A multimodal biocomputing-driven poultry health intelligent monitoring system and method

Abstract

The present invention discloses a multimodal biocomputing-driven intelligent health monitoring system for poultry, which relates to the field of intelligent poultry breeding technology, including a hardware terminal device module, a data processing module and a front-end module. The hardware terminal device module collects poultry audio data in real time and uploads it to the data processing module; the data processing module performs health recognition and classification analysis on the received poultry audio data by applying an independently developed audio artificial intelligence recognition algorithm to determine whether the poultry is sick and the type of disease, and feeds back the recognition results to the front-end module in real time. The system realizes comprehensive monitoring of the health status of poultry by integrating front-end, back-end and hardware technologies; the system adopts SE‑ResNet network and continuous learning technology, and continuously improves the algorithm to provide more accurate health monitoring.

Description

Technical Field

[0001] The present invention relates to the technical field of smart poultry farming, and in particular to a poultry health intelligent monitoring system and method driven by multimodal biocomputing. Background Art

[0002] Respiratory diseases in chickens can damage the respiratory system of chickens and reduce their own disease resistance, which in turn leads to reduced feed intake, decreased physical fitness, reduced production performance, and even the elimination of a large number of chickens, causing varying degrees of economic losses to egg-laying and broiler farmers. It is of great significance to monitor the calls of chickens in the chicken house, analyze the audio data, and determine whether the chickens have respiratory diseases and the types of diseases. The existing method of judging respiratory diseases in chickens uses manual identification, and professionals conduct on-site inspections in the chicken house to manually locate the location of the chicken disease, which requires huge labor and time costs. In addition, since there is no time limit for the onset of chicken diseases and professional inspectors do not work 24 / 7, there are omissions in capturing the respiratory diseases of chickens. Summary of the invention

[0003] In view of the problems of low efficiency and missed diagnosis in the existing chicken respiratory disease judgment, the present invention is proposed.

[0004] Therefore, the problem to be solved by the present invention is how to efficiently, accurately and automatically monitor and diagnose respiratory diseases in chickens to reduce labor costs and improve the accuracy and timeliness of disease identification.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] In the first aspect, an embodiment of the present invention provides a multimodal biocomputing-driven intelligent health monitoring system for poultry, which includes a hardware terminal device module, including a Raspberry Pi 4b and a small microphone, wherein the hardware terminal device module collects poultry audio data in real time and uploads the poultry audio data to a data processing module; the data processing module includes a poultry smart cloud, which is used to receive and process the poultry audio data sent by the front-end module, and the poultry smart cloud performs health identification and classification analysis on the received poultry audio data by applying an independently developed audio artificial intelligence recognition algorithm to determine whether the poultry is sick and the type of disease, and feeds back the identification results to the front-end module in real time; the front-end module includes a mobile application and a web page, which is used for veterinarians to remotely monitor the health status of chickens in a chicken house, and the front-end module parses and organizes the identification results fed back by the data processing module, and displays them to the user through a page or a mobile application interface.

[0007] As a preferred solution of the multimodal biocomputing-driven poultry health intelligent monitoring system described in the present invention, the processing flow of the data processing module is as follows: in the initial stage, a detailed key factor analysis is performed to determine the key factors affecting the audio data quality and the target detection effect, as well as the environmental and physiological indicators affecting the health of poultry; a target detection algorithm model is developed, integrating signal processing, pattern recognition, and biometric recognition technology to perform comprehensive analysis and target detection on the collected multimodal data; in the algorithm model training process, the random changes and noise conditions of the actual environment are simulated, and real scene data is introduced as training material; convolutional neural networks, recurrent neural networks, and generative adversarial networks are integrated into the network model, and the feature learning ability is enhanced in combination with the attention mechanism; cross-validation is used to evaluate the model performance, and different hyperparameter settings are extensively tested to find the optimal network configuration; comprehensive model verification and evaluation are performed to ensure that the performance meets expectations, and the optimized model is deployed in practical applications.

[0008] As a preferred solution of the multimodal biocomputing-driven poultry health intelligent monitoring system described in the present invention, the development of the target detection algorithm model includes the following contents: drawing on the information processing mechanism of biological systems, constructing an algorithm framework based on biological computing models including DNA computing, cell computing, and neuron computing, and giving the model the biological characteristics of self-adaptation, self-organization, and self-repair; using multi-source heterogeneous data recorded in a real breeding room as model input, and annotating, cleaning, and slicing the original multi-source heterogeneous data; designing a feature extraction engine based on DNA molecular computing, using DNA sequences and their pairing rules to perform parallel filtering and mapping operations on input data for efficient feature extraction; fusing features from different sources to construct a multimodal feature representation, and converting each data slice into a three-dimensional tensor (batch, fea, feature_dim); selecting a model architecture based on DNA / cell / neuron computing including a molecular convolutional network and a dendritic pulse network;

[0009] Simulate the biological antigen-antibody recognition mechanism, use the labeled data as the antigen pattern, and perform pattern matching classification through immune calculation; use cross entropy loss and other loss functions to calculate the gap between the predicted label and the true label; introduce evolutionary algorithm to optimize model parameters so that the model has evolutionary adaptability, and train the model to survive and evolve in an artificial life virtual environment; after the trained model is deployed, introduce online learning strategy, continuously obtain data from the actual environment, and dynamically update model parameters; where batch represents batch size, fea represents time step or frequency step, and feature_dim represents feature dimension.

