Intelligent animal breeding management system and method

Abstract

The invention discloses an intelligent animal breeding management system and method, and relates to the technical field of animal breeding. A farm registration filing module automatically identifies a breeding scale and verifies farm filing information; the breeding file management module obtains environment information of a farm and updates a breeding file based on the environment information; the immunization and archive intelligent management module intelligently verifies vaccines, associates animal ear tags and generates electronic immunization records; the epidemic prevention material management module constructs a prediction model based on the epidemic prevention material historical use data, and automatically pushes a purchase prompt when the material inventory is lower than a preset threshold value; the epidemic situation analysis module receives animal symptom photos uploaded by workers, matches suspected epidemic diseases, pushes troubleshooting guidance, and predicts epidemic disease transmission risks in combination with breeding files and immune files. According to the invention, automation, standardization and precision of breeding management are realized, the problems of passive decision making, non-standard process, difficult record tracing and the like are solved, and the management efficiency and epidemic prevention level of the breeding industry are improved.

Description

Technical Field

[0001] This invention relates to the field of animal husbandry technology, and more specifically to an intelligent animal husbandry management system and method. Background Technology

[0002] Currently, my country's animal husbandry industry is accelerating its transformation towards large-scale, intensive, and standardized operations. In particular, the industry scale of specialty breeds such as poultry and Tan sheep continues to expand, becoming an important growth point for the agricultural economy. With consumers' increasing demands for food safety, traceability of the breeding process, standardized disease prevention, and data-driven management have become core requirements for the industry's high-quality development. To address these industry development needs, some research institutions are indirectly collecting vital signs and breeding information by implanting chips in animals ("targeting"), or introducing specialized medical equipment such as veterinary color Doppler ultrasound and digital X-ray imaging (DR) systems to assist in health monitoring. Simultaneously, IoT monitoring and network management platforms are beginning to be applied in breeding scenarios, attempting to achieve unified control over the breeding process.

[0003] However, most existing IoT management platforms operate in a passive "command-execution" mode, relying on experienced managers to lead breeding decisions. Incorrect decisions can lead to significant losses. Furthermore, the breeding process lacks standardized SOPs (Standard Operating Procedures), resulting in insufficient standardization of production processes and hindering unified management. In the implementation of the mandatory immunization "vaccinate first, subsidize later" policy, the link between vaccines and animals requires manual reporting, making immunization records prone to errors and difficult to trace. The cumbersome subsidy application process also negatively impacts farmers' enthusiasm.

[0004] Therefore, in view of the shortcomings of existing technologies, how to provide an intelligent animal breeding management system and method is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides an intelligent animal breeding management system and method, which realizes the automation, standardization and precision of breeding management, solves problems such as passive decision-making, non-standard processes and difficulty in tracing records, and improves the management efficiency and disease prevention level of the breeding industry.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent animal husbandry management system, comprising: The farm registration and filing module is used to automatically identify the scale of farming and verify the farm filing information; The aquaculture record management module is used to acquire environmental information of the aquaculture farm and update the aquaculture records based on the environmental information. The intelligent immunization and record management module is used for intelligent verification of vaccines, as well as linking them to animal ear tags, generating electronic immunization records, and updating immunization records synchronously. The epidemic prevention materials management module is used to build a prediction model based on historical usage data of epidemic prevention materials using an improved deep learning algorithm. The improved deep learning algorithm captures the sparse features of the data through a dynamic sparse connection network and optimizes the feature representation by combining a variational information bottleneck mechanism. The prediction model predicts the usage trend of epidemic prevention materials and automatically pushes a purchase reminder when the material inventory is lower than a preset threshold. The epidemic analysis module receives photos of animal symptoms uploaded by staff, matches suspected diseases using intelligent recognition technology, and pushes investigation guidelines. It also combines breeding records and immunization records to generate regional immunization rate heat maps and disease risk warning models through big data algorithms to predict the risk of disease transmission.

[0007] Preferably, the farm registration and filing module includes: a panoramic image upload and intelligent recognition unit, used to receive panoramic images of farms uploaded by farmers, and to construct a farm recognition model based on the YOLOv8n optimization algorithm. The YOLOv8n optimization algorithm replaces the ordinary convolutions in the backbone network with DSXConv convolution modules, and automatically identifies the scale of farming and facility type by combining the EAPIOU loss function.

[0008] The intelligent verification unit is used to automatically verify the access business license and breeding permit information of the farm through the blockchain cross-departmental data sharing mechanism; The filing and review unit is used to automatically link the farm identification results with the verification results and manually filled information, generate a filing and review list, and store it in the blockchain.

[0009] Preferably, the aquaculture record management module includes: An automatic environmental data acquisition unit is used to connect to the environmental sensor network of the farm to acquire environmental data of the farm in real time. The intelligent record update unit is used to automatically generate weekly reports on the aquaculture environment based on the collected environmental data and to update the aquaculture records on the blockchain simultaneously. The record traceability and query unit is used to store the breeding records in a chain according to the time dimension.

[0010] Preferably, the intelligent management module for immunization and records includes: a vaccine intelligent verification unit, which is used to connect to the vaccine traceability platform through blockchain and intelligently verify the authenticity, expiration date, and batch legality of the vaccine based on the scanned QR code on the vaccine packaging; The electronic immunization record generation unit is used to scan the animal ear tag ID, associate the verified vaccine information with the animal's identity, and generate an electronic immunization record; An immunization record update unit is used to upload the electronic immunization record to a blockchain immunization record repository.

