A method and system for evaluating the health status of free-range cattle based on ear tags

Abstract

The application discloses a free-range cattle health condition evaluation method and system based on ear tags, and belongs to the technical field of intelligent animal husbandry. The method first binds individual identification information of cattle, collects normal operation data within a set time period, and constructs an individual health baseline library containing multi-dimensional baselines; collects head inertia data, near-infrared spectrum data, position data and body temperature data through a collection terminal worn on the ear of the cattle, and performs time synchronization processing on all collected data; obtains multi-dimensional health feature data based on the collected data; compares the multi-dimensional health feature data with the individual health baseline library, generates a health grade result according to a time sequence cascading inference rule, outputs corresponding early warning information, and synchronously updates the individual health baseline library. The method can realize continuous monitoring and grading evaluation of the health state of free-range cattle individuals, and adapt to the health management needs of large-scale free-range scenes.

Description

Technical Field

[0001] This application relates to the field of smart animal husbandry technology, specifically to a method and system for assessing the health status of free-range cattle based on cattle ear tags. Background Technology

[0002] In free-range beef cattle farming, timely monitoring of livestock health is crucial for ensuring profitability. Existing health monitoring solutions based on cattle ear tags mostly only achieve location tracking, body temperature collection, and simple activity level statistics. They cannot obtain core physiological and behavioral data reflecting early health changes, such as feed intake and rumination, making it difficult to detect abnormalities in the subclinical stages of disease.

[0003] Current health assessment methods generally employ a single-dimensional judgment logic with fixed thresholds, failing to consider the temporal cascade patterns of health deterioration in ruminants or establish specific normalized benchmarks for individual differences. This leads to a high false alarm rate, easily misjudging behavioral fluctuations caused by environmental factors such as grassland changes and weather effects as health abnormalities. Furthermore, existing data collection schemes often employ continuous high-frequency sampling modes. In scenarios involving large-scale free-range grazing and insufficient public network coverage, the terminal's battery life and data transmission capabilities cannot support long-term stable deployment, making it difficult to achieve continuous health assessments at the individual level. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for assessing the health status of free-range cattle based on ear tags.

[0005] In a first aspect, embodiments of this application provide a method for assessing the health status of free-range cattle based on cow ear tags, comprising the following steps:

[0006] S1. Bind individual cattle identification information, collect routine operation data within a set time period, and build an individual health baseline database, which includes feed intake baseline, rumination baseline, behavior baseline, physical baseline and body temperature baseline;

[0007] S2. Collect head inertial data, near-infrared spectral data, position data and body temperature data through a data acquisition terminal worn on the ears of cattle, and perform time synchronization processing on all collected data;

[0008] S3. Based on head inertial data and near-infrared spectral data, calculate feeding characteristic data and rumination characteristic data, combine location data to calculate activity characteristic data and social characteristic data, calculate physical characteristic data based on feeding characteristic data and activity characteristic data, and generate multi-dimensional health characteristic data;

[0009] S4. Compare the multi-dimensional health feature data with the individual health baseline database, generate health level results according to the time-series cascade inference rules, output corresponding early warning information, and update the individual health baseline database synchronously.

[0010] In some embodiments, the process of solving the feeding characteristic data and rumination characteristic data in step S3 includes the following sub-steps:

[0011] S3.1. Perform time-series segmentation on the head inertial data, identify feeding behavior segments and rumination behavior segments, and extract feeding duration, feeding frequency, chewing number, rumination duration, rumination frequency, and chewing rhythm data.

[0012] S3.2. Match and process the near-infrared spectral data corresponding to the feeding behavior segments, and output the dietary nutrient parameters data and the risk value data of poisonous weeds. The dietary nutrient parameters include the proportion of crude protein, the proportion of neutral detergent fiber, and the proportion of acid detergent fiber.

[0013] S3.3. Combine dietary nutrient parameters and chewing frequency data to calculate dry matter intake and metabolizable energy intake data, and use rumination duration data to verify and correct the dry matter intake data.

[0014] In some embodiments, the physical feature data calculation process in step S3 includes the following sub-steps:

[0015] S3.4. Extract gait feature data of walking segments from head inertial data. Gait feature data includes stride frequency, stride variation coefficient, ground impact amplitude, and head swing symmetry.

[0016] S3.5. Calculate daily energy balance data based on metabolizable energy intake data and activity characteristic data, including activity distance, standing time and lying time;

[0017] S3.6. Combine gait characteristic data and daily energy balance data to estimate body condition score data and weight change trend data, and generate physical characteristic data.

[0018] In some embodiments, the data acquisition process in step S2 includes the following sub-steps:

[0019] S2.1. Maintain continuous acquisition of head inertial data at a set frequency, and perform real-time behavior recognition on the head inertial data;

[0020] S2.2. When feeding behavior is detected, increase the frequency of head inertial data acquisition and simultaneously start the interval acquisition of near-infrared spectral data;

[0021] S2.3. When rumination behavior is detected, adjust the head inertial data to the corresponding acquisition frequency and keep the near-infrared spectroscopy module in the off state.