[0010] As a preferred solution of the multimodal biocomputing-driven poultry health intelligent monitoring system described in the present invention, the design of a feature extraction engine based on DNA molecular computing includes the following steps: obtaining multi-source heterogeneous data after annotation, cleaning and slicing; constructing a DNA molecular filter that simulates the structure of the cochlea to perform parallel frequency band filtering on the audio signal; designing a DNA concentration encoding mechanism to map the energy of each frequency band to the DNA concentration to simulate the nonlinear perception of the human ear; compressing and removing redundancy of DNA-encoded speech features by designing a DNA enzyme cutting pattern; using a deep learning model to extract visual features including poultry posture and mouth movement from video data, and combining the visual features with speech features based on biocomputing; The method is to fuse features and construct multimodal feature representation; construct a virtual poultry farming scenario based on the metaverse and collect large-scale simulation data from multiple perspectives and modalities; automatically annotate simulation data using semi-supervised or unsupervised learning by comparing with real data and expert knowledge, and transform and amplify the annotated simulation data; use different poultry voiceprint features as antigen patterns, and use immune computational modeling to build an antibody library for voiceprint recognition and health status classification; place the model in the virtual environment of the metaverse for survival and evolution, and introduce evolutionary algorithms to optimize the model architecture and parameters to adapt to the changing virtual and real environment; deploy the trained model to the actual application environment, and introduce online learning strategies to continuously update the antibody library to adapt to the new data distribution.

[0011] As a preferred solution of the multimodal biocomputing driven poultry health intelligent monitoring system of the present invention, the construction of a DNA molecular filter simulating the cochlear structure includes the following steps: based on the frequency range and frequency band division of the human cochlea, a set of DNA sequence sets {S1, S2, ..., S n}, where each sequence S i The length L i and nucleotide sequence, corresponding to the center frequency f of the ith band filter i and bandwidth B i ; Using DNA coding algorithm, the sampling frequency is f s The input speech signal x(t) is encoded into a DNA sequence X of length L; extract each X i The concentration value C i , forming the filtered frequency band component set {C1,C2,...,C n}; Perform DNA molecular computation on DNA sequence X→{X1,X2,...,X n},in Represents DNA sequence hybridization operation; calculate each X i The number of A, T, C, and G nucleotides in i , T i , C i , Gi , then X i DNA concentration C i =A i +T i +C i +G i ; Set the concentration values ​​{C1,C2,...,C n} is normalized to obtain the relative energy E of the i-th frequency band i =C i / ∑C j i,j=1,2,...,n;using formula F i =aln(1+bE i ) i Perform nonlinear mapping to obtain the nonlinear characteristics F of the i-th frequency band i , where a and b are constants; {F1, F2, ..., F n}Re-encode into a DNA sequence Y of length M; design a set of K DNA enzymes E = {E1, E2, ..., E K}, cut Y according to the encoding rules to get Y j , calculate each Y j Middle A i , T i , C i , G i The number of compressed features is output as the compressed feature; if the compressed feature is not within the preset threshold range, the re-involved DNA sequence set {S1, S2, ..., S n}, otherwise the compressed features are output.

[0012] As a preferred solution of the multimodal biocomputing-driven poultry health intelligent monitoring system described in the present invention, the simulation of random changes and noise conditions in the actual environment includes the following steps: using 3D modeling software to create a realistic poultry farming scene to simulate the structure and material of the chicken house, and setting the environmental parameter adjustment function to reflect the dynamic changes of the actual farming environment; adding a poultry 3D model, simulating the movement and physiological reactions of the poultry through a programming algorithm, and generating random environmental events; using measured noise data to create a noise model, and simulating the spatial distribution and temporal changes of noise according to different areas and conditions of the farm; designing a noise propagation and attenuation model, taking into account the propagation characteristics of sound in different materials and structures, and the influence of spatial geometry on the sound field; introducing real scene data including historical meteorological data, farm environmental monitoring records and poultry behavior logs; combining the simulated data with the real data, and increasing the diversity and coverage of the training set through data augmentation technology.

