Infrared temperature measurement filtering method and wearable system for pet living body detection

Abstract

The present application relates to the technical field of pet infrared temperature measurement, and specifically discloses an infrared temperature measurement filtering method for pet living body detection and a wearable system. The method comprises: obtaining multi-dimensional monitoring data of a pet's neck with a preset temperature measurement window as a period, wherein the multi-dimensional monitoring data comprises infrared object temperature original values, environmental temperature values, and three-axis acceleration and three-axis angular velocity data collected synchronously through a unified time base and containing multiple sampling points. The technical scheme of the present application synchronously acquires infrared temperature original values and inertial measurement data, extracts motion, posture and temperature features therefrom to objectively quantitatively evaluate the fitting level of the device and the pet, and dynamically adjusts candidate interval threshold values, calibration offsets and target state machine parameters according to the level, so as to adaptively control the temperature measurement filtering standard in an actual wearing scene.

Description

Technical Field

[0001] This invention relates to the field of infrared thermometry technology for pets, specifically to an infrared thermometry filtering method and wearable system for live pet detection. Background Technology

[0002] Currently, pet health monitoring often uses wearable smart collars, which continuously and non-contactly measure the surface temperature of the pet's neck skin through a built-in infrared temperature sensor. This method relies on maintaining a relatively stable, close alignment between the sensor's temperature measurement window and the pet's neck skin to obtain valuable physiological temperature data.

[0003] However, pets frequently engage in activities such as running, jumping, splashing water, or scratching during their daily lives, causing the collar to shift, flip, or loosen to varying degrees. In these specific application scenarios, the infrared sensor may deviate from its effective temperature measurement area, thus detecting the temperature of the pet's outer fur or the surrounding air.

[0004] Existing infrared thermometers typically use fixed temperature screening thresholds and static outlier rejection logic. They cannot objectively determine the tightness of the fit by combining the pet's real-time movement status, nor can they effectively distinguish between real body temperature changes and invalid detection data caused by device displacement. This makes it easy for the device to record temperature data that is interfered with during the pet's activity, resulting in the final output body temperature result deviating from the pet's true physiological state. Summary of the Invention

[0005] This invention aims to at least partially solve one of the technical problems in related technologies. Therefore, the objective of this invention is to propose an infrared thermometry filtering method and wearable system for pet liveness detection, so as to accurately reflect the actual physiological state of the pet.

[0006] To achieve the above objectives, a first aspect of the present invention provides an infrared thermometry filtering method for pet liveness detection, comprising:

[0007] With a preset temperature measurement window as the cycle, multi-dimensional monitoring data of the pet's neck is acquired. The multi-dimensional monitoring data includes the original infrared object temperature value, ambient temperature value, and triaxial acceleration and triaxial angular velocity data, which are collected synchronously with a unified time base and contain multiple sampling points.

[0008] Based on the original values ​​of triaxial acceleration, triaxial angular velocity, and infrared object temperature, motion intensity features, posture stability features, and temperature fluctuation features are extracted. Based on the motion intensity features, posture stability features, and temperature fluctuation features, a comprehensive deviation score of fit is calculated, and fit levels are classified.

[0009] Based on the current fit level, the temperature measurement parameters are dynamically adjusted, including candidate interval threshold, calibration offset, target state machine count threshold, and number of valid value holding times.

[0010] Using the dynamically adjusted temperature measurement parameters, the original temperature values ​​of the infrared objects within the current temperature measurement window are sequentially subjected to physical validity filtering, interval filtering based on candidate interval thresholds, calibration correction based on calibration offset, and target state machine jump control based on target state machine count thresholds to identify valid pet body surface temperature samples.

[0011] The effective pet body surface temperature sample is combined with the ambient temperature value to perform accuracy correction, and the pet body surface temperature value and the corresponding fit level label are output.

[0012] To achieve the above objectives, a second aspect of the present invention provides an infrared thermometry wearable system for pet liveness detection, comprising:

[0013] The infrared temperature acquisition module is used to acquire the raw temperature value of the infrared object and the ambient temperature value;

[0014] Inertial measurement module, used to acquire triaxial acceleration and triaxial angular velocity data;

[0015] The main control processing module is communicatively connected to both the infrared temperature acquisition module and the inertial measurement module, and is configured to perform the following steps:

[0016] With a preset temperature measurement window as the cycle, multi-dimensional monitoring data of the pet's neck is acquired. The multi-dimensional monitoring data includes the original temperature value of the infrared object, the ambient temperature value, and the triaxial acceleration and triaxial angular velocity data, which are collected synchronously with a unified time base and contain multiple sampling points.