[0022] In some embodiments, the individual health baseline database construction process in step S1 includes the following sub-steps:

[0023] S1.1. Enter the basic attribute information of the cattle, including breed information, age information and initial body condition information;

[0024] S1.2. Collect health status operation data within a set period, extract individual normal characteristic intervals through clustering processing, and generate individual dimensional baseline data;

[0025] S1.3. Obtain group characteristic data of cattle of the same age in the same herd, generate group reference baseline data, and integrate individual dimension baseline data with group reference baseline data into an individual health baseline library.

[0026] In some embodiments, a multi-dimensional consistency check is performed before generating the health level result; when a single-dimensional feature data deviates from the corresponding interval of the individual health baseline database while the other dimensional feature data are within the corresponding interval, it is determined to be due to environmental factors and no warning level upgrade is triggered; when multi-dimensional feature data deviates from the corresponding interval synchronously and conforms to the time-series cascading pattern, the corresponding warning level upgrade operation is performed.

[0027] In some embodiments, each set of dietary nutrient parameter data is associated with the corresponding location data, and pasture nutrient distribution map data is generated through spatial interpolation; the pasture nutrient distribution map data is then used to correct the dry matter intake calculation parameters for different regions.

[0028] In some embodiments, behavioral abnormality index data are calculated based on feeding characteristic data, activity characteristic data, social characteristic data, and diurnal rhythm data; social characteristic data is generated by topological calculation using the location data of all cattle, and includes the average distance of the herd, the duration of separation from the herd, and the synchronization rate of group activities.

[0029] In some embodiments, the temporal cascade inference rule performs judgment according to a three-level warning level; the first level warning corresponds to a state in which the feeding rhythm or behavioral index deviates in a single dimension; the second level warning corresponds to a state in which abnormal feeding continues for a set duration and is accompanied by a negative energy balance and a downward trend in body condition; the third level warning corresponds to a state in which body temperature deviates on the basis of the second level warning.

[0030] Secondly, this application provides a health status assessment system for free-range cattle based on cow ear tags, used to perform the methods described above. The system includes a smart ear tag terminal, an edge gateway device, and a cloud management platform. The smart ear tag terminal is worn on the ears of cattle and integrates an inertial sensing module, a near-infrared spectroscopy module, a positioning and communication module, a body temperature sensing module, and a main control module, used to collect multi-source raw data and perform preliminary behavior recognition. The edge gateway device is deployed in the pasture area to receive data transmitted by the smart ear tag terminal, perform feature calculation and health assessment processing locally, and output early warning information. The cloud management platform is used to store all data, update the baseline library and algorithm model, and display health status data and pasture distribution data.

[0031] The beneficial effects of this application are as follows: By constructing an individual health baseline library containing multi-dimensional data, it can adapt to the differences in breed and age of different cattle, avoiding assessment bias caused by universal fixed thresholds. Through the coupled decomposition of multi-source raw data, multi-dimensional health characteristics such as feeding, rumination, activity, and physique can be extracted from the information collected from ear tags, achieving a comprehensive representation of health status. The assessment rules using temporal cascaded inference conform to the physiological development law of changes in the health status of ruminants. Combined with a multi-dimensional consistency verification mechanism, it can effectively distinguish between environmental fluctuations and individual health abnormalities, reducing the probability of false alarms in health assessment. The use of event-triggered data acquisition logic can reduce terminal power consumption while ensuring the accuracy of feature acquisition, adapting to the long-term deployment needs of large-scale free-range scenarios; coupled with the dynamic update mechanism of the baseline, it can maintain the long-term adaptability of the assessment system and improve the stability and reliability of health monitoring. Attached Figure Description

[0032] Figure 1 This is a flowchart from this application;

[0033] Figure 2 This is a module diagram in this application. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0035] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0036] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0037] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0038] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0039] In free-range beef cattle farming, timely monitoring of livestock health is crucial for ensuring profitability. Existing ear-tagged health monitoring solutions mostly only achieve location tracking, body temperature collection, and simple activity level statistics, failing to acquire core physiological and behavioral data reflecting early health changes, such as feed intake and rumination, making it difficult to detect abnormalities in the subclinical stages of disease. Current health assessment methods generally employ a fixed threshold-based, single-dimensional judgment logic, failing to consider the temporal cascade patterns of health deterioration in ruminants or establish specific normative benchmarks for individual differences. This easily leads to misjudging behavioral fluctuations caused by environmental factors such as pasture changes and weather effects as health abnormalities, resulting in a high false alarm rate. Furthermore, existing data collection schemes often use continuous high-frequency sampling modes. In scenarios with large-scale free-range farming and insufficient public network coverage, terminal battery life and data transmission capabilities cannot support long-term stable deployment, making it difficult to achieve continuous health assessment at the individual level. Therefore, the first aspect of this application provides a method and system for assessing the health status of free-range cattle based on ear tags.

