Pig individual identification and health monitoring system based on biological characteristics

Abstract

The invention discloses a pig individual identification and health monitoring system based on biological characteristics, and relates to the technical field of medical treatment, and the system comprises a multi-mode biological characteristic collection module which is used for obtaining pig 3D facial characteristics, ear blood vessel distribution characteristics, gait dynamic characteristics and physiological data; the biological feature fusion recognition module is used for fusing multi-modal features based on a Transform architecture and realizing individual unique identification and feature template dynamic updating; and the digital twinborn health monitoring module is used for constructing a pig full-life-cycle digital twinborn model and realizing health state evaluation and disease early warning in combination with an LSTM neural network. According to the pig individual recognition and health monitoring system based on the biological characteristics, 1024-dimensional standardized feature vectors are constructed through cooperation of multiple devices such as an RGB-D camera, a near-infrared camera and a binocular vision sensor, modal weights are dynamically distributed in combination with a Transform network, and high-precision recognition in a complex scene is achieved.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and specifically to a pig individual identification and health monitoring system based on biometric characteristics. Background Art

[0002] Research on the group behavior of animals shows that healthy live pigs have specific living habits, such as excretion and feeding behaviors. By observing these behaviors, suspected pig diseases can be determined in a timely manner, reducing the likelihood of swine epidemics. In order to have a very accurate and timely understanding of the health status of live pigs, breeders need to monitor their behaviors such as feeding, drinking, and excretion. However, the behaviors, social interactions, and psychological activities of pigs are often intertwined, which often affect their health status. Therefore, during the breeding process of live pigs, breeders need to promptly detect and diagnose their abnormal behaviors. For large and medium-sized live pig breeding bases, relying on the individual behaviors of breeders for monitoring is quite inefficient. Moreover, in large-scale live pig breeding, there are too many individuals, and breeders have limited energy and cannot monitor each pig. This is likely to result in missed inspections. If diseased pigs are not promptly inspected and treated, it may cause spread, affecting other normal pigs and resulting in immeasurable property losses.

[0003] The invention patent with the patent number CN115777571A discloses a monitoring system and method for identifying pig behaviors. The system includes a camera for capturing real-time picture information in the pigsty. It is characterized by further comprising: an audio acquisition module for acquiring the sound information of live pigs; an infrared temperature detector for following and acquiring the body temperature information of a specified pig; and a data processing module for acquiring picture information, sound information, and the body temperature information of pigs. The method includes: S1: acquiring the image information of individual live pigs in the pigsty; S2: acquiring and combining and saving the external shape characteristics, sound characteristics, and body temperature characteristics of each live pig individual, and independently numbering each pig.

[0004] However, the existing monitoring systems for identifying pig behaviors rely on a single sensor and cannot fuse multi-dimensional biometric characteristics such as sound, body temperature, and gait. As a result, individual identification is vulnerable to occlusion and lighting effects, and it is difficult to distinguish individuals with similar appearances. In addition, traditional methods do not establish a dynamic feature template library, have poor adaptability to the external shape changes during the growth process of pigs, and require frequent manual recalibration, which is time-consuming and laborious. Moreover, the existing systems only evaluate the health status through a few indicators such as body temperature and food intake, lack real-time analysis of key characteristics such as gait stability and body surface temperature distribution, resulting in delayed disease warning. They do not construct an individualized digital twin model, cannot dynamically fit the growth curve, and the warning threshold is fixed and not adjusted dynamically according to the age and breed of pigs, making it difficult to achieve precise health management. Summary of the Invention

[0005] In view of the deficiencies of the prior art, the present invention provides a pig individual identification and health monitoring system based on biometrics, which solves the existing problems.

[0006] To achieve the above objectives, the present invention is realized through the following technical solutions: A pig individual identification and health monitoring system based on biometrics, comprising: A multi-modal biometric acquisition module for acquiring 3D facial features, ear blood vessel distribution features, gait dynamic features, and physiological data of pigs; A biometric fusion and recognition module that fuses multi-modal features based on the Transformer architecture to achieve individual unique identification and dynamic update of feature templates; A digital twin health monitoring module that constructs a digital twin model of the entire life cycle of pigs and combines the LSTM neural network to achieve health status assessment and early disease warning.

