A system for disease prevention, traceability, and risk management throughout the entire egg-laying chicken breeding process.

Abstract

This invention relates to the field of poultry farming technology, specifically to a full-process disease prevention traceability and risk management system for egg-laying breeder chickens. The system includes: a multimodal data acquisition module, an edge computing preprocessing module, and an intelligent risk management module. The multimodal data acquisition module collects multimodal data during the chicken flock's rearing process and transmits it to the edge computing preprocessing module. The edge computing preprocessing module receives the multimodal data, preprocesses it, and sends the standardized data to the intelligent risk management module for further processing. The intelligent risk management module includes a fusion prediction model, an abnormal behavior recognition system, and a multi-level dynamic disease prevention strategy engine. The fusion prediction model predicts the probability of disease outbreaks in the flock. The abnormal behavior recognition system outputs the disease risk level of the flock. The multi-level dynamic disease prevention strategy engine responds accordingly based on the disease risk level. This invention's system can solve the problems of insufficient multimodal fusion, inaccurate prediction, and delayed response.

Description

Technical Field

[0001] This invention relates to the field of poultry farming technology, specifically to a whole-process disease prevention traceability and risk management system for egg-laying breeder chickens. Background Technology

[0002] Disease prevention in egg-laying breeder chickens is a key technology system that achieves disease control and production performance assurance throughout the growth cycle of breeder chickens through systematic management of aspects such as monitoring breeding environment parameters, assessing the health status of the flock, and implementing disease prevention measures.

[0003] There are significant technical bottlenecks in the current disease prevention and control of egg-laying breeder chickens: traditional monitoring systems rely on single sensors or independent models to process data, lack multimodal fusion analysis capabilities, generally have low disease prediction accuracy, and lack real-time linkage between environmental and flock physiological indicators in dynamic disease prevention strategies. Therefore, we propose a whole-process disease prevention traceability and risk management system for egg-laying breeder chickens. Summary of the Invention

[0004] To address the aforementioned issues, the present invention aims to provide a full-process disease prevention traceability and risk management system for egg-laying breeder chicken farming. This system possesses the advantages of achieving precise disease prevention, full-process traceability, and dynamic management of the farming process through multi-dimensional data fusion, intelligent model construction, and other means, thereby resolving significant technical bottlenecks in current egg-laying breeder chicken farming disease prevention.

[0005] This invention discloses a whole-process disease prevention traceability and risk management system for egg-laying chicken breeding, comprising: a multimodal data acquisition module, an edge computing preprocessing module, and an intelligent risk management module;

[0006] The multimodal data acquisition module is used to collect multimodal data during the chicken farming process and transmit the multimodal data to the edge computing preprocessing module; the multimodal data includes environmental parameter data, images of chicken behavior, and voiceprint data;

[0007] The edge computing preprocessing module receives multimodal data collected by the multimodal data acquisition module, preprocesses the multimodal data according to its data type, obtains standardized data, and sends it to the intelligent risk management module for processing. The standardized data includes standardized images of chicken flock behavior, valid environmental parameter data, and valid voiceprint data.

[0008] The intelligent risk management module includes a fusion prediction model, an abnormal behavior recognition system, and a multi-level dynamic epidemic prevention strategy engine. The fusion prediction model predicts the probability of disease in chicken flocks based on effective environmental parameter data. The abnormal behavior recognition system outputs the disease risk level of chicken flocks based on standardized images of chicken flock behavior and effective voiceprint data. The multi-level dynamic epidemic prevention strategy engine sets up a multi-level response mechanism and responds accordingly based on the disease risk level output by the abnormal behavior recognition system.

[0009] Furthermore, the abnormal behavior recognition system includes an image detection branch, a voiceprint detection branch, and a fusion output layer. The image detection branch outputs image recognition results to the fusion output layer based on standardized images of chicken flock behavior. The voiceprint detection branch outputs voiceprint detection results to the fusion output layer based on valid voiceprint data. The fusion output layer fuses the image recognition results and voiceprint detection results based on the DS evidence theory and outputs the disease risk level of the chicken flock.

[0010] Furthermore, the image detection branch includes a data augmentation module, a backbone section, a neck section, and a detection head;

[0011] The data augmentation module performs Mosaic data augmentation and adaptive anchor box calculation on standardized chicken flock behavior images, and then passes the processed data to the Backbone part.

[0012] The Backbone section performs multi-scale feature fusion on the standardized chicken behavior images processed by the data augmentation module, and then passes the processed images to the Neck section.

[0013] The Neck section uses a BiFPN bidirectional feature pyramid to perform cross-level feature interaction on the standardized chicken flock behavior image processed by the Backbone section, and then transmits it to the detection head after processing.

[0014] The detection head outputs image recognition results based on standardized chicken flock behavior images processed from the Neck portion.