[0011] Preferably, the prediction model is constructed using an improved deep learning algorithm, including: Collect historical usage data of epidemic prevention materials; Preprocess the collected historical usage data of epidemic prevention materials; Extract the core features of epidemic prevention materials from the preprocessed data to form a feature dataset X. feat ; The feature dataset X feat The data is input into a dynamic sparse connection network. By dynamically adjusting the network structure and connection coefficients, the sparse features in the data on the consumption of epidemic prevention materials are captured, and a preliminary feature representation is generated. Based on the preliminary feature representation, a variational information bottleneck mechanism is introduced to generate an optimized feature representation by balancing information compression and feature extraction. The optimized feature representation is input into the neural evolution strategy optimization algorithm, a population containing P neural network individuals is initialized, high-quality individuals are selected by calculating fitness, a new population is generated through crossover and mutation operations, the process is iterated to the preset number of generations, and the individual with the highest fitness is selected as the final prediction model.

[0012] Preferably, based on the preliminary feature representation, a variational information bottleneck mechanism is introduced to generate an optimized feature representation by balancing information compression and feature extraction, including: The initial feature representation Y DSCN As input, the mean function μ and variance function λ of the preliminary feature representation are computed through a fully connected layer: ; ; Among them, W μ and W λ Let b be the weight matrix. μ and b λ For bias terms; Introducing standard normal distribution noise Generate feature outputs containing random noise; A VIB loss function is constructed by combining reconstruction loss and KL divergence, and the loss function is then optimized. The VIB layer parameters are optimized by minimizing the loss function through gradient descent, and the optimized feature representation is output.

[0013] Preferably, a farm identification model is constructed based on the YOLOv8n optimization algorithm, including: For recognition targets of different sizes in panoramic images, dynamic snake convolution of different sizes is used to extract features from the input panoramic image through multiple branches, capturing feature information at different scales. The features extracted by each branch of the serpentine convolution are fused with background information through multi-scale feature cross-calculation to enhance the model's ability to perceive targets in complex scenes and generate attention maps. The attention map is concatenated by the Concat operation to obtain the fused feature value, which is then added to the original input feature value. After convolution, the information is compressed and fused to output the optimized feature value.

[0014] Preferably, an intelligent animal husbandry management method includes: Automatically identify the scale of breeding and verify the registration information of breeding farms; Obtain environmental information from the farm and update the aquaculture records based on the environmental information; Intelligent verification of vaccines, linking them to animal ear tags, generating electronic immunization records, and synchronously updating immunization profiles; Based on historical usage data of epidemic prevention materials, an improved deep learning algorithm is used to build a prediction model. The improved deep learning algorithm captures the sparse features of the data through a dynamic sparse connection network and optimizes the feature representation by combining a variational information bottleneck mechanism. The prediction model predicts the usage trend of epidemic prevention materials and automatically pushes a purchase reminder when the inventory of materials is lower than a preset threshold. The system receives photos of animal symptoms uploaded by staff, matches suspected diseases using intelligent identification technology, and pushes investigation guidelines. It also combines breeding records and immunization records to generate regional immunization rate heat maps and disease risk warning models through big data algorithms to predict the risk of disease transmission.

[0015] As can be seen from the above technical solution, compared with the prior art, this invention discloses an intelligent animal breeding management system and method. It uses blockchain technology as its core support, integrating improved YOLO algorithms, improved deep learning algorithms, and IoT sensing technology to construct an intelligent management system covering the entire lifecycle of a farm. The system leverages the distributed ledger, immutability, and trusted execution characteristics of blockchain to ensure the authenticity, security, and traceability of breeding data. Through intelligent algorithms, it achieves automated identification of breeding scenarios, data prediction, and intelligent early warning, replacing traditional manual operations and improving the efficiency, accuracy, and scientific nature of animal breeding management decisions. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of an intelligent animal breeding and management system provided by the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] This invention discloses an intelligent animal husbandry management system, such as... Figure 1 As shown, it includes: The farm registration and filing module is used to automatically identify the scale of farming and verify the farm filing information; The aquaculture record management module is used to acquire environmental information of the aquaculture farm and update the aquaculture records based on the environmental information. The intelligent immunization and record management module is used for intelligent verification of vaccines, as well as linking them to animal ear tags, generating electronic immunization records, and updating immunization records synchronously. The epidemic prevention materials management module is used to build a prediction model based on historical usage data of epidemic prevention materials using an improved deep learning algorithm. The improved deep learning algorithm captures the sparse features of the data through a dynamic sparse connection network and optimizes the feature representation by combining a variational information bottleneck mechanism. The prediction model predicts the usage trend of epidemic prevention materials and automatically pushes a purchase reminder when the material inventory is lower than a preset threshold. The epidemic analysis module receives photos of animal symptoms uploaded by staff, matches suspected diseases using intelligent recognition technology, and pushes investigation guidelines. It also combines breeding records and immunization records to generate regional immunization rate heat maps and disease risk warning models through big data algorithms to predict the risk of disease transmission.