[0013] As a preferred solution of the multimodal biocomputing-driven poultry health intelligent monitoring system described in the present invention, determining whether poultry is sick and the type of disease includes the following steps: obtaining poultry audio data, and key acoustic features extracted from the poultry audio data by a feature extraction engine of DNA molecular computing; using a deep learning model to extract poultry behavior features from the video, including posture, gait, and mouth movement, as auxiliary feature input; using semi-supervised or unsupervised learning to automatically annotate simulation data, and transforming and amplifying the annotated simulation data; based on the principle of biological immune computing, using the annotated normal poultry voiceprint as an antigen pattern, and training an antibody library through an evolutionary algorithm as a baseline model; preliminarily matching the normal voiceprint pattern in the baseline antibody library with the input audio, and calculating the similarity score; for audio with a score higher than a preset confidence threshold, using an attention mechanism to focus on the spatiotemporal context from the multimodal features The information is combined with prior knowledge including the age and breed of poultry to accurately determine whether it is a normal behavior pattern; for suspected abnormal audio, a generative adversarial network is used for anomaly detection to identify deviant samples from the normal voiceprint distribution; for the detected abnormal audio, the predefined disease voiceprint library, environmental noise library and non-chicken call library are matched respectively, and their respective similarity score vectors are calculated; the similarity score vector is input into an attention fusion module, and the abdominal cavity video and environmental perception information are introduced for weighted fusion; the weighted fused comprehensive score vector is output to a classification decision module, and the posterior probability is calculated according to the Bayesian decision theory to make the final health status judgment; for the audio judged to be sick, it is carefully compared with the disease voiceprint library to identify the specific disease type and its confidence; the decision results including health, non-chicken call, disease and its type and confidence score are sent to the front-end module in real time through the message queue.

[0014] In the second aspect, an embodiment of the present invention provides a multimodal biocomputing-driven intelligent poultry health monitoring method, which includes deploying Raspberry Pi 4b and a small microphone in multiple chicken houses, and placing the microphone in the center of the cage top; continuously recording the sound environment of the chickens through a hardware terminal to collect various mixed sound data; uploading the collected audio data to a data processing module of the poultry smart cloud in real time, and performing noise filtering, segmentation and preprocessing on the uploaded audio data; using a feature extraction engine of DNA molecular computing to extract key acoustic features from the audio data, and combining the poultry behavior features extracted from the video to construct a multimodal feature representation; using a deep learning model and bioimmune computing principles to extract key acoustic features from the audio data, and constructing a multimodal feature representation of the poultry behavior features extracted from the video; It can perform preliminary matching of normal behavior patterns of poultry, identify abnormal behaviors, and perform anomaly detection and disease identification through attention mechanism and generative adversarial network; display terminal device status, sound list and poultry health monitoring results on the page and mobile application front end; veterinarians can view monitoring data including equipment status and health status in a specific chicken farm in real time on internal computers or smartphone applications; use trained audio artificial intelligence recognition algorithms to analyze audio data, judge and classify chicken sounds in real time to identify whether there are signs of illness; once the system identifies sick sounds, it automatically locates and marks the corresponding terminal device and the building where it is located, and provides real-time feedback to the front end to assist veterinarians in responding quickly.

[0015] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the multimodal bio-computing driven poultry health intelligent monitoring system as described in the first aspect of the present invention are implemented.

[0016] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the multimodal bio-computing driven poultry health intelligent monitoring system as described in the first aspect of the present invention are implemented.

[0017] The beneficial effects of the present invention are as follows: the system can accurately identify the health status and disease types of poultry by integrating multimodal data and applying advanced biocomputing and deep learning technologies; by early identification and timely treatment of poultry health problems, the system helps to reduce mortality and productivity loss caused by diseases, and reduce the overall risk and potential economic losses of farms; the system provides supporting Web visualization pages and APPs, allowing users to view the health status of chickens in the chicken house in real time; health monitoring is achieved by entering the system using mobile applications or Web pages, allowing inspectors to conveniently monitor the health status of poultry at any time and place. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 Diagram for a multimodal biocomputing-driven smart poultry health monitoring system.

[0020] Figure 2 Terminal equipment layout diagram of the poultry health intelligent monitoring system driven by multimodal biocomputing.

[0021] Figure 3 Target detection for a multimodal biocomputing-driven poultry health intelligent monitoring system as healthy call outcomes.

[0022] Figure 4 Target detection of diseased acoustic outcomes for a multimodal biocomputing-driven smart poultry health monitoring system.

[0023] Figure 5 Target detection for non-chicken-calling outcomes for a multimodal biocomputing-driven smart monitoring system for poultry health.

[0024] Figure 6 The sick sounds page is an example of a smart poultry health monitoring system driven by multimodal biocomputing.

[0025] Figure 7 Modifying audio types in an example of a multimodal biocomputing-driven smart monitoring system for poultry health.

[0026] Figure 8 Original sound for a smart monitoring system for poultry health driven by multimodal biocomputing.

[0027] Fig. 9 Preprocessed sounds for a multimodal biocomputing-driven smart poultry health monitoring system.