[0040] Example 1

[0041] This embodiment provides a specific implementation process for a method of assessing the health status of free-range cattle based on ear tags. This method utilizes a multimodal sensor data coupling solution and a temporal cascade inference mechanism to achieve continuous dynamic monitoring and graded assessment of the individual health status of free-range cattle. It adapts to the needs of contactless livestock and poultry health management in large-scale free-range scenarios. (Refer to...) Figure 1 The specific implementation process is as follows:

[0042] Step 1. Individual Health Baseline Construction: Bind individual cattle identification information, collect routine operational data over a set period, and construct an individual health baseline database. This database includes feed intake baseline, rumination baseline, behavioral baseline, physical baseline, and body temperature baseline, specifically:

[0043] Step 1.1. Input of individual basic attribute information:

[0044] The basic attribute information of the cattle is entered, including breed information, age information, and initial body condition information. A unique ear tag code is assigned to each cattle, and the ear tag code is linked to the corresponding basic attribute information to establish an individual data file. The initial body condition information is entered after manual assessment and serves as the initial calibration basis for estimating physical characteristics.

[0045] Step 1.2. Extraction of Individual Normal Feature Intervals:

[0046] Health status data was collected over a set period. Clustering was used to extract individual normative characteristic intervals, generating individual baseline data. The health status data consisted of multi-source raw data collected from cattle in a healthy state, including inertial data, spectral data, location data, and body temperature data. The K-means clustering algorithm was used to divide the feature data for each dimension into clusters, selecting the numerical interval corresponding to the cluster with the highest data percentage as the individual normative characteristic interval. Individual baseline data included feed intake baseline, rumination baseline, behavioral baseline, physical baseline, and body temperature baseline. Each baseline corresponded to a set of normative characteristic intervals, used for deviation comparison in subsequent health assessments.

[0047] Step 1.3. Group Reference Baseline Integration:

[0048] Group characteristic data of cattle of the same age and within the same herd are obtained to generate group reference baseline data. Individual baseline data and group reference baseline data are then integrated into an individual health baseline database. Cattle of the same age and within the same herd refer to a group of cattle of the same breed and age range within the same breeding area. Group characteristic data is calculated by taking the median value of the normal characteristic data of all cattle within the herd; the corresponding numerical range is the group reference baseline data. The group reference baseline data is used to distinguish between individual abnormalities and herd-wide environmental changes. When an individual's characteristics deviate from its own baseline but remain within the herd baseline range, it is determined to be due to herd-wide environmental factors and is not considered an individual health abnormality.

[0049] In some embodiments, the generation of the herd reference baseline may introduce spatial location weights, assigning higher weights to cattle feature data that have a higher degree of overlap with the target cattle's activity area, thereby improving the adaptability of the herd reference.

[0050] Step 2. Multi-source data synchronous acquisition: Using a data acquisition terminal worn on the cow's ear, head inertial data, near-infrared spectral data, position data, and body temperature data are collected. Time synchronization processing is performed on all collected data, specifically including:

[0051] Step 2.1. Low-frequency routine data collection:

[0052] Head inertial data is continuously collected at a set frequency for real-time behavior recognition. The head inertial data, collected by an inertial measurement unit (IMU) worn on the cow's ear, includes three-axis acceleration and three-axis angular velocity data. An IMU is a sensing device capable of detecting the linear acceleration and rotational angular velocity of an object; in this embodiment, different behavioral states are identified by collecting the cow's head motion data. Under normal conditions, head inertial data is collected at a low frequency to maintain basic behavior recognition capabilities and reduce terminal power consumption. Body temperature data and position data are collected at set intervals; all collected data are appended with a unified timestamp for subsequent data time synchronization processing.

[0053] Step 2.2. Feeding behavior triggers high-frequency data collection:

[0054] Upon detecting feeding behavior, the frequency of head inertial data acquisition is increased, and intermittent acquisition of near-infrared spectral data is simultaneously initiated. Feeding behavior is identified through head posture and chewing vibration characteristics. When a forward-leaning head posture with high-frequency chewing vibration waveforms is detected, feeding behavior is determined. Increasing the acquisition frequency of head inertial data allows for the acquisition of more refined chewing waveform features, improving the accuracy of chewing frequency and rhythm recognition. Near-infrared spectral data is acquired by a miniature near-infrared spectral sensor, with the acquisition direction directed towards the pasture area in front of the cattle's mouth, to detect the spectral reflectance characteristics of the forage. Near-infrared spectroscopy is a detection technology that analyzes the chemical composition of substances based on their reflection characteristics of near-infrared light; in this embodiment, it is used for non-contact identification of the nutritional components and types of forage. The near-infrared spectral sensor is activated only intermittently during feeding behavior to avoid power waste caused by continuous acquisition.

