Method for identifying early performance behavior of pig epidemic disease based on three-axis attitude angle

Abstract

The invention discloses a method for identifying an early performance behavior of a pig epidemic disease based on a three-axis attitude angle, and belongs to the technical field of intelligent breeding and livestock and poultry health monitoring. According to the method, a wearable device integrated with a gyroscope is worn on the head of a pig, and three-axis attitude angle signals such as a yaw angle (Yaw), a pitch angle (Pitch) and a roll angle (Roll) of the pig are collected in real time. And carrying out sliding processing on the continuous attitude angle data according to a set time window, extracting behavior characteristic parameters, and carrying out comprehensive identification on behavior performance in the early stage of the suspected epidemic disease in combination with behavior chain pattern identification, individual-group comparative analysis, continuous trend monitoring and an abnormal scoring mechanism. The method does not depend on other sensors such as an accelerometer, an image, a sound or a body temperature sensor, only takes the attitude angle as a unique input signal source, has the advantages of low power consumption, light model weight, convenient deployment and suitability for edge recognition, can quickly find out an individual with abnormal behavior in the early stage of epidemic diseases, realizes intelligent early warning earlier than definite diagnosis, and has a good application prospect. The method is suitable for epidemic disease prevention and control and fine management of various large-scale pig farms.

Description

Technical Field

[0001] This invention relates to the field of intelligent health monitoring technology for livestock and poultry, specifically to a method for identifying early signs of disease in pigs based on three-axis attitude angle (Yaw, Pitch, Roll) signals from a gyroscope. This method is applicable to health early warning, disease prevention and control, and remote intelligent breeding management of pigs during fattening, breeding, and nursery stages. Background Technology

[0002] In modern pig farm management systems, timely detection of early disease signals is a core element in ensuring herd health and preventing the spread of infection. Especially in the prevention and control of major animal diseases such as African swine fever, porcine reproductive and respiratory syndrome (PRRS), porcine circovirus, and classical swine fever, the principle of "early detection and early isolation" is more practically valuable than simply relying on confirmed diagnosis results.

[0003] Currently, the technologies used for disease surveillance mainly include body temperature monitoring, video behavior analysis, and multimodal sensor fusion (such as accelerometers, infrared, and sound), but they generally have the following shortcomings: Delayed sampling or diagnosis: By the time the body temperature rises and a confirmed diagnosis is made, the optimal isolation window has already been missed; High equipment cost and deployment difficulty: Video and multi-sensor fusion systems rely on complex deployment, resulting in high costs and difficult maintenance; Poor scene adaptability: The accuracy of behavior recognition drops significantly in dense enclosures, low-light environments, and when there is severe occlusion. Reliance on cloud computing and heavy model learning: It is highly dependent on the network and cannot adapt to edge deployment and offline scenarios.

[0004] In contrast, lightweight identification methods based on three-axis attitude angle (yaw, pitch, and roll) signals are becoming a new key path in intelligent livestock management due to their strong noise resistance, low power consumption, lightweight models, and suitability for long-term operation on wearable devices such as ear tags.

[0005] Studies have shown that pigs often exhibit a series of non-specific but significant behavioral changes in the early stages of disease, such as reduced activity, prolonged lying down, pen-licking restlessness, decreased feed intake, frequent jumping attempts, and abnormal nocturnal activity. These behaviors are often manifested as significant deviations in parameters such as the amplitude, frequency, directionality, and rhythm of posture angle fluctuations, and can be continuously observed without intervention.

[0006] However, there is currently a lack of a recognition method specifically designed for "early behavioral manifestations of disease" with attitude angle as the sole input source, which can run locally on edge devices and is suitable for large-scale deployment and remote management.

[0007] This invention addresses the aforementioned problems by proposing a method for identifying and issuing early warnings of early-stage swine diseases using only three-axis attitude angles as input signals, without relying on external sensors such as accelerometers, images, or body temperatures. This method combines behavioral modeling and risk scoring mechanisms. It has the advantages of clear identification logic, strong real-time performance, and low deployment costs, making it suitable for widespread use in various types of pig farms. Summary of the Invention

[0008] This invention provides a method for identifying early signs of disease in pigs based on three-axis attitude angles. The core of this method is to identify characteristic behavioral patterns and combinations that pigs may exhibit before the onset of disease by relying solely on the three-axis attitude angle signals (Yaw, Pitch, Roll) collected from smart ear tags, thereby forming an early warning mechanism and improving the efficiency of disease prevention and control in farms.