[0015] Furthermore, the voiceprint detection branch includes an MFCC feature extraction module, a Mel filter bank, and an SVM classifier connected in sequence;

[0016] The MFCC feature extraction module performs MFCC feature extraction on the valid voiceprint data, generates a multi-dimensional feature vector, and inputs the multi-dimensional feature vector into the Mel filter bank;

[0017] The Mel filter bank reduces the dimensionality of the input multidimensional feature vector, and then inputs the dimensionality-reduced multidimensional feature vector into the SVM classifier.

[0018] The SVM classifier outputs voiceprint detection results based on the dimensionality-reduced multidimensional feature vectors.

[0019] Furthermore, the fusion output layer includes a decision fusion module and a risk level output unit;

[0020] The decision fusion module fuses the image recognition results and voiceprint detection results based on the DS evidence theory to obtain multimodal detection results, which are then input to the risk level output unit. The risk level output unit outputs the disease risk level of the chicken flock based on the multimodal detection results.

[0021] Furthermore, the fusion prediction model is an LSTM-Transformer fusion prediction model, which includes an input layer, a time series feature extraction layer, an encoder, and an output layer that are connected in sequence.

[0022] Furthermore, it also includes a blockchain traceability module; the blockchain traceability module is connected to the multimodal data acquisition module, the edge computing preprocessing module, and the intelligent risk management module respectively; the blockchain traceability module uses Hyperledger Fabric to build a consortium blockchain architecture to store the relevant data generated by the multimodal data acquisition module, the edge computing preprocessing module, and the intelligent risk management module during the processing.

[0023] Furthermore, the disease risk level is divided into Level I, Level II, and Level III, and the multi-level dynamic epidemic prevention strategy engine sets up Level I response mechanism, Level II response mechanism, and Level III response mechanism corresponding to the disease risk level.

[0024] The beneficial effects of this invention are:

[0025] 1. This full-process disease prevention traceability and risk management system for egg-laying breeder chickens uses a multimodal data acquisition module to collect images of the breeding environment, flock behavior, and abnormal sound signals. After noise reduction, feature extraction, and outlier filtering by the edge computing preprocessing module, the data is input into the intelligent risk management module. The LSTM-Transformer fusion prediction model extracts time series and multi-dimensional features to output the disease probability of the flock. The improved YOLOv5 combined with voiceprint detection outputs the risk level through the decision fusion module and risk level output unit, driving the multi-level dynamic disease prevention strategy engine to link in real time, solving the problems of insufficient multimodal fusion, inaccurate prediction, and delayed response.

[0026] 2. This full-process disease prevention traceability and risk management system for egg-laying chicken breeding utilizes a blockchain traceability module built on a consortium blockchain using Hyperledger Fabric. It employs an improved PBFT consensus algorithm, introduces a timestamp verification mechanism, removes the pre-confirmation stage, and sets a message waiting threshold. Blocks contain data such as environmental parameter hash values, vaccination records, and transportation trajectories, stored in CouchDB at the underlying layer. It provides dual-mode query interfaces for QR Code and RFID, enabling data storage and precise full-chain traceability, thus solving the problems of easily tampered data, coarse traceability granularity, and low query efficiency in centralized databases. Attached Figure Description

[0027] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0028] Figure 1 This is a flowchart of an abnormal behavior recognition system;

[0029] Figure 2 Structure diagram of an abnormal behavior recognition system

[0030] Figure 3 To improve the timing diagram of the PBFT consensus algorithm;

[0031] Figure 4 This is a diagram of the consortium blockchain architecture of the present invention. Detailed Implementation

[0032] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Many specific details are set forth in the following description to provide a thorough understanding of the present invention; however, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the spirit of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0033] The purpose of this invention is to provide a whole-process epidemic prevention traceability and risk management system for egg-laying breeder chicken farming. It has the advantages of achieving precise epidemic prevention, whole-process traceability and dynamic management of the farming process through multi-dimensional data fusion, intelligent model construction and distributed ledger technology, and solves the significant technical bottlenecks in the current epidemic prevention of egg-laying breeder chicken farming.

[0034] This invention discloses a whole-process disease prevention traceability and risk management system for egg-laying chicken breeding, comprising: a multimodal data acquisition module, an edge computing preprocessing module, an intelligent risk management module, a blockchain traceability module, and an intelligent decision support module.

[0035] The intelligent risk management module, blockchain traceability module, and intelligent decision support module are all deployed on a cloud server; the cloud server is the core computing and data hub of the system of this invention.

[0036] The edge computing preprocessing module at the farm site performs preliminary cleaning, filtering, and standardization of the raw data collected on-site, such as the chicken coop environment, chicken activity footage, and chicken calls. Then, it transmits the standardized data to the cloud server through an encrypted network channel.

[0037] The cloud server, leveraging its computing and storage capabilities, supports the operation of all three core modules. The intelligent risk management module calculates the probability of disease outbreaks in the flock, determines the disease risk level, and then issues corresponding disease prevention instructions. The blockchain traceability module manages the consortium blockchain nodes, verifies data, and generates blocks in the cloud, storing all key data throughout the entire farming process to ensure data integrity and support full-chain information traceability. The intelligent decision support module optimizes feed formulations and assesses the biosecurity level of the farm; all calculation results are simultaneously stored on the blockchain.

