Livestock intelligent management system and method that deeply integrates predictive virtual fence control and automated physique assessment

Abstract

20 26 20 51 88 30 J un 2 02 6 A B S T R A C T 2 0 2 6 2 0 5 1 8 8 3 0 J u n 2 0 2 6 A B S T R A C T 2 0 2 6 2 0 5 1 8 8 3 0 J u n 2 0 2 6

Description

TECHNICAL FIELD 5

[0001] The present invention relates to the field of intelligent management of livestock husbandry technologies, and in particular, to a livestock intelligent management system and a method that deeply integrates predictive virtual fence control and automated physique assessment, achieving integrated management of precise identification, behavior prediction, dynamic constraints, and health assessment of livestock individuals. 10                        BACKGROUND

[0002] The large-scale development of livestock husbandry has put forward higher requirements for livestock management, and the core challenges focus on two dimensions: regional control and individual monitoring.

[0003] At the level of regional control, traditional physical fences have problems such as 15 high construction costs, poor flexibility, and susceptibility to livestock casualties. Virtual fence technology has become an important direction for industry development by replacing physical barriers with electronic boundaries. However, the existing virtual fence technology has the following limitations: firstly, most systems adopt a trigger based response mechanism, which only intervenes when livestock touch or approach the boundary, resulting in delayed 20 intervention timing and difficulty in achieving effective guidance; secondly, the stimulation strategy lacks dynamic adaptability and usually adopts a fixed hierarchical stimulation sequence (such as a progressive pattern of sound, vibration and electric shock), which makes livestock prone to adaptation and reduces intervention effectiveness; thirdly, setting stimulation parameters uniformly without considering individual physiological differences may result in 25 excessive or insufficient stimulation; fourthly, lack of predictive analysis ability, unable to 2026205188   30 Jun 2026 adjust management and control strategies in advance according to livestock behavior trends.

[0004] At the individual monitoring level, traditional Body Condition Score (BCS) and weight measurement rely on manual operation, which has problems such as strong subjectivity, poor consistency, and low efficiency. In recent years, computer vision technology has been 5 introduced into this field, but existing solutions mostly rely on static image analysis and cannot achieve continuous monitoring; although some solutions use 3D vision technology, they lack precise correlation mechanisms with individual identity recognition, rendering it difficult to form traceable health records.

[0005] Besides that, the existing gate fence systems based on radio frequency identification 10 devices, (RFID) and other radio frequency identification technologies have efficiency limitations in single target recognition: traditional RFID gate fences are limited by radio frequency signal conflicts and overlapping reading ranges, rendering it difficult to accurately distinguish and match individual livestock with their corresponding physique data when multiple livestock pass through simultaneously or rapidly continuously, resulting in high rates 15 of missed and false detections and low passage efficiency; the existing single passage path with a narrow gate design can avoid such problems, but it seriously restricts the detection throughput in large-scale breeding scenarios and cannot meet the needs of efficient management of modern pastures.

[0006] More importantly, the existing virtual fence system and physique assessment system 20 have isolated data, rendering it impossible to dynamically optimize control strategies based on individual health state, resulting in a lack of system synergy. SUMMARY

[0007] The present invention provides a livestock intelligent management system and a method that deeply integrates predictive virtual fence control and automated physique 25 assessment, overcoming the current deficiency of being unable to provide dynamic and accurate control strategies based on individual health state due to data isolation between virtual fence systems and physique assessment systems.

[0008] The technical solution of the present invention to solve the above technical problems 2026205188   30 Jun 2026 is as follows.

[0009] On the one hand, the present invention provides a livestock intelligent management system that deeply integrates predictive virtual fence control and automated physique assessment, including: a wearable terminal subsystem, a gate fence assessment subsystem, and 5 a cloud management subsystem, where the wearable terminal subsystem is worn on a livestock’ neck and includes a data acquisition module configured to collect multidimensional perception data of the livestock, a processor module configured to generate a risk quantification indicator of the livestock crossing a boundary, and a stimulus execution device configured to stimulate the livestock to 10 regulate a movement behavior, where the processor module is provided with a trained temporal prediction model configured to predict a movement trajectory of the livestock in a future time window and a probability distribution of intersection with a preset virtual boundary, where the gate fence assessment subsystem is deployed above a livestock passage path and includes a visual acquisition module and a data processing module; the visual 15 acquisition component is configured to collect image data of the livestock passing below, and the data processing module is configured to process the image data to obtain posture evaluation indicators and a mass estimation value of the livestock; where the cloud management subsystem is connected to the wearable terminal subsystem and the gate fence assessment subsystem through communication, and includes a 20 collaborative optimization module; where the collaborative optimization module generates a personalized adjustment factor based on the posture evaluation indicators and mass estimation value, and synchronizes them to the wearable terminal subsystem; where the processor module dynamically selects a stimulation strategy from a simulation strategy set based on the risk quantification indicator and applies the personalized 25 adjustment factor to modify stimulation parameters of the stimulation strategy (exposing animals with smaller mass or poorer body shape to lower intensity stimuli); then, a stimulation output is executed by the stimulus execution device.

[0010] On the basis of the above technical solution, the present invention can also be improved as follows. 30

[0011] Furthermore, the wearable terminal subsystem further includes a terminal body with 2026205188   30 Jun 2026 waterproof and weather resistant functions, and a flexible fixing strap is configured for the terminal body to detachedly worn on the livestock’ s neck.

[0012] Furthermore, a solar panel is fixedly provided outside the terminal body, and an identity recognition unit for individual livestock identification is further provided on the 5 terminal body, and the identity recognition unit includes one or more of a numerical number, a two-dimensional code, and a Radio Frequency Identification, (RFID) tag. The identity identification information is presented in the form of optical visual codes (two-dimensional codes) and / or radio frequency identification tags, and the individual identification step includes parsing the optical visual codes based on image recognition algorithms or obtaining radio 10 frequency identification information based on radio frequency reading devices.