[0028] Fig.10 Partial representation of MFCC feature vectors for a multimodal biocomputing-driven smart poultry health monitoring system. DETAILED DESCRIPTION

[0029] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0030] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0031] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0032] Example 1

[0033] Reference Figure 1 to Figure 7 , which is the first embodiment of the present invention, provides a multimodal biocomputing driven poultry health intelligent monitoring system, comprising:

[0034] The hardware terminal device module includes a Raspberry Pi 4b and a small microphone. The hardware terminal device integrates an artificial intelligence algorithm module to achieve interconnection with a front-end interface and a back-end server.

[0035] The front-end module, including a mobile application and a web page, is used by veterinarians to remotely monitor the health status of chickens in the chicken house.

[0036] It should be noted that the front-end module contains the terminal equipment status information, sound list, building division and statistical information in each breeding room in all chicken farms.

[0037] Specifically, the WEB page is mainly used on the company's internal computers for veterinarians to monitor the chicken coop in real time during working hours; it allows a specific chicken farm to be selected in the chicken farm list to view the monitoring data of the chicken farm, and displays the status information of the terminal equipment in the specific chicken farm, including the building where the equipment is located, the equipment name, the equipment nickname, the current status of the equipment, the equipment IP address, the disk usage, the memory usage, the CPU usage, the CPU temperature, etc.; click the "Enter Device" button on the terminal device to view the health status of the area responsible for the equipment on the current date; click the "Real-time Monitoring" button to conduct real-time monitoring of the area responsible for this specific equipment.

[0038] Furthermore, by clicking on the sound, you can view the audio list collected by this chicken farm. The audio is identified by the data processing module to determine the audio type, including non-chicken calls, healthy and sick. If the audio type is sick, the system locates the terminal position of the sick sound according to the information of the audio data file to locate the sick chicken.

[0039] It should be noted that the mobile application can be installed on a smartphone and provides a user-friendly graphical user interface, allowing veterinarians to monitor the health of chickens in the chicken house anytime and anywhere. Its functions are similar to those of the WEB page, including equipment status viewing for a specific chicken farm, real-time monitoring of the health status of the equipment's responsible area, and audio list viewing.

[0040] The data processing module includes a poultry smart cloud, which processes the audio recorded by the radio equipment and uses the independently developed audio artificial intelligence recognition algorithm to extract the spectral characteristics of the audio for intelligent recognition and classification. It also determines whether the poultry is sick and the type of disease, and provides real-time feedback to the front-end module for veterinary review.

[0041] Specifically, at the initial stage, a detailed key factor analysis is carried out to determine the key factors that affect the quality of audio data and the effect of target detection, as well as the environmental and physiological indicators that affect the health of poultry; a target detection algorithm model is developed, integrating signal processing, pattern recognition, and biometric recognition technologies to conduct comprehensive analysis and target detection on the collected multimodal data; in the process of algorithm model training, the random changes and noise conditions of the actual environment are simulated, and real scene data is introduced as training material; convolutional neural networks, recurrent neural networks, and generative adversarial networks are integrated into the network model, and the attention mechanism is combined to enhance the feature learning ability; cross-validation is used to evaluate model performance, and different hyperparameter settings are extensively tested to find the optimal network configuration; comprehensive model verification and evaluation are carried out to ensure that the performance meets expectations, and the optimized model is deployed in practical applications.

[0042] Preferably, the information processing mechanism of biological systems is used to build an algorithm framework based on biological computing models including DNA computing, cell computing, and neuron computing, and the model is endowed with biological characteristics of self-adaptation, self-organization, and self-repair; the multi-source heterogeneous data recorded in the real breeding room is used as the model input, and the original multi-source heterogeneous data is annotated, cleaned, and sliced; a feature extraction engine based on DNA molecular computing is designed, and the input data is subjected to parallel filtering and mapping operations using DNA sequences and their pairing rules for efficient feature extraction; features from different sources are fused to construct a multimodal feature representation, and each data slice is converted into a three-dimensional tensor; selection based on The model architecture of DNA / cell / neuron computing includes molecular convolutional networks and dendritic pulse networks; it simulates the biological antigen-antibody recognition mechanism, uses the labeled data as the antigen pattern, and performs pattern matching classification through immune computing; uses cross entropy loss and other loss functions to calculate the gap between the predicted label and the true label; introduces evolutionary algorithms to optimize model parameters so that the model has evolutionary adaptability, and trains the model to survive and evolve in an artificial life virtual environment; after the trained model is deployed, it introduces an online learning strategy to continuously obtain data from the actual environment and dynamically update the model parameters; among them, represents the batch size, represents the time step or frequency step, and represents the feature dimension.

[0043] It should be noted that the recognition accuracy and data processing capabilities of poultry health monitoring have been significantly improved by introducing biocomputing-based methods (such as DNA computing, cell computing, and neuron computing) and multimodal data analysis. Using evolutionary algorithms and online learning strategies, the system not only has strong adaptive and evolutionary capabilities to cope with changes in the breeding environment and new types of diseases, but also accelerates model development and optimization by combining the application of virtual and real environments, providing a safe and efficient method to predict and respond to health problems. These features give the solution significant advantages over existing technologies in improving breeding efficiency and ensuring poultry welfare.