[0038] The multimodal data acquisition module uses deployed temperature and humidity sensors, ammonia concentration sensors, visual monitoring equipment, and voiceprint acquisition equipment to collect environmental parameter data, images of chicken behavior, and abnormal sound signals (voiceprint data) in real time during the chicken farming process; all of this data is collectively referred to as multimodal data. The multimodal data acquisition module then transmits the collected multimodal data to the edge computing preprocessing module for preprocessing.

[0039] The aforementioned sensors were deployed following a uniform coverage principle: the temperature and humidity sensor (SHT30) was installed at a 4-meter spacing and a vertical height of 1.2 meters (the height of the chicken's back), with a sampling frequency of 15Hz. The data was converted by a 12-bit ADC and the noise was ≤0.1LSB. The ammonia concentration sensor (TGS2602) used differential sampling and was equipped with an activated carbon filter layer (response time ≤10 seconds), with a range of 0-200ppm and an accuracy of ±1.5%. The visual monitoring equipment used a Hikvision DS-2CD3T26WD-I3 camera, which supports H.265 encoding (bitrate 2Mbps). The ROI area, which occupies 70% of the screen, was set to be the chicken activity area. The video was transmitted in real time through the GStreamer pipeline at a frame rate of 25fps. The voiceprint acquisition device used a Knowles SPM1423 four-channel microphone array with integrated AGC automatic gain control (gain range 0-40dB). It was connected to the edge computing node of the edge computing preprocessing module via an I2S interface at a sampling rate of 48kHz.

[0040] The edge computing preprocessing module receives multimodal data collected by the multimodal data acquisition module, preprocesses it according to its data type to obtain standardized data, and then sends it to the intelligent risk management module for further processing. The standardized data includes standardized images of chicken behavior, valid environmental parameter data, and valid voiceprint data. The edge computing nodes of the edge computing preprocessing module consist of an NVIDIA Jetson Nano (quad-core A57, 1.43GHz) and a custom FPCA coprocessor (Xilinx Artix-7), configured with 4GB of LPDDR4 memory and 32GB of eMMC storage. Edge computing nodes connect to sensors in the multimodal data acquisition module via an RS485 bus (115200bps baud rate). The RS485 bus uses the Modbus RTU protocol (slave addresses 0x01-0x10), and each sensor is configured with a 10ms timeout retransmission mechanism. The edge computing preprocessing module communicates with the cloud server via a 4G / 5G industrial router (Huawei ME909s-821), supports TCP long connections (30-second heartbeat packets), and uses TLS1.3 protocol encryption (ECDHE-ECDSA-CHACHA20-POLY1305 cipher suite).

[0041] Specifically, the edge computing preprocessing module preprocesses the environmental parameter data as follows: First, it performs wavelet transform noise reduction on the environmental parameter data. Then, it uses the sym8 wavelet basis to decompose the environmental parameter data into five levels. Next, it uses a soft thresholding method to remove noise, with the threshold dynamically calculated based on three times the median absolute deviation, thus effectively filtering environmental interference and equipment acquisition noise. Finally, it uses the 3σ criterion to independently calculate the mean and standard deviation of each feature dimension in the environmental parameter data, filtering out outliers that exceed the range of μ±3σ. The outliers are repaired by linear interpolation between adjacent frames, keeping the interpolation error within 5%, thereby obtaining effective environmental parameter data and ensuring the validity and accuracy of the environmental parameter data.

[0042] The edge computing preprocessing module preprocesses the voiceprint data as follows: First, it performs wavelet transform noise reduction on the voiceprint data (implemented using Matlab Wavelet Toolbox). Then, it uses the sym8 wavelet basis to decompose the voiceprint data into 5 levels. Next, it uses a soft thresholding method to remove noise, with the threshold dynamically calculated based on 3 times the median absolute deviation, effectively filtering environmental interference and equipment acquisition noise. Finally, it uses the 3σ criterion to independently calculate the mean and standard deviation of each feature dimension in the voiceprint data, filtering out outliers exceeding the μ±3σ range. The outliers are repaired by linear interpolation between adjacent frames, keeping the interpolation error within 5%, thus obtaining effective voiceprint data and ensuring its validity and accuracy.

[0043] The edge computing preprocessing module preprocesses the chicken behavior image as follows: The edge computing preprocessing module divides the image into 8×8 pixel cells, groups them into 2×2 cells, calculates histograms in 9 gradient directions, generates a 3780-dimensional feature vector, and then reduces the dimensionality to 200 dimensions through PCA to complete the HOG algorithm feature extraction, thereby obtaining the feature vector of the chicken behavior image (also known as the standardized chicken behavior image).

[0044] The intelligent risk management module includes a fusion prediction model, an abnormal behavior recognition system, and a multi-level dynamic epidemic prevention strategy engine.