[0020] Specifically, the farm registration and filing module includes a panoramic image upload and intelligent recognition unit, which receives panoramic images of farms uploaded by farmers and constructs a farm recognition model based on the YOLOv8n optimization algorithm. The YOLOv8n optimization algorithm replaces the ordinary convolution of the backbone network with the DSXConv convolution module to enhance the feature extraction capabilities of small facilities and scattered farming areas. Combined with the EAPIOU loss function, it automatically identifies the farming scale (such as the number of Tan sheep and the area of ​​the farming area) and facility type (such as pen structure, feeding equipment, and disease prevention facilities). The intelligent verification unit is used to automatically verify the business license and breeding permit information of the farm through the blockchain cross-departmental data sharing mechanism; it connects to the official interface of electronic certificates such as business licenses and breeding permits, and automatically verifies the authenticity and validity of the filing information through the blockchain cross-departmental data sharing mechanism to prevent false filings.

[0021] The registration and review unit automatically links farm identification results with verification results and manually entered information to generate a registration and review checklist, which is then stored on the blockchain. The blockchain permanently stores the review process and results, ensuring traceability and improving review efficiency and accuracy.

[0022] Specifically, the aquaculture record management module includes: The automatic environmental data acquisition unit is used to connect to the environmental sensor network of the farm to acquire environmental data of the farm in real time; it captures environmental data such as temperature, humidity, ammonia concentration, and light intensity of the farm in real time, and realizes encrypted transmission and storage of data through the blockchain subnet to ensure that the data is not tampered with.

[0023] The intelligent record update unit is used to automatically generate a weekly aquaculture environment report based on the collected environmental data. The weekly aquaculture environment report includes environmental data trend analysis and outlier annotation; and is synchronously updated to the aquaculture record on the blockchain; replacing traditional manual filling and reducing human error.

[0024] The record traceability and query unit is used to store the breeding records in a chain according to the time dimension. It supports searching by breeding stage, animal batch, environmental indicators, etc. All modification records are logged to ensure the integrity and reliability of the records.

[0025] Specifically, the immunization and record intelligent management module includes: a vaccine intelligent verification unit, which is used to connect to the vaccine traceability platform through blockchain and intelligently verify the authenticity, expiration date, and batch legality of the vaccine based on the scanned vaccine packaging QR code; The electronic immunization record generation unit is used to scan the animal's ear tag ID, associate the verified vaccine information with the animal's identity, and generate an electronic immunization record; the electronic immunization record includes core information such as: immunization time, vaccine type, vaccination personnel, and ear tag ID; The immunization record update unit is used to upload the electronic immunization records to the blockchain immunization record repository. A Trusted Execution Environment (TEE) is used to ensure data privacy, while simultaneously synchronizing the data with the livestock management department's monitoring platform to achieve real-time sharing and traceability of immunization data.

[0026] Specifically, improved deep learning algorithms are used to build predictive models, including: Collect historical usage data of epidemic prevention materials; the historical usage data of epidemic prevention materials includes related data such as material type (vaccines, disinfectants, etc.), usage time, usage quantity, consumption scenario (sheep farms / poultry farms, etc.), breeding scale, immunization cycle, regional epidemic level, and seasonal factors.

[0027] The collected historical usage data of epidemic prevention materials are preprocessed. The preprocessing includes data cleaning and data normalization. The data cleaning includes: removing duplicate records and data with missing key fields. For abnormal consumption values ​​caused by sudden outbreaks (such as extreme values ​​that far exceed the normal usage), the IQR method (interquartile range) is used to identify them and the data is replaced or truncated by the median to avoid outliers interfering with model training.

[0028] The data normalization includes: using the Min-Max normalization method to map feature data of different magnitudes (such as breeding scale, usage quantity) to the [0,1] interval to eliminate dimensional differences. The formula is: Xnorm=(X-Xmin) / (Xmax-Xmin), where X is the original data, and Xmax and Xmin are the maximum and minimum values ​​of the feature, respectively.

[0029] Extract the core features of epidemic prevention materials from the preprocessed data to form a feature dataset X. feat The feature dataset X feat include: Time characteristics: the month and quarter corresponding to the time of use, whether it is during the peak period of immunity, and whether it is during the season with a high incidence of disease; Scene characteristics: type of breeding (sheep / poultry, etc.), scale of breeding (number of animals in stock), and regional epidemic risk level; Material characteristics: material type, unit usage, immunization cycle (for vaccines), and expiration date; Related features: historical consumption growth rate, remaining inventory, and last purchase quantity.

[0030] The feature dataset X feat The data is input into a dynamic sparse connection network. By dynamically adjusting the network structure and connection coefficients, the sparse features in the consumption data of epidemic prevention materials are captured. The sparse features refer to the consumption of some materials only during the immunization cycle and the non-periodic consumption during sudden outbreaks, generating a preliminary feature representation. The computational process of dynamic sparse connected networks includes: ; Among them, Y DSCN This is the initial feature representation of the DSCN output, where σ is the ReLU activation function and α is... i The sparse connection coefficients are dynamically adjusted through network training; when the sparse coefficients approach 0, the corresponding feature connections are suppressed. Let i be the sparse weight matrix. For the i-th part of the feature dataset, b DSCN This is a bias term.

[0031] By using a dynamic sparsity mechanism to ignore redundant features and noise, the model focuses on key influencing factors such as the immunization cycle and the scale of aquaculture, thereby improving the model's ability to process sparse data.

[0032] Based on the preliminary feature representation, a variational information bottleneck mechanism is introduced to generate an optimized feature representation by balancing information compression and feature extraction. The optimized feature representation is input into the neural evolution strategy optimization algorithm, and a population containing P neural network individuals (each individual corresponds to a different network structure and parameter settings) is initialized. High-quality individuals are selected by calculating fitness (prediction error), and a new population is generated through crossover and mutation operations. The process is iterated to the preset number of generations, and the individual with the highest fitness is selected as the final prediction model.