[0044] Specifically, the design of a feature extraction engine based on DNA molecular computing includes the following steps: obtaining multi-source heterogeneous data after annotation, cleaning and slicing; constructing a DNA molecular filter that simulates the structure of the cochlea to perform parallel frequency band filtering on the audio signal; designing a DNA concentration encoding mechanism to map the energy of each frequency band to the DNA concentration to simulate the nonlinear perception of the human ear; compressing and removing redundancy of DNA-encoded speech features by designing a DNA enzyme cutting pattern; using a deep learning model to extract visual features from video data, including poultry posture and mouth movements, and integrating the visual features with speech features based on biological computing to construct a multimodal feature representation; constructing a basic In the virtual poultry farming scene of the Metaverse, large-scale multi-perspective and multi-modal simulation data is collected; by comparing with real data and expert knowledge, semi-supervised or unsupervised learning is used to automatically annotate the simulation data, and the annotated simulation data is transformed and amplified; the voiceprint characteristics of different poultry are used as antigen patterns, and an antibody library is modeled through immune computational modeling for voiceprint recognition and health status classification; the model is placed in the virtual environment of the Metaverse to survive and evolve, and an evolutionary algorithm is introduced to optimize the model architecture and parameters to adapt to the changing virtual and real environment; the trained model is deployed in the actual application environment, and an online learning strategy is introduced to continuously update the antibody library to adapt to the new data distribution.

[0045] Specifically, the construction of a DNA molecular filter simulating the cochlear structure includes the following steps: based on the frequency range and frequency band division of the human cochlea, a set of DNA sequence sets {S1, S2, ..., S n}, where each sequence S i The length L i and nucleotide sequence, corresponding to the center frequency f of the ith band filter i and bandwidth B i ; Using DNA coding algorithm, the sampling frequency is f s The input speech signal x(t) is encoded into a DNA sequence X of length L; extract each X i The concentration value C i , forming the filtered frequency band component set {C1,C2,...,C n}; Perform DNA molecular computation on DNA sequence X→{X1,X2,...,X n},in Represents DNA sequence hybridization operation; calculate each X i The number of A, T, C, and G nucleotides in i , T i , C i , G i , then X i DNA concentration C i =A i +Ti +C i +G i ; Set the concentration values ​​{C1,C2,...,C n} is normalized to obtain the relative energy E of the i-th frequency band i =C i / ∑C j i,j=1,2,...,n;using formula F i =aln(1+bE i ) i Perform nonlinear mapping to obtain the nonlinear characteristics F of the i-th frequency band i , where a and b are constants; {F1, F2, ..., F n}Re-encode into a DNA sequence Y of length M; design a set of K DNA enzymes E = {E1, E2, ..., E K}, cut Y according to the encoding rules to get Y j , calculate each Y j Middle A i , T i , C i , G i The number of compressed features is output as the compression feature; if the compression feature is not within the preset threshold range, the redesigned DNA sequence set {S1, S2, ..., S n}, otherwise the compressed features are output.

[0046] Preferably, simulating random changes and noise conditions in the actual environment includes the following steps: using 3D modeling software to create a realistic poultry farming scene to simulate the structure and material of the chicken house, and setting the environmental parameter adjustment function to reflect the dynamic changes of the actual farming environment; adding a poultry 3D model, simulating the movement and physiological response of the poultry through a programming algorithm, and generating random environmental events; using measured noise data to create a noise model, and simulating the spatial distribution and temporal changes of noise according to different areas and conditions of the farm; designing a noise propagation and attenuation model, considering the propagation characteristics of sound in different materials and structures, and the influence of spatial geometry on the sound field; introducing real scene data including historical meteorological data, farm environmental monitoring records and poultry behavior logs; combining the simulated data with the real data, and increasing the diversity and coverage of the training set through data augmentation technology.

[0047] Furthermore, determining whether poultry is sick and the type of disease it has includes the following steps: obtaining poultry audio data and key acoustic features extracted from the poultry audio data by a feature extraction engine of DNA molecular computing; using a deep learning model to extract poultry behavioral features from the video, including posture, gait, and mouth movement, as auxiliary feature input; using semi-supervised or unsupervised learning to automatically annotate simulation data, and transforming and amplifying the annotated simulation data; based on the principle of biological immune computing, using the annotated normal poultry voiceprint as an antigen pattern, and training an antibody library through an evolutionary algorithm as a baseline model; preliminarily matching the normal voiceprint pattern in the baseline antibody library with the input audio, and calculating a similarity score; for audio with a score higher than a preset confidence threshold, using an attention mechanism to focus on spatiotemporal context information from multimodal features, combining prior knowledge including poultry age and species, accurately determine whether it is a normal behavior pattern; for suspected abnormal audio, use the generative adversarial network for anomaly detection, and identify deviant samples from the normal voiceprint distribution; for the detected abnormal audio, match the predefined disease voiceprint library, environmental noise library and non-chicken call library respectively, and calculate their respective similarity score vectors; input the similarity score vector into an attention fusion module, and introduce abdominal cavity video and environmental perception information for weighted fusion; output the weighted fused comprehensive score vector to a classification decision module, calculate the posterior probability according to the Bayesian decision theory, and make the final health status judgment; for the audio judged to be sick, compare it with the disease voiceprint library in detail to identify the specific disease type and its confidence; send the decision results including healthy, non-chicken call, sick and its type and confidence score to the front-end module in real time through the message queue.

