Method for identifying abnormal behaviors and death states of chickens based on attitude angles

Abstract

The invention provides a method for identifying abnormal behaviors and death states of chickens based on attitude angles, which comprises the following steps of: continuously acquiring three-axis attitude angle signals (namely a pitch angle Pitch, a yaw angle Yaw and a roll angle Roll) by using an inertial sensing device worn on the body surface of a chicken, constructing a low-frequency behavior track sequence, and identifying the abnormal behaviors and the death states of the chicken according to the low-frequency behavior track sequence; and through a set posture change threshold value, posture stability judgment, frequency distribution analysis and a posture imbalance recognition algorithm, whether the chicken is in the states of dysphoria, abnormally static, prostration, stiff head raising or death or not is judged. The method does not depend on acceleration or six-axis fusion signals, has the advantages of being low in power consumption, high in adaptability, easy and convenient to deploy and the like, and is suitable for continuous health monitoring and death early warning management of chicken individuals in a large-scale farm. The method has the characteristics of clear signal input boundary, accurate recognition mechanism, economical system deployment and the like, and the digital management efficiency and the animal welfare guarantee level of poultry breeding are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of livestock intelligent monitoring, in particular to a method for identifying abnormal behaviors and death states of chickens based on three-axis attitude angle signals, which belongs to the cross-application of animal health monitoring, breeding management and behavior recognition technology. BACKGROUND

[0002] With the development of poultry large-scale breeding, chicken health monitoring and death warning become a crucial link in the breeding process. Traditional manual inspection methods have problems such as high labor intensity, delayed identification, high omission rate, etc., and are difficult to meet the real-time monitoring needs of high-density farms. Therefore, in recent years, various chicken behavior recognition technologies based on sensors, video analysis or fusion algorithms have emerged, mainly including the following categories:

[0003] 1. Recognition method based on video image analysis: This method relies on cameras to collect chicken activity images, combines target detection and behavior classification algorithms, and realizes chicken state analysis. However, this method is sensitive to light, shading and deployment environment, and has problems such as high energy consumption, high maintenance cost, privacy risk, etc., and is difficult to adapt to closed or night breeding environment.

[0004] 2. Recognition method based on six-axis inertial sensor (IMU) fusion: Use accelerometer and gyroscope fusion to obtain chicken motion characteristics such as tremor frequency, movement distance, activity intensity, etc., and build a recognition model. However, the acceleration signal is easily affected by foot shaking, feeding tremor and other factors on the chicken, with high noise, and the fusion algorithm has high requirements for power consumption and computing resources, which is not conducive to long-term deployment of small devices such as foot rings.

[0005] 3. Abnormal judgment method based on activity score and group comparison: This method judges individual abnormalities by calculating the activity intensity or change trend of the chicken, combined with the average value of the group, but lacks clear criteria for specific behaviors, which may lead to misjudgment or delayed recognition, especially in the judgment of death state.

[0006] In summary, the existing chicken behavior recognition technology generally has the problems of complex sensor input, high computing resource consumption, high deployment threshold or fuzzy recognition boundary, etc., and there is an urgent need for a simple structure, accurate recognition, low power consumption, and individual-level continuous monitoring technology, especially in the identification of abnormal behaviors and death states of chickens, which should have high accuracy and clear judgment mechanism. SUMMARY

[0007] The present application relates to the technical field of poultry behavior monitoring, and specifically relates to a method for identifying abnormal behaviors and death states of chickens based on posture angles.

[0008] The present application performs time series analysis on the posture angle data continuously collected by the device worn by the chicken, extracts representative feature indicators, and uses the indicators to identify abnormal behavior states (such as posture angle remaining unchanged for a long time, minimum fluctuation amplitude, abnormal yaw angle deviation, etc.), and accordingly determines whether the chicken is in a syncope, forced, weak, abnormal behavior or suspected death state, thereby achieving early warning.

[0009] Further, the present application has broken through to construct an identification mechanism for the "turning circle" behavior of the chicken. The mechanism is based on the change frequency and amplitude threshold characteristics of the yaw angle (Yaw) in a short time window to identify whether the chicken has a behavior pattern of continuously rotating in place. This type of behavior has typical significance in pathological states such as vestibular dysfunction, central nervous infection or early disease behavior of poultry, and is difficult to identify but has significant diagnostic value. The automatic identification method for the turning circle behavior proposed by the present application fills the gap in the field of intelligent behavior recognition of poultry in modeling such abnormal behaviors, and significantly improves the sensitivity and specificity of identification.

[0010] The method proposed by the present application is suitable for daily health monitoring, abnormal behavior identification and death state warning of chickens and other poultry, and can be widely deployed in intensive farms, research monitoring platforms and intelligent poultry breeding systems, and has good engineering implementation and popularization value.

[0011] I. Invention purposes

[0012] The present application aims to provide a method for identifying abnormal behaviors and death states of chickens based on posture angles, to solve the technical problems of complex sensor dependence, unclear behavior recognition boundary, high power consumption and inadaptability to foot ring deployment in existing poultry monitoring technology.

[0013] The present application uses a low-power posture sensing terminal worn on the feet of the chicken to collect only three-axis posture angle signals (including pitch angle Pitch, yaw angle Yaw and roll angle Roll), completely abandoning the dependence on accelerometer data. On this basis, a behavior discrimination model based on posture angle change trend and time window evaluation mechanism is constructed to realize automatic identification and intelligent warning of typical states such as lowering head and not moving, continuous stillness, and death by falling over.