[0009] The method includes the following steps:

[0010] Step 1: Attitude Angle Data Acquisition

[0011] A smart ear tag with an embedded gyroscope sensor is worn at the base of the pig's ear to collect three-axis attitude angle data in real time. Yaw angle: reflects the left and right yaw of the head; Pitch: Reflects the actions of looking down, looking up, and looking up; Roll angle: reflects the tilt, lying, and rolling posture.

[0012] The sampling frequency is recommended to be 1Hz–5Hz, and the sampled data should be segmented by a set sliding time window (e.g., 5 minutes, 10 minutes).

[0013] Step 2: Behavioral Feature Extraction and Scoring Modeling

[0014] Within each time window, the following feature indicators are extracted: Mean, standard deviation, and coefficient of variation of attitude angles; Counting rapid mutation points (e.g., the Yaw angle oscillates rapidly more than N times within ±15°); Attitude angle dominant frequency analysis (main frequency peak extracted by FFT); The amplitude of fluctuations is consistent with the rhythm; Duration of behavior and frequency of switching.

[0015] Based on the above characteristics, a general scoring model for abnormal pig behavior is constructed, using the following weighted scoring formula:

[0016] in: f i: The i-th attitude feature (such as pitch fluctuation coefficient, roll change amplitude, etc.); w i The weight value of this feature (which can be set through training or empirical values); n: The number of features involved in the scoring.

[0017] When the score exceeds the set threshold T, it is judged as "abnormal behavioral trend".

[0018] Step 3: Behavioral Sequence and Combination Recognition

[0019] If multiple abnormal behavior segments are detected consecutively, the system further identifies combinations of pre-morbid behavioral features, such as: Restless movement (high-frequency yaw oscillation) → multiple jumps and head movements (rapid pitch increase) → lying still (long-term roll bias); Frequent unusual activity at night; Reduced feeding postures + prolonged periods of maintaining the same posture.

[0020] The system uses a finite state transition model and abnormal rhythm judgment logic to identify "early behavioral sequences of the epidemic".

[0021] Step 4: Comparative Analysis of Individuals and Groups

[0022] By constructing a reference model of group behavior, we can determine the degree of deviation of individuals in the following dimensions: Differences in behavioral patterns (whether they deviate from the average behavioral template of the group); Has the ranking of abnormal behavior scores been rising continuously? Premature appearance of a specific behavioral sequence (the first person in the entire group to exhibit "anxiety + restlessness").

[0023] Individuals with significant deviations will be classified as "suspected pre-illness individuals" and marked for further observation.

[0024] Step 5: Risk Level Assessment and Early Warning Output

[0025] The system comprehensively calculates the early disease risk level based on parameters such as behavioral score, combined behavioral confidence, and population comparison bias. Low risk: Score slightly above the normal range, with only a single abnormal behavior; Medium risk: Two or more consecutive abnormal behavioral characteristics with combined rhythms; High risk: Consists of typical pre-epidemic behavioral sequences, with significant deviations from the population pattern.

[0026] The identification results can be sent to the management platform via Bluetooth, LoRa, Wi-Fi, etc., triggering a visual warning interface, isolation suggestions, or voice prompts.

[0027] This technical solution emphasizes the "behavioral change first" approach to early disease detection, abandoning reliance on multiple sensors such as body temperature, images, and sound. It can complete pre-disease risk identification solely based on attitude angle signals, and has advantages such as flexible deployment, timely response, low cost, and strong algorithm interpretability. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the overall structure of the system described in this invention; Figure 2 A schematic diagram of a structure for attaching smart ear tags to pigs; Figure 3 Flowchart for the acquisition and sliding window processing of three-axis attitude angle data; Figure 4 The plot shows the trends of pitch and yaw angles in typical behaviors (taking "feeding behavior" and "head-up alert behavior" as examples); Figure 5 A three-axis attitude angle change curve for "jump escape behavior"; Figure 6 This is a frequency domain feature map of "licking the fence behavior". Detailed Implementation

[0029] To make the objectives, technical solutions, and beneficial effects of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. This invention is not limited to the specific embodiments described below; any equivalent substitutions or improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

[0030] Example 1: Early Warning of "Lying Still" Behavior Based on Pitch Fluctuation Recognition

[0031] In a large fattening farm, intelligent ear tags with three-axis attitude angle acquisition capabilities were fitted to fattening pigs. The system samples the pitch angle every 5 seconds and performs continuous analysis within a 5-minute sliding window.