[0007] Preferably, the 3D facial features include facial contour point cloud and nose spot distribution, the ear blood vessel features include blood vessel curvature and branch angle, and the gait dynamic features include step length and forelimb swing angle.

[0008] Preferably, the loss function of the biometric fusion and recognition module is: ; where is the dynamic weight coefficient, , , are the facial, ear, and gait feature comparison loss functions respectively, and .

[0009] Preferably, the health status assessment formula of the digital twin health monitoring module is: ; where, is the activity score, is the weight gain score, is the abnormal body surface temperature score, , , are dynamically adjusted according to the age and breed of pigs.

[0010] Preferably, the multi-modal biometric acquisition module includes: A multi-angle RGB-D camera for collecting 3D facial point cloud and contour features; A near-infrared camera (wavelength 850nm) for collecting ear blood vessel distribution images; A binocular vision sensor (baseline distance 120mm) for collecting gait sequence videos (frame rate 30fps).

[0011] Preferably, the biometric fusion recognition module includes an online learning unit that updates the individual feature template every 30 minutes according to the latest collected data to adapt to the feature changes during the growth process of pigs.

[0012] Preferably, the digital twin health monitoring module has a built-in disease prediction model that analyzes the growth curve, gait stability index (CSI), and body surface temperature distribution based on an LSTM neural network to identify limb diseases or growth abnormalities 72 hours in advance.

[0013] Preferably, the multimodal biometric acquisition module further includes: A weight sensor for real-time acquisition of pig weight data; An infrared thermal imager for collecting body surface temperature distribution and identifying fever areas.

[0014] Preferably, the digital twin health monitoring module supports a hierarchical warning mechanism: Yellow warning (health score H < 60): Trigger an alarm for abnormal activity level or abnormal weight gain; Red warning (health score H < 40): Trigger a suspected disease alarm and automatically associate historical health data to generate a diagnosis suggestion.

[0015] The present invention also discloses a health monitoring method for a pig individual recognition and health monitoring system based on biometrics, including the following steps: Step 1, data acquisition: An RGB-D camera from multiple angles, a near-infrared camera, and a binocular vision sensor collect 3D facial, ear blood vessel, and gait data of pigs in real time, and a weight sensor and an infrared thermal imager synchronously collect weight and body temperature data; Step 2, preprocessing: The edge computing node performs point cloud denoising and image enhancement on the original data to generate a 1024-dimensional standardized feature vector; Step 3, individual recognition: The biometric fusion recognition module fuses multimodal features through a Transformer architecture, outputs an individual number, and updates the individual feature template in the digital twin model; Step 4, health assessment: The digital twin platform calculates the health score H according to the health status assessment formula, analyzes the growth curve and gait data in combination with an LSTM neural network, and identifies abnormal states; Step 5, hierarchical warning: When H < 60 or H < 40, trigger a yellow or red warning respectively, synchronize it to the breeding management system and generate a processing suggestion (such as adjusting the feed formula, isolating and observing), and record the processing effect to optimize the warning model.

[0016] Beneficial effects The present invention provides a pig individual recognition and health monitoring system based on biometrics. Compared with the prior art, it has the following beneficial effects: 1. The pig individual identification and health monitoring system based on biometrics collaborates with multiple devices such as RGB-D cameras, near-infrared cameras, and binocular vision sensors to construct a 1024-dimensional standardized feature vector, combines a Transformer network to dynamically allocate modal weights, realizes high-precision identification in complex scenarios, supports fast adaptation across breeds, and the dynamic template update mechanism can adapt to the shape changes during the growth process of pigs, solving the occlusion and light dependence problems of traditional single-modal identification, and providing a data basis for precise individual management.