[0014] By the present application, the recognition accuracy and response timeliness of chicken death and abnormal state can be significantly improved, the artificial inspection pressure is reduced, and the automation and intelligent level of the breeding link is enhanced. It is especially suitable for continuous, non-contact, low-power individual state monitoring of chickens in intensive breeding scenes, and has good engineering feasibility and large-scale deployment value.

[0015] II. Technical solutions

[0016] (I) The method of the present application comprises the following steps:

[0017] Step one, attitude angle acquisition

[0018] In the present application, attitude angle acquisition is the basic link for identifying the behavior state of chickens. Specifically, an integrated attitude recognition device is worn on the feet (preferably the tarsal joint or tarsometatarsal position) of the chicken individual, which is embedded with a gyroscope sensor module and can collect three-axis attitude angle information of the chicken during natural activity, i.e.: Yaw: reflects the change in the turning angle of the chicken's body in the horizontal plane; Pitch: reflects the degree of up and down inclination of the head and tail direction of the chicken; Roll: reflects the attitude angle of the left and right side of the chicken's body.

[0019] The recognition device has a low-power design and can operate continuously without external power supply. The sampling strategy adopts a timed low-frequency sampling mode, i.e. automatically collecting three-axis attitude angle data once per set period, and the sampling period ranges from 10 seconds to 60 seconds, which can be flexibly adjusted according to different breeding environments and recognition accuracy requirements.

[0020] In order to balance data integrity and energy consumption control, the device remains in a dormant state during the non-sampling period, and wakes up quickly, collects, caches and writes to the storage unit or reports to the platform through the wireless communication module during the sampling period. Short sampling time and low energy consumption help to realize long-term continuous monitoring of chicken individuals in large-scale breeding farms.

[0021] The collected three-axis attitude angle data will be used as the only input signal for subsequent behavior feature extraction and abnormality recognition, and does not rely on accelerometers, temperature and humidity sensors or camera equipment, ensuring the simplicity, stability and easy deployment of the system.

[0022] Step two, attitude angle feature extraction

[0023] After completing the attitude angle data acquisition, the system will analyze the Yaw, Pitch, Roll three-axis angle data continuously obtained within a set time window (such as 5 minutes, 10 minutes or custom length) to identify whether the chicken has a behavior abnormality or a body position abnormality.

[0024] The following three core features are selected as the basis for analysis:

[0025] 1. Yaw angle change frequency feature (turning behavior recognition) This feature is a key criterion for chicken behavior recognition, especially for determining whether there are frequent turns, turning walks, and other restless behaviors. The recognition method is as follows: Calculate the Yaw angle difference ΔYaw between consecutive sampling points; If ΔYaw exceeds the set threshold (such as 15°~30°, adjusted according to breed and behavior characteristics), it is recorded as a direction change; Count the total number of direction changes in the unit time window to obtain the Yaw change frequency index; If this frequency is significantly higher than the individual historical mean or the normal change frequency of the group average, or its Z-score deviation exceeds the set threshold (such as Z>2), it is determined to be in a "high-frequency turning state"; This state is commonly seen in pre-onset restlessness, central symptoms such as encephalitis, environmental strong stimulation, or stress behaviors such as group compression.

[0026] This feature is one of the most representative core features in the recognition of chicken abnormal behaviors in the invention. The Yaw frequency signal of the turning behavior has independent invention value.

[0027] 2. Posture angle change stability feature This feature is used to identify whether the chicken is in a state of continuous stillness, low activity, or even paralysis, reflecting its behavior activity level. The specific method is as follows: Calculate the average change amplitude between sampling points for each posture angle (Yaw, Pitch, Roll); If the posture angle change amplitudes of multiple consecutive sampling points are all below the stillness threshold (such as 2°~5°), and the duration exceeds the set threshold (such as 10 minutes), it is recognized as a high stability state; The high stability state may represent that the chicken is in: weak lying down; long-term sleep (if abnormal and persistent); or has died; Combined with individual historical activity trajectories and group activity levels, it can be used to eliminate short-term rest and other non-abnormal behaviors.

[0028] 3. Orientation distribution feature (Pitch / Roll abnormal body position recognition) This feature is used to identify whether the chicken is in an abnormal spatial body position, such as falling over, lying on the side, head facing the ground, and other common high-risk postures for death. The specific method is as follows: Count the frequency and distribution range of Pitch and Roll angles in each time window; If the Pitch angle of consecutive sampling points is in the extreme range (such as <-75° or >+75°), and the Roll angle is close to ±90°, it is considered as an abnormal orientation; If the orientation state lasts more than a set time (such as 5 minutes) without returning to the normal range, it is determined as a "continuous abnormal body position"; Combined with the attitude angle stability feature, if it also shows "attitude angle almost unchanged + extreme abnormal orientation", it is highly suspected that the chicken has died or is in a paralyzed state.

[0029] 4. Feature combination judgment and graded early warning The above three core features can be used alone for abnormal identification, or combined to form a multi-dimensional joint judgment index. The system supports setting different combination rules and logical relationships, including but not limited to: If the Yaw change frequency significantly increases + the attitude angle fluctuates sharply, it can be identified as restless wandering and circling behavior; If the Yaw change frequency is 0, and the attitude angle almost does not change, combined with the extreme abnormal orientation angle, it is highly suspected to be in a dead state; If the Pitch / Roll maintains an abnormal posture + the attitude angle change rate is close to 0, it is judged as an abnormal body position and triggers a high-level early warning.