[0013] Furthermore, a bottom of the terminal body is further provided with a counterweight block configured to balance the center of gravity of the terminal body.

[0014] Furthermore, the data acquisition module includes a main control unit, a satellite positioning unit, an inertial measurement unit, a geomagnetic induction unit, and an 15 environmental temperature monitoring unit that are provided on the terminal body.

[0015] Furthermore, the stimulus execution device includes one or more of an electric shock unit, a vibration unit, and an audio unit provided on the terminal body. Each unit is preferably arranged with asymmetric distribution in spatial layout.

[0016] Furthermore, the terminal body is provided with a rechargeable battery module that 20 supplies power to the data acquisition module, the processor module, and the stimulus execution device.

[0017] Furthermore, the visual acquisition module includes a depth camera that captures images of the livestock passing through the gate fence from above.

[0018] Furthermore, the cloud management subsystem further includes a data storage 25 module, a model training module, and a human-computer interaction module, where the data storage module is configured to store livestock archives, multidimensional perception data, physique data, and system logs; where the model training module trains and optimizes a prediction model based on accumulated data, and the human-computer interaction module is configured to provide access interfaces to management personnel. 30

[0019] It should be noted that a temporal prediction model to be trained by the model 2026205188   30 Jun 2026 training module adopts a deep learning architecture based on self-attention mechanism, which captures temporal dependencies through multi head attention mechanism, including input layer, position encoding layer, encoder layer, decoder layer, and output layer. The encoder layer captures temporal dependencies through multi head attention mechanism, and the model is 5 embedded in the wearable terminal subsystem for real-time operation.

[0020] Furthermore, the wearable terminal subsystem, the gate fence assessment subsystem, and the cloud management subsystem are connected by a network communication subsystem signal.

[0021] Furthermore, the network communication subsystem includes a wide area network 10 communication module and a local area communication module, where the wide area network communication module includes a cellular mobile network or a satellite communication network, and the local area communication network includes a wired local area network, a wireless local area network, or a low-power Internet of Things, (IoT) based on Long Range Radio, (LORA). 15

[0022] On the other hand, the present invention further provides a livestock intelligent management method that deeply integrates predictive virtual fence control and automated physique assessment, including the following steps: S1: data collection and generation of a risk quantification indicator: using the wearable terminal subsystem to collect multidimensional perception data of livestock at a preset 20 sampling frequency, preprocessing and extracting features from the data, constructing highdimensional temporal feature vectors, using the prediction model to predict a movement trajectory of livestock in a future time window and a probability distribution of intersection with the preset virtual boundary, comprehensively evaluating a risk level of the livestock crossing the boundary based on prediction results, and generating a risk quantification indicator; 25              S2: intelligent physique assessment and personalized adjustment factor generation: using the gate fence assessment subsystem to collect images of the livestock passing through the livestock passage path below, establishing a one-to-one data association between the images and the identity information of the identity recognition unit, performing three-dimensional reconstruction processing on the images, and calculating posture data of the livestock, including 30 posture evaluation indicators, geometric parameters, and mass estimation value; receiving and 2026205188   30 Jun 2026 storing the posture data, by the cloud management subsystem, generating the personalized adjustment factor based on the posture data, and then transmitting the personalized adjustment factor to the wearable terminal subsystem; S3: stimulus selection output: dynamically selecting the stimulation strategy from 5 the stimulation strategy set based on the risk quantification indicator, by the wearable terminal subsystem, and applying the personalized adjustment factor to modify stimulation parameters of the stimulation strategy; then, executing the stimulus output, by the stimulus execution device; S4: response monitoring: continuously collecting actual behavioral response data of 10 the livestock, by the wearable terminal subsystem, after the stimulus is applied, and updating prediction models and risk quantification indicator based on the response data.

[0023] Specifically, the multidimensional perception data in S1 includes latitude and longitude coordinates obtained through the satellite positioning unit, acceleration and angular velocity data obtained through the inertial measurement unit, azimuth data obtained through the 15 geomagnetic induction unit, and temperature data obtained through the environmental temperature monitoring unit.

[0024] Specifically, preprocessing of data in S1 includes filtering denoising, outlier removal, and coordinate system transformation; extracting features of data in S1 includes: motion speed features, calculating instantaneous velocity based on continuous position data; motion direction 20 characteristics, integrating acceleration, angular velocity, and azimuth data to calculate the motion direction angle; acceleration characteristics: obtaining linear acceleration and angular acceleration based on the inertial measurement unit; behavioral state characteristics: based on acceleration features, identifying behavioral states including stillness, walking, running, and bowing; and constructing the high-dimensional temporal feature vectors in S1 refers to merge 25 the above characteristics.

[0025] Specifically, the prediction results in S1 include intersection time, intersection position, and intersection probability value, where the risk quantification indicator is calculated by weighing the following factors: 30            R=wi x D + W2 x T + W3 x B + W4 x E + ws x H, 2026205188   30 Jun 2026 where D is a distance factor, calculated based on a distance between a current position and the virtual boundary; T is a time factor, calculated based on the predicted intersection time; B is a behavioral factor, calculated based on a current behavioral state of the livestock; E is an environmental factor, calculated based on environmental temperature; H is a 5 historical factor, calculated based on historical response patterns of the livestock to stimuli; wi to ws are weight coefficients of each factor.

[0026] In other words, the risk quantification indicator is weighted based on the following factors: predicted distance from the virtual boundary, predicted intersection time, livestock behavior state, environmental conditions (temperature), and historical response patterns. 10

[0027] Specifically, in S2, performing three-dimensional reconstruction processing on the image and calculating the posture data of the livestock, including posture evaluation indicators, geometric parameters, and mass estimation value in S2 includes the following steps: S21: performing coordinate transformation on the image to obtain an original point cloud data, followed by denoising and filtering original point cloud of the image to remove 15 environmental noise and outliers; S22: preforming voxel structure and semantic segmentation of filtered point cloud to extract livestock subject regions; S23: converting the segmented point cloud into a continuous 3D mesh model using triangulation algorithm; 20              S24: calculating physique parameters of the livestock based on three-dimensional point cloud data: posture evaluation indicators - extracting a contour curve of the livestock’s back, calculating curvature features of the curve, and matching it with a standard template to obtain posture scores; geometric parameters of body shape - calculating spatial dimensions of the livestock such as length, width, and height; mass estimation value - calculating the mass 25 based on point cloud volume estimation and density coefficient.