[0055] Step 2.3. Rumination behavior triggers adaptation data collection:

[0056] When rumination is detected, the head inertial data is adjusted to the corresponding acquisition frequency, while the near-infrared spectroscopy module remains off. Rumination is identified through the static head posture and slow, regular chewing vibrations. During rumination, the cattle's head posture is stable, and the chewing rhythm is uniform with a lower frequency than during foraging. The head inertial data is adjusted to an acquisition frequency suitable for rumination, controlling power consumption while ensuring the accuracy of rumination duration and chewing rhythm recognition. Since the cattle are not grazing during rumination, there is no need to collect forage spectral data; therefore, the near-infrared spectroscopy module is turned off to further reduce terminal power consumption.

[0057] In some embodiments, the location data acquisition frequency can be dynamically adjusted according to the behavior state. When walking is detected, the location data acquisition frequency is increased, and when lying down is detected, the location data acquisition frequency is decreased, thereby further optimizing the overall power consumption level of the terminal.

[0058] Step 3. Multi-dimensional health characteristic calculation: Based on head inertial data and near-infrared spectral data, feed intake characteristic data and rumination characteristic data are calculated. Combined with location data, activity characteristic data and social characteristic data are calculated. Based on the feed intake characteristic data and activity characteristic data, physical characteristic data are calculated to generate multi-dimensional health characteristic data, specifically including:

[0059] Step 3.1. Behavior temporal segmentation and basic feature extraction:

[0060] Temporal segmentation of head inertial data is performed to identify feeding and rumination behavior segments, extracting feeding duration, feeding frequency, chewing count, and rumination duration, rumination frequency, and chewing rhythm data. The temporal segmentation process employs a sliding window mechanism, truncating head inertial data segments according to a set window length, and extracting temporal and frequency domain features for each window. Temporal features include the number of peaks, mean amplitude, and standard deviation of amplitude; frequency domain features include the dominant frequency and spectral energy distribution. The extracted features are input into a pre-trained temporal classification model, which outputs the behavior category corresponding to each window. Windows of consecutive categories are merged into a complete behavior segment. The temporal classification model uses a lightweight one-dimensional convolutional neural network, comprising an input layer, three convolutional layers, two pooling layers, one fully connected layer, and an output layer. The input layer receives feature vectors of a set dimension; the convolutional layers extract local patterns of temporal features; the pooling layers reduce dimensionality; the fully connected layer integrates features; and the output layer outputs probability values ​​for each behavior category, selecting the category with the highest probability as the final recognition result. During the model training phase, labeled behavioral inertia data was used as the training set. The network parameters were optimized using the backpropagation algorithm, and the cross-entropy loss function was adopted. Based on the segmented behavioral fragments, the number of feeding behavior fragments per unit time was counted to obtain the feeding frequency, the duration of a single feeding behavior fragment was counted to obtain the single feeding duration, and the number of chewing peaks within a feeding behavior fragment was counted to obtain the number of chewing times. Similarly, the rumination frequency, single rumination duration, and rumination chewing rhythm data were obtained.

[0061] Step 3.2. Forage spectrum matching and nutrient parameter calculation:

[0062] Near-infrared spectral data corresponding to grazing behavior segments are matched and processed to output dietary nutrient parameters and toxic weed risk values. Dietary nutrient parameters include crude protein percentage, neutral detergent fiber percentage, and acid detergent fiber percentage. The average reflectance spectrum of multiple near-infrared spectral data collected within a single grazing behavior segment is obtained by averaging the values. The average reflectance spectrum is then matched with a built-in reference forage spectrum library to obtain the matching results. Spectral matching employs a spectral angle matching algorithm, which measures spectral similarity by calculating the angle between two spectral vectors. The spectral angle matching value is calculated for each entry in the reference forage spectrum library, and all reference entries with matching values ​​below a set threshold are selected. A weighted calculation is performed based on the reciprocal of the matching value to obtain the average dietary nutrient parameters for that grazing period. The measured spectrum is then matched with toxic weed reference spectrum entries. When the matching value is below a set threshold, the corresponding level of toxic weed risk value data is output.

[0063] Step 3.3. Calculation of dry matter intake and metabolizable energy intake:

[0064] By combining dietary nutrient parameters and chewing frequency data, dry matter intake and metabolizable energy intake were calculated. Rumination duration data were used to validate and correct the dry matter intake data. Dry matter intake was calculated using the fiber-corrected chewing energy method. Dry matter intake refers to the total weight of dry matter in the forage consumed by cattle after removing moisture. It is a fundamental indicator for assessing nutrient intake levels. In this embodiment, it was calculated by coupling chewing frequency with dietary fiber content, and used for subsequent energy balance analysis and body condition assessment. The formula for calculating dry matter intake is:

[0065]

[0066] DMI refers to dry matter intake per unit time period. This represents the total number of chews within the corresponding time period. The basic chewing dry matter conversion factor (BCC) is given by α, the fiber content correction factor is given by α, and NDF is the percentage of neutral detergent fiber in the diet. This formula reflects the physiological pattern of ruminants: the higher the dietary fiber content, the more chewing is required per unit weight of dry matter. Therefore, the BCC is corrected by the percentage of neutral detergent fiber to improve the accuracy of feed intake calculation. Metabolizable energy intake is obtained by multiplying the dry matter intake by the metabolizable energy content per unit dry matter.