[0048] Furthermore, the present embodiment also provides a multimodal biocomputing-driven intelligent poultry health monitoring method, including deploying Raspberry Pi 4b and small microphones in multiple chicken houses, and placing the microphones in the center of the cage tops; continuously recording the sound environment of the chickens through a hardware terminal to collect various mixed sound data; uploading the collected audio data to the data processing module of the poultry smart cloud in real time, and performing noise filtering, segmentation and preprocessing on the uploaded audio data; using the feature extraction engine of DNA molecular computing to extract key acoustic features from the audio data, and combining the poultry behavior features extracted from the video to construct a multimodal feature representation; through deep learning models and bioimmune computing principles, Perform a preliminary match on the normal behavior patterns of poultry, identify abnormal behaviors, and perform anomaly detection and disease identification through attention mechanisms and generative adversarial networks; display the terminal device status, sound list and poultry health monitoring results on the page and mobile application front end; veterinarians can view the monitoring data of a specific chicken farm, including equipment status and health status, in real time on internal computers or smartphone applications; use trained audio artificial intelligence recognition algorithms to analyze audio data, judge and classify chicken sounds in real time to identify whether there are signs of illness; once the system identifies a sick sound, it automatically locates and marks the corresponding terminal device and the building where it is located, and provides real-time feedback to the front end to assist veterinarians in responding quickly.

[0049] This embodiment also provides a computer device, which is suitable for the case of a poultry health intelligent monitoring system driven by multimodal biological computing, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the poultry health intelligent monitoring system driven by multimodal biological computing as proposed in the above embodiment.

[0050] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0051] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the multimodal biological computing-driven intelligent poultry health monitoring system proposed in the above embodiment is implemented.

[0052] In summary, this system can accurately identify the health status and disease types of poultry by integrating multimodal data and applying advanced biocomputing and deep learning technologies; by early identification and timely treatment of poultry health problems, the system helps to reduce mortality and productivity loss caused by diseases, and reduce the overall risk and potential economic losses of farms; the system provides supporting Web visualization pages and APPs, allowing users to view the health status of chickens in the chicken house in real time; health monitoring is achieved by entering the system using mobile applications or Web pages, allowing inspectors to conveniently monitor the health status of poultry at any time and place.

[0053] Example 2

[0054] Reference Figure 8 to Figure 10 , which is the second embodiment of the present invention, provides a multimodal biocomputing-driven poultry health intelligent monitoring system. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0055] Specifically, for example, the health of poultry near a terminal device in a breeding room is monitored for a period of time during the day, such as from 2 a.m. to 4 a.m. According to data provided by poultry farmers, the most obvious vocalizations of sick chickens occur from 2 a.m. to dawn, and this is also the most common time for manual inspections.

[0056] Preferably, during these 2 hours, a terminal device records a total of 7200 seconds of WAV audio files, and the sampling rate of the audio files is 22050Hz to ensure that the sound data of the poultry calls are completely recorded. However, long WAV files cannot be directly used as input for artificial intelligence recognition algorithms, so some preprocessing operations are performed on the sound.

[0057] Specifically, the audio data is preprocessed by high-pass filtering to remove low-frequency sounds and filter out noise in the breeding room, such as chickens flapping their wings, walking noises, and natural sounds in the breeding room that interfere with recognition. Figure 8 As shown, the preprocessed sound is Fig. 9 As shown in the figure. The pre-processed sound is sliced, and the 7200-second sound is sliced ​​into 1440 slices with a fixed length of 5 seconds for subsequent feature extraction. Based on the sliced ​​audio data, the feature extraction engine based on DNA molecular computing extracts audio data features. The partial representation of the feature vector is as follows Fig.10 shown.

[0058] It should be noted that Figure 8 and Fig. 9 The horizontal axis represents time, and the vertical axis represents amplitude.

[0059] Finally, by comparing the prediction results with those of experts, the accuracy rate reached 98%. Compared with the traditional manual inspection method, the introduction of the poultry health monitoring system significantly improves the detection speed, and the terminal equipment can work 24 hours a day, greatly saving time and labor costs. This efficient monitoring system is of great significance to the poultry farming industry.