[0032] If the pitch angle fluctuation is found to be below a set threshold (e.g., ±3°) for two consecutive windows, and the roll angle remains biased to one side for more than 10 minutes, and the behavior is significantly different from the pig's previous activity pattern, it is marked as "prolonged resting". The system introduces a "behavioral variation scoring mechanism" (based on the pitch coefficient of variation, CV) as a quantitative standard. If the abnormal score is ≥ 0.6, an early warning is triggered.

[0033] Formula example:

[0034] in: μ Pitch Mean value of the pitch angle; σ Pitch Standard value for the Pitch angle; CV Pitch Coefficient of variation: The higher the value, the more volatile the fluctuation.

[0035] CV Pitch >T CV This is marked as a high-fluctuation state and included in the anomaly scoring item.

[0036] Example 2: Identifying "Licking the Bar - Annoyance" Behavior Based on Yaw High-Frequency Spectrum

[0037] This embodiment is used in a sow pen to analyze the frequency distribution of the yaw angle over the past 3 minutes using the FFT method. When a significant energy peak is detected in the yaw angle within the 1.5Hz–3Hz frequency band and lasts for more than 15 seconds, it is determined that "pen-licking" behavior has occurred.

[0038] If this behavior occurs more than 3 times within 3 consecutive hours, the system will automatically determine that the pig may be in the early stage of stress or the stage of disease-related behavior, mark it as a medium-risk object, and the system will record and push it to the management platform.

[0039] Example 3: "Disease Behavior Pattern Chain" Based on Behavioral Chain Sequence Recognition

[0040] The system analyzes the behavioral event sequence of a single pig over a 12-hour period. If the following behavioral chain is continuously identified: anxiety (high-frequency yaw) → jumping attempt (sudden pitch increase + rapid roll change) → lying still (stable angle) → not eating (no downward pitch movement), it matches the preset "typical chain behavior pattern before disease".

[0041] When the integrity of the behavioral chain reaches 75% or more, the system identifies the individual as a high-risk individual, directly adds them to the key monitoring list, and tracks their trajectory and records it in logs.

[0042] Example 4: Individual Movement Recognition Based on Group Deviation Detection

[0043] The system models the average posture angle behavior of all pigs in the group daily. If a pig's Roll angle activity range deviates from the group average by more than 25% within 6 consecutive hours, and its Pitch angle downward frequency decreases by more than 50% and its Yaw angle fluctuates drastically, it is considered to have a behavioral pattern that is out of sync with the group rhythm.

[0044] By combining an individual's past behavior model, it is determined that if their "behavioral deviation score" exceeds a threshold, an early disease manifestation marker is triggered, which is used to assist managers in making differentiated interventions.

[0045] Example 5: Integrated Early Warning System Based on a Scoring Synthesis Model

[0046] The system scores abnormal behaviors independently according to the following five categories: Pitch angle variability; High-frequency disturbance spectrum; Intensity of behavioral mutation; Completeness of the behavioral chain; Group deviation score.

[0047] The comprehensive early warning index is calculated as follows:

[0048] Among them, each S i All scores are standardized scores of 0-1.

[0049] When the overall score is greater than 0.65 and the duration exceeds 3 hours, the early warning label for the epidemic is triggered.

[0050] The management system supports batch display and sorting by risk level, making it easy to identify potentially infected individuals as soon as possible.

[0051] Example 6: Identifying "Abnormal Lying Still" Behavior Based on Sliding Window Coefficient of Variation

[0052] The system sets a sliding window W = 5 minutes, and calculates the coefficient of variation (CV) for the pitch angle sequence {p1, p2, ..., pn} within each window:

[0053] in: μ Pitch Mean value of the pitch angle; σ Pitch : Standard value for the Pitch angle.

[0054] When CV in three consecutive windows Pitch If the value is less than 0.05 and the Roll angle is fixed to one side, it is determined to be an "abnormal resting" state. If the duration exceeds 30 minutes, an early risk warning is triggered.