[0067] The metabolizable energy content per unit dry matter is calculated from the proportions of crude protein, neutral detergent fiber, and acid detergent fiber in the diet using a pre-defined regression equation. The verification and correction process involves calculating the theoretical rumination time based on the proportion of neutral detergent fiber in the diet using a rumination duration prediction model. The theoretical rumination time is then compared with the measured rumination time. When the deviation exceeds a set range, the dry matter intake calculation is corrected in the opposite direction of the deviation; the larger the deviation exceeds the range, the greater the correction. The rumination duration prediction model is obtained through fitting a large amount of measured data and reflects the positive correlation between dietary fiber content and rumination duration.

[0068] In some embodiments, the calculation of dry matter intake can incorporate a forage height correction term. By using the grassland vegetation height data corresponding to the location data, the basic chewing dry matter conversion coefficient can be adjusted to further adapt to forage foraging scenarios at different growth stages.

[0069] Step 3.4. Calculation of Activity Features and Social Features:

[0070] Activity and social characteristic data are calculated by combining location data. Activity characteristic data includes activity mileage, standing time, and lying time. Activity mileage is calculated by path integration of trajectory points from location data. Standing and lying time are obtained by combining the posture characteristics of head inertia data with the movement status statistics of location data. A standing state is defined as a head posture that is upright with no significant movement, while a lying state is defined as a head posture that is horizontal with no movement for an extended period. Social characteristic data is generated through topological calculations using the location data of all cattle. Social characteristic data includes average herd distance, time spent away from the herd, and herd activity synchronization rate. The average herd distance is calculated by: calculating the herd's center of gravity coordinates based on the location coordinates of all cattle at the same time, calculating the distance between a single cattle's position and the herd's center of gravity coordinates, and taking the average value over a unit of time. The time spent away from the herd is calculated by: counting the total time within a unit of time when the distance between a single cattle and the herd's center of gravity exceeds a set threshold. The group activity synchronization rate is calculated as follows: calculate the Pearson correlation coefficient between the activity time series curve of an individual cow and the average activity time series curve of the group. The correlation coefficient value is the group activity synchronization rate. The higher the value, the higher the consistency between the activity pattern of an individual cow and the group.

[0071] Step 3.5. Physical characteristic data calculation:

[0072] Based on metabolic energy intake data and activity characteristic data, daily energy balance values ​​are calculated. Gait characteristic data for walking segments are extracted from head inertia data, including gait frequency, stride variation coefficient, ground impact amplitude, and head sway symmetry. Combining gait characteristic data and daily energy balance value data, body condition score data and weight change trend data are estimated to generate physical characteristic data. The daily energy balance value is calculated based on the principle of energy balance, reflecting the difference between the cattle's daily energy intake and expenditure.

[0073] The maintenance metabolic energy requirement is calculated based on the cattle's current weight, breed coefficient, and ambient temperature; it represents the energy needed to maintain the cattle's basic life activities. The activity metabolic energy expenditure is calculated based on the activity coefficient derived from daily activity distance, standing time, and lying time; it represents the additional energy consumed by the cattle during activity. A positive daily energy balance value indicates that energy intake exceeds expenditure, and the cattle are on a weight gain trend; a negative daily energy balance value indicates that energy intake is less than expenditure, and the cattle are on a weight loss trend. Body condition scores are estimated using a multiple linear regression model. Body condition scoring is a standardized grading method used in animal husbandry to assess the body fat reserves and nutritional status of cattle. The score results reflect the long-term nutritional level and health status of the cattle. In this embodiment, body condition scores are estimated by coupling gait characteristics with energy balance data, enabling continuous dynamic monitoring without manual measurement. The body condition score estimation formula is:

[0074]

[0075] Where BCS is the estimated body condition score, SF is cadence, SV is the stride variation coefficient, IA is the impact amplitude, HS is the head swing symmetry, and EB is the daily energy balance value. to These are regression coefficients obtained through pre-training. Step frequency is calculated as the ratio of peak acceleration counts within a walking segment to the walking duration; stride variation coefficient is calculated as the ratio of the standard deviation of consecutive strides to the mean stride length; ground impact amplitude is calculated as the peak longitudinal acceleration at the moment of foot strike during walking; head sway symmetry is calculated as the ratio of the left and right amplitudes of the left-right angular velocity during walking. Weight change trend data is calculated using the cumulative daily energy balance value and the weight-energy conversion coefficient; combined with initial weight data, current weight can be estimated.

[0076] In some embodiments, image acquisition devices and tag reading devices can be installed at the necessary passageways of the pasture. When cattle pass by, images of their body sides are acquired and ear tags are read. Calibration body condition score data is obtained through image recognition. The coefficients of the body condition score regression model are updated regularly using this calibration data to improve the accuracy of long-term estimation.