[0033] ; in, Let Y be the optimal model parameters, L be the MSE loss function, and Y be the optimal model parameters. true This represents the actual consumption of materials.

[0034] Construct a variational Bayesian validation model, dynamically adjust the validation set distribution through variational inference, and maximize the variational lower bound L. ELBO Optimize model parameters and variational parameters to ensure the model's predictive stability on unknown data: ; Generate adversarial examples (adding small perturbations to the original feature data), evaluate the model's prediction accuracy under perturbed data, adjust model parameters to improve robustness, and ensure prediction reliability in extreme scenarios such as sudden outbreaks of epidemics.

[0035] Input the characteristic data of the period to be predicted (such as the scale of breeding, immunization plan and seasonal factors in the next month) into the trained model, and output the predicted total consumption, average weekly consumption and peak consumption time of various epidemic prevention materials in the period.

[0036] By combining the forecast results with the preset inventory threshold, when the inventory falls below the threshold after the forecast consumption, a purchase reminder is automatically triggered, and a suggested purchase quantity is pushed. The suggested purchase quantity is calculated based on the forecast consumption trend and the safety stock level.

[0037] This invention addresses the need for predicting the consumption trends of epidemic prevention materials. It constructs a prediction model based on an improved deep learning algorithm, capturing sparse features from historical usage data of these materials through a Dynamic Sparse Connection Network (DSCN). This model then optimizes feature representation using a Variational Information Bottleneck (VIB) mechanism, accurately predicting future usage trends and providing data support for intelligent inventory early warning and procurement decisions. The model balances data sparsity handling, key feature extraction, and generalization capabilities, making it suitable for predicting the consumption of various epidemic prevention materials such as vaccines, disinfectants, and protective clothing.

[0038] Dynamic sparse connection networks can adaptively ignore redundant information, focusing on key influencing factors of epidemic prevention material consumption and adapting to scenarios with sparse data. The variational information bottleneck mechanism filters core features through information compression, improving model generalization ability and reducing the risk of overfitting. This invention integrates scenario features such as breeding type and immunization cycle, and can be specifically adapted to the prediction needs of epidemic prevention materials in different scenarios such as sheep farms and poultry farms.

[0039] Specifically, based on the preliminary feature representation, a variational information bottleneck mechanism is introduced to generate an optimized feature representation by balancing information compression and feature extraction, including: The initial feature representation Y DSCN As input, the mean function μ and variance function λ of the preliminary feature representation are computed through a fully connected layer: ; ; Among them, W μ and W λ Let b be the weight matrix. μ and b λ For bias terms; Introducing standard normal distribution noise This enhances the model's generalization ability and generates feature outputs containing random noise. Calculate the output of the VIB layer: .

[0040] The VIB loss function is constructed by combining reconstruction loss and KL divergence, and then optimized; a balance is struck between information compression and feature fidelity. ; Among them, L recon The reconstruction loss is used to measure the similarity between the output and the input, where β is the weight parameter, and q(Z|Y) = 1 / 2. DSCN Let p(Z) be the variational posterior distribution, p(Z) be the prior distribution, and D be the variational posterior distribution. KL This is the KL divergence, used to measure the difference between two distributions.

[0041] The VIB layer parameters are optimized by minimizing the loss function through gradient descent, and the optimized feature representation is output.

[0042] Specifically, a farm identification model is constructed based on the YOLOv8n optimization algorithm, including: The C2f-DSM module is formed by replacing the ordinary convolution in the YOLOv8n backbone network C2f with the DSXConv convolution module, which enhances the ability to extract features from small facilities (such as waterers and disinfection equipment), scattered breeding areas (such as multiple independent pens and outdoor breeding areas) and individual Tan sheep in the farm scene.

[0043] Multi-scale serpentine convolution feature extraction: For recognition targets of different sizes in the panoramic image, four different sizes of dynamic snake convolution (DsConv) of 3×3, 5×5, 7×7 and 9×9 are used to extract the feature quantity x of the input panoramic image in multiple branches to capture feature information at different scales. ; Among them, DsConv i This represents a dynamic serpentine convolution with a kernel size of i×i. The kernel is optimized through continuity constraints and iterative strategies to enhance feature capture of slender structures (such as fences and feeding troughs) and scattered targets (such as free-range sheep).

[0044] Multidimensional Collaborative Attention Module (MCA) Fusion: The features extracted by each branch of the serpentine convolution are input into the MCA module. Through multi-scale feature cross-computation and fusion with background information, the model's ability to perceive targets in complex scenes is enhanced, and attention maps xc1, xc2, xc3, and xc4 are generated. ; Feature fusion and output: The attention map is concatenated through a concat operation to obtain a fused feature quantity Y. This Y is then added to the original input feature quantity x, followed by 1×1 convolution for information compression and fusion, outputting the optimized feature quantity. This improves the completeness of feature representation. Y = Concat(xc1 + xc2 + xc3 + xc4); Output = Conv1(Y+x).

[0045] Multi-scale convolutional kernels are adapted to extract features from individual sheep (small targets), pen structures (medium targets), and breeding area boundaries (large targets); the MCA module solves the problem of scattered targets and complex backgrounds in panoramic images, reducing interference from irrelevant backgrounds such as trees and terrain.