[0030] The system can further output abnormal labels and score weights based on the above criteria, support model training sample accumulation and abnormal behavior library construction, and be linked with the platform health early warning system to realize automatic monitoring and timely response of chicken individual behavior risk.

[0031] Step three, behavior comparison and judgment

[0032] After completing the attitude angle feature extraction, the invention compares the attitude angle feature values of the chicken individual in the set time window with its historical behavior pattern or the collective behavior characteristics of the chicken group, judges whether there is abnormal deviation, and then identifies the behavior abnormality or body position abnormality state. The comparison process includes two links of reference construction and deviation judgment:

[0033] 1. Reference behavior construction The system can construct the behavior reference based on the following two ways: Individual historical reference: record the feature value statistical information of the current individual in the past normal state, such as mean, standard deviation; Group average reference: select the average behavior characteristics of healthy individuals in the current batch of chickens as a reference; Let the feature value be x t, whose historical mean is μ and standard deviation is σ, the standardized deviation (Z-score) is calculated as follows:

[0034] When Z|>Z th (threshold value, recommended value 2~3), it can be determined that the feature has significant deviation.

[0035] 2. Deviation determination rules The system determines three feature dimensions independently or jointly:

[0036] a. Yaw change frequency deviation determination: If the number of direction changes f yaw satisfies:

[0037] Then it is determined that the chicken direction change is too frequent, suspected of being agitated or circling behavior.

[0038] b. Attitude angle stability deviation determination: Let the angle change amplitude Δθ i of each point in the continuous N sampling points, if:

[0039] Where δ min is the static judgment threshold (e.g. 3°), and N·T>T table (T is the sampling period, T table is the minimum static duration), it is determined as an abnormal static state.

[0040] c. Pitch / Roll orientation abnormality determination: Let the Pitch angle be θ pitch , and the Roll angle be θ roll , when:

[0041] Where these two intervals are the preset abnormal orientation intervals (e.g. the overturning interval), and the state duration t>T risk , it is determined that the chicken is in a high-risk body position.

[0042] 3. Joint determination strategy (optional) To improve the recognition accuracy, the system can give weighted scores to the above three dimensions to build an abnormal comprehensive score S:

[0043] Where: Z yawFor direction frequency deviation degree; S stable ∈ {0,1} indicates whether it is static; S posture ∈{0,1} indicates whether it is abnormal; w1, w2, w3 are weight coefficients.

[0044] When S>S th (set threshold), output high confidence anomaly label, enter the next step of risk identification and early warning mechanism.

[0045] Step four, risk identification and early warning push

[0046] After completing the comparison and analysis of individual posture angle behavior characteristics and reference benchmark, the system determines the abnormal behavior and identifies the risk level according to the set criteria, and triggers the corresponding early warning push mechanism according to the identification result, realizes the real-time response and management intervention to the abnormal state of the chicken.

[0047] 1. Abnormal behavior identification mechanism The system identifies the risk type according to the single-dimensional or joint determination result extracted in the foregoing steps, combined with the following strategies: If only the Yaw angle frequency deviation is significant, but the posture stability and orientation distribution are normal, it is identified as behavior abnormalities such as restlessness, wandering, and turning, which may be related to environmental changes, stress, or slight pathology; If the posture angle change stability is abnormal for a long time, but does not fall into the abnormal orientation interval, it is preliminarily judged as suspected lying or physical weakness state; If the chicken is continuously in the abnormal orientation interval (such as Pitch close to -90°, Roll close to ±90°), and the posture is long-term unchanged, it is considered as a highly suspected death state; If multiple feature dimensions simultaneously meet the abnormal condition, and the duration exceeds the threshold (for example, more than 3 abnormal sampling points within 10 minutes), a high risk level label can be directly output.

[0048] 2. Early warning classification and determination standard In the present application, in order to improve the practicability and response efficiency of the identification result, the system divides the early warning into three levels according to the severity of the chicken posture angle abnormality, each level corresponds to different determination conditions and management suggestions, which is convenient for the manager to carry out targeted disposal.

[0049] The first level of early warning is suitable for the situation of single dimension deviation, for example, only the Yaw angle change frequency is significantly higher than the historical or group reference benchmark, which indicates that the chicken may have behavior abnormal state such as restlessness, frequent wandering, and mild stress, but no obvious body position abnormality or static feature appears. In this case, the breeder is suggested to observe carefully, but no immediate intervention is required.

[0050] Secondary warning is suitable for the simultaneous occurrence of any one of the two dimensions deviating, that is, one of the attitude angle change stability anomaly (long time almost no movement) and the orientation distribution anomaly (Pitch or Roll angle falls into the preset high-risk body position interval) is established, the system will determine that the chicken has a strong abnormal signal, which may be manifested as paralysis, lethargy, weakness or physical dysfunction. Such state suggests that the breeder should conduct artificial verification or take appropriate intervention measures.

[0051] Tertiary warning is the highest level of warning of the system, which is suitable for meeting the abnormal conditions of three dimensions at the same time within the same time period: that is, Yaw angle change frequency anomaly, attitude angle change stability high anomaly and orientation distribution falling into the dangerous interval. Such comprehensive deviation usually represents that the chicken falls on the ground, the body cannot recover, remains stationary for a long time and the direction is abnormal, which is most likely to be dead or lose the ability of autonomous movement. At this time, the system will trigger the highest level of risk prompt, suggesting immediate on-site confirmation and processing.

[0052] The system supports user-defined response mechanisms corresponding to each warning level, including data reporting methods, local alarm prompts, sampling strategy adjustments, individual marking, and inspection scheduling linkages, to achieve intelligent closed-loop management from abnormal identification to response and disposal.