[0028] Specifically, generating the personalized adjustment factor based on posture data by the cloud management subsystem in S2 includes calculating an adjustment factor for an upper limit of an electric shock intensity based on the mass estimation value: the smaller the mass, the smaller the adjustment factor, and reducing the upper limit of the electric shock intensity; 30 calculating an adjustment factor of a stimulus trigger threshold based on the posture evaluation 2026205188   30 Jun 2026 indicators: the worse the posture, the larger the adjustment factor, and increasing the stimulus trigger threshold.

[0029] Specifically, the stimulation strategy set in S3 includes stimulation modes, stimulation intensity, and stimulation application timing; where the stimulation modes at least 5 include: a single stimulation mode, that is, applying acoustic energy stimulation, mechanical vibration stimulation, or electric shock stimulation separately; a combination stimulation mode, that is a combination of multiple stimulation modes applied simultaneously; a skip mode, that is, skipping low-intensity stimuli and directly applying high-intensity stimuli based on the risk quantification indicator. The stimulation strategy set includes multiple stimulation modalities 10 and their combinations, and the selection process is not constrained by a predetermined fixed order.

[0030] Specifically, the behavioral response data in S4 includes: immediate response features, acceleration changes, direction changes, and posture adjustments of the livestock at the moment of stimulus application; trajectory deviation characteristics, a degree of deviation 15 between an actual motion trajectory of the livestock and the predicted trajectory; boundary proximity change: the trend of changes in the distance between the livestock and virtual boundary after stimulation; behavioral state transition: a transition time and a transition quality of the livestock from its current behavioral state to its target behavioral state; response data effectiveness index E that quantifies a stimulus effect is calculated as 20 follows: E=a x AV + p x AD + y x AB, where AV is a change in velocity, AD is a change in distance of the boundary, AB is an improvement in behavior state, and a, p, and y are weight coefficients; when the E value is below a preset threshold, marking the stimulus as an invalid response. 25

[0031] Compared with the existing technology, the beneficial effects of the present invention are as follows.

[0032] The livestock intelligent management system and method provided by the present invention, which deeply integrate predictive virtual fence control and automated physique assessment, utilize the wearable terminal subsystem and the gate fence assessment subsystem 30 to obtain multidimensional perception data and posture data of livestock, and further process 2026205188   30 Jun 2026 them to obtain the risk quantification indicator and the personalized adjustment factor. The fusion of the two provides dynamic optimization control strategies based on different individual specific factors, thereby overcoming the shortcomings of the existing technology that virtual fence control systems and individual physique data are isolated from each other, stimulation 5 strategy lacks adaptive changes, and prediction models cannot be continuously updated automatically. It realizes closed-loop collaboration of individual recognition, behavior prediction, accurate constraints, and health monitoring, which improves management efficiency and reducing livestock stress and injury risks. BRIEF DESCRIPTION OF DRAWINGS 10

[0033] FIG. 1 is an overall architecture schematic diagram of a livestock intelligent management system provided by the present invention.

[0034] FIG. 2 is a detailed diagram of an overall architecture of the system shown in FIG. 1.

[0035] FIG. 3 is a schematic diagram of a hardware layout of a wearable terminal subsystem 15 within the system shown in FIG. 1.

[0036] FIG. 4 is a schematic diagram of a hardware layout of an assessment gateway of a gate fence assessment subsystem in the system shown in FIG. 1.

[0037] FIG. 5 is a schematic diagram of a multimodal data fusion and trajectory prediction process of the livestock intelligent management system and method provided by the present 20 invention.

[0038] FIG. 6 is a logical schematic diagram for stimulus decision of the system and method provided by the present invention.

[0039] FIG. 7 is a process flow schematic diagram for the gate fence assessment subsystem in the system to perform 3D point cloud reconstruction and grid processing on captured cow 25 images provided by the present invention. DESCRIPTION OF EMBODIMENTS

[0040] The following will provide a clear and complete description of the technical solution 2026205188   30 Jun 2026 provided by the present invention in combination with the accompanying drawings and specific embodiments. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the 5 protection scope of the present invention.

[0041] In the description of the present invention, if terms indicating orientation such as “up”, “down”, “left”, “right”, “top”, “bottom”, “inside”, “outside” are used, the orientation or position relationship indicated is based on the orientation or position relationship shown in the accompanying drawings, only for the convenience of describing the present invention and 10 simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention.

[0042] Embodiment 1

[0043] Referring to FIG.s 1 and 2, an embodiment of the present invention provides a 15 livestock intelligent management system that deeply integrates predictive virtual fence control and automated physique assessment. It includes four functional subsystems: a wearable terminal subsystem 200, a gate fence assessment subsystem 100, a network communication subsystem (including LORA nodes 210 and LORA base stations 230), a cloud management subsystem 300, and a mobile APP 400 / 401. 20

[0044] The wearable terminal subsystem 200 is worn on a livestock’ neck (livestock), responsible for collecting multi-dimensional perception data of the livestock, executing stimulus output, managing energy supply, and communicating with external nodes. This subsystem includes a data acquisition module, a processor module, a stimulus execution device, and a battery module; the data acquisition module includes a main control unit, a satellite 25 positioning unit, an inertial measurement unit, a geomagnetic induction unit, an environmental temperature monitoring unit, and an identity identification unit. Where the main control unit coordinates the operation of various functional modules; the satellite positioning unit obtains geographical coordinates of animals; the inertial measurement unit perceives motion posture; the geomagnetic induction unit determines the direction and orientation; the environmental 30 temperature monitoring unit monitors environmental parameters; the stimulus execution device 2026205188   30 Jun 2026 executes multimodal stimuli; the battery module provides power support; and the identity identification unit carries individual identification information.