[0060] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A multimodal biocomputing-driven poultry health intelligent monitoring system, characterized by: include, A hardware terminal device module, including a Raspberry Pi 4b and a small microphone, wherein the hardware terminal device module collects poultry audio data in real time and uploads the poultry audio data to a data processing module; A data processing module, including a poultry smart cloud, is used to receive and process poultry audio data sent by the front-end module. The poultry smart cloud performs health recognition and classification analysis on the received poultry audio data by applying a self-developed audio artificial intelligence recognition algorithm to determine whether the poultry is sick and the type of disease, and feeds back the recognition results to the front-end module in real time; A front-end module, including a mobile application and a web page, is used for veterinarians to remotely monitor the health status of chickens in a chicken house. The front-end module parses and organizes the recognition results fed back by the data processing module and displays them to users through a web page or a mobile application interface; The processing flow of the data processing module is as follows: At the start-up stage, a detailed key factor analysis was conducted to identify key factors affecting audio data quality and target detection, as well as environmental and physiological indicators that affect poultry health; Develop target detection algorithm models, integrate signal processing, pattern recognition, and biometrics technologies, and conduct comprehensive analysis and target detection on collected multimodal data; During the algorithm model training process, the random changes and noise conditions of the actual environment are simulated, and real scene data is introduced as training material; Integrate convolutional neural networks, recurrent neural networks, and generative adversarial networks into the network model, and combine them with attention mechanisms to enhance feature learning capabilities; Use cross-validation to evaluate model performance and extensively test different hyperparameter settings to find the optimal network configuration; Conduct comprehensive model validation and evaluation to ensure performance meets expectations, and deploy optimized models into real-world applications; The target detection algorithm model developed includes the following contents: Drawing on the information processing mechanism of biological systems, we build an algorithm framework based on biological computing models, including DNA computing, cell computing, and neuron computing, and endow the model with biological characteristics of self-adaptation, self-organization, and self-repair; The multi-source heterogeneous data recorded in the real breeding room is used as the model input, and the original multi-source heterogeneous data is annotated, cleaned and sliced; Design a feature extraction engine based on DNA molecular computing, which uses DNA sequences and their pairing rules to perform parallel filtering and mapping operations on input data for efficient feature extraction; Features from different sources are fused to construct a multimodal feature representation, and each data slice is converted into a three-dimensional tensor batch,fea,feature_dim; Selected model architectures based on DNA / cell / neuron computing include molecular convolutional networks and dendritic spike networks; Simulate the biological antigen-antibody recognition mechanism, use the labeled data as antigen patterns, and perform pattern matching classification through immune computing; Cross entropy loss is used as the loss function to calculate the gap between the predicted label and the true label; Introducing evolutionary algorithms to optimize model parameters enables the model to have evolutionary adaptability, and train the model to survive and evolve in an artificial life virtual environment; After the trained model is deployed, an online learning strategy is introduced to continuously obtain data from the actual environment and dynamically update the model parameters; Among them, batch represents the batch size, fea represents the time step or frequency step, feature_dim represents the feature dimension; The design of a feature extraction engine based on DNA molecular computing comprises the following steps: Obtain multi-source heterogeneous data after annotation, cleaning and slicing; Construct a DNA molecular filter that simulates the structure of the cochlea and performs parallel band filtering on audio signals; Design a DNA concentration encoding mechanism to map the energy of each frequency band to DNA concentration, simulating the nonlinear perception of the human ear; By designing the DNA enzyme cutting pattern, the DNA encoded speech features are compressed and de-redundant; Use deep learning models to extract visual features including poultry posture and mouth movements from video data, and fuse the visual features with speech features based on biocomputing to construct a multimodal feature representation; Construct a virtual poultry farming scenario based on the Metaverse and collect large-scale simulation data with multiple perspectives and modalities; By comparing with real data and expert knowledge, semi-supervised or unsupervised learning is used to automatically annotate simulated data, and the annotated simulated data is transformed and amplified; The voiceprint characteristics of different poultry are used as antigen patterns, and the antibody library is modeled through immune computation for voiceprint recognition and health status classification; The model is placed in the virtual environment of the metaverse to survive and evolve, and an evolutionary algorithm is introduced to optimize the model architecture and parameters to adapt to the changing virtual and real environment; Deploy the trained model to the actual application environment and introduce online learning strategies to continuously update the antibody library to adapt to the new data distribution; The construction of a DNA molecular filter simulating the cochlear structure comprises the following steps: Based on the frequency range and band division of the human cochlea, a set of DNA sequences {S1, S2, ..., S n }, where each sequence S i Length L i and nucleotide sequence, corresponding to the center frequency f of the ith band filter i and bandwidth B i ; Using DNA coding algorithm, the sampling frequency is f s The input speech signal x(t) is encoded into a DNA sequence X of length L; Extract each X i The concentration value C i , forming the filtered frequency band component set {C1,C2,...,C n }; Perform DNA molecular computation on a DNA sequence X→{X1,X2,...,X n },in Represents DNA sequence hybridization operation; Calculate each X i The number of A, T, C, and G nucleotides in i , T i , C i , G i , then X i DNA concentration C i =A i +T i +C i +G i ; The concentration values ​​{C1,C2,...,C n } is normalized to obtain the relative energy E of the i-th frequency band i =C i / ∑C j i,j=1,2,...,n; Using Formula F i =aln(1+bE i ) i Perform nonlinear mapping to obtain the nonlinear characteristics F of the i-th frequency band i , where a and b are constants; {F1,F2,...,F n }Re-encode into a DNA sequence Y of length M; Design a group of K DNA enzymes E = {E1, E2, ..., E K }, cut Y according to the encoding rules to get Y j , calculate each Y j Middle A i , T i , C i , G i The number of is output as compressed features; If the compression feature is not within the preset threshold range, the re-involved DNA sequence set {S1, S2, ..., S n }, otherwise the compressed features are output; The simulation of random changes and noise conditions in the actual environment includes the following steps: Use 3D modeling software to create realistic poultry farming scenes to simulate the structure and material of the chicken house, and set environmental parameter adjustment functions to reflect the dynamic changes of the real farming environment; Add poultry 3D models, simulate poultry movements and physiological responses through programming algorithms, and generate random environmental events; Use measured noise data to create a noise model and simulate the spatial distribution and temporal variation of noise according to different areas and conditions of the farm; Design noise propagation and attenuation models, taking into account the propagation characteristics of sound in different materials and structures, as well as the impact of spatial geometry on the sound field; The introduction of real-world scenario data includes historical meteorological data, farm environmental monitoring records, and poultry behavior logs; Combine simulated data with real data to increase the diversity and coverage of the training set through data augmentation techniques; The method of determining whether poultry is sick and the type of sickness comprises the following steps: Obtain poultry audio data and key acoustic features extracted from the poultry audio data by a feature extraction engine of DNA molecular computing; Use deep learning models to extract poultry behavior features from videos, including posture, gait, and mouth movement, as auxiliary feature inputs; Use semi-supervised or unsupervised learning to automatically annotate simulation data, and transform and amplify the annotated simulation data; Based on the principle of biological immune computing, the annotated normal poultry voiceprints are used as antigen patterns, and the antibody library is trained by evolutionary algorithms as the baseline model; Preliminarily match the normal voiceprint pattern in the baseline antibody library with the input audio and calculate the similarity score; For audio with a score higher than a preset confidence threshold, the attention mechanism is used to focus on spatiotemporal context information from multimodal features, combined with prior knowledge including poultry age and breed, to accurately determine whether it is a normal behavior pattern; For suspicious abnormal audio, we use generative adversarial networks to detect anomalies and identify deviant samples from normal voiceprint distribution; For the detected abnormal audio, match the predefined disease voiceprint library, environmental noise library and non-chicken call library respectively, and calculate their respective similarity score vectors; The similarity score vector is input into an attention fusion module, and the abdominal video and environmental perception information are introduced for weighted fusion; Output the weighted fusion comprehensive score vector to a classification decision module, calculate the posterior probability according to Bayesian decision theory, and make the final health status judgment; For audio that is judged to be diseased, it is carefully compared with the disease voiceprint library to identify the specific disease type and its confidence level; The decision results, including healthy, non-crowning, sick, their types and confidence scores, are sent to the front-end module in real time through the message queue.