[0077] Step 3.6. Calculation of Behavioral Abnormality Index:

[0078] Based on feed intake characteristics, activity characteristics, social characteristics, and circadian rhythm data, a behavioral abnormality index was calculated. The behavioral abnormality index quantifies the degree of behavioral deviation in cattle, reflecting the severity of behavioral syndromes. It is calculated using a multi-dimensional weighted summation method, and the formula is as follows:

[0079]

[0080] BAI stands for Abnormal Behavior Index. The weighting coefficients for the four dimensions are as follows: F1 is the score for feeding motivation, F2 is the score for activity willingness, F3 is the score for social connection, and F4 is the score for circadian rhythm. The score for each dimension is calculated by comparing the corresponding feature data with the corresponding baseline in the individual's health baseline database. The greater the deviation, the lower the score for the corresponding dimension, and the lower the overall value of the behavioral abnormality index. The feeding motivation score is calculated based on feeding duration, feeding frequency, and feeding rhythm stability; the activity willingness score is calculated based on activity distance, standing / lying ratio, and average moving speed; the social connection score is calculated based on average herd distance, time spent away from the herd, and group activity synchronization rate; and the circadian rhythm score is calculated based on the degree of consistency between the diurnal distribution pattern of feeding, activity, and rest and the baseline.

[0081] Step 3.7. Generation and parameter correction of grassland nutrient distribution map:

[0082] By associating each set of dietary nutrient parameters with corresponding location data, spatial interpolation is used to generate pasture nutrient distribution map data. This pasture nutrient distribution map data is then used to correct the dry matter intake calculation parameters for different areas. Dietary nutrient parameters for each feeding period are bound to the coordinates of the corresponding feeding location, forming nutrient sample points with spatial coordinates. Spatial interpolation is then performed on all nutrient sample points to generate a nutrient parameter raster map covering the entire breeding area, which is the pasture nutrient distribution map data. The pasture nutrient distribution map data includes crude protein distribution maps, neutral detergent fiber distribution maps, and toxic weed risk distribution maps. Spatial interpolation uses an inverse distance weighted interpolation method, with weights calculated based on the reciprocal of the distance between the sample point and the interpolation point; closer sample points have higher weights. Parameter correction process: Based on the current location coordinates of the cattle, the corresponding nutritional parameters in the pasture nutrient distribution map are called, and the fiber content correction coefficient and unit dry matter metabolizable energy content in the dry matter feed intake calculation formula are replaced with the measured parameters of the corresponding location to improve the adaptability of feed intake and energy calculation in different areas.

[0083] In some embodiments, spatial interpolation can be performed using Kriging interpolation, which improves the smoothness and accuracy of the interpolation results by considering the spatial autocorrelation of sample points, thus adapting to free-range ranch scenarios with complex terrain conditions.

[0084] Step 4. Time-Series Cascaded Health Assessment and Baseline Update: Compare multi-dimensional health characteristic data with the individual health baseline database, generate health level results according to the time-series cascaded inference rules, output corresponding early warning information, and synchronously update the individual health baseline database. Specifically, this includes:

[0085] Step 4.1. Multi-dimensional feature baseline comparison:

[0086] Multidimensional health characteristic data is compared with an individual health baseline database to calculate the deviation data for each dimension. The multidimensional health characteristic data includes feeding characteristics, rumination characteristics, activity characteristics, social characteristics, physical characteristics, body temperature data, and behavioral abnormality index data. The deviation data is calculated as follows: the absolute value of the difference between the characteristic value and the median value of the corresponding baseline interval is divided by the width of the baseline interval to obtain the standardized deviation; the larger the standardized deviation value, the greater the degree of deviation from the normal state.

[0087] Step 4.2. Multi-dimensional consistency verification:

[0088] Before generating health level results, a multi-dimensional consistency check is performed. If a single-dimensional feature data deviates from the corresponding interval of the individual's health baseline while the other dimensional feature data remain within their respective intervals, it is determined to be due to environmental factors, and no warning level upgrade is triggered. If multiple dimensional feature data simultaneously deviate from their corresponding intervals and conform to a cascading temporal pattern, the corresponding warning level upgrade operation is performed. The consistency check is used to distinguish between individual health abnormalities and feature fluctuations caused by environmental factors. For example, if only the feed intake dimension deviates from the baseline, while rumination, activity, body condition, and body temperature are all within the baseline interval, it is determined to be a fluctuation in feed intake caused by changes in the forage quality of the grazing area, and does not belong to individual health abnormalities. When the feed intake dimension deviates along with abnormal rumination, a decrease in behavioral indices, and a negative energy balance, it is determined to be a multi-dimensional consistent abnormality, conforming to the cascading pattern of health deterioration, and an upgrade of the warning level is allowed.