[0045] The gate fence assessment subsystem 100 is deployed above a livestock passage path (such as a gate fence 120), responsible for visual collection, identity recognition, and 5 physique assessment of a passing livestock; this subsystem includes a visual acquisition module, a data processing module, and a communication module; the visual acquisition module collects depth and color images of the livestock from a top-down perspective; the data processing module executes image processing, point cloud reconstruction, feature extraction, and physique analysis algorithms; and the communication module uploads analysis results to the cloud 10 management subsystem 300 through a communication transmission 110.

[0046] The network communication subsystem includes a wide area network communication link and a local area network communication link; the wide area network communication link adopts a cellular mobile network or a satellite communication network to achieve remote data transmission; and the local area network communication link adopts low-15 power wide area network technology (such as LoRa); the LORA node 210 is provided in the wearable terminal subsystem 200, and the LORA base station 230 is provided in the gate fence assessment subsystem 100 or independently deployed to achieve short-range low-power communication.

[0047] The cloud management subsystem 300 is deployed on a cloud server and includes a 20 data storage module, a model training module, a collaborative optimization module, and a human-computer interaction module; the data storage module stores animal archives, perception data, physique data, and system logs; the model training module trains and optimizes a prediction model based on accumulated data; the collaborative optimization module executes personalized parameter adjustment strategy; the human-computer interaction module provides 25 access interfaces to management personnel; the mobile app 400 is used to display BCS indicators and trends, and the mobile app 401 is used to display the position and area of the cow, and supports adjusting the cow’s location through the app.

[0048] Embodiment 2

[0049] Referring to FIG. 3, an embodiment provides an external hardware layout of the 30 wearable terminal subsystem (LIVESTOCK WEARABLE MULTI-MODAL SENSOR 2026205188   30 Jun 2026 TERMINAL). The wearable terminal subsystem also integrates functional modules such as satellite positioning unit, inertial measurement unit, geomagnetic induction unit, environmental perception unit, communication unit, and energy supply unit (battery module) internally.

[0050] The wearable terminal subsystem 200: made of flexible materials with a curvature 5 that adapts to contours of animal necks. In an implementation mode, medical grade silicone material is used, which has waterproof and dustproof properties.

[0051] The satellite positioning unit: used to obtain geographic coordinates of animal. In an implementation mode, a multi-mode satellite positioning chip is used to support joint positioning of multiple constellations, with a positioning accuracy of up to meters. 10

[0052] The inertial measurement unit: used to perceive motion acceleration and angular velocity and calculate attitude information. In an implementation mode, a six-axis inertial sensor is used, including a three-axis accelerometer and a three-axis gyroscope.

[0053] The geomagnetic induction unit: used to sense a direction of a geomagnetic field and assist in determining the direction of motion. In an implementation mode, a three-axis 15 magnetometer is used.

[0054] The environmental perception unit: used to monitor environmental temperature. In an implementation mode, a digital temperature sensor is used.

[0055] The communication unit: used for data exchange with external nodes. In an implementation mode, a low-power wide area network communication module (LORA node 20   210) is used to support long-distance low-power transmission.

[0056] In this embodiment, the stimulus execution device includes an electric shock unit, a vibration unit, and an audio unit electrically installed on a terminal body.

[0057] The electric shock unit 201: which includes at least one electrode for applying current stimulation. In an implementation mode, the electrode is located at a bottom center 25 position of the terminal body 200, ensuring effective contact with the skin of the animal’s neck.

[0058] The vibration unit 202: which includes one or more vibration generating devices for generating tactile stimuli. In an implementation mode, the vibration generating device is built into the terminal body 200, adopting a triangular asymmetric layout in space, staggered from an audio module 203, to enhance the directional discrimination of stimuli. 30

[0059] The audio unit 203: which includes one or more sound generating devices for 2026205188   30 Jun 2026 generating audible sound or ultrasound stimuli. In an implementation mode, the sound generation device is built into the terminal body 200 and distributed in a triangular asymmetric layout in the 0 degree, 120 degrees, and 240 degrees directions of the terminal body 200 in space. 5

[0060] The wearable terminal subsystem provided in this embodiment further includes a counterweight block 204 that is provided at a bottom of the terminal body 200, used to balance the center of gravity of the terminal and ensure stable contact between each functional module and the animal’s neck.

[0061] The wearable terminal subsystem provided in this embodiment further includes a 10 flexible fixing band (elastic band) 205: used to fix the terminal body 200 to the animal’s neck. In an implementation mode, a flexible strap with adjustable length is used.

[0062] The wearable terminal subsystem provided in this embodiment further includes a solar panel 206 that is provided on an outer surface of the terminal body 200, which can be flexible and configured to collect solar energy to supply power or supplement energy to the 15 terminal.

[0063] The wearable terminal subsystem provided in this embodiment further includes an identity recognition unit provided on the terminal body 200, which is also attached to an outer surface of the terminal body 200, carrying encoded information that can be used for individual identification. In an implementation mode, an optical visual code (such as two-dimensional QR 20 code 207) is used. By combining the QR code 207 with the camera (or depth camera) that is fixed on the gate fence assessment subsystem 100, it is possible to accurately match the data obtained by the wearable terminal with the gate fence evaluation data, thereby effectively solving the problem of difficult matching of RFID or LORA when multiple animals pass through the gate fence 120 at the same time. 25

[0064] The wearable terminal subsystem provided in this embodiment further includes a main battery: a rechargeable lithium battery that supplies power to each functional module is built into the terminal body 200.