[0046] The EAPIOU loss function is adopted to replace the default YOLOv8n loss function, which enhances the matching penalty mechanism for target position, aspect ratio and angle, and improves the accuracy of breeding scale statistics and facility type identification, especially for scenarios such as counting the number of Tan sheep and calculating the area of ​​irregular pens.

[0047] The loss function expression is as follows: ; Among them: B p and B g These are the predicted bounding box and the ground truth bounding box (e.g., the bounding box of an individual sheep or the boundary box of a pen); d is the Euclidean distance between the center coordinates of the predicted bounding box and the ground truth bounding box, and c is the diagonal distance of the minimum bounding rectangle, which improves the accuracy of position matching. , Penalties for width and height differences are applied to the size recognition of regular / irregular facilities such as pens and feeding equipment. An angle penalty term is used to address matching errors caused by target tilting in panoramic images (such as obliquely placed epidemic prevention equipment or the posture differences of free-range sheep).

[0048] This invention can reduce the problems of missed detection and duplicate counting when counting Tan sheep; improve the accuracy of area measurement for irregular breeding areas (such as mountain Tan sheep breeding areas); and optimize the classification accuracy of different types of facilities (such as open pens and closed pens).

[0049] In one specific embodiment of the present invention, (i) a dataset is constructed to collect panoramic images of farms of different breeding types (sheep, poultry, livestock, etc.) and different scenes (plains, mountains, sheds), covering different lighting conditions (sunny days, cloudy days) and time periods (early morning, afternoon), and a total of 15,000 original images are collected, of which 5,000 are panoramic images of sheep farms.

[0050] Use the LabelImg tool to annotate the image. The annotation content includes: Related to breeding scale: individual Tan sheep (labeled as a single animal), breeding area boundary (labeled as a polygon); Related to facility type: 8 categories of facility labels are defined, including pen structure (categorized as open / closed), feeding equipment, disease prevention facilities (disinfection equipment, isolation pens), and drinking water equipment.

[0051] The collected data undergoes preprocessing, specifically including: Image enhancement: Randomly crop (simulate panoramic image segments from different shooting angles), flip, rotate, and adjust the brightness of the training set images to expand the dataset to 100,000 images and improve the model's generalization ability; Normalization: The image size is uniformly adjusted to 640×640, and the pixel values ​​are normalized to the [0,1] range to eliminate dimensional differences; Dataset partitioning: Divide the dataset into training, validation, and test sets in a 7:1:2 ratio to ensure even data distribution.

[0052] (II) Model Training and Optimization

[0053] Training environment: Operating system Ubuntu 20.04, GPU model NVIDIA RTX A6000, deep learning framework PyTorch 1.11.0, optimizer Adam, learning rate dynamically adjusted (initial learning rate 0.001, decaying by 10% every 50 rounds).

[0054] Training parameters: number of iterations (epochs) 200, training batch (Batch Size) 8, weight decay coefficient 0.0005, to prevent model overfitting.

[0055] Optimization strategy: During training, monitor the precision, recall, and mean average precision (mAP 0.5) of the validation set. If there is no improvement in mAP 0.5 on the validation set for 10 consecutive rounds, trigger the early stopping mechanism to save the optimal model parameters.

[0056] In one specific embodiment of the present invention, the intelligent recognition process includes: Farmers can upload panoramic photos of their farms via a mobile app, supporting JPG and PNG formats, with an image resolution of at least 1920×1080.

[0057] The server resizes and normalizes the uploaded image to generate a feature map that adapts to the model input.

[0058] The preprocessed feature map is input into the optimized YOLOv8n model, features are extracted using the C2f-DSM module, target matching is optimized using the EAPIOU loss function, and the recognition result is output. The scale of breeding includes: the number of Tan sheep in stock, based on the count of individual detection boxes; and the area of ​​the breeding area, calculated based on the area of ​​the bounding box polygon. Facility types include: identifying the location coordinates, category, and confidence level of various facilities, with a confidence threshold set to 0.7, and filtering out low-confidence identification results.

[0059] The identification results are linked with the electronic certificate verification information to generate a filing and review list, which is then synchronized to the blockchain storage to ensure that the identification results are traceable.

[0060] This invention addresses the characteristics of Tan sheep—small individual size and free-range distribution—by using multi-scale convolutional kernels to improve individual detection accuracy and reduce background interference from mountains and grasslands. It supports the identification of facilities and scales for different breeding types such as Tan sheep, poultry, and livestock, and adaptively adjusts feature extraction weights through labels. Automatic identification results replace manual statistics, reducing the single farm registration review time from 2 hours to 15 minutes and lowering review costs.

[0061] The panoramic image upload and intelligent recognition unit is the core functional unit of the farm registration and filing module. It supports farmers to upload panoramic images of their farms (including scenes specific to Tan sheep farms) via mobile devices. Based on the lightweight YOLOv8n model, it retains its efficient inference advantages and optimizes for the large-scene, multi-target, and small-sample features of farm panoramic images. Core improvements include replacing the DSXConv convolution module and introducing the EAPIOU loss function, balancing recognition speed and accuracy, and adapting to the processing needs of images uploaded from mobile devices. An intelligent recognition model for farms is built based on the optimized YOLOv8n algorithm. The DSXConv convolution module enhances the feature extraction capabilities of small facilities and scattered farming areas, and the EAPIOU loss function improves recognition accuracy, enabling automatic recognition of farming scale (such as the number of Tan sheep and the area of ​​the farming area) and facility type (such as pen structure, feeding equipment, and disease prevention facilities). This provides data support for filing and review, replacing the basic information statistics work in manual on-site verification.