[0045] The fusion prediction model predicts the probability of disease outbreaks in chicken flocks based on effective environmental parameter data. Specifically, the fusion prediction model is an LSTM-Transformer fusion prediction model, including an input layer, a time series feature extraction layer, an encoder, and an output layer. The input layer processes the effective environmental parameter data and chicken flock health index data, and then feeds it to the time series feature extraction layer. After processing, the time series feature extraction layer outputs a time series feature vector, which is then fed to the encoder for multi-dimensional feature interaction. Finally, the output layer outputs the probability of disease outbreaks in the chicken flock for future time periods.

[0046] The edge computing preprocessing module inputs valid environmental parameter data and flock health index data into the input layer. Valid environmental parameter data includes temperature and humidity gradient change rate, peak ammonia concentration and daily average value. Flock health index data includes feeding frequency, drinking time and egg production. The relevant data of flock health index can also be collected and statistically analyzed by the multimodal data acquisition module and processed by the edge computing preprocessing module.

[0047] The time-series feature extraction layer is constructed using an 8-layer bidirectional LSTM network with 256 neurons per layer. A Dropout rate of 0.3 is configured to suppress overfitting, and the activation function is tanh. The encoder is a 4-layer Transformer encoder with a single-head attention dimension of 64 and a multi-head attention dimension of 8. The feedforward network uses the ReLU activation function with a dimension of 512. After layer normalization, multi-dimensional feature interaction is achieved through residual connections.

[0048] The output layer is trained using the cross-entropy loss function, and the optimizer used is AdamW with a learning rate of 1e-4 and weight decay of 0.01. The time window for the output layer can be set to 7 days, and the sliding step size is set to 1 day. The data input to the encoder is standardized by Z-score and then mapped to the [0,1] interval through a fully connected layer to predict the probability of an epidemic in the future time period (3 days in this example). During prediction, the warning threshold can be dynamically adjusted by combining the confidence interval of historical data. It should be noted that the time window, sliding step size, and duration of the future time period can be adaptively adjusted as needed.

[0049] The abnormal behavior recognition system outputs the disease risk level of chicken flocks based on standardized images of flock behavior and effective voiceprint data. This system is an improvement upon the YOLOv5 abnormal behavior recognition system. The system includes an image detection branch, a voiceprint detection branch, and a fusion output layer, such as... Figure 1 , Figure 2 As shown, the image detection branch outputs image recognition results to the decision fusion module of the fusion output layer based on standardized chicken flock behavior images; the voiceprint detection branch outputs voiceprint detection results to the decision fusion module of the fusion output layer based on valid voiceprint data; the fusion output layer uses DS evidence theory to fuse the image recognition results and voiceprint detection results, and outputs the disease risk level of the chicken flock.

[0050] Specifically, the image detection branch includes a data augmentation module (Mosaic), a backbone section, a neck section, and a detection head connected in sequence, such as... Figure 2 As shown.

[0051] The data augmentation module is used to optimize standardized chicken behavior images. Specifically, the module performs Mosaic data augmentation and adaptive anchor box calculation on the standardized chicken behavior images, adjusts the input resolution to 640×640, uses a batch size of 32, employs cosine annealing (initially 1e-3), and iterates for 500 epochs. This yields optimized standardized chicken behavior images, which are then input into the backbone.

[0052] The Backbone section performs multi-scale feature fusion on the optimized, standardized chicken behavior image to enhance the weights of key features. Specifically, the Backbone section includes a multi-receptive field feature extraction module and a CBMA attention module. The CBMA attention module consists of a C3 module and an SPPF structure. The multi-receptive field feature extraction module uses parallel 3×3 (stride 2) and 5×5 (padding=2) convolutional kernels to extract features from the optimized, standardized chicken behavior image, outputting a feature map. This feature map is then concatenated through channels and input to the C3 module (with a bottleneck layer ratio set to 0.5). After processing by the C3 module, it undergoes multi-scale fusion using an SPPF structure (5×5 pooling kernel) to enhance the weights of key features. After processing by the Backbone section, this data is input to the Neck section.

[0053] The Neck section employs a BiFPN bidirectional feature pyramid to perform cross-level feature interaction on the standardized chicken flock behavior images processed by the Backbone section. After processing, the Neck section is input into the detection head, which outputs image recognition results. These results include identification of abnormal behaviors such as huddling and feather pecking. The detection head uses a CIoU+FocalLoss loss function (α=0.25, γ=2) to improve detection accuracy and optimizes the anchor box size for abnormal behaviors; the optimized anchor box sizes are (80, 60) and (120, 90).

[0054] The voiceprint detection branch comprises a sequentially connected MFCC feature extraction module, a Mel filter bank for dimensionality reduction, and an SVM classifier. Valid voiceprint data (44.1kHz audio signal) undergoes MFCC feature extraction by the MFCC feature extraction module, and the resulting 40-dimensional feature vector is input to the Mel filter bank for dimensionality reduction. The Mel filter bank reduces the 40-dimensional feature vector to 20 dimensions, and the SVM classifier (using an RBF kernel, γ=0.1, C=10) outputs the voiceprint detection result. The SVM classifier is a cough classifier.