[0017] 2. The pig individual identification and health monitoring system based on biometrics constructs a digital twin health record based on the Gompertz growth model and the LSTM neural network, real-time fits parameters such as daily weight gain and activity level, and realizes multi-dimensional health assessment through the gait stability index and the body surface temperature anomaly score. The disease discovery speed is significantly improved through a hierarchical early warning mechanism, the fever symptoms are identified in advance, the limb diseases are warned in advance, the missed detection rate is reduced, and combined with dynamic weight adjustment and automated processing suggestions, the average slaughter time is shortened and the veterinary drug usage is reduced, reducing the breeding risk from the source and realizing precise and intelligent health management. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the system architecture of the present invention; Figure 2 It is a schematic diagram of the biometric fusion algorithm process of the present invention; Figure 3 It is a schematic diagram of the health monitoring and early warning logic of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0020] Refer to Figures 1-3 The present invention provides multiple technical solutions: The first embodiment: The pig individual identification and health monitoring system based on biometrics includes: A multi-modal biometric acquisition module for obtaining 3D facial features, ear blood vessel distribution features, gait dynamic features and physiological data of pigs. The multi-modal biometric acquisition module includes: A multi-angle RGB-D camera for collecting 3D facial point clouds and contour features; Near-infrared camera (wavelength 850nm), used to collect images of the blood vessel distribution in the ear; Binocular vision sensor (baseline distance 120mm), used to collect gait sequence videos (frame rate 30fps).

[0021] The 3D facial features include facial contour point cloud and nose spot distribution, the ear blood vessel features include blood vessel curvature and branch angle, and the gait dynamic features include step length and forelimb swing angle.

[0022] The multi-modal biometric acquisition module further includes: Weight sensor, used to obtain real-time pig weight data; Infrared thermal imager, used to collect body surface temperature distribution and identify fever areas.

[0023] Biometric fusion and recognition module, which fuses multi-modal features based on the Transformer architecture to achieve individual uniqueness identification and dynamic update of feature templates. The loss function of the biometric fusion and recognition module is: ; where is the dynamic weight coefficient, 、 、 are the facial, ear, and gait feature comparison loss functions respectively, and .

[0024] The biometric fusion and recognition module contains an online learning unit, which updates the individual feature template every 30 minutes according to the latest collected data to adapt to the feature changes during the growth of pigs.

[0025] Digital twin health monitoring module, which constructs a digital twin model for the entire life cycle of pigs, and combines the LSTM neural network to achieve health status evaluation and early disease warning. The health status evaluation formula of the digital twin health monitoring module is: ; where, is the activity score, is the weight gain score, is the body surface temperature abnormality score, 、 、 are dynamically adjusted according to the pig's age and breed.

[0026] The digital twin health monitoring module has a built-in disease prediction model, which analyzes the growth curve, gait stability index (CSI), and body surface temperature distribution based on the LSTM neural network to identify limb diseases or growth abnormalities 72 hours in advance.

[0027] The digital twin health monitoring module supports a hierarchical warning mechanism: Yellow warning (health score H < 60): Trigger an alarm for abnormal activity or abnormal weight gain; Red warning (health score H < 40): Trigger a suspected disease alarm and automatically associate historical health data to generate a diagnosis suggestion.

[0028] The second implementation method: A health monitoring method for a pig individual identification and health monitoring system based on biometrics, including the following steps: Step 1. Data collection: The multi-angle RGB-D camera, near-infrared camera, and binocular vision sensor collect 3D facial, ear blood vessel, and gait data of pigs in real time, and the weight sensor and infrared thermal imager synchronously collect weight and body temperature data; Step 2. Preprocessing: The edge computing node performs point cloud denoising and image enhancement on the original data to generate a 1024-dimensional standardized feature vector; Step 3. Individual identification: The biometric fusion identification module fuses multi-modal features through the Transformer architecture, outputs the individual number, and updates the individual feature template in the digital twin model; Step 4. Health assessment: The digital twin platform calculates the health score H according to the health status assessment formula, analyzes the growth curve and gait data in combination with the LSTM neural network, and identifies abnormal states; Step 5. Hierarchical warning: When H < 60 or H < 40, trigger a yellow or red warning respectively, synchronize it to the breeding management system and generate processing suggestions (such as adjusting the feed formula, isolating and observing), and record the processing effect to optimize the warning model.

[0029] The third implementation method: System deployment I. System hardware and data collection Multi-modal biometric collection module 3D facial and ear feature collection RGB-D camera: Deploy 2 - 4 depth cameras (such as Intel RealSense D435i) on the ceiling of the feeding area and rest area in the pigsty, with an installation height of 2.5 - 3m, a coverage range of 5m × 5m, collect the frontal and side facial point cloud data of pigs every 5 minutes, and synchronously obtain RGB images for nose spot feature extraction.