[0062] In one specific embodiment of the present invention, an intelligent animal husbandry management method includes: Automatically identify the scale of breeding and verify the registration information of breeding farms; Obtain environmental information from the farm and update the aquaculture records based on the environmental information; Intelligent verification of vaccines, linking them to animal ear tags, generating electronic immunization records, and synchronously updating immunization profiles; Based on historical usage data of epidemic prevention materials, an improved deep learning algorithm is used to build a prediction model. The improved deep learning algorithm captures the sparse features of the data through a dynamic sparse connection network and optimizes the feature representation by combining a variational information bottleneck mechanism. The prediction model predicts the usage trend of epidemic prevention materials and automatically pushes a purchase reminder when the inventory of materials is lower than a preset threshold. The system receives photos of animal symptoms uploaded by staff, matches suspected diseases using intelligent identification technology, and pushes investigation guidelines. It also combines breeding records and immunization records to generate regional immunization rate heat maps and disease risk warning models through big data algorithms to predict the risk of disease transmission.

[0063] In a specific embodiment of the present invention, taking the breeding of Tan sheep in a certain region (including scenarios such as free-range breeding in mountains and intensive breeding in sheds) as an example, intelligent management and control of the entire process of registration, archives, immunization, materials, and epidemics is realized, which is suitable for the breeding characteristics of Tan sheep, such as small individual size, free-range distribution, and clear epidemic prevention needs.

[0064] I. System Overall Architecture

[0065] Hardware layer: mobile APP (for farmers / staff), environmental sensor network, RFID ear tags and reading / writing devices, server cluster, blockchain alliance node devices.

[0066] Software layer: The core algorithm model is built based on PyTorch 1.11.0, using the Ubuntu 20.04 operating system, and data transmission is achieved through Zigbee / GPRS. The trusted execution environment (TEE) is built using ARMTrustZone.

[0067] Data Layer: Integrates blockchain-stored filing / archive / immunization data, real-time sensor environmental data, historical data on epidemic prevention materials, and regional epidemic data, and connects with the national veterinary drug traceability system and local government data chains.

[0068] II. Implementation Details of Core Modules

[0069] (I) Farm Registration and Filing Module

[0070] 1. Panoramic image upload and intelligent recognition

[0071] Farmers can upload panoramic images of their farms via a mobile app, covering key areas such as pens, outdoor farming areas, and disease prevention facilities, adapting to different terrains such as mountains and plains.

[0072] The optimized YOLOv8n model was adopted, and the C2f-DSM module was used to enhance the feature extraction of individual Tan sheep (small target), scattered pens (medium target), and breeding area boundaries (large target). The EAPIOU loss function was combined to reduce the counting error caused by mountain background interference.

[0073] The system automatically outputs recognition results, including the number of Tan sheep, the area of ​​the breeding area, and the type of facilities (8 categories of labels such as open / closed pens, feeding troughs, disinfection equipment, and isolation pens).

[0074] 2. Cross-departmental intelligent verification

[0075] Establish a regional agricultural blockchain alliance chain to connect with local agricultural and rural affairs bureaus, market supervision bureaus, government service centers, and other departments to share electronic license interfaces such as business licenses and breeding permits.

[0076] The system automatically verifies the consistency between the uploaded information and the blockchain-stored evidence, focusing on verifying whether the scope of the breeding permit includes Tan sheep breeding and the validity period of the permit, thus preventing false registration.

[0077] 3. Filing review and evidence preservation

[0078] The system automatically associates identification results, verification results, and manually entered information (such as the person in charge of breeding and contact information) to generate a standardized filing and review checklist.

[0079] The review process and results are permanently stored on the blockchain, supporting traceability and ensuring that the registration information cannot be tampered with.

[0080] (II) Aquaculture Record Management Module

[0081] 1. Automatic collection of environmental data

[0082] Deploy a low-power sensor network for different environments: every 50m within the enclosure 2 Install one temperature, humidity, and ammonia concentration sensor in the outdoor breeding area, with a coverage area of ​​500m. 2 Light intensity / temperature and humidity sensors are deployed at intervals, supporting Zigbee wireless transmission (mountainous scenarios) and GPRS backup transmission.

[0083] Real-time data collection: temperature, humidity, ammonia concentration, light intensity, and encrypted transmission and storage via a blockchain subnet.

[0084] 2. Intelligent file update

[0085] The system automatically generates weekly reports on the aquaculture environment, including data trend curves and outlier annotations (such as triggering an alert when ammonia concentration is >25ppm), and updates them synchronously to the blockchain aquaculture archive.

[0086] Files are categorized and stored according to the breeding stages of Tan sheep (lamb stage, fattening stage, and breeding stage), and linked to batch information to replace manual entry and reduce human error.

[0087] 3. Archival traceability and inquiry

[0088] Supports multi-dimensional retrieval: Search archives by breeding stage, batch number, environmental indicator threshold, time range, etc. All modification records are kept, ensuring traceability of the entire cycle of Tan sheep breeding.

[0089] (III) Intelligent Management Module for Immunization and Records

[0090] 1. Intelligent verification of vaccines

[0091] By connecting to the national veterinary drug traceability system via blockchain, staff can scan the QR code on vaccine packaging to automatically verify authenticity, expiration date, and batch legality, with a focus on mandatory immunization vaccines for Tan sheep (foot-and-mouth disease vaccine, small ruminant disease vaccine, Brucella vaccine) and commonly used vaccines (trivalent and quadrivalent vaccine, sheep pox vaccine).