[0053] 3. Warning pushing methods When the system identifies abnormal behavior or risk state, it immediately performs warning pushing operation, including but not limited to: Upload warning data to local control terminal or cloud platform through wireless module (such as Bluetooth, LoRa, WiFi); Including chicken number, location coordinates, abnormal type, abnormal level, trigger time, etc. Supporting automatic generation of abnormal logs, behavior backtracking records and alarm reports; Platform can set pushing strategies, such as: Push notifications to breeder APP in real time; Link alarm lamp, buzzer and other on-site prompts; Automatically mark individual as "need to focus on monitoring".

[0054] 4. Subsequent intervention and closed-loop processing After receiving the warning information, the platform can perform one of the following actions according to the set strategy: Record abnormal behavior trajectory for backtracking analysis; Adjust the sampling frequency (such as from 60 seconds to 10 seconds) to enhance monitoring; Start the inspection task or prompt manual intervention; Multi-modal cross-validation is performed in combination with other data (e.g., body temperature, feeding amount).

[0055] Through the above closed-loop mechanism, the continuous state monitoring, intelligent risk identification and active intervention management of individual chickens are realized, and the biological safety level and disease prevention and control ability of the farm are improved.

[0056] (II) Method flow summary

[0057] The present application provides a chicken abnormal behavior and death state identification method based on three-axis attitude angle data, which uses a wearable attitude recognition device as a data collection terminal, combines a low-power sampling strategy and a multi-dimensional behavior feature extraction mechanism, realizes real-time monitoring, intelligent analysis and automatic early warning of chicken state. This method has the advantages of convenient deployment, accurate identification and controllable power consumption, and is suitable for poultry health management scenarios in medium and large-scale farms.

[0058] The entire method flow includes the following five core steps:

[0059] 1. Attitude angle acquisition: Wear an attitude recognition device on the chicken's feet, collect three-axis attitude angle data, including Yaw (yaw angle), Pitch (pitch angle) and Roll (roll angle), at a sampling period of 10 to 60 seconds to achieve low-power continuous sampling.

[0060] 2. Feature extraction: Analyze the attitude angle data within the set time window, extract the Yaw direction change frequency, attitude angle change stability and Pitch / Roll orientation distribution features, and use them to identify chicken behavior patterns and spatial attitude states.

[0061] 3. Behavior comparison and determination: Compare the current individual features collected with historical behavior data or group behavior benchmarks, and identify abnormal deviations through Z-score, threshold judgment, etc. Determine whether the chicken has restlessness, physical weakness or death risk by independently or combined judging each feature dimension.

[0062] 4. Risk identification and early warning push: According to the number and severity of abnormal features, the risk level is classified (first, second, third), and the early warning information is uploaded to the platform or local management terminal through the wireless communication module, realizing automatic alarm and remote response.

[0063] 5. Data closed loop and continuous monitoring: The system supports continuous tracking of abnormal individuals, dynamically adjusts the sampling frequency according to abnormal conditions, generates behavior trajectory records, and builds individual health records, providing data support for early detection and early intervention of diseases.

[0064] The above steps can be continuously executed in the smart terminal worn by the chicken, real-time data is acquired through the built-in attitude angle sensing module, and recognition and state judgment are completed on the terminal or edge gateway, so that intelligent recognition of abnormal behaviors and death states of the chickens is realized.

[0065] It is worth noting that the method can be deployed in an actual breeding environment in combination with a matching attitude sensing hardware and a wireless communication module to form a complete intelligent monitoring system.

[0066] The method does not need to rely on video recognition, image processing or multi-sensor fusion, and can realize key state recognition of poultry based on only three-axis attitude angles as the only input signal source, has wide adaptability and practical popularization value, and is especially suitable for intelligent supervision scenes of large-scale poultry breeding.

[0067] (Three) Input boundary limitation and fusion inertia path avoidance prevention declaration based on gyroscope attitude angle signals

[0068] To ensure that the poultry behavior recognition method described in the application has a clear and definite signal input boundary, and effectively prevents bypassing the core mechanism of the application through an inertia fusion path, the data types and recognition paths collected by the method are limited and declared as follows: In the application, behavior recognition only relies on three-axis attitude angle signals as the only criterion source. The three-axis attitude angle signals refer to three-dimensional angle data directly output by a gyroscope module, i.e. pitch angle (Pitch), yaw angle (Yaw) and roll angle (Roll), hereinafter referred to as "three-axis attitude angle data". All abnormal frequency recognition, static state judgment and head orientation analysis of the method are based on the original change trend of the angle data. The method does not collect, use or fuse acceleration data, nor does it construct an inertia fusion model or a posture reconstruction algorithm based on acceleration and angular velocity.

[0069] To prevent others from performing avoidance operations through a six-axis inertia fusion path (i.e. inertia measurement unit output containing three-axis acceleration and three-axis angular velocity, hereinafter referred to as "six-axis inertia data"), the following patent boundaries are explicitly limited: 1. The application does not accept an identification path based on a six-axis fusion model (such as Kalman filter, complementary filter, Madgwick algorithm, etc.); 2. The application does not allow acceleration data to be used as input, and the attitude angle change trajectory is fitted, reconstructed or compensated before being used for behavior judgment; 3. Even if the identification device itself has six-axis sensor capability, as long as its behavior recognition process only relies on true collected three-axis attitude angle features for judgment, it is still considered to fall within the protection scope of the application. 4. Any behavior that changes the sensor input path (such as using the acceleration signal as the main input) or infers non-acquired attitude angle features instead of directly acquiring data points shall be regarded as an evasion behavior and included in the protection boundary of this invention.