[0089] Step 4.3. Timing Cascade Early Warning Determination:

[0090] Health level results are generated according to the temporal cascade inference rules, and corresponding warning information is output. The temporal cascade inference rules are executed according to a three-level warning level; the first level warning corresponds to a single-dimensional deviation in the feeding rhythm or behavioral index; the second level warning corresponds to a state where abnormal feeding persists for a set duration and is accompanied by a negative energy balance and a declining trend in body condition; the third level warning corresponds to a state where body temperature deviates from the second level warning level. The warning level determination is implemented using a graded threshold function, and the calculation formula is as follows:

[0091]

[0092] Where Level represents the warning level. The deviation of the feeding dimension. This represents the deviation of the behavioral abnormality index. For the degree of deviation from energy balance, Degree of deviation from physical condition The values ​​represent the body temperature deviation, with Duration representing the duration of the abnormality, T1 being the first-level deviation threshold, T2 the second-level deviation threshold, and T3 the abnormal body temperature threshold. The duration threshold is used. The temporal cascade reasoning judgment rule follows the temporal pattern of ruminant disease development. Health abnormalities initially manifest as subtle changes in feeding and behavior, which, after continued development, lead to energy metabolism disorders and a decline in body condition, eventually resulting in clinical symptoms such as elevated body temperature. Determining the warning level step-by-step according to this temporal pattern effectively reduces the probability of false alarms and improves the reliability of health assessments. When outputting warning information, the dimensions involved in the abnormality and the possible types of abnormalities are simultaneously labeled, providing a reference direction for on-site inspections.

[0093] Step 4.4. Dynamically updating the individual health baseline database:

[0094] The individual health baseline database is updated synchronously using a sliding window rolling update mechanism. Feature data indicating a healthy state within a set time period is selected, and the individual's normal characteristic interval is recalculated, replacing the original individual-dimensional baseline data. During the update process, abnormal data under warning conditions is automatically excluded; only data judged to be in a normal state is used in baseline calculation, ensuring that the baseline always reflects the normal health status of the cattle. Simultaneously, the baseline interval is adaptively adjusted according to the cattle's growth stage and seasonal changes; growth stage adjustments are based on breed growth curve calculations, and seasonal adjustments are based on historical population baseline change trends for the same period.

[0095] This embodiment achieves continuous dynamic monitoring and graded assessment of the health status of free-range cattle through coupled computation of multimodal sensor data and a temporal cascaded inference mechanism. Following the physiological patterns of ruminants and the temporal logic of disease development, this method effectively reduces false alarms caused by fluctuations in single-dimensional data by cross-validating multi-dimensional data such as feed intake, rumination, energy metabolism, physical condition, and behavior, thereby improving the reliability of health assessments. The method employs an event-triggered data acquisition mechanism, which reduces terminal power consumption while ensuring data accuracy, adapting to the long-term deployment needs of large-scale free-range scenarios. Using cattle as mobile sensing nodes, this method generates pasture nutrient distribution maps from nutrient data at feeding points, reverse-optimizes feed intake calculation parameters, and simultaneously achieves individual health monitoring and pasture resource perception, enhancing the overall application value of the system. This method eliminates the need for dense deployment of fixed monitoring equipment, reducing deployment costs and maintenance difficulties in free-range pastures, enabling early detection of abnormal health conditions, and providing stable data support for livestock health management on pastures.

[0096] Example 2

[0097] This embodiment provides a health status assessment system for free-range cattle based on cow ear tags, used to implement the above-mentioned assessment method, referring to... Figure 2 The assessment system consists of smart ear tag terminals, edge gateway devices, and a cloud management platform. The smart ear tag terminals, worn on the ears of cattle, integrate inertial sensing modules, near-infrared spectroscopy modules, positioning and communication modules, body temperature sensing modules, and a main control module. The inertial sensing module collects inertial motion data of the cattle's heads; the near-infrared spectroscopy module collects near-infrared reflectance spectral data of pasture during grazing; the positioning and communication module acquires the cattle's location information and completes wireless data transmission; the body temperature sensing module collects the cattle's body temperature data; and the main control module coordinates the acquisition sequence of each sensor module, performing preliminary behavior recognition and data preprocessing. The edge gateway devices are deployed in the pasture area, receiving data transmitted from all smart ear tag terminals within the pasture, performing feature calculation and health assessment processing locally, outputting corresponding early warning information, and synchronizing all data to the cloud management platform when network connectivity is maintained. The cloud management platform stores all operational data for all individuals, updates the individual health baseline database, iteratively optimizes the algorithm model, and visualizes health status data and pasture distribution data.

[0098] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0099] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within this application. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for assessing the health status of free-range cattle based on cow ear tags, characterized in that, Includes the following steps: S1. Bind individual cattle identification information, collect routine operation data within a set time period, and build an individual health baseline database, which includes feed intake baseline, rumination baseline, behavior baseline, physical baseline and body temperature baseline; S2. Collect head inertial data, near-infrared spectral data, position data and body temperature data through a data acquisition terminal worn on the ears of cattle, and perform time synchronization processing on all collected data; S3. Based on head inertial data and near-infrared spectral data, calculate feeding characteristic data and rumination characteristic data, combine location data to calculate activity characteristic data and social characteristic data, calculate physical characteristic data based on feeding characteristic data and activity characteristic data, and generate multi-dimensional health characteristic data; S4. Compare the multi-dimensional health feature data with the individual health baseline database, generate health level results according to the time-series cascade inference rules, output corresponding early warning information, and update the individual health baseline database synchronously.