[0030] Near-infrared camera: Install a near-infrared imaging device (wavelength 850nm, such as FLIR A35) 30cm above the feeding trough in the feeding area, focus on the ear area of the pigs, collect the blood vessel distribution images, with a daily collection frequency of ≥20 times, covering different lighting conditions (automatic switching between day / night modes).

[0031] Gait and physiological data collection Binocular vision sensor: Install a binocular camera 1.8 m above the pigsty passage (baseline distance 120 mm, such as ZED2i) to capture videos of pigs walking, and synchronously record dynamic features such as step length and swing angle of the front limbs. For a single pig walking once, ≥10 gait cycles are collected.

[0032] Weight and body temperature sensors: Embed 4 groups of pressure sensors (such as Honeywell weighing modules) on the ground in the feeding area to automatically obtain weight data when pigs are feeding; install an infrared thermal imager (such as NEC R300) on the top of the pigsty to scan the entire area every 10 minutes and identify abnormal temperatures in areas such as the ears and back (such as local temperature increase ≥1°C).

[0033] (2) Edge computing node preprocessing process Point cloud data processing Use voxel filtering (voxel size 5 mm) to remove 3D facial point cloud noise, separate the pig's face from the background through the RANAC plane segmentation algorithm, and retain valid point cloud data (number of points ≥5000).

[0034] Extract key points of the facial contour (15 feature points such as the tip of the nose and the base of the ear), and calculate 50-dimensional geometric features such as contour curvature and symmetry.

[0035] Image and video processing Perform adaptive histogram equalization on the near-infrared ear images to enhance blood vessel contrast, use the U-Net network to segment the blood vessel area, and extract 20-dimensional features such as blood vessel branch angle and curvature radius.

[0036] For the gait video, track the motion trajectory of the joint points through the optical flow method, smooth the trajectory by combining with the Kalman filter, and calculate 12-dimensional dynamic features such as step length, step frequency, and center of gravity offset.

[0037] Data fusion Concatenate 3D facial features (4096 dimensions), ear blood vessel features (512 dimensions), gait features (128 dimensions), weight (1 dimension), and body temperature (1 dimension) into a 1024-dimensional standardized feature vector, and upload it to the central server through the MQTT protocol.

[0038] II. Biometric fusion recognition module Transformer network architecture Input layer: Perform positional encoding on different modality features (facial feature encoding length 50, gait sequence encoding length 10) to generate feature vectors with positional information.

[0039] Multi-head self-attention layer: 8 attention heads respectively learn the interaction relationships of different modalities, such as the association weights between facial features and ear features (formula: ), dynamically allocate weights (e.g., the weight of ear features in the night scene is increased by 30%).

[0040] Fusion layer: Map the attention output to the individual recognition space through a fully connected layer, and output a 128-dimensional feature vector for individual matching.

[0041] Dynamic template update mechanism Build an individual feature template library, and the initial template is generated by averaging the data collected 3 times at birth.

[0042] Online learning algorithm: When the matching degree between the newly collected features and the template is <95%, trigger template update, and use exponential moving average (EMA) to fuse the new features (update formula: ), ensure that the template update error ≤ 2%.

[0043] (2) Individual recognition process Real-time matching: When a pig enters the monitoring area, the edge computing node generates a feature vector and uploads it. The central server matches the individual number through the nearest neighbor algorithm (cosine distance threshold 0.8), and the matching time ≤ 200ms.

[0044] Cross-breed adaptability: For different breeds such as Landrace and Duroc, pre-train the model parameters through the meta-learning algorithm (MAML). Only 50 samples are required to complete the adaptation when deploying a new pig herd.

[0045] III. Digital twin health monitoring module (1) Construction of individualized health records Digital twin model parameters Growth curve: Fit the daily weight gain data based on the Gompertz growth model (formula: , where is the mature weight, , are breed parameters), and the prediction error ≤ 3%.