[0070] By defining and declaring the above technical path, this invention effectively blocks a variety of possible circumvention methods such as "pseudo-decoupling input substitution", "fusion data compensation", and "attitude angle virtual inversion", thereby achieving complete protection for the three-axis attitude angle low-power identification mechanism and clearly distinguishing it from six-axis inertial fusion schemes in terms of technical implementation path.

[0071] III. Beneficial Effects

[0072] (I) This invention provides a method for identifying abnormal behavior and mortality status in chickens based on three-axis attitude angle signals. It relies solely on the pitch, yaw, and roll angles output by the gyroscope for behavior discrimination, and has the following beneficial effects:

[0073] 1. The data source is clear and unique, making it suitable for low-cost terminal deployment. This method does not rely on accelerometers or fusion six-axis algorithms. It uses only the three-axis attitude angle signal collected by the gyroscope as the sole input, thus avoiding the computational complexity and power consumption problems caused by fusion algorithms. It is particularly suitable for deployment in low-power terminals such as chicken ankle bracelets to achieve long-term continuous operation.

[0074] 2. The recognition algorithm is simple and efficient, facilitating edge deployment and low-power operation. By constructing multi-dimensional behavioral features such as abnormal frequency, stationary time, attitude angle amplitude and turning frequency, rapid judgment can be achieved without complex model training. It has portability, edge processing capabilities and real-time response effects, and is suitable for distributed monitoring needs in large-scale aquaculture scenarios.

[0075] 3. Possesses a clear boundary for distinguishing between a state of death and severely abnormal behavior. By combining and analyzing features such as the continuous extreme range of roll angle (e.g., >90°), pitch angle imbalance, and yaw angle stagnation, we can accurately identify the mortality states of chickens such as "lying on their backs" and "lying on their stomachs with stiffness." Compared with existing methods that rely on lack of activity or group difference comparison, this method is more deterministic and stable.

[0076] 4. An innovative "chicken circling" behavior recognition mechanism was introduced to fill the gap in abnormal behavior monitoring. The application first proposes a "turning behavior recognition" method based on the yaw angle change frequency and amplitude threshold, which recognizes the abnormal pattern of high-frequency one-way deflection or rotation in place of the chicken within a unit of time. This behavior is common in vestibular dysfunction, poisoning reaction or early manifestation of nervous system damage in poultry, which has important disease warning and stress response value, and makes up for the lack of existing recognition system in the perception of abnormal behaviors of the nervous system.

[0077] 5. Signal recognition strategy suitable for chicken ring wearing shape The method combines the actual shape of the chicken foot wearing device, constructs a recognition logic based on the trend of the drooping angle, the change of the rolling amplitude and the long-time stillness determination, avoids the poor fixity and recognition failure of traditional ear tags and back clip devices in the chicken scene, and improves the signal stability and behavior expression consistency.

[0078] (II) Advantages compared with prior art

[0079] 1. Different from six-axis fusion recognition scheme: the existing behavior recognition method mainly adopts acceleration and gyroscope fusion, the application actively excludes acceleration interference, enhances the independence and interpretability of the attitude angle as a single behavior criterion.

[0080] 2. Different from video image recognition technology: no need to deploy camera, image acquisition or complex video analysis system, avoid high energy consumption, high cost, weak adaptation (night, shielding) and other problems, more suitable for large-scale breeding farm environment.

[0081] 3. Different from the method based on activity score or group comparison: the application provides individual-level and regularized attitude angle behavior feature index, supports fine management and accurate positioning of abnormal individuals, without relying on historical model or group average value.

[0082] 4. It has the ability to determine the clear cause of death: compared with the method of only identifying "low activity" or "lying on the ground" state, the application can further identify whether the chicken enters abnormal rolling, rigid lying and other death postures, and improve the accuracy and timeliness of death cause speculation and death alarm. BRIEF DESCRIPTION OF DRAWINGS

[0083] Figure 1 : the overall flowchart of the method of the application; Figure 2 : structure diagram of the attitude angle acquisition and processing module; Figure 3 : comparison diagram of attitude angle time series under different behavior states; Figure 4 : logic discrimination diagram of attitude angle change trend and stability judgment; Figure 5 : score calculation flowchart of continuous abnormal behavior scoring mechanism. Detailed Implementation

[0084] 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.

[0085] Example 1: Method and Verification for Identifying and Validating Circling Behavior in Chickens

[0086] This embodiment aims to illustrate how to identify whether chickens exhibit abnormal "circling" behavior based on the Yaw angle sequence in three-axis attitude angle data.

[0087] In actual testing, the posture recognition device described in this invention (which only collects yaw, pitch, and roll angle data) was fitted to the foot of a breeder chicken. The data was sampled every 15 seconds, with each sample lasting 10 seconds, for a total collection time of 10 minutes.

[0088] The following yaw angle sequence samples (unit: degrees) were extracted from the sampled data: {30°, 75°, 125°, 170°, -140°, -90°, -45°, 0°, 45°, 90°} Analysis revealed that the yaw angle difference ΔY between adjacent sampling points is: {45°, 50°, 45°, 50°, 50°, 45°, 45°, 45°, 45°}.