[0017] Based on the original values ​​of triaxial acceleration, triaxial angular velocity, and infrared object temperature, motion intensity features, posture stability features, and temperature fluctuation features are extracted. Based on the motion intensity features, posture stability features, and temperature fluctuation features, a comprehensive deviation score of fit is calculated, and fit levels are classified.

[0018] Based on the current fit level, the temperature measurement parameters are dynamically adjusted, including candidate interval threshold, calibration offset, target state machine count threshold, and number of valid value holding times.

[0019] Using the dynamically adjusted temperature measurement parameters, the original temperature values ​​of the infrared objects within the current temperature measurement window are sequentially subjected to physical validity filtering, interval filtering based on candidate interval thresholds, calibration correction based on calibration offset, and target state machine jump control based on target state machine count thresholds to identify valid pet body surface temperature samples.

[0020] The effective pet body surface temperature sample is combined with the ambient temperature value to perform accuracy correction, and the pet body surface temperature value and the corresponding fit level label are output.

[0021] To achieve the above objectives, a third aspect of the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the above-described infrared temperature measurement and filtering method for pet liveness detection.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0023] The technical solution of this invention simultaneously acquires raw infrared temperature values ​​and inertial measurement data, extracting motion, posture, and temperature features to objectively quantify and evaluate the fit level between the device and the pet. Based on this level, it dynamically adjusts the candidate interval threshold, calibration offset, and target state machine parameters, thereby adaptively controlling the temperature measurement filtering standard in actual wearing scenarios. In practical use, when the pet's activity causes the device to shift or loosen, this method automatically narrows the temperature screening interval and tightens the state machine's judgment threshold, promptly intercepting interference data caused by the sensor deviating from the skin. When the fit is stable, it ensures the continuity of temperature measurement. Simultaneously, it performs accuracy correction in conjunction with ambient temperature, reducing the superimposed impact of physical movement displacement and external environmental fluctuations on infrared temperature measurement, enabling the final output surface temperature to more objectively and accurately reflect the pet's actual physiological state. Attached Figure Description

[0024] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0025] Figure 1 This is a flowchart illustrating the infrared thermometry filtering method for live pet detection provided by the present invention.

[0026] Figure 2 This is a waveform diagram of the original multimodal sensor data acquisition and signal denoising reconstruction in the infrared thermometry filtering method for pet liveness detection provided by this invention;

[0027] Figure 3 This is a nonlinear response surface plot of fit deviation score and core temperature measurement parameters in the infrared thermometry filtering method for live pet detection provided by this invention.

[0028] Figure 4 This is a waveform diagram of the target state machine under dynamic interference in the infrared thermometry filtering method for pet liveness detection provided by this invention.

[0029] Figure 5 This is a parameter convergence and internal hysteresis temperature prediction diagram of the first-order thermal equivalent model in the infrared thermometry filtering method for pet liveness detection provided by this invention under cross-temperature thermal shock.

[0030] Figure 6 This is a thermodynamic pseudo-color mapping diagram of the transient drift compensation amount changing with the ambient temperature gradient in the infrared thermometry filtering method for live pet detection provided by this invention.

[0031] Figure 7 This is a scatter plot of the three-dimensional spatial distribution of transient acceleration vectors under long-term resting state in the infrared thermometry filtering method for pet liveness detection provided by this invention.

[0032] Figure 8 This is a nonlinear exponential saturation curve of the artificial temperature rise of the local heat island in the closed microenvironment in the infrared thermometry filtering method for live pet detection provided by this invention.

[0033] Figure 9 This is a schematic diagram illustrating the implementation of the infrared temperature measurement wearable system for pet liveness detection provided by the present invention;

[0034] Figure 10 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0035] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0036] The following describes, with reference to the accompanying drawings, an infrared temperature measurement filtering method, an infrared temperature measurement wearable system, and an electronic device for pet liveness detection according to embodiments of the present invention.

[0037] Example 1:

[0038] like Figure 1 As shown, this embodiment provides an infrared temperature measurement filtering method for pet liveness detection. In practical applications, this method typically operates within the microprocessor or digital signal processor of a smart health collar worn around the pet's neck.

[0039] Because pets cannot cooperate with medical equipment measurements like humans, their daily activities such as running, eating, and scratching cause frequent relative displacement of wearable devices, introducing significant environmental temperature interference and motion artifacts. The method provided in this embodiment aims to achieve precise stripping and correction of medical-grade body surface temperature in complex dynamic scenarios through low-level fusion of multi-sensor data and adaptive dynamic control logic.