[0055] Image recognition and voiceprint detection results are input into the decision fusion module of the fusion output layer. The decision fusion module constructs a basic probability allocation function based on DS evidence theory, defining the focal elements of clustering / feather pecking / coughing sounds as {risk, safety} (triggered when joint confidence ≥ 0.75). It fuses multimodal detection results through orthogonal summation. Then, it performs fusion judgment by combining preset confidence thresholds (clustering 0.8, feather pecking 0.7, coughing sound 0.85), and finally, the risk level output unit of the fusion output layer outputs the disease risk level (Level I / II / III), achieving accurate identification of abnormal behavior and disease risk in laying hens.

[0056] The multi-level dynamic epidemic prevention strategy engine responds accordingly based on the disease risk level output by the abnormal behavior recognition system. The multi-level dynamic epidemic prevention strategy engine sets up a three-level response mechanism:

[0057] When the disease risk level reaches Level III (yellow alert), a positive pressure ventilation system with a filtration efficiency of not less than 99.9% will be automatically activated, and the ventilation rate will be dynamically adjusted according to the real-time ammonia concentration. The positive pressure ventilation system uses a PID control algorithm (Kp=0.8, Ki=0.3, Kd=0.1) to adjust the fan speed, with a sensor feedback frequency of 10Hz and an actuator response delay of ≤50ms.

[0058] When the disease risk level reaches Level II (orange alert), the peracetic acid atomizing spray device is triggered, with the atomized particle diameter controlled at 5-10 μm and the disinfectant concentration set at 0.2%. The peracetic acid atomizing spray device controls the solution ratio through a high-precision plunger pump (flow accuracy ±1%), and the atomizing nozzle adopts ultrasonic vibration (frequency 1.7MHz). The particle size distribution is calibrated in real time by a laser particle size analyzer (5-10 μm accounts for ≥95%).

[0059] When the disease risk level reaches Level I (red alert), the RT-PCR testing equipment will be activated to initiate nucleic acid testing for all personnel. The testing process includes sample collection, RNA extraction, and quantitative PCR amplification, with a single sample testing cycle not exceeding 4 hours. The RT-PCR testing equipment integrates a microfluidic chip and uses quantitative PCR technology (excitation wavelength 480nm, emission wavelength 520nm). The thermal cycler has a temperature control accuracy of ±0.1℃, and the fluorescence signal acquisition frequency is 10Hz. Data is uploaded via the Modbus TCP protocol, and the test results are hashed and uploaded to the blockchain using SHA-256.

[0060] The hardware layer of the multi-level dynamic epidemic prevention strategy engine uses the STM32H743 main control chip, integrating a 16-bit ADC and CAN bus interface to achieve real-time monitoring of device status. The multi-level dynamic epidemic prevention strategy engine communicates with the device executing the response mechanism via the CAN bus (2.0B protocol, bit rate 500kbps). A custom frame format (ID range 0x100-0x200) contains an 8-byte data field; the first 4 bytes are control instructions, and the last 4 bytes are parameter values. Error frame handling uses Automatic Retransmission Request (ARQ), with a maximum of 3 retransmissions.

[0061] The blockchain traceability module stores key data from each node throughout the entire aquaculture process. The blockchain traceability module uses a consortium blockchain architecture, such as... Figure 3 As shown.

[0062] The consortium blockchain architecture uses smart contracts to manage the data read / write permissions of each node and execute interaction rules. Through the orderly connection of each layer, the entire architecture achieves secure storage, accurate traceability, and compliant supervision of data throughout the entire egg-laying hen breeding process. The smart contracts are written in Go and manage the data read / write permissions of farm nodes and the read-only permissions of regulatory agency nodes. The consortium blockchain architecture is built on Hyperledger Fabric version 2.5.

[0063] The consortium blockchain architecture consists of an application query layer, a node layer, and a consensus layer. Data is transmitted between the application query layer and the node layer, and between the node layer and the consensus layer, via a network. The network uses the gRPC protocol and employs TLS encryption (certificates are valid for one year).

[0064] The application query layer includes QR Code and RFID query interfaces. The QR Code query interface allows users to scan QR codes to obtain full-chain data including breeding hen hatching records, vaccination dates, egg-laying dates, and transportation routes. This full-chain data is uploaded by the node layer. The RFID query interface allows for quick location of historical environmental parameter curves and disease prevention implementation records for specific breeding batches by reading RFID tags. The QR Code query interface conforms to the ISO / IEC 18004 standard, employs Level L error correction (7% recovery capability), and encodes information such as breeding hen hatching records and vaccination dates into Base64 format JSON objects. Offline decoding is achieved using the ZXing library, with a response time ≤800ms. The RFID tags for the RFID query interface are UHF band tags conforming to the EPCClass 1 Gen 2 standard (operating frequency 865-868MHz). The user area occupies 496KB of the 512KB memory, divided into an environmental data area (50B per record, retaining 180 days of data) and a disease prevention record area (30B per record). The reader uses the ImpinjR2000 chip, is equipped with an 8dBi gain antenna, and supports reading at a distance of 5 meters and the ALOHA anti-collision algorithm (16 time slots).