[0089] This sequence exhibits the following behavioral characteristics: All ΔY directions show an "increasing" or unidirectional trend of cumulative angle. Within approximately 90 seconds, the total change in the Yaw angle exceeded 360°, forming a complete circle; The fact that the pitch and roll angles did not fluctuate significantly indicates that the chickens did not move forward but only circled in place. The occurrence of similar phenomena three times in a row, combined with on-site video verification, was determined to be a phenomenon of repeatedly circling around the edge of the fence, which was identified as an early manifestation of stress-induced neurological disorder.

[0090] This behavior is highly correlated with disease precursors, nervous system imbalances, and stress responses, indicating that this method can accurately identify early abnormal activity in chickens and has good field adaptability and early warning value.

[0091] Example 2: Application Scenario of Automatic Recognition of Death Status in Chicken Coops

[0092] In a closed and standardized layer chicken farm, 50 chickens were selected as experimental objects, and a posture recognition device with low-power Bluetooth communication function was worn on their right foot. The device embedded a three-axis gyroscope to collect the attitude angle data (Yaw, Pitch, Roll) of the chickens. The main parameters are as follows: Sampling period: collect every 30 seconds; Data transmission method: broadcast to the nearest edge gateway through Bluetooth; Posture angle change threshold setting: Yaw angle abnormal frequency threshold: ≥8 times / 5 minutes; Posture angle static judgment threshold: three-axis angle change is less than 3° within 5 minutes; Abnormal orientation determination range: Pitch angle <-75°; Roll angle >+70° or <-70°.

[0093] During continuous monitoring, a chicken (ID-037) began to show the following data characteristics at 2:43 am: Yaw angle: almost no direction change within 5 minutes; Pitch angle: continuously maintained at -82° ~ -88°, stable and no recovery; Roll angle: stable at +85°; Total attitude angle variation: less than 1.5° within 5 minutes; Continuous abnormal time: more than 15 minutes.

[0094] The system made judgments according to the set criteria: Meet the high static criterion of attitude angle; Pitch / Roll angle falls into the pre-set high-risk body position interval; Direction change frequency is zero.

[0095] Therefore, the system comprehensively determines that the chicken is highly suspected to have died, automatically marks it as "level three warning", and uploads the following warning information to the platform through the edge gateway at 2:59 am: Individual number: ID-037; Judgment result: suspected death; Risk level: three; Judgment time: July 22, 2025 02:59; Current coordinate position: shed B1-08; Attitude angle data snapshot and risk score.

[0096] The chicken farm manager receives the platform notification and conducts manual inspection, on-site confirms that the chicken has died, and marks the completion through the platform. The system automatically returns to the normal monitoring state and classifies the individual data into the death behavior sample library for subsequent model training and risk identification parameter optimization.

[0097] Implementation effect description: The overall detection false positive rate is less than 1.5%, effectively identifying most abnormal death individuals; Compared with manual inspection, the death identification time is shortened by an average of 2 hours; The system running power consumption is controlled within 0.3Wh per day, supporting continuous operation for more than 30 days; The method is highly adaptable and can be extended to multiple varieties such as laying hens, broilers, and breeding hens.

[0098] Example three: chicken restless behavior identification and early warning application

[0099] In the same batch of experimental chicken population, the system continuously collects the three-axis attitude angle data (Yaw, Pitch, Roll) of each chicken, and performs feature extraction and behavior comparison analysis according to the set sampling period (every 30 seconds).

[0100] The individual with ID-019, between 6:15 and 6:30 in the morning, exhibits the following behavior characteristics: Yaw angle change frequency: 42 times of direction change were recorded within 15 minutes, with an average of 2.8 changes every 30 seconds; Mean of historical Yaw frequency: about 9 times / 15 minutes, standard deviation 3.2; Z-score calculation:

[0101] Pitch angle and Roll angle fluctuation: within the normal range, no abnormal orientation; Attitude angle stability: there is slight fluctuation, non-stationary state, consistent with the characteristics of active individuals; Gait and displacement trajectory: through the synchronous data of the platform, it can be seen that the chicken frequently moves back and forth in the shed at short distances, forming a non-linear loop trajectory.

[0102] The system determines according to the abnormal behavior identification logic set by the present application: Yaw angle frequency is significantly abnormal (Z-score exceeds 10); Attitude is unstable but does not fall into the death risk interval; No stationary state, no overturning orientation.

[0103] Accordingly, the system determines that the individual exhibits high-frequency directional change behavior, in combination with trajectory patterns and non-target behavior characteristics, and identifies it as "suspected restless or mild stress state", and triggers a first-level warning.

[0104] The platform pushes the following information: Chicken number: ID-019; Determination time: July 22, 2025 06:33; Abnormal type: high-frequency turning behavior, suspected restlessness; Risk level: first-level warning; Related indicators: Yaw angle change frequency = 42 times / 15 minutes; Suggested measures: Observe its drinking and feeding behavior, and isolate and investigate environmental stress factors if necessary.

[0105] The manager then reviews the monitoring records in this area and finds that the light flickering in the chicken house where chicken ID-019 is located is abnormal, and the fan makes intermittent noise, which is suspected to cause local chicken unrest. After subsequent environmental repair, the chicken's behavior returned to normal within 30 minutes, the Yaw change frequency gradually fell, and the platform marked that the warning had been lifted.

[0106] Implementation effect description: Real-time identification of restless behavior effectively prompts environmental stress; No video analysis is required, and early behavior monitoring can be achieved with attitude angles alone; The warning prompt is accurate, making it easy for managers to intervene and optimize the environment; A complete closed loop of "abnormal behavior → risk prompt → environmental check → recovery tracking" is established.