[0040] Specifically, this embodiment provides an infrared temperature measurement filtering method for pet liveness detection, which includes the following data acquisition steps. The system acquires multi-dimensional monitoring data of the pet's neck at a preset temperature measurement window period. The multi-dimensional monitoring data includes the original infrared object temperature value, ambient temperature value, and triaxial acceleration and triaxial angular velocity data, which are synchronously acquired through a unified time base and contain multiple sampling points.

[0041] It should be noted that the aforementioned preset temperature measurement window refers to the time span during which the system conducts a complete data evaluation and decision-making process.

[0042] Under the scheduling of the embedded operating system, the system is equipped with a high-precision hardware timer to establish a unified time base. The purpose of this unified time base is to ensure that sensor data with different sampling frequencies can be strictly aligned on the time axis.

[0043] Within each temperature measurement window, the infrared temperature acquisition module and the inertial measurement module transfer the underlying digital signals to the memory buffer through the direct memory access mechanism, thereby generating the raw infrared object temperature value containing multiple consecutive sampling points, the ambient temperature value read by the thermistor inside the infrared sensor, and the triaxial acceleration and triaxial angular velocity data output by the six-axis inertial sensor.

[0044] This synchronous acquisition mechanism eliminates the data phase difference between different sensors, laying the data foundation for subsequent joint extraction of spatiotemporal features.

[0045] Furthermore, after acquiring the aforementioned multi-dimensional monitoring data, the system performs feature extraction and status assessment steps.

[0046] Based on the original values ​​of triaxial acceleration, triaxial angular velocity, and infrared object temperature, the system extracts motion intensity features, attitude stability features, and temperature fluctuation features, and calculates a comprehensive deviation score for fit based on these features, thereby classifying fit levels.

[0047] Specifically, the steps described above for extracting motion intensity features, attitude stability features, and temperature fluctuation features based on the triaxial acceleration, triaxial angular velocity, and the original infrared object temperature include the following mathematical statistics and signal processing procedures:

[0048] The system first calculates the standard deviation of the sum of triaxial acceleration values ​​at all sampling points within a single temperature measurement window, using this as the motion intensity characteristic. In this process, the system calculates this motion intensity characteristic value using the following formula:

[0049] ;

[0050] In the formula, Defined as the calculated motion intensity characteristic value; Defined as the total number of sampling points within a single temperature measurement window; Defined as the first The sum of the triaxial accelerations at a sampling point is the square root of the sum of the squares of the acceleration components at that sampling point on the three orthogonal spatial axes. Defined as the arithmetic mean of the sum of all triaxial accelerations within a single temperature measurement window.

[0051] In a biophysical sense, when a pet is at rest, the sum of its three-axis accelerations is approximately equal to the Earth's gravitational acceleration constant. By calculating the standard deviation, the algorithm can automatically filter out the DC component of gravitational acceleration, thereby accurately extracting the dynamic high-frequency acceleration component generated by the pet's own muscle exertion. The larger this value, the more intense the transient impact force and amplitude of movement on the pet's neck, and the higher the risk of equipment slippage.

[0052] For example, the system calculates the mean of the triaxial angular velocities of all sampling points within a single temperature measurement window in parallel, using this as the attitude stability feature. In this process, the system calculates this attitude stability feature value using the following formula:

[0053] ;

[0054] In the formula, Defined as the attitude stability feature value of the calculated output; Defined as the first The triaxial angular velocity of each sampling point is Components on the axis; Defined as the first The triaxial angular velocity of each sampling point is Components on the axis; Defined as the first The triaxial angular velocity of each sampling point is Components on the axis.

[0055] Three-axis angular velocity primarily reflects the rotational angular rate of the collar device in three-dimensional space. Compared to linear acceleration, the mean of the sum of angular velocities provides a more intuitive indication of whether the device has rolled or yawed around the pet's neck. If the attitude stability characteristic value is high, it indicates that the device is in a high-frequency flipping state. At this time, the orientation of the infrared temperature measurement window is extremely unstable, possibly pointing directly at the outside air or away from the hairless area.

[0056] like Figure 2 The display shows the waveforms of multimodal sensor data during a 10-second observation period. The horizontal axis represents time in seconds, the left vertical axis represents temperature in degrees Celsius, and the right vertical axis represents dynamic acceleration characteristics in meters per second squared.

[0057] Figure 2 It contains three curves: the gray curve represents the original temperature signal, the red curve represents the reconstructed temperature signal, and the blue curve represents the dynamic acceleration characteristics.