[0055] Example 7: Identifying "Individual Deviation Behavior" Based on Z-score and Group Comparison

[0056] Let R be the average Roll angle of pig individual i over 1 hour. i The population mean during this period was μ. G The standard deviation is σ G The calculated Z-score is as follows:

[0057] When |Z iIf the behavior persists for more than 3 hours, it indicates a significant deviation from the group's behavior. Combined with the co-occurrence of decreased feed intake or pen-licking behavior, this indicates a severe deviation from the group's behavior, triggering a medium-level warning.

[0058] Example 8: Identifying "Irritable Fence-Licking" Behavior Based on High-Frequency Power Spectral Density

[0059] The system performs a Fast Fourier Transform on the Yaw angle sequence within a 2-minute sliding window to obtain the spectral energy P(f). The integral energy is then calculated in the 1-3Hz frequency band.

[0060] If E 1-3Hz >T lick If the behavior occurs more than 3 times (based on experience thresholds), the system will mark it as "annoying bar-licking behavior".

[0061] If we combine this with the individual's increased resting time and decreased food intake over the past 24 hours, it forms a logical chain of "potential early-stage stress response to an epidemic," triggering a combined early warning system.

[0062] The parameters used in the above embodiments (such as window length, duration, threshold, etc.) can be adjusted according to the type of pigsty and management strategy.

Claims

1. A method for identifying early signs of disease in pigs based on three-axis attitude angles, characterized in that, Includes the following steps: Wearable devices with integrated gyroscopes are worn on the heads of pigs to collect three-axis attitude angle data, including yaw, pitch and roll. The collected attitude angle data is divided into time sliding windows, and behavioral features are extracted within each window; Based on temporal changes, frequency domain features, trend changes, mutation detection, or stability analysis of attitude angles, identify abnormal behavioral manifestations related to the early stages of an epidemic. Output the results of suspected early-stage disease behavior identification and upload them wirelessly to the management platform for recording, analysis and early warning; The abnormal behaviors mentioned include, but are not limited to: not eating, prolonged lying stillness, abnormal postural fluctuations, excessive anxiety, frequent turning of the head, and disordered behavioral rhythms. In the behavior recognition process, only three-axis attitude angle data is used as the input signal source, without relying on accelerometer, image, sound, temperature or other sensor inputs.

2. The method according to claim 1, wherein the attitude angle data sampling period is 1–10 seconds and the sliding window length is 30 seconds to 10 minutes.

3. The method according to claim 1, wherein the behavioral feature includes one or more of the following: Mean, standard deviation, and coefficient of variation of the attitude angle sequence; The average magnitude or directional offset of attitude angle changes; Frequency distribution of attitude angle changes or concentration of dominant frequency energy; The number and amplitude of abrupt changes in the attitude angle signal; The range of fluctuation or the period of static stability of the attitude angle curve.

4. The method of claim 1, wherein the identified suspected behavioral chain includes one of the following behavioral patterns: Restlessness → jumping and shaking head → prolonged lying still; Frequent turning around → loss of appetite → abnormal postural stability; Drastic changes in behavioral frequency → Plateau-like low activity.

5. The method according to claim 1, wherein the behavior recognition algorithm includes, but is not limited to: rule matching, sliding window mutation detection, frequency domain energy distribution analysis, state transition modeling, change trend analysis, and statistical anomaly scoring.

6. The method according to claim 1, wherein the output behavior recognition result is transmitted via low-power wireless communication, including Bluetooth, LoRa, Wi-Fi or cellular communication, etc.

7. The method according to claim 1, wherein the early identification results of the epidemic can be combined with the individual's historical behavioral baseline and the average behavior of the group for difference analysis to enhance the confidence of anomaly detection.

8. The method according to claim 1, wherein the behavior determination does not include a scoring mechanism as a necessary condition, and the identification can be achieved independently based on threshold discrimination, frequency domain features, or state change path, etc.

9. The method according to claim 1, wherein the input signal used in the recognition process is limited to three-axis attitude angle data, and regardless of the presence of other sensor modules, as long as the key criterion for behavior recognition depends on the three-axis attitude angle, it is considered to fall within the protection scope of this invention.

Images