[0107] Example Four: Suspected Early Disease Behavior Identification and Intervention Warning of Chickens

[0108] In a grower house, the management platform monitors the attitude angles of all chickens for 24 hours without interruption through the method described in the invention. Each chicken wears a low-power attitude recognition device, and Yaw, Pitch, and Roll data are collected every 30 seconds.

[0109] A chicken (number ID-045) exhibited the following abnormal behavior characteristics between 10:00 and 12:00 on July 21, 2025: Yaw angle change frequency: decreased from the historical average of about 18 times / 30 minutes to 5 times, with a decrease of more than 72%; Attitude angle stability characteristics: Roll angle and Pitch angle change amplitude decreased significantly; In the next 10 minutes, the three-axis angle change amplitudes were all less than 3°; Orientation feature: No rollover state; Pitch angle continuously between -20° and -40°, as a low head and neck posture; Duration: posture low active state lasts about 90 minutes; Z-score calculation example (take Yaw angle frequency as an example):

[0110] (According to historical mean 18, standard deviation 4.5)

[0111] System identification and decision-making process: Yaw frequency significantly decreased, reaching low active critical; Posture angle change stability is abnormal, indicating that the individual activity significantly decreases; Orientation does not roll over, but continues to maintain an abnormal low head posture; The comprehensive score reaches the secondary warning standard.

[0112] The system therefore generates the following warning information: Individual number: ID-045; Decision type: low active behavior + suspected disease precursor; Decision time: July 21, 2025 12:03; Risk level: secondary warning; Suggested measures: Suggest observing feces, water intake, respiratory performance, and isolating quarantine if necessary.

[0113] Subsequent intervention and result verification: The breeding personnel conducted on-site review at 14:00 after the warning prompt and found that ID-045 chicken had a small amount of loose stool, reduced water intake but not completely stopped food intake, and slightly faster breathing. The chicken was isolated and treated with antibacterial drugs. Within 24 hours, the symptoms were significantly relieved, and the subsequent behavior returned to normal, and the system automatically removed the warning mark.

[0114] After review, the chicken was diagnosed with mild E. coli infection, and if not identified in time, it could have caused the risk of spread in the group.

[0115] Implementation effect description: Through the change of posture angle behavior, the system identifies the decrease in activity and abnormal posture of the chicken in the early stage of the disease; The warning trigger time is about 6 hours earlier than the obvious clinical symptoms; It realizes non-contact, low-cost, and high-frequency early warning of epidemic diseases; Effectively support the front movement and digitization of the disease prevention and control system of the farm.

[0116] Example Five: Verification of Chicken Low Head Behavior Recognition Mechanism Based on Attitude Angle Linkage Characteristics

[0117] This embodiment aims to illustrate that even if the posture recognition device for identifying chicken behavior in the present application is installed on the chicken's foot instead of directly collecting head position data, the system can still achieve effective identification of suspected pathological behaviors such as "low head and neck" through the attitude angle linkage characteristics.

[0118] Specifically, chickens are a compact and agile bird species, and there is significant linkage and coordination between their head, torso, and legs. In daily behaviors such as standing, walking, and low head foraging, the movement of the chicken's head is usually accompanied by forward movement and forward inclination of the center of gravity. This body posture change will be synchronized to the feet in certain situations, resulting in significant changes in the three-axis attitude angles (pitch angle, roll angle, and yaw angle) on the posture recognition device worn on the chicken's foot.

[0119] In one example of a chicken individual monitored in this embodiment, it showed abnormal performance such as long inactivity and continuous low head in a certain observation period. The system analyzed the attitude angle data collected during this period and found the following characteristics: 1. Pitch angle: from -10° to -20° in normal standing, down to -30° to -45°, and remains stable for multiple sampling periods, reflecting the chicken's obvious forward inclination or body posture; 2. Roll angle: presents a slight left-right deviation with a small change range, indicating that it does not fall or perform violent actions; 3. Yaw angle: almost no change in consecutive sampling periods, indicating that the chicken did not adjust the direction or move the position during this period.

[0120] This combination of characteristics is highly consistent with the chicken's "low head, neck, and neck" state, which is often seen in the early stages of infectious diseases or other health abnormalities.

[0121] To verify the universality and feasibility of this recognition logic, the system conducted a comparative analysis of the three-axis attitude angle data of a large number of chicken samples in different behavior states, which is summarized as follows: Normal standing state: Pitch angle maintains -10° to -20°, Roll angle presents slight fluctuations, and Yaw angle occasionally adjusts; Active walking state: all three-axis attitude angles show dynamic fluctuations, Pitch angle changes significantly, Roll angle alternates left and right, and Yaw angle frequently changes; Low head static state: the pitch angle drops to the interval of -30° to -45° and remains stable; the roll angle changes significantly narrow or tends to be static; the yaw angle is almost unchanged in continuous multiple sampling periods; Turnover / death state: the pitch angle drops to the limit angle of -75° to -90°, the roll angle deviates to more than ±70°, and the three-axis angle basically no longer changes, showing "complete static".

[0122] Based on the above data rules, the following determination mechanism is proposed: By continuously sinking the pitch angle, narrowing the roll angle change, and keeping the yaw angle direction unchanged, the chicken can be identified with high confidence in the "low head and shrinkage" behavior state; This state can be clearly distinguished from the "normal standing", "active activity" and "turnover death" state in the posture angle data layer; The system can combine this mechanism with other characteristic indicators to establish a suspected abnormal behavior identification model for early detection of chicken individuals that may be infected with diseases or in a stress state.