[0058] During the resting state from 0 to 3 seconds, the original temperature signal fluctuated steadily around 38.5 degrees Celsius, and the blue dynamic acceleration feature was at a low level close to 0 meters per second squared.

[0059] When the pet is moving violently from the 3rd to the 6th second, the blue dynamic acceleration characteristic curve rises rapidly and exceeds the threshold of 2 meters per second squared. At this time, due to the displacement of the collar and the introduction of cold air, the gray original temperature signal waveform oscillates violently and drops sharply to below 37 degrees Celsius.

[0060] Faced with this interference, the red reconstructed temperature signal promptly triggered the state machine's hold and filtering mechanism, maintaining the temperature within the physiologically reasonable range of above 38 degrees Celsius during the interference period. This intercepted the abnormal cooling data caused by physical displacement, demonstrating the technical effect of adaptively adjusting temperature measurement parameters to filter interference signals in this embodiment.

[0061] Simultaneously, it can be observed that at the end of the sustained high-frequency interference from the 5th to the 6th second, the red reconstructed temperature signal curve shows a slow decline of approximately 0.6 degrees Celsius. Furthermore, after the action ceases at the 6th second, there is a delay of about 1 second in the waveform's regression to the basal body temperature. This indicates that the current numerical preservation strategy, in dealing with interference of extremely long duration, adopts a more conservative smoothing and filtering mechanism to ensure the absolute reliability of the data. This effectively avoids the temperature jump that may occur immediately after the interference ends, ensuring extremely high confidence in the recovered temperature measurement data.

[0062] In addition, the system calculates the variance of the original temperature values ​​of the infrared object within multiple consecutive temperature measurement windows, which is used as the temperature fluctuation feature.

[0063] Since the physiological surface temperature of a pet is a slowly changing physical quantity in a short period of time, if the original value of the infrared object temperature shows drastic variance fluctuations in multiple consecutive temperature measurement windows, it usually means that the sensor's field of view is switching back and forth between the pet's skin, dense fur, and the cold air in the environment. This is also a significant electrophysiological characteristic of poor physical fit of the device.

[0064] Optionally, after extracting the features from the above three dimensions, the system needs to integrate them into a unified evaluation index. The system performs extreme value normalization on the motion intensity feature, posture stability feature, and temperature fluctuation feature respectively, and then performs a weighted summation using preset weighting coefficients to obtain the comprehensive deviation score of the fit. In this process, the system uses the following formula for calculation:

[0065] ;

[0066] In the formula, Defined as the final calculated overall deviation score of fit; Defined as the eigenvalue that is mapped to the interval between 0 and 1 after performing extreme value normalization on the eigenvalue of motion intensity; Defined as the eigenvalue after performing extreme value normalization on the attitude stability eigenvalue; Defined as the characteristic value after performing extreme value normalization on the temperature fluctuation characteristics; Defined as the preset first weighting coefficient assigned to the normalized motion intensity feature value; Defined as a preset second weighting coefficient assigned to the normalized attitude stability feature value; Defined as a preset third weighting coefficient assigned to the normalized temperature fluctuation characteristic value.

[0067] The system compares the overall deviation score of the fit with a preset first grading threshold and a second grading threshold, wherein the first grading threshold is less than the second grading threshold, and outputs the fit level as excellent, good, or poor.

[0068] The specific logic is as follows: when the score is less than the first grade threshold, it is judged as excellent fit; when the score is greater than or equal to the first grade threshold and less than the second grade threshold, it is judged as good fit; when the score is greater than or equal to the second grade threshold, it is judged as poor fit.

[0069] It should also be noted that after obtaining the accurate physical fit level, the system dynamically adjusts the temperature measurement parameters according to the current fit level. The temperature measurement parameters include the candidate interval threshold, calibration offset, target state machine count threshold, and number of valid value holding times.

[0070] Subsequently, the system uses the dynamically adjusted temperature measurement parameters to sequentially perform physical validity filtering, interval filtering based on candidate interval thresholds, calibration correction based on calibration offset, and target state machine jump control based on target state machine count thresholds on the original temperature values ​​of the infrared objects within the current temperature measurement window, ultimately identifying valid pet body surface temperature samples.

[0071] Specifically, the aforementioned target state machine counting thresholds include the target entry counting threshold and the target loss counting threshold.

[0072] The specific execution process of dynamically adjusting the temperature measurement parameters according to the current fit level is as follows: the system establishes a mapping relationship between the fit level and the temperature measurement parameters, queries the mapping relationship based on the currently acquired fit level, and extracts the corresponding temperature measurement parameters.