[0065] The node layer includes farm nodes (configured with Intel Xeon E-2278G processors, 32GB RAM, and 1TB NVMe SSDs), feed supplier nodes, slaughterhouse nodes, and regulatory agency nodes (configured with dual Intel Xeon Silver 4210 processors, 64GB RAM, and 2TB PCIe SSDs). Farm node data includes data collected and generated by modules such as the multimodal data acquisition module, edge computing preprocessing module, and intelligent risk management module, encompassing the entire data chain including breeder hen hatching records, vaccination dates, egg-laying dates, and transportation routes. The breeder hen hatching records, vaccination dates, egg-laying dates, and transportation routes involved in the entire data chain can also be obtained by the multimodal data acquisition module.

[0066] The blockchain traceability module utilizes an improved PBFT consensus algorithm to achieve data storage and traceability. The improved PBFT consensus algorithm employs a core process of "transaction proposal submission - VDF timestamp sorting - message verification and filtering - consensus confirmation - block generation - block upload," as detailed below:

[0067] The transaction proposals are derived from key data on the entire breeding process generated by other modules of the system, including multimodal data from the multimodal data acquisition module, standardized data from the edge computing preprocessing module, disease risk analysis results (disease risk level and disease probability) from the intelligent risk management module, and execution records of epidemic prevention measures (response mechanism).

[0068] Transaction proposal submission: Each node in the node layer sends a transaction proposal to the consensus node of the consortium blockchain through a client, such as... Figure 4 As shown.

[0069] VDF timestamp sorting: After receiving a transaction proposal, the 5 consensus nodes (3 master and 2 slave) in the consortium blockchain first calculate the VDF proof of the transaction proposal, and then sort and verify the transaction proposals according to their timestamps; calculating the VDF proof of the transaction proposal is to ensure that the timestamp error is ≤20ms.

[0070] Message verification filtering: Transaction proposals are transmitted between consensus nodes in the form of messages for consensus confirmation. Message transmission is optimized through the BFT-SMaRt protocol. The maximum waiting threshold for messages is 300ms. Messages exceeding the maximum waiting threshold are automatically filtered.

[0071] Consensus confirmation and block generation on-chain: More than 2 / 3 of the consensus nodes complete consensus confirmation. Compared to the existing PBFT consensus algorithm, the improved PBFT consensus algorithm removes the pre-confirmation phase. Once consensus confirmation is complete, the corresponding node's client is notified.

[0072] Block Generation: Transaction proposals are generated into blocks. The block generation interval is accurate to 10 minutes based on GPS clock synchronization, with a deviation of ≤100ms. Each block contains a 128-byte SHA-256 hash of environmental parameters, a 64-byte vaccine traceability code conforming to GB / T20014.3, and 32 bytes of GPS trajectory data with an accuracy of ±1m in the WGS84 coordinate system. The transaction proposal is organized using a Merkle Patricia tree.

[0073] Block On-Chain: After being packaged and organized, the generated blocks are stored in the underlying CouchDB database, completing the on-chain data notarization. Before being uploaded to the blockchain, blocks are encrypted using AES-256 (the key is updated every 7 days). Hash notarization employs a double SHA-256 algorithm: first, a hash operation is performed on the original data, and then a second hash operation is performed on the generated hash value, ensuring a 100% tamper detection rate for the uploaded data. The CouchDB database uses a composite indexing mechanism based on timestamps, device IDs, and vaccine batches, which can improve index scanning efficiency by 60% during range queries.

[0074] In summary, the participants, such as farms, feed suppliers, slaughterhouses, and regulatory agencies, which are the various nodes in the node layer, will organize all the key data generated throughout the entire breeding process into transaction proposals and submit them to the consensus nodes in the consortium blockchain that are specifically responsible for data verification. In this embodiment, there are 5 consensus nodes, of which 3 are master nodes and 2 are slave nodes. After receiving a transaction proposal, the consensus nodes first verify the time sequence of each proposal to ensure accuracy. Then, the proposals are transmitted between the consensus nodes as messages, completing the consensus verification process. During this process, an automatic message timeout filtering rule is implemented, directly discarding invalid timeout messages. The redundant pre-confirmation step in traditional algorithms is also eliminated, significantly improving verification efficiency. Once more than two-thirds of the consensus nodes have verified and confirmed a transaction proposal, the system packages these verified proposals into blocks with a fixed format. Finally, the generated blocks are encrypted and stored in the underlying CouchDB database, completing the entire process of on-chain data storage. The data on-chain possesses immutability and allows for full-process traceability and querying.

[0075] Example 2:

[0076] This embodiment also includes an intelligent decision support module. The intelligent decision support module is connected to the edge computing preprocessing module and the blockchain traceability module.