[0123] In summary, even if the posture acquisition device is installed on the chicken's feet, the suspected disease behavior such as "low head and inaction" can still be indirectly but reliably identified through the structural linkage characteristics of the chicken itself, which provides technical support for the construction of a low-cost, high-availability, and scalable poultry health monitoring system.

[0124] Example six: judging abnormal restless behavior of chickens based on night activity level

[0125] Applicable background: chickens should be in a relatively static state at night. If a certain individual has frequent movements or repeated low head and turning behaviors at night, it may be due to environmental stress, parasitic disturbance or early pathological reaction.

[0126] Data characteristics: Frequent changes in yaw angle; Pitch angle up and down disturbance (such as low head and head switching); The activity score is more than twice the standard deviation of the group mean at the same period.

[0127] Technical value: support for night disease early warning, especially suitable for monitoring "restless-quiet" regular changes in intensive farming.

[0128] Example seven: early determination mechanism of turnover and death

[0129] Applicable background: once the chicken turns over, it will quickly develop into death if it cannot recover itself. Turnover and natural prostration can be distinguished in posture angle.

[0130] Posture angle characteristics: Pitch angle absolute value greater than 75°, or close to -90°; Roll angle close to ±90°; Three-axis angles show no significant change within several minutes.

[0131] Additional indicators: Repeated posture angle mutation → stationary state, indicating that the chicken cannot recover after a short struggle; Compared with the group activity heat map, it is on the edge of the low activity area or an isolated point.

[0132] Technical value: suitable for early identification of chicken rollover, trampling, injury and other serious abnormal states, and timely intervention.

[0133] Example eight: active degree sudden drop behavior identification in feeding period

[0134] Applicable background: chickens should show high feeding activity in a specific period (such as after morning feeding). If an individual is significantly less active during this period, it may be a health warning signal.

[0135] Identification mechanism: During the feeding peak period (e.g. 06:00-07:30), a chicken is continuously in a low activity state; Yaw angle is stable, pitch angle is low but stationary, combined with heat map showing in the corner or isolated area.

[0136] Judgment method: Compare the individual's activity with the group average activity to assess the deviation; If this phenomenon exists for two consecutive days, a health warning will be issued.

Claims

1. A method for identifying abnormal behavior and mortality status in chickens based on posture angle, characterized in that, The method includes the following steps: (1) By using a gyroscope sensor worn on the body of the chicken, three-axis attitude angle signals, including pitch, yaw and roll, are collected in real time. (2) Record the collected attitude angle signals in time series and extract the attitude angle feature data within a set time period; (3) Analyze the feature data to identify one or more of the following abnormal behavioral features: The attitude angle remains stable. The attitude angle fluctuation is extremely small; The Yaw angle continues to shift or undergoes periodic abrupt changes; High-frequency oscillation or continuous rotation of the Yaw angle within a short time window characterizes the circling behavior of chickens. (4) Based on the abnormal behavior characteristics, determine whether the chicken is in an abnormal behavior state or a suspected death state, and generate the corresponding warning result.

2. The method according to claim 1, characterized in that: The three-axis attitude angle signals are pitch, yaw, and roll angles acquired by the gyroscope, excluding acceleration signals or their derived parameters.

3. The method according to claim 2, characterized in that: The behavior recognition method excludes the use of acceleration signals, angular velocity signals, or attitude angles calculated by a six-axis fusion algorithm, which includes, but is not limited to, Kalman filtering, complementary filtering, Mahony algorithm, or Madgwick algorithm.

4. The method according to claim 1, characterized in that: In the abnormal static dimension, by setting a threshold for attitude angle change, the number of times the attitude angle change amplitude is lower than the threshold within a continuous time window is determined, which is used to identify immobile, rigid, or behaviorally degenerate states.

5. The method according to claim 1, characterized in that: In the abnormal orientation dimension, a death judgment range for attitude angle is set, including but not limited to a roll angle greater than 90° or a pitch angle less than -60°, and a death warning is triggered when the orientation continues for more than a preset time (e.g., 20 minutes).

6. The method according to claim 1, characterized in that, The method identifies "circling behavior" by analyzing the periodic variation trend of chickens' yaw angles, and specifically includes the following steps: (1) Within a set time window, extract the continuous yaw angle sequence {Y1,Y2,...,Y}. n }; (2) Calculate the change in the Yaw angle ΔY between adjacent sampling points. i =|Y i+1 -Y i | and determine its direction (clockwise or counterclockwise); (3) When ΔY i If the unidirectional cumulative trend is observed over multiple consecutive sampling periods, and the cumulative change in the yaw angle exceeds 360° within the time window, then it is determined that the chicken exhibits continuous unidirectional circling behavior. Further analysis of the chicken's historical behavioral patterns or group reference baselines can help identify whether the circling behavior is abnormal, serving as one of the criteria for diagnosing disease stress, neurological disorders, or other behavioral abnormalities.

7. The method according to claim 6, wherein, The determination of the "continuous change trend of unidirectional yaw angle" includes: (1) ΔY of three or more consecutive sampling points i The directions are consistent (i.e., all positive or all negative), and ΔY i All are not less than the set threshold (e.g., 10°); (2) The set rotation judgment time window T is 30~120 seconds. When the total cumulative change of the Yaw angle within T is ∑|ΔY i |≥360° is considered a complete rotation; (3) If two or more complete circling behaviors occur consecutively, and no forward movement is observed (as evidenced by changes in pitch angle / acceleration), then output the "continuous circling in place" behavior status label.

Images