[0092] 2. Generation of electronic immunohistographs

[0093] Using RFID ear tags (compliant with national animal identification standards), scanning the ear tag ID automatically links to verified vaccine information and generates an electronic record containing core information such as immunization time, vaccine type, recipient, ear tag ID, and immunization site.

[0094] In response to the concentrated immunization of Tan sheep in spring and autumn, batch scanning and data entry are supported to improve immunization efficiency.

[0095] 3. Immunization record updates and sharing

[0096] Electronic immunization records are encrypted and uploaded to a blockchain-based immunization archive via a TEE built by ARMTrustZone, ensuring data privacy.

[0097] Real-time synchronization with the local animal husbandry management department's monitoring platform automatically marks the expiration date of vaccines with a long immunity period (3 years), such as peste des petits ruminants, and supports supplementary immunization early warning.

[0098] (iv) Epidemic Prevention Materials Management Module

[0099] 1. Data Acquisition and Preprocessing

[0100] Collect historical data on epidemic prevention materials for more than 3 years: material type (foot-and-mouth disease vaccine, triple and quadruple vaccine, disinfectant, protective clothing, etc.), usage time, quantity, breeding scale, immunization cycle (e.g., foot-and-mouth disease is immunized once every 6 months), regional epidemic level, and seasonal factors (spring and autumn immunization peak season, winter high incidence of enterotoxemia).

[0101] Data preprocessing: Duplicate / missing data were removed, and IQR was used to identify abnormal consumption values ​​caused by sudden outbreaks of disease (such as the surge in vaccine usage during outbreaks of peste des petits ruminants), which were then replaced by the median. Min-Max normalization was used to map the scale of breeding, usage, etc., to the [0,1] interval.

[0102] 2. Feature Extraction and Model Training

[0103] Construction of core feature dataset: Time characteristics: month, quarter, whether it is the peak immunization period in March-April / September-October, whether it is the peak season for winter diseases; Scene characteristics: type of breeding (Tan sheep), number of sheep in stock, regional epidemic risk level; Material characteristics: vaccine type (mandatory / common), unit dosage, immunization cycle, and expiration date; Related features: historical consumption growth rate, remaining inventory, and last purchase quantity.

[0104] Model training: The feature dataset is input into the Dynamic Sparse Connection Network (DSCN) to capture sparse features (such as vaccines that are only consumed during the immune period). The feature representation is optimized through the variational information bottleneck (VIB) mechanism. The optimal prediction model is generated through 200 iterations of the neuroevolutionary strategy algorithm. Variational Bayesian validation is used to ensure generalization ability.

[0105] 3. Material Forecasting and Procurement Reminders

[0106] Input data such as the scale of breeding and immunization plan for the next 1-3 months, and the model will output the predicted total consumption, average weekly consumption and peak time of various materials.

[0107] Set a safety stock threshold. When the predicted inventory is lower than the threshold, automatically send a purchase reminder to the farm and regional material suppliers, and clearly suggest the purchase quantity.

[0108] (v) Epidemic Analysis Module

[0109] 1. Symptom Identification and Screening Guidelines

[0110] Staff members upload photos of sheep symptoms (such as oral necrosis, diarrhea, and fever) via the APP. The system uses an intelligent identification model to match suspected diseases (such as peste des petits ruminants, enterotoxemia, and brucellosis) and pushes screening guidelines (symptom comparison, sampling methods, and isolation requirements).

[0111] The model training dataset was supplemented with photos of local Tan sheep disease cases, covering different lighting and shooting angles to improve recognition accuracy.

[0112] 2. Regional risk early warning

[0113] By combining breeding records and immunization records, a township-level immunization rate heat map is generated using big data algorithms, which intuitively displays the distribution of Tan sheep that have not been immunized or whose immunization has expired.

[0114] Construct an epidemic risk early warning model, input regional immunization rate, environmental data, and historical epidemic data, and output high / medium / low risk levels. Trigger a transmission risk warning for areas with an immunization rate of less than 80% and push enhanced immunization recommendations.

[0115] This invention utilizes blockchain technology as its core support, integrating improved YOLO algorithms, improved deep learning algorithms, and IoT sensing technology to construct an intelligent management system covering the entire lifecycle of a livestock farm. The system leverages the distributed ledger, immutability, and trusted execution characteristics of blockchain to ensure the authenticity, security, and traceability of livestock data. Intelligent algorithms enable automated identification of livestock scenarios, data prediction, and intelligent early warning, replacing traditional manual operations and improving the efficiency, accuracy, and scientific basis of animal husbandry management, especially providing customized support for specialty livestock breeds such as Tan sheep.

[0116] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0117] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent animal breeding management system, characterized in that, The application comprises: a farm registration and filing module for automatically identifying the scale of breeding and verifying the farm filing information; a breeding archive management module for obtaining environmental information of the farm and updating the breeding archive based on the environmental information; an immune and archive intelligent management module for intelligently verifying vaccines, associating animal ear tags, generating electronic immune records, and synchronously updating immune archives; an epidemic prevention material management module for constructing a prediction model using an improved deep learning algorithm based on historical use data of epidemic prevention materials, wherein the improved deep learning algorithm captures data sparsity features through a dynamic sparse connection network and optimizes feature representation by combining a variational information bottleneck mechanism; predicting the use trend of epidemic prevention materials through the prediction model and automatically pushing procurement reminders when the material inventory is below a preset threshold; an epidemic situation analysis module for receiving animal symptom photos uploaded by staff, matching suspected epidemics through intelligent recognition technology and pushing investigation guidelines, and generating a regional immune rate heat map and an epidemic risk warning model through big data algorithms based on breeding archives and immune archives to predict epidemic transmission risks.