[0077] The intelligent decision support module includes a reinforcement learning-based feed formulation optimization system and a biosafety assessment model. The reinforcement learning-based feed formulation optimization system uses standardized data preprocessed by an edge computing preprocessing module to output an optimized feed formulation, which is then uploaded to a blockchain traceability module for storage. The biosafety assessment model uses standardized data preprocessed by the edge computing preprocessing module to output a biosafety index. This biosafety index triggers a multi-level dynamic epidemic prevention strategy engine to implement corresponding epidemic prevention upgrades. Data from the biosafety assessment model is also stored in the blockchain traceability module.

[0078] Specifically, the reinforcement learning-based feed formulation optimization system defines a state space and an action space. The action space (feed formulation) is adaptively adjusted based on the state space. The state space consists of four dimensions: ambient temperature (accuracy ±0.1℃), relative humidity (accuracy ±1%RH), laying hen age (integer), and egg production rate (resolution 0.5%). The action space consists of adjustments to the proportions of six ingredients: corn, soybean meal, and fishmeal (step size 0.5%, range -5% to +5%).

[0079] The reinforcement learning-based feed formulation optimization system uses the PPO algorithm for policy optimization, with a clip parameter of 0.2, a value function coefficient of 0.5, and an entropy coefficient of 0.01. The neural network structure of the reinforcement learning-based feed formulation optimization system includes a 4-dimensional input layer, a first fully connected layer (128 neurons, ReLU activation), a second fully connected layer (64 neurons, ReLU activation), and a 6-dimensional output layer (tanh activation function) connected in sequence.

[0080] The state normalization of the feed formulation optimization system based on reinforcement learning adopts the mean-variance method with a 30-day sliding window. The experience replay buffer capacity is 100,000 records. The strategy is updated every 5,000 data records collected. The batch size during training is 256. The initial learning rate is 3e-4, which decays by 0.98 every 100 iterations. The goal is to improve the feed conversion rate during the laying period by ≥18% compared to the baseline value of 1.8.

[0081] The random forest biosafety assessment model outputs a biosafety index based on data collected by the multimodal data acquisition module, including personnel movement data, vehicle disinfection records, and vaccine potency data.

[0082] Specifically, during the feature engineering phase, the random forest biosafety assessment model performs WOE encoding on 15-dimensional features such as personnel flow data, disinfection data, and vaccine data collected by the multimodal data acquisition module, and missing values ​​are processed using K-nearest neighbor (K=3) interpolation.

[0083] The parameters of the random forest biosafety assessment model are: 100 decision trees, maximum depth of a single tree of 12, minimum number of leaf node samples of 10, and feature selection method using Gini index as the criterion, with 10-dimensional features of importance ≥0.05 selected by SHAP value.

[0084] The random forest biosafety assessment model employs a bagging strategy, an 80% sample sampling ratio, a 70% feature sampling ratio, and a hard-vote averaging strategy during training. It outputs a biosafety index ranging from 0 to 100, with a threshold of 70 triggering an early warning. If the biosafety index falls below 70, the intelligent risk management module's multi-level dynamic epidemic prevention strategy engine will automatically trigger corresponding epidemic prevention measures, making epidemic prevention more precise and effective.

[0085] The random forest biosafety assessment model is deployed on NVIDIA Jetson AGX Orin edge computing nodes with an inference latency of ≤20ms. It supports loading the latest 100 data points every morning for incremental learning (updating the tree model using the random subspace method). The confusion matrix is ​​calibrated regularly (once a week) based on labeled data from actual epidemic prevention events to ensure an F1-score ≥0.85.

[0086] In summary, the egg-laying hen breeding full-process disease prevention traceability and risk management system of the present invention uses a multimodal data acquisition module to collect images of the breeding environment, flock behavior, and abnormal sound signals. After noise reduction, feature extraction, and outlier filtering by the edge computing preprocessing module, the data is input into the intelligent risk management module. The LSTM-Transformer fusion prediction model extracts time series and multi-dimensional features and outputs the probability of disease occurrence. The improved YOLOv5 combined with voiceprint detection outputs the risk level through the decision fusion module, driving the multi-level dynamic disease prevention strategy engine to link in real time, thus solving the problems of insufficient multimodal fusion, inaccurate prediction, and delayed response.

[0087] Furthermore, this full-process epidemic prevention traceability and risk management system for egg-laying chicken breeding utilizes a blockchain traceability module built on a consortium blockchain using Hyperledger Fabric. It employs an improved PBFT consensus algorithm, introduces a timestamp verification mechanism, removes the pre-confirmation stage, and sets a message waiting threshold. The blocks contain data such as environmental parameter hash values, vaccination records, and transportation trajectories, stored in CouchDB at the underlying layer. It provides dual-mode query interfaces for QR Code and RFID, enabling data storage and precise traceability across the entire chain, thus solving the problems of easily tampered data, coarse traceability granularity, and low query efficiency in centralized databases.