[0046] Activity pattern: Calculate the daily activity duration through gait data (threshold: piglets ≥ 8 hours, fattening pigs ≥ 5 hours), and construct the activity score Sa (0 - 100 points, a warning is triggered when it is less than 60 points) in combination with the feeding frequency (≥ 3 times / day).

[0047] Health status assessment model Dynamic weight adjustment: Automatically adjust the weights according to the pig's age (e.g., 1 - 20 days old: ; over 60 days old: ), ensure that the evaluation focus is different at different growth stages.

[0048] Body surface temperature abnormality score: Identify the temperatures of 5 key areas such as the ear and abdomen through thermal imaging data. When the temperatures of ≥3 areas exceed the mean + 1°C, points.

[0049] (2) Early warning mechanism for diseases LSTM prediction model Input features: Daily weight gain, activity duration, and body temperature data (total 21 dimensions) for the past 7 days, and output the growth trend prediction for the next 3 days.

[0050] Abnormality determination: When the daily weight gain decreases by ≥10% for 3 consecutive days and the predicted trend continues to deteriorate, trigger a growth abnormality warning.

[0051] Gait stability index (CSI) Calculation method: , when and the weight-bearing time of a single limb > 15 minutes, it is determined as limb abnormality, and the disease probability is output in combination with historical health data (such as the probability of arthritis ≥70%).

[0052] Hierarchical warning response Yellow warning (H < 60): Push abnormal information (such as "The activity level of pig No. 0012 is insufficient") through the breeding management system, and automatically adjust the feed formula (increase the energy feed by 10%).

[0053] Red warning (H < 40): Trigger an audible and visual alarm and generate an isolation suggestion, synchronously retrieve historical data to generate a diagnostic report (such as "The body temperature has increased + abnormal gait in the past 3 days, suspected of Streptococcus suis infection"), and the response time for handling suggestions ≤ 5 minutes.

[0054] The fourth implementation method: Examples, application verification in large-scale pig farms (1) Scenario configuration Pig farm scale: 1000 fattening pigs, divided into 10 pig houses, 100 pigs in each house, density 1.2㎡ / pig.

[0055] Hardware deployment: Install 4 RGB-D cameras, 2 near-infrared cameras, 1 set of binocular vision sensors, and 8 groups of weight sensors in each house.

[0056] Algorithm parameters: 5000 pigs (covering 3 breeds) for Transformer model training data, LSTM prediction window 7 days, online learning update interval 30 minutes.

[0057] (2) Operating effects Individual recognition performance Recognition accuracy under normal light is 99.2%, accuracy at night (infrared supplementary light) is 98.5%, and accuracy in the ear tag occlusion scenario is 97.3%.

[0058] The deployment time of the new pig group (Duroc pigs) is 24 hours, while the traditional method takes 2 weeks, and the cross-breed identification accuracy rate is 95%.

[0059] Health monitoring effect Early warning lead time for diseases: The fever symptom can be identified 48 hours in advance, and limb diseases can be warned 72 hours in advance. The missed detection rate is reduced from 15% to 3%.

[0060] Improvement in breeding efficiency: The average slaughter time is shortened by 7 days, the usage of veterinary drugs is reduced by 30%, and the intensity of manual inspection is reduced by 50%.

[0061] (III) Abnormal handling examples Case 1: Early warning of piglet diarrhea For piglet No. 0035, the daily weight gain decreased by 15% for two consecutive days, CSI = 0.78, the body surface temperature was normal but the activity level decreased. The system triggered a yellow warning, indicating "abnormal growth, it is recommended to conduct fecal detection". After the veterinarian's detection and confirmation of mild diarrhea, it recovered within 48 hours after timely medication, avoiding mass infection.

[0062] Case 2: Early warning of adult pig arthritis For adult pig No. 1021, the gait CSI = 0.65, the ear temperature increased by 1.2 °C, and the health score H = 35. The system triggered a red warning and recommended isolation. After X-ray examination, knee joint inflammation was confirmed, and the treatment cycle was shortened by 5 days, avoiding the deterioration of joint damage.

[0063] Through the above specific implementation manners, the present invention realizes the full-process implementation from hardware deployment, data processing to intelligent decision-making, verifies the significant advantages of multi-modal biometric fusion and digital twin technology in pig individual identification and health monitoring, meets the needs of large-scale pig farms for precision breeding and intelligent management, and has significant practical value.

