditions to obtain the posterior probability of the chicken being diseased in the current time window.

[0203] In one exemplary embodiment, the epidemic evolution simulation module 505 is used for:

[0204] Based on the temperature and humidity data of the current time window, the parameters affected by temperature and humidity in the preset SIR-Coccidia transmission dynamics model of chicken coccidiosis are corrected to obtain the corrected SIR-Coccidia transmission dynamics model of chicken coccidiosis; the epidemic evolution of future time steps is simulated based on the population infection index and the corrected SIR-Coccidia transmission dynamics model of chicken coccidiosis.

[0205] In one exemplary embodiment, the policy processing module 506 is configured to:

[0206] A multi-objective optimization function is obtained, aiming to minimize the spread rate of the epidemic, minimize drug costs, and minimize the risk of drug residues. The decision variables in the multi-objective optimization function include fan speed, drug type, and drug concentration. Based on the multi-objective optimization function, the Pareto optimal frontier is searched in the solution space using the NSGA-II algorithm. The fan speed, drug type, and drug concentration values ​​corresponding to the optimal solution are selected to obtain the preliminary prevention and control strategy.

[0207] In an exemplary embodiment, the instruction set generation module 507 is configured to:

[0208] A second simulation is performed based on the revised prevention and control strategy and the SIR-Coccidia transmission dynamics model. When the results of the second simulation meet the preset conditions, a target early warning instruction set containing environmental control instructions and drug administration recommendations is obtained based on the revised prevention and control strategy.

[0209] The modules in the aforementioned early warning device for coccidiosis in chickens can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, 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.

[0210] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational 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 a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores the data involved in the aforementioned methods. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a method for early warning of coccidiosis in chickens.

[0211] In one exemplary embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in the various method embodiments described above.

[0212] In one exemplary embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the various method embodiments described above.

[0213] In one exemplary embodiment, a computer program product is provided having a computer program stored thereon, the computer program being executed by a processor of the steps described in the various method embodiments above.

[0214] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0215] Those skilled in the art will understand that all or part of the processes in 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, and when executed, it can include the processes of the embodiments of the above methods. The above embodiments only illustrate several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for early warning of coccidiosis in chickens, characterized in that, The method includes: Time alignment and spatial coordinate mapping were performed on acoustic and video data in the chicken farming environment to obtain a standardized monitoring data set with spatiotemporal consistency, so as to determine the acoustic and video data of chickens in the current time window; Based on the acoustic and video data of the chicken within the current time window, the voiceprint feature vector and body posture feature vector of the chicken are obtained. The abnormal voiceprint index is obtained based on the voiceprint feature vector of the chicken, and the abnormal body posture index is obtained based on the body posture feature vector. Based on the voiceprint feature vector and the body posture feature vector, a multimodal joint feature representation is obtained; The multimodal joint feature representation, the voiceprint anomaly index, and the body posture anomaly index are used as observational evidence, and the environmental parameter data within the current time window are used as prior probability conditions. Based on the observational evidence and the prior probability conditions, Bayesian inference is performed to obtain the posterior probability of the chickens being sick in the current time window. The population infection index of the monitoring area in the current time window is obtained based on the posterior probability of disease of each chicken in the monitoring area in the current time window. The population infection index is input into a pre-set SIR-Coccidia transmission dynamics model for avian coccidiosis to simulate the epidemic evolution at future time steps; When the simulation results of the epidemic evolution exceed the preset safety threshold, a multi-objective optimization function is obtained with the objectives of minimizing the epidemic spread rate, minimizing drug costs, and minimizing drug residue risks; the decision variables in the multi-objective optimization function include fan speed, drug type, and drug concentration. Based on the multi-objective optimization function, the Pareto optimal frontier is searched in the solution space using the NSGA-II algorithm. The fan speed, drug type, and drug concentration values ​​corresponding to the optimal solution are selected to obtain a preliminary prevention and control strategy. The preliminary prevention and control strategy is then matched with a pre-set veterinary prevention and control knowledge base for compliance processing to obtain a revised prevention and control strategy. A second simulation was conducted based on the revised prevention and control strategy and the aforementioned SIR-Coccidia transmission dynamics model for chicken coccidiosis. When the results of the second simulation meet the preset conditions, a target early warning instruction set containing environmental control instructions and drug administration recommendations is obtained according to the revised prevention and control strategy.

2. The method according to claim 1, characterized in that, Time alignment and spatial coordinate mapping were performed on acoustic and video data from chicken farming environments to obtain a standardized monitoring data set with spatiotemporal consistency, including: After time alignment of acoustic and video data from the chicken farming environment, noise reduction is performed on the acoustic data, and keyframe extraction is performed on the video data. Based on the transformation matrix between the spatial positioning coordinate system and the physical world coordinate system of the microphone array, and the transformation matrix between the pixel coordinate system and the physical world coordinate system of the camera, the denoised acoustic data and the video data after keyframe extraction are mapped to a unified three-dimensional spatial grid to obtain a standardized monitoring data set containing spatiotemporal consistency; wherein, the acoustic data is acquired through the microphone array, and the video data is acquired through the camera.