[0065] Embodiment 3

[0066] Referring to FIG. 4, an embodiment provides a hardware layout diagram of the gate 30 fence assessment subsystem deployed above the animal passage path. 2026205188   30 Jun 2026

[0067] The gate fence assessment subsystem includes gate fence assessment gateway, which integrates visual acquisition, data processing, and communication functions. In an implementation mode, gateway housing is provided with a solar panel 102 and an antenna 103 for independent power supply and wireless communication. 5

[0068] A 3D depth camera 101, used to capture depth and color images of the animal. In an implementation mode, a depth camera based on structured light or stereo vision principles is installed vertically above the gate fence 120, at a height of 2-2.5 meters from the ground, with a field of view covering the entire cross-section of the passage path.

[0069] The gate fence 120: adopts a wide passage path design, allowing multiple animals 10 to pass through at the same time. Unlike traditional RFID based gate fence systems, this system uses the 3D depth camera 101 to capture depth and color images of the animal, combined with the QR code 207 on the terminal body of the wearable terminal subsystem, to achieve a visual and QR code joint recognition mechanism. When multiple animals pass through the gate fence at the same time, the system accurately associates the depth point cloud data of each animal 15 with the QR code identification based on spatial position information. Even if there is obstruction or cross movement between animals, it can still accurately identify the threedimensional physique parameters of each individual. This design breaks through the recognition bottleneck of traditional RFID technology in multi-target scenarios, thereby significantly improving the detection efficiency of gate fences and the accuracy of individual identification. 20

[0070] The communication transmission 110: the gate fence assessment gateway sends the collected data and processing results to the cloud management subsystem, such as a server through wired or wireless means.

[0071] The workflow of the gate fence assessment subsystem 100 is as follows: when the animal enters an area of the gate fence 120, the 3D depth camera 101 is 25 triggered to start capturing image sequences; the collected depth image provides spatial distance information for each pixel point, and the color image provides appearance texture information; based on the color images, identifying the identity information on the wearable terminals; in an implementation mode, the two-dimensional QR code 207 attached to the terminal body 200 is parsed using image recognition algorithms to obtain the unique identity 30 code of the animal; the use of QR code 207 can effectively solve the problem of difficult RFID 2026205188   30 Jun 2026 or LORA matching when multiple animals pass through the gate fence at the same time; reconstruction of three-dimensional point cloud data of animal body surface based on the deep image, referring to FIG. 7, the three-dimensional reconstruction process includes the following steps: firstly, denoising and filtering original point cloud of the image to remove 5 environmental noise and outliers; the second step is to preform voxel structure and semantic segmentation of filtered point cloud to extract animal subject regions; the third step is to convert the segmented point cloud into a continuous 3D mesh model using triangulation algorithm, which facilitates subsequent volume calculation and contour extraction; Specifically, point cloud data is obtained by converting pixel coordinates and depth values into world coordinates 10 through camera interpolation; subsequently, the point cloud data is filtered for denoising, background segmentation, and registration fusion to construct a complete 3D model; based on three-dimensional point cloud data, calculating physique parameters of the animal: posture evaluation indicators - extracting a contour curve of the animal’s back, calculating curvature features of the curve, and matching it with a standard template to obtain 15 posture scores; geometric parameters of body shape - calculating spatial dimensions of animal such as length, width, and height; mass estimation value - calculating the mass based on point cloud volume estimation and density coefficient. finally, associating the physique analysis results with animal identity and storing them. 20

[0072] An embodiment provides a method for livestock management using the livestock intelligent management system that deeply integrates predictive virtual fence control and automated physique assessment as described above, referring to FIG. 5, which includes the following steps: S1. data collection and generation of a risk quantification indicator: 25              firstly, utilizing the wearable terminal subsystem of the system (animal wearable multimodal sensor terminal) to collect data; the data acquisition module of the wearable terminal subsystem continuously collects multidimensional perception data at a preset sampling frequency; the multidimensional perception data includes: geographic spatial location data -latitude and longitude coordinates obtained through the satellite positioning unit; motion 30 posture data - acceleration and angular velocity data obtained through the inertial measurement 2026205188   30 Jun 2026 unit; direction and orientation data, the azimuth data obtained through the geomagnetic induction unit; environmental perception data - temperature data obtained through the environmental perception unit; preprocessing and extracting features from collected raw perception data, and 5 constructing high-dimensional temporal feature vectors; the preprocessing includes filtering and denoising, outlier removal, coordinate system transformation, and other operations; extracting features includes: motion speed feature: calculating instantaneous velocity based on continuous position data; motion direction characteristics: fusion of magnetometer and gyroscope data to calculate the motion direction angle; acceleration characteristics: calculating 10 linear acceleration and angular acceleration based on gyroscope data; behavioral state characteristics: based on acceleration features, identifying behavior states such as stillness, walking, running, and bowing; merging the above characteristics to construct the highdimensional temporal feature vectors; inputting the high-dimensional temporal feature vectors into a trained temporal 15 prediction model to predict a motion trajectory of the animal within a future time window; the temporal prediction model adopts a deep learning architecture based on selfattention mechanism; the model includes: an input layer that receives the high-dimensional temporal feature vectors from the past N time steps; position encoding layer - adds position encoding information to an input sequence; encoder layer - includes multiple layers of self- 20 attention encoding blocks to extract temporal feature representations; decoder layer - based on the encoder output, predict the motion trajectory of M future time steps; output layer - outputs predicted trajectory coordinates and confidence scores.

[0073] In an implementation mode of the present invention, a length of the input sequence N=30 (corresponding to 3-second historical data) and the predicted time window M=30 25 (corresponding to 3-second future data).

[0074] In an implementation mode of the present invention, it has been tested that the root means square error of a test model on a test set is less than 0.5 meters, and a prediction accuracy of intersection time is higher than 85%.

[0075] Based on the predicted trajectory, calculate a probability distribution of the 30 intersection between the animal and a preset virtual boundary, including the predicted 2026205188   30 Jun 2026 intersection time, intersection location, and intersection probability value.