[0064] Meanwhile, the content not described in detail in this specification belongs to the prior art well-known to those skilled in the art, and the model parameters of each electrical appliance are not specifically limited, and conventional equipment can be used.

[0065] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0066] Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A pig individual identification and health monitoring system based on biometrics, characterized in that: include: Multimodal biometric acquisition module, used to obtain pigs' 3D facial features, ear vascular distribution characteristics, gait dynamic characteristics and physiological data; The biometric fusion recognition module integrates multimodal features based on the Transformer architecture to achieve individual unique identification and dynamic update of feature templates; The digital twin health monitoring module builds a digital twin model of the entire life cycle of pigs and combines it with the LSTM neural network to realize health status assessment and early disease warning.

2. The pig individual identification and health monitoring system based on biometrics according to claim 1 is characterized by: The 3D facial features include facial contour point cloud and nose spot distribution, the ear blood vessel features include blood vessel curvature and branching angle, and the gait dynamic features include step length and forelimb swing angle.

3. The pig individual identification and health monitoring system based on biometrics according to claim 1, characterized in that: The loss function of the biometric fusion recognition module is: ； Among them, is the dynamic weight coefficient, , , are the facial, ear and gait feature contrast loss functions respectively, and .

4. The pig individual identification and health monitoring system based on biometrics according to claim 1 is characterized by: The health status evaluation formula of the digital twin health monitoring module is: ； in, Rate the activity level. Score weight gain, To score abnormal skin temperature, , , Dynamically adjust according to the age and breed of pigs.

5. The pig individual identification and health monitoring system based on biometrics according to claim 1 is characterized by: The multimodal biometric feature acquisition module comprises: Multi-angle RGB-D camera for collecting 3D facial point cloud and contour features; A near-infrared camera is used to collect images of the blood vessels in the ear; Binocular vision sensor, used to collect gait sequence videos.

6. The pig individual identification and health monitoring system based on biometrics according to claim 1 is characterized by: The biometric fusion recognition module includes an online learning unit, which updates the individual feature template every 30 minutes based on the latest collected data to adapt to the feature changes during the growth process of the pigs.

7. The pig individual identification and health monitoring system based on biometrics according to claim 1 is characterized by: The digital twin health monitoring module has a built-in disease prediction model, which analyzes growth curves, gait stability index and body surface temperature distribution based on the LSTM neural network to identify limb diseases or growth abnormalities in advance.

8. The pig individual identification and health monitoring system based on biometrics according to claim 1, characterized in that: The multimodal biometric feature acquisition module also includes: Weight sensor, used to obtain pig weight data in real time; Infrared thermal imager, used to collect body surface temperature distribution and identify fever areas.

9. The pig individual identification and health monitoring system based on biometrics according to claim 1, characterized in that: The digital twin health monitoring module supports a hierarchical early warning mechanism: Yellow warning: triggers an alarm for abnormal activity or weight gain; Red alert: Triggers a suspected disease alert and automatically associates historical health data to generate diagnostic recommendations.

10. A health monitoring method based on the pig individual identification and health monitoring system based on biometrics according to any one of claims 1 to 9, characterized in that: The following steps are involved: Step 1: Data collection: Multi-angle RGB-D cameras, near-infrared cameras and binocular vision sensors collect 3D facial, ear blood vessel and gait data of pigs in real time, and weight sensors and infrared thermal imagers synchronously collect weight and temperature data; Step 2: Preprocessing: The edge computing node performs point cloud denoising and image enhancement on the original data to generate a 1024-dimensional standardized feature vector; Step 3: Individual identification: The biometric fusion identification module fuses multimodal features through the Transformer architecture, outputs the individual number, and updates the individual feature template in the digital twin model; Step 4: Health assessment: The digital twin platform calculates the health score H according to the health status assessment formula, and combines the LSTM neural network to analyze the growth curve and gait data to identify abnormal conditions; Step 5. Graded warning: When H < 60 or H < 40, a yellow or red warning is triggered respectively, which is synchronized to the breeding management system and generates treatment suggestions, and the treatment effects are recorded to optimize the warning model.

Images