3. The method according to claim 1, characterized in that, The population infection index is input into a pre-set SIR-Coccidia transmission dynamics model for avian coccidiosis to simulate the epidemic evolution at future time steps, including: Based on the temperature and humidity data of the current time window, the parameters affected by temperature and humidity in the preset SIR-Coccidia transmission dynamics model of chicken coccidiosis are corrected to obtain the corrected SIR-Coccidia transmission dynamics model of chicken coccidiosis. The future time step of the epidemic evolution was simulated based on the aforementioned population infection index and the modified SIR-Coccidia transmission dynamics model.

4. The method according to any one of claims 1 to 3, characterized in that, Based on the posterior probability of disease in each chicken within the monitoring area during the current time window, the population infection index of the monitoring area during the current time window is obtained, including: Calculate the weighted average of the posterior probability of disease in all chickens within the monitoring area during the current time window to obtain the population infection index of the monitoring area during the current time window.

5. An early warning device for coccidiosis in chickens, characterized in that, The device includes: The data processing module is used to perform time alignment and spatial coordinate mapping on the acoustic and video data in the chicken farming environment to obtain a standardized monitoring data set with spatiotemporal consistency, so as to determine the acoustic and video data of chickens in the current time window. The feature vector acquisition module is used to obtain the voiceprint feature vector and body posture feature vector of the chicken based on the sound wave data and video data of the chicken in the current time window. The disease probability prediction module is used to obtain a voiceprint abnormality index based on the voiceprint feature vector of the chicken, and a body posture abnormality index based on the body posture feature vector; to obtain a multimodal joint feature representation based on the voiceprint feature vector and the body posture feature vector; to use the multimodal joint feature representation, the voiceprint abnormality index, and the body posture abnormality index as observational evidence, and to use environmental parameter data within the current time window as prior probability conditions; and to perform Bayesian inference based on the observational evidence and the prior probability conditions to obtain the posterior probability of the chicken being diseased in the current time window. The herd infection index determination module is used to obtain the herd infection index of the monitoring area in the current time window based on the posterior probability of disease of each chicken in the monitoring area in the current time window; The epidemic evolution simulation module is used to input the population infection index into a preset chicken coccidiosis SIR-Coccidia transmission dynamics model to simulate the epidemic evolution at future time steps; The strategy processing module is used to obtain a multi-objective optimization function with the objectives of minimizing the epidemic spread rate, minimizing drug costs, and minimizing drug residue risks when the simulation results of the epidemic evolution exceed a preset safety threshold. The decision variables in the multi-objective optimization function include fan speed, drug type, and drug concentration. Based on the multi-objective optimization function, the NSGA-II algorithm is used to search for the Pareto optimal frontier in the solution space, and the fan speed, drug type, and drug concentration values ​​corresponding to the optimal solution are selected to obtain a preliminary prevention and control strategy. The preliminary prevention and control strategy is then matched with a pre-set veterinary prevention and control knowledge base for compliance processing to obtain a revised prevention and control strategy. The instruction set generation module is used to perform a secondary simulation based on the modified prevention and control strategy and the SIR-Coccidia transmission dynamics model of chicken coccidiosis. When the secondary simulation results meet the preset conditions, a target early warning instruction set containing environmental regulation instructions and drug administration recommendations is obtained according to the modified prevention and control strategy.

6. The apparatus according to claim 5, characterized in that, The data processing module is used for: After time alignment of acoustic and video data from the chicken farming environment, noise reduction is performed on the acoustic data, and keyframe extraction is performed on the video data. Based on the transformation matrix between the spatial positioning coordinate system and the physical world coordinate system of the microphone array, and the transformation matrix between the pixel coordinate system and the physical world coordinate system of the camera, the denoised acoustic data and the video data after keyframe extraction are mapped to a unified three-dimensional spatial grid to obtain a standardized monitoring data set containing spatiotemporal consistency; wherein, the acoustic data is acquired through the microphone array, and the video data is acquired through the camera.

7. The apparatus according to claim 5, characterized in that, The epidemic evolution simulation module is used for: Based on the temperature and humidity data of the current time window, the parameters affected by temperature and humidity in the preset SIR-Coccidia transmission dynamics model of chicken coccidiosis are corrected to obtain the corrected SIR-Coccidia transmission dynamics model of chicken coccidiosis. The future time step of the epidemic evolution was simulated based on the aforementioned population infection index and the modified SIR-Coccidia transmission dynamics model.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

Images