[0076] Based on the predicted results, comprehensively evaluating a risk level of the livestock crossing the boundary and generating the risk quantification indicator R. The risk quantification indicator is calculated by weighing the following factors: 5            R = wix D + w2x T + w3x B + w4x E + w5x H where D is a distance factor, calculated based on a distance between a current position and the virtual boundary; T is a time factor, calculated based on the predicted intersection time; B is a behavioral factor, calculated based on a current behavioral state of the animal; E is an environmental factor, calculated based on environmental conditions such as 10 temperature; H is a historical factor, calculated based on historical response patterns of the animal to stimuli; wi to w5 are weight coefficients of each factor.

[0077] S2. Intelligent physique assessment and personalized adjustment factor generation: using the gate fence assessment subsystem to collect images of the animal passing through the livestock passage path below, establishing a one-to-one data association between 15 the images and the identity information of the identity recognition unit, performing threedimensional reconstruction processing on the images (refer to embodiment 3 for the threedimensional reconstruction process), and calculating the posture data of the livestock, including posture evaluation indicators, geometric parameters, and mass estimation value; receiving and storing the posture data, by the cloud management subsystem, generating the personalized 20 adjustment factor based on the posture data, and transmitting the personalized adjustment factor to the wearable terminal subsystem.

[0078] S3. Stimulus selection output: dynamically selecting stimulation patterns from a stimulation strategy set based on the risk quantification indicator R; the stimulation strategy set includes: a single stimulation 25 mode: applying acoustic energy stimulation, mechanical vibration stimulation, or electric shock stimulation separately; a combination stimulation mode: a combination of multiple stimulation modes applied simultaneously; a skip mode: based on a risk level, skipping low-intensity stimuli and directly applying high-intensity stimuli. in an implementation mode of the present invention, a stimulus decision follows the 30 following rule: when R<Ri, no stimulus is applied; when Ri < R<R , apply acoustic stimulation; 2026205188   30 Jun 2026 when R2 < R<R3, apply mechanical vibration stimulation or a combination of acoustic energy and mechanical vibration stimulation; when R3< R<R4, apply electric shock stimulation or a combination stimulation of mechanical vibration and electric shock; when R > R4, apply full modal combined stimulation or skip and direct apply the electric shock stimulation; 5               referring to FIG. 6, a stimulus decision logic has the following characteristics: the selection of stimulus types has no predetermined fixed order and can dynamically skip based on the risk quantification indicator; the stimulation intensity can be adjusted within a preset range; the timing of stimulus application is dynamically determined based on the predicted intersection time; 10               finally, based on the selected stimulation mode, the personalized adjustment factor obtained from the posture data of the gate fence assessment subsystem and cloud management subsystem are combined to modify stimulation parameters of the stimulation strategy, and then the wearable terminal is controlled to execute the corresponding stimulation output.

[0079] S4. Response monitoring: 15               after the stimulus is applied, continuously collecting behavioral response data from the animal, by the wearable terminals, monitoring their immediate response to the stimulus and subsequent behavioral changes; the behavioral response data includes: immediate response features - acceleration changes, direction changes, and posture adjustments of the animal at the moment of stimulus application; trajectory deviation characteristics - a degree of deviation 20 between an actual motion trajectory of the animal and the predicted trajectory; boundary proximity change - the trend of changes in the distance between the animal and the virtual boundary after stimulation; behavioral state transition - a transition time and a transition quality of the animal from its current behavioral state to its target behavioral state; in an implementation mode, response monitoring is achieved by comparing the 25 motion feature vectors before and after stimulation, and the system calculates a response effectiveness index E to quantify the stimulation effect: E = a x AV + p x AD + y x AB where AV is a change in velocity, AD is a change in distance of the boundary, AB is an improvement in behavior state, and a, p, and y are weight coefficients; when the E value is 30 below a preset threshold, marking the stimulus as an invalid response; 2026205188   30 Jun 2026 based on the actual animal behavior response data obtained from response monitoring, updating the temporal prediction model and risk quantification parameters to achieve adaptive optimization of the system;

[0080] Updating the temporal prediction model includes the following sub steps: 5               training data accumulation: saving input features (pre stimulus state, stimulus type, stimulus intensity) and output response (response effectiveness index E, actual trajectory) of this stimulus decision as training samples in a local buffer; online fine-tuning: when the local buffer accumulates enough samples (such as more than 100), the incremental learning of the temporal prediction model is triggered, and the model 10 parameters are fine-tuned based on new data to improve the prediction accuracy for that specific animal; historical factor update: based on the response effectiveness index E, updating a historical response pattern H of the animal for subsequent risk quantification calculations; when the animal consistently demonstrate high response effectiveness to specific types of stimuli, 15 reducing the risk weight of that stimulus; on the contrary, increasing the risk weight or trying other stimulus modes; individualized strategy optimization: associating and storing successful stimulus response patterns to form an individualized stimulation strategy library for the animal, which can be prioritized for subsequent stimulus decisions. 20

[0081] In an implementation mode, the model update adopts a federated learning mechanism: each wearable terminal subsystem fine tunes the model locally based on individual data, and only updates the model parameters (rather than the original data) and uploads them to the cloud management subsystem; cloud based aggregation of multi terminal model updates and trains global optimization models; the global model is regularly distributed to each terminal 25 as a basic model, thereby achieving group intelligence sharing while protecting data privacy.