2. The intelligent animal breeding management system according to claim 1, characterized in that, The farm registration and filing module comprises a panoramic image uploading and intelligent recognition unit for receiving panoramic images of the farm uploaded by breeders, constructing a farm identification model based on a YOLOv8n optimization algorithm, and automatically identifying the scale of breeding and the type of facilities by replacing the normal convolution of the backbone network with a DSXConv convolution module and combining an EAPIOU loss function; an intelligent verification unit for automatically verifying the access business license and breeding license information of the farm through a blockchain cross-department data sharing mechanism; a filing review unit for automatically associating the identification results and verification results of the farm with manually reported information, generating a filing review list, and storing it in the blockchain.

3. The intelligent animal breeding management system according to claim 1, wherein, The breeding archive management module comprises: an environmental data automatic collection unit for interfacing with the environmental sensor network of the farm and obtaining environmental data of the farm in real time; an intelligent archive update unit for automatically generating a breeding environment weekly report based on the collected environmental data and synchronously updating the breeding archive on the blockchain; an archive traceability query unit for chain-storing the breeding archive by time dimension.

4. The intelligent animal breeding management system according to claim 1, characterized in that, The immune and archive intelligent management module comprises: a vaccine intelligent verification unit for connecting a vaccine traceability platform through a blockchain, intelligently verifying the authenticity, validity period, and batch legality of vaccines according to the scanned vaccine package QR code; an electronic immune record generation unit for scanning animal ear tag IDs, associating the verified vaccine information with animal identities, and generating electronic immune records; 5. The intelligent animal breeding management system according to claim 1, wherein, an immune archive update unit for uploading the electronic immune records to the blockchain immune archive library. The improved deep learning algorithm for constructing the prediction model comprises: collecting historical use data of epidemic prevention materials; Extract the core features of epidemic prevention materials from the preprocessed data to form a feature dataset X feat ; The feature dataset X feat is input to a dynamic sparse connection network, sparse features in the epidemic prevention material consumption data are captured by dynamically adjusting the network structure and the connection coefficient, and a preliminary feature representation is generated. preprocessing the collected historical use data of epidemic prevention materials; based on the preliminary feature representation, introducing a variational information bottleneck mechanism to balance information compression and feature extraction, and generating an optimized feature representation; The optimized feature representation is input into a neural evolution strategy optimization algorithm, a population containing P neural network individuals is initialized, high-quality individuals are screened by calculating the fitness, a new population is generated through crossover and mutation operations, and the iteration is performed to the preset evolution generation, and the individual with the highest fitness is selected as the final prediction model.

6. The intelligent animal breeding management system according to claim 5, wherein, Based on the preliminary feature representation, a variational information bottleneck mechanism is introduced, and through the balance of information compression and feature extraction, an optimized feature representation is generated, including: Compute the preliminary feature representation Y DSCN As input, compute the mean function μ and the variance function λ of the preliminary feature representation through a fully connected layer: ； ； where W μ and W λ are weight matrices, b μ and b λ are bias terms; introducing standard normal distribution noise to generate a feature output containing random noise; The VIB loss function is constructed by combining the reconstruction loss and the KL divergence, and the loss function is optimized; The VIB layer parameters are optimized by gradient descent to minimize the loss function, and the optimized feature representation is output.

7. The intelligent animal breeding management system according to claim 2, wherein, Based on the YOLOv8n optimization algorithm, a farm identification model is constructed, including: For different sizes of identification targets in the panoramic image, dynamic snake convolution of different sizes is used to extract multi-branch features of the input panoramic image, and capture feature information of different scales; The feature quantities extracted by each branch snake convolution are fused through multi-scale feature cross calculation and background information, and the model's perception ability to targets in complex scenes is enhanced, and the attention map is generated by Concat operation to concatenate the channels of the attention map to obtain the fused feature quantity, which is then added to the original input feature quantity, and the information is compressed and fused through convolution to output the optimized feature quantity.

8. The intelligent animal breeding management method is applied to the intelligent animal breeding management system of any one of claims 1-7, characterized in that, Including: Automatically identify the scale of breeding, and verify the record information of the farm; Obtain the environmental information of the farm, and update the breeding file based on the environmental information; Intelligently verify the vaccine and associate the animal ear tag to generate an electronic immune record and update the immune file synchronously; Based on the historical use data of epidemic prevention materials, an improved deep learning algorithm is used to construct a prediction model, which captures data sparse features through a dynamic sparse connection network and optimizes feature representation through a variational information bottleneck mechanism; The use trend of epidemic prevention materials is predicted through the prediction model, and when the material inventory is lower than the preset threshold, an automatic procurement reminder is pushed; Receive the animal symptom photos uploaded by the staff, match the suspected epidemic through intelligent recognition technology and push the investigation guide, and at the same time, generate a regional immune rate heat map and an epidemic risk early warning model through big data algorithm based on the breeding file and the immune file, and predict the epidemic spread risk.

Images