[0088] The relevant modules involved in this system are all hardware system modules or functional modules that combine computer software programs or protocols with hardware in the prior art. The computer software programs or protocols involved in these functional modules are technologies known to those skilled in the art and are not improvements to this system. The improvement of this system lies in the interaction or connection between the modules, that is, in improving the overall structure of the system to solve the corresponding technical problems that this system aims to address.

[0089] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A system for disease prevention, traceability, and risk management throughout the entire process of egg-laying chicken breeding, characterized in that, include: Multimodal data acquisition module, edge computing preprocessing module, and intelligent risk management module; The multimodal data acquisition module is used to collect multimodal data during the chicken farming process and transmit the multimodal data to the edge computing preprocessing module; the multimodal data includes environmental parameter data, images of chicken behavior, and voiceprint data; The edge computing preprocessing module receives multimodal data collected by the multimodal data acquisition module, preprocesses the multimodal data according to its data type, obtains standardized data, and sends it to the intelligent risk management module for processing. The standardized data includes standardized images of chicken flock behavior, valid environmental parameter data, and valid voiceprint data. The intelligent risk management module includes a fusion prediction model, an abnormal behavior recognition system, and a multi-level dynamic epidemic prevention strategy engine; The fusion prediction model predicts the probability of disease outbreaks in chicken flocks based on effective environmental parameter data; The abnormal behavior recognition system outputs the disease risk level of the flock based on standardized images of flock behavior and effective voiceprint data. The multi-level dynamic epidemic prevention strategy engine sets up a multi-level response mechanism, and responds accordingly based on the disease risk level output by the abnormal behavior recognition system.

2. The whole-process disease prevention traceability and risk control system for egg-laying chicken breeding as described in claim 1, characterized in that, The abnormal behavior recognition system includes an image detection branch, a voiceprint detection branch, and a fusion output layer. The image detection branch outputs image recognition results to the fusion output layer based on standardized images of chicken flock behavior. The voiceprint detection branch outputs voiceprint detection results to the fusion output layer based on valid voiceprint data. The fusion output layer fuses the image recognition results and voiceprint detection results based on the DS evidence theory and outputs the disease risk level of the chicken flock.

3. The whole-process disease prevention traceability and risk control system for egg-laying chicken breeding as described in claim 2, characterized in that, The image detection branch includes a data augmentation module, a backbone section, a neck section, and a detection head; The data augmentation module performs Mosaic data augmentation and adaptive anchor box calculation on standardized chicken flock behavior images, and then passes the processed data to the Backbone part. The Backbone section performs multi-scale feature fusion on the standardized chicken behavior images processed by the data augmentation module, and then passes the processed images to the Neck section. The Neck section uses a BiFPN bidirectional feature pyramid to perform cross-level feature interaction on the standardized chicken flock behavior image processed by the Backbone section, and then transmits it to the detection head after processing. The detection head outputs image recognition results based on standardized chicken flock behavior images processed from the Neck portion.

4. The whole-process disease prevention traceability and risk control system for egg-laying chicken breeding as described in claim 2, characterized in that, The voiceprint detection branch includes an MFCC feature extraction module, a Mel filter bank, and an SVM classifier connected in sequence; The MFCC feature extraction module performs MFCC feature extraction on the valid voiceprint data, generates a multi-dimensional feature vector, and inputs the multi-dimensional feature vector into the Mel filter bank; The Mel filter bank reduces the dimensionality of the input multidimensional feature vector, and then inputs the dimensionality-reduced multidimensional feature vector into the SVM classifier. The SVM classifier outputs voiceprint detection results based on the dimensionality-reduced multidimensional feature vectors.

5. The whole-process disease prevention traceability and risk control system for egg-laying breeder chickens according to claim 2, characterized in that, The fusion output layer includes a decision fusion module and a risk level output unit; The decision fusion module fuses image recognition results and voiceprint detection results based on DS evidence theory to obtain multimodal detection results, which are then input to the risk level output unit. The risk level output unit outputs the disease risk level of the chicken flock based on the multimodal detection results.

6. The whole-process disease prevention traceability and risk control system for egg-laying breeder chickens according to claim 1, characterized in that, The fusion prediction model is an LSTM-Transformer fusion prediction model, which includes an input layer, a time series feature extraction layer, an encoder, and an output layer that are connected in sequence.

7. The whole-process disease prevention traceability and risk control system for egg-laying chicken breeding as described in claim 1, characterized in that, It also includes a blockchain traceability module; the blockchain traceability module is connected to the multimodal data acquisition module, the edge computing preprocessing module and the intelligent risk management module respectively; the blockchain traceability module uses Hyperledger Fabric to build a consortium blockchain architecture to store the relevant data generated by the multimodal data acquisition module, the edge computing preprocessing module and the intelligent risk management module during the processing.

8. The whole-process disease prevention traceability and risk control system for egg-laying breeder chickens according to claim 1, characterized in that, Disease risk levels are divided into Level I, Level II, and Level III. The multi-level dynamic epidemic prevention strategy engine sets up Level I response mechanisms, Level II response mechanisms, and Level III response mechanisms corresponding to the disease risk levels.

Images