[0082] Embodiment 5

[0083] An embodiment provides specific steps for achieving data collaboration between virtual fence control and physique assessment: physique data acquisition: obtaining posture evaluation indicators and mass 30 estimation value of the animal from the gate fence assessment subsystem 100; 2026205188   30 Jun 2026 parameter mapping: based on physique data, generating personalized adjustment factor for virtual fence control parameters; specifically, it includes: calculating an adjustment factor for an upper limit of electric shock intensity based on the mass estimation value - the smaller the mass, the smaller the adjustment factor, and reducing the upper limit of the electric 5 shock intensity; calculating an adjustment factor of a stimulus trigger threshold based on the posture evaluation indicators - the worse the posture, the larger the adjustment factor, and increasing the stimulus trigger threshold (delayed trigger); parameter distribution: the personalized adjustment factor is synchronized to the wearable terminal subsystem 200 through the communication network (such as LORA node 10   210, LORA base station 230); dynamic control: the wearable terminal subsystem 200 applies the personalized adjustment factor to modify default stimulation parameters during stimulus decision, thereby achieving differentiated protection for weak or large animal; in an implementation mode, an updating mechanism of the collaborative optimization 15 includes: updating physique data every time when the animal passes through the gate fence 120; the parameter adjustment factor is synchronized in real-time to the wearable terminal subsystem 200 after the physique data is updated; collaborative optimization records are saved on the server 300, and management personnel can view the BCS trend and corresponding parameter adjustment history of each animal through the mobile APP 400. 20

[0084] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A livestock intelligent management system that deeply integrates predictive virtual fence control and automated physique assessment, comprising: a wearable terminal subsystem, a gate fence assessment subsystem, and a cloud management subsystem,5        wherein the wearable terminal subsystem is worn on a livestock’ neck and comprises adata acquisition module configured to collect multidimensional perception data of the livestock, a processor module configured to generate a risk quantification indicator of the livestock crossing a boundary, and a stimulus execution device configured to stimulate the livestock to regulate a movement behavior,10       wherein the processor module is provided with a trained temporal prediction modelconfigured to predict a movement trajectory of the livestock in a future time window and a probability distribution of intersection with a preset virtual boundary,wherein the gate fence assessment subsystem is deployed above a livestock passage path and comprises a visual acquisition module and a data processing module;15        the visual acquisition component is configured to collect image data of the livestockpassing below, and the data processing module is configured to process the image data to obtain posture evaluation indicators and a mass estimation value of the livestock;wherein the cloud management subsystem is connected to the wearable terminal subsystem and the gate fence assessment subsystem through communication, and comprises a20 collaborative optimization module;wherein the collaborative optimization module generates a personalized adjustment factor based on the posture evaluation indicators and mass estimation value, and synchronizes them to the wearable terminal subsystem;wherein the processor module dynamically selects a stimulation strategy from a25 simulation strategy set based on the risk quantification indicator and applies the personalized adjustment factor to modify stimulation parameters of the stimulation strategy; then, a stimulation output is executed by the stimulus execution device.2026205188   30 Jun 20262. The livestock intelligent management system that deeply integrates predictive virtual fence control and automated physique assessment according to claim 1, wherein the wearable terminal subsystem further comprises a terminal body with waterproof and weather resistant functions, and a flexible fixing strap is configured for the terminal body to detachedly worn on 5 the livestock’ s neck.

3. The livestock intelligent management system that deeply integrates predictive virtual fence control and automated physique assessment according to claim 2, wherein a solar panel is fixedly provided outside the terminal body, and an identity recognition unit for individual livestock identification is further provided on the terminal body, and the identity recognition 10 unit comprises one or more of a numerical number, a two-dimensional code, and a Radio Frequency Identification, (RFID) tag.

4. The livestock intelligent management system that deeply integrates predictive virtual fence control and automated physique assessment according to claim 2, wherein a bottom of the terminal body is further provided with a counterweight block configured to balance the 15 center of gravity of the terminal body.

5. The livestock intelligent management system that deeply integrates predictive virtual fence control and automated physique assessment according to claim 2, wherein the data acquisition module comprises a main control unit, a satellite positioning unit, an inertial measurement unit, a geomagnetic induction unit, and an environmental temperature monitoring 20 unit that are provided on the terminal body.

6. The livestock intelligent management system that deeply integrates predictive virtual fence control and automated physique assessment according to claim 2, wherein the stimulus execution device comprises one or more of an electric shock unit, a vibration unit, and an audio unit provided on the terminal body.25        7. The livestock intelligent management system that deeply integrates predictive virtualfence control and automated physique assessment according to claim 2, wherein the terminal body is provided with a rechargeable battery module that supplies power to the data acquisition module, the processor module, and the stimulus execution device.

8. The livestock intelligent management system that deeply integrates predictive virtual 30 fence control and automated physique assessment according to claim 1, wherein the visual2026205188   30 Jun 2026acquisition module comprises a depth camera that captures images of the livestock passing through the gate fence from above.

9. The livestock intelligent management system that deeply integrates predictive virtual fence control and automated physique assessment according to claim 1, wherein the cloud 5 management subsystem further comprises a data storage module, a model training module, and a human-computer interaction module,wherein the data storage module is configured to store livestock archives, multidimensional perception data, physique data, and system logs;wherein the model training module trains and optimizes a prediction model based on 10 accumulated data, and the human-computer interaction module is configured to provide access interfaces to management personnel.

10. The livestock intelligent management system that deeply integrates predictive virtual fence control and automated physique assessment according to any one of claims 1 to 9, wherein the wearable terminal subsystem, the gate fence assessment subsystem, and the cloud 15 management subsystem are connected by a network communication subsystem signal.

11. The livestock intelligent management system that deeply integrates predictive virtual fence control and automated physique assessment according to claim 10, wherein the network communication subsystem comprises a wide area network communication module and a local area communication module,20       wherein the wide area network communication module comprises a cellular mobilenetwork or a satellite communication network, and the local area communication network comprises a wired local area network, a wireless local area network, or a low-power Internet of Things, (IoT) based on Long Range Radio, (LORA).