2. The method according to claim 1, characterized in that, The calculation process for feeding and rumination characteristics in step S3 includes the following sub-steps: S3.

1. Perform time-series segmentation on the head inertial data, identify feeding behavior segments and rumination behavior segments, and extract feeding duration, feeding frequency, chewing number, rumination duration, rumination frequency, and chewing rhythm data. S3.

2. Match and process the near-infrared spectral data corresponding to the feeding behavior segments, and output the dietary nutrient parameters data and the risk value data of poisonous weeds. The dietary nutrient parameters include the proportion of crude protein, the proportion of neutral detergent fiber, and the proportion of acid detergent fiber. S3.

3. Combine dietary nutrient parameters and chewing frequency data to calculate dry matter intake and metabolizable energy intake data, and use rumination duration data to verify and correct the dry matter intake data.

3. The method according to claim 1, characterized in that, The physical characteristic data calculation process in step S3 includes the following sub-steps: S3.

4. Extract gait feature data of walking segments from head inertial data. Gait feature data includes stride frequency, stride variation coefficient, ground impact amplitude, and head swing symmetry. S3.

5. Calculate daily energy balance data based on metabolizable energy intake data and activity characteristic data, including activity distance, standing time and lying time; S3.

6. Combine gait characteristic data and daily energy balance data to estimate body condition score data and weight change trend data, and generate physical characteristic data.

4. The method according to claim 1, characterized in that, The data acquisition process in step S2 includes the following sub-steps: S2.

1. Maintain continuous acquisition of head inertial data at a set frequency, and perform real-time behavior recognition on the head inertial data; S2.

2. When feeding behavior is detected, increase the frequency of head inertial data acquisition and simultaneously start the interval acquisition of near-infrared spectral data; S2.

3. When rumination behavior is detected, adjust the head inertial data to the corresponding acquisition frequency and keep the near-infrared spectroscopy module in the off state.

5. The method according to claim 1, characterized in that, The process of building the individual health baseline database in step S1 includes the following sub-steps: S1.

1. Enter the basic attribute information of the cattle, including breed information, age information and initial body condition information; S1.

2. Collect health status operation data within a set period, extract individual normal characteristic intervals through clustering processing, and generate individual dimensional baseline data; S1.

3. Obtain group characteristic data of cattle of the same age in the same herd, generate group reference baseline data, and integrate individual dimension baseline data with group reference baseline data into an individual health baseline library.

6. The method according to claim 1, characterized in that, In step S4, a multi-dimensional consistency check is performed before generating the health level result. When the single-dimensional feature data deviates from the corresponding interval of the individual health baseline database while the other dimensional feature data are within the corresponding interval, it is determined to be due to environmental factors and the warning level upgrade is not triggered. When the multi-dimensional feature data deviates from the corresponding interval synchronously and conforms to the time-series cascading pattern, the corresponding warning level upgrade operation is performed.

7. The method according to claim 2, characterized in that, In step S3, each set of dietary nutrient parameter data is associated with the corresponding location data, and pasture nutrient distribution map data is generated through spatial interpolation. The pasture nutrient distribution map data is then used to correct the dry matter intake calculation parameters for different regions.

8. The method according to claim 1, characterized in that, In step S3, behavioral abnormality index data is calculated based on feeding characteristic data, activity characteristic data, social characteristic data, and diurnal rhythm data; Social characteristic data is generated by topological calculations using the location data of all cattle. The social characteristic data includes the average distance of the herd, the duration of separation from the herd, and the synchronization rate of group activities.

9. The method according to claim 1, characterized in that, In step S4, the temporal cascade inference rule is executed according to the three-level warning level; the first level warning corresponds to a single-dimensional deviation in the feeding rhythm or behavioral index; The Level 2 warning corresponds to a state where abnormal feeding continues for a set duration and is accompanied by a negative energy balance and a declining body condition. The third-level warning corresponds to a state of body temperature deviation that occurs on the basis of the second-level warning.

10. A system for assessing the health status of free-range cattle based on ear tags, used to perform the method as described in any one of claims 1-9, characterized in that, The system includes a smart ear tag terminal, an edge gateway device, and a cloud management platform. The smart ear tag terminal is worn on the ears of cattle and integrates an inertial sensing module, a near-infrared spectroscopy module, a positioning and communication module, a body temperature sensing module, and a main control module. It is used to collect multi-source raw data and perform preliminary behavior recognition. The edge gateway device is deployed in the pasture area to receive data transmitted by the smart ear tag terminal, perform feature calculation and health assessment processing locally, and output early warning information. The cloud management platform is used to store all data, update the baseline library and algorithm model, and display health status data and grassland distribution data.

Images