2. A multimodal biocomputing-driven poultry health intelligent monitoring method, based on the multimodal biocomputing-driven poultry health intelligent monitoring system according to claim 1, characterized in that: include, Deploy Raspberry Pi 4b and small microphones in multiple chicken houses, and place the microphone in the center of the cage roof; Continuously record the chickens’ sound environment through a hardware terminal to collect various mixed sound data; The collected audio data is uploaded to the data processing module of the Poultry Smart Cloud in real time, and the uploaded audio data is subjected to noise filtering, segmentation and preprocessing; The feature extraction engine of DNA molecular computing is used to extract key acoustic features from audio data, and combined with poultry behavioral features extracted from the video to construct a multimodal feature representation; Through deep learning models and biological immune computing principles, the normal behavior patterns of poultry are preliminarily matched, abnormal behaviors are identified, and anomaly detection and disease identification are performed through attention mechanisms and generative adversarial networks; Display terminal device status, sound list and poultry health monitoring results on the web page and mobile application front end; Veterinarians can view monitoring data including equipment status and health status in real time on a specific chicken farm on an internal computer or smartphone application; Apply trained audio artificial intelligence recognition algorithms to analyze audio data, judge and classify chicken sounds in real time to identify whether there are signs of illness; Once the system recognizes the sound of illness, it automatically locates and marks the corresponding terminal device and the building where it is located, and provides real-time feedback to the front end to assist the veterinarian in responding quickly.

3. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the multimodal biocomputing-driven poultry health intelligent monitoring system according to claim 1 are implemented.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the multimodal biocomputing-driven poultry health intelligent monitoring system of claim 1 are implemented.

Images