12. A livestock intelligent management method that deeply integrates predictive virtual25 fence control and automated physique assessment, wherein the livestock intelligent management system according to any one of claims 1 to 11 is used, and the livestock intelligent management method comprises the following steps:S1: data collection and generation of a risk quantification indicator: using the wearable terminal subsystem to collect multidimensional perception data of livestock at a preset sampling 30 frequency, preprocessing and extracting features from the data, constructing high-dimensional2026205188   30 Jun 2026temporal feature vectors, using the prediction model to predict a movement trajectory of livestock in a future time window and a probability distribution of intersection with the preset virtual boundary, comprehensively evaluating a risk level of the livestock crossing the boundary based on prediction results, and generating a risk quantification indicator;5        S2: intelligent physique assessment and personalized adjustment factor generation: usingthe gate fence assessment subsystem to collect images of the livestock passing through the livestock passage path below, establishing a one-to-one data association between the images and the identity information of the identity recognition unit, performing three-dimensional reconstruction processing on the images, and calculating posture data of the livestock, including10 posture evaluation indicators, geometric parameters, and mass estimation value; receiving and storing the posture data, by the cloud management subsystem, generating the personalized adjustment factor based on the posture data, and then transmitting the personalized adjustment factor to the wearable terminal subsystem;S3: stimulus selection output: dynamically selecting the stimulation strategy from the 15 stimulation strategy set based on the risk quantification indicator, by the wearable terminal subsystem, and applying the personalized adjustment factor to modify stimulation parameters of the stimulation strategy; then, executing the stimulus output, by the stimulus execution device;S4: response monitoring: continuously collecting actual behavioral response data of the20   livestock, by the wearable terminal subsystem, after the stimulus is applied, and updatingprediction models and risk quantification indicator based on the response data.

13. The livestock intelligent management method that deeply integrates predictive virtual fence control and automated physique assessment according to claim 12, wherein the multidimensional perception data in S1 comprises latitude and longitude coordinates obtained25 through the satellite positioning unit, acceleration and angular velocity data obtained through the inertial measurement unit, azimuth data obtained through the geomagnetic induction unit, and temperature data obtained through the environmental temperature monitoring unit.

14. The livestock intelligent management method that deeply integrates predictive virtual fence control and automated physique assessment according to claim 12, wherein preprocessing2026205188   30 Jun 2026of data in S1 comprises filtering denoising, outlier removal, and coordinate system transformation;extracting features of data in S1 comprises: motion speed features, calculating instantaneous velocity based on continuous position data; motion direction characteristics, 5 integrating acceleration, angular velocity, and azimuth data to calculate the motion direction angle; acceleration characteristics: obtaining linear acceleration and angular acceleration based on the inertial measurement unit; behavioral state characteristics: based on acceleration features, identifying behavioral states including stillness, walking, running, and bowing; and constructing the high-dimensional temporal feature vectors in S1 refers to merge the above 10 characteristics.

15. The livestock intelligent management method that deeply integrates predictive virtual fence control and automated physique assessment according to claim 12, wherein the prediction results in S1 comprise intersection time, intersection position, and intersection probability value, wherein the risk quantification indicator is calculated by weighing the following factors:15       R=wi x D + W2 x T + W3 x B + W4 x E + ws x H,wherein D is a distance factor, calculated based on a distance between a current position and the virtual boundary; T is a time factor, calculated based on the predicted intersection time; B is a behavioral factor, calculated based on a current behavioral state of the livestock; E is an environmental factor, calculated based on environmental temperature; H is a historical factor, 20 calculated based on historical response patterns of the livestock to stimuli; wi to ws are weight coefficients of each factor.

16. The livestock intelligent management method that deeply integrates predictive virtual fence control and automated physique assessment according to claim 12, wherein performing three-dimensional reconstruction processing on the image and calculating the posture data of 25 the livestock, including posture evaluation indicators, geometric parameters, and mass estimation value in S2 comprises the following steps:S21: performing coordinate transformation on the image to obtain an original point cloud data, followed by denoising and filtering original point cloud of the image to remove environmental noise and outliers;2026205188   30 Jun 2026S22: preforming voxel structure and semantic segmentation of filtered point cloud to extract livestock subject regions;S23: converting the segmented point cloud into a continuous 3D mesh model using triangulation algorithm;5        S24: calculating physique parameters of the livestock based on three-dimensional pointcloud data: posture evaluation indicators - extracting a contour curve of the livestock’s back, calculating curvature features of the curve, and matching it with a standard template to obtain posture scores; geometric parameters of body shape - calculating spatial dimensions of the livestock such as length, width, and height; mass estimation value - calculating the mass based10 on point cloud volume estimation and density coefficient.

17. The livestock intelligent management method that deeply integrates predictive virtual fence control and automated physique assessment according to claim 12, wherein generating the personalized adjustment factor based on posture data by cloud management subsystem in S2 comprises:15        calculating an adjustment factor for an upper limit of an electric shock intensity based onthe mass estimation value: the smaller the mass, the smaller the adjustment factor, and reducing the upper limit of the electric shock intensity;calculating an adjustment factor of a stimulus trigger threshold based on the posture evaluation indicators: the worse the posture, the larger the adjustment factor, and increasing the20 stimulus trigger threshold.

18. The livestock intelligent management method that deeply integrates predictive virtual fence control and automated physique assessment according to claim 12, wherein the stimulation strategy set in S3 comprises stimulation modes, stimulation intensity, and stimulation application timing,25        wherein the stimulation modes at least comprise: a single stimulation mode, that is,applying acoustic energy stimulation, mechanical vibration stimulation, or electric shock stimulation separately; a combination stimulation mode, that is a combination of multiple stimulation modes applied simultaneously; a skip mode, that is, skipping low-intensity stimuli and directly applying high-intensity stimuli based on the risk quantification indicator.2026205188   30 Jun 202619. The livestock intelligent management method that deeply integrates predictive virtual fence control and automated physique assessment according to claim 12, wherein the behavioral response data in S4 comprises: immediate response features, acceleration changes, direction changes, and posture adjustments of the livestock at the moment of stimulus application;5 trajectory deviation characteristics, a degree of deviation between an actual motion trajectory of the livestock and the predicted trajectory; boundary proximity change: the trend of changes in the distance between the livestock and virtual boundary after stimulation; behavioral state transition: a transition time and a transition quality of the livestock from its current behavioral state to its target behavioral state;10        response data effectiveness index E that quantifies a stimulus effect is calculated as follows:E=a x AV + p x AD + y x AB,wherein AV is a change in velocity, AD is a change in distance of the boundary, AB is an improvement in behavior state, and a, p, and y are weight coefficients; when the E value is below a preset threshold, marking the stimulus as an invalid response.

Images