[0073] A multi-dimensional parameter mapping table is pre-programmed into the non-volatile memory at the system's core. The fit levels are sequentially divided into excellent fit, good fit, and poor fit. As the fit level transitions from excellent to poor, it means that the signal-to-noise ratio of the measurement environment gradually deteriorates. At this time, the range of the candidate interval threshold corresponding to the mapping relationship decreases, the corresponding calibration offset decreases, the corresponding target entry count threshold increases, the corresponding target loss count threshold decreases, and the corresponding number of valid value retentions increases.

[0074] For example, the dynamic adjustment of the aforementioned temperature measurement parameters has clear physical and engineering significance. Taking calibration offset as an example, when an infrared thermopile sensor performs non-contact temperature measurement, the thermal radiation energy it receives is inversely proportional to the square of the detection distance. When the fit level transitions from excellent to poor, it indicates that a large air gap has formed between the device and the pet's skin. Air, as a poor thermal conductor, will cause a significant decrease in the effective thermal radiation flux received by the sensor, resulting in a system bias of lower temperature readings. Since the underlying temperature reconstruction logic of the system subtracts this calibration offset from the original reading temperature, reducing the calibration offset by querying the mapping relationship (i.e., reducing the subtracted value) can directly increase the final calculated temperature output. The system thereby performs reverse compensation at the underlying algorithm level to offset the radiation attenuation caused by the increased gap.

[0075] Similarly, when the environment deteriorates, increasing the target entry counting threshold raises the confidence threshold for the algorithm to confirm that a live target has entered the field of view, effectively preventing false triggering; while decreasing the target loss counting threshold allows the system to more decisively discard invalid drift data when the signal is severely interfered with.

[0076] Optionally, the aforementioned candidate interval thresholds include strict candidate intervals for target entry determination and soft candidate intervals for target retention determination.

[0077] The specific steps for dynamically adjusting the candidate interval thresholds in the temperature measurement parameters are as follows: The system obtains a preset lower and upper limit benchmark, and maps the excellent, good, and poor fit to a first, second, and third level code, respectively, with decreasing numerical values. The system uses the interval parameter corresponding to the first level code as a benchmark; when the current fit level is not excellent, the system calculates the level difference between the level code corresponding to the current fit level and the first level code.

[0078] Furthermore, to achieve smooth scaling of the candidate interval, the system uses the following formula to calculate the dynamically adjusted upper limit:

[0079] ;

[0080] In the formula, Defined as the current upper limit after dynamic adjustment; Defined as a preset baseline upper limit; Defined as a preset adjustment step size; Defined as a first-level code; Defined as the level code corresponding to the current fit level.

[0081] Simultaneously, the system adds the product of the preset adjustment step size and the level difference to the baseline lower limit, using this as the dynamically adjusted current lower limit, thereby generating the strict candidate interval and soft candidate interval for the current temperature measurement window. Through the aforementioned linear contraction algorithm, when the fit deteriorates, the algorithm allows the candidate interval range of the temperature to adaptively contract, forcibly blocking edge-distorted temperature points that are significantly deviated from the physiological range due to cold air interference.

[0082] like Figure 3 The nonlinear response surface plot of fit deviation score and core temperature measurement parameters is displayed. The first horizontal axis of the bottom surface is the normalized motion intensity feature, with the unit being dimensionless. The second horizontal axis of the bottom surface is the normalized temperature fluctuation feature, also with the unit being dimensionless. The vertical axis represents the upper limit of the candidate interval threshold, with the unit being degrees Celsius.

[0083] Figure 3 The three-dimensional response surface in the image intuitively reflects the dynamic adjustment process of the temperature measurement parameters through pseudo-color gradients. The dark blue area represents a lower threshold upper limit, while the bright yellow area represents a higher threshold upper limit.

[0084] From the overall trend of the surface, it can be seen that when the normalized motion intensity feature and the normalized temperature fluctuation feature are both near 0, that is, when the pet is in a stable resting state, the upper limit of the candidate interval threshold is maintained in the bright yellow high area of ​​39.5 degrees Celsius. At this time, the system has a relatively wide receiving window for temperature data.

[0085] As the two feature values ​​gradually increase towards 1, indicating that the pet's movement intensifies and the signal stability decreases, the surface exhibits a clear non-linear concave trend. The upper limit of the threshold gradually and smoothly shrinks to the dark blue area around 37.5 degrees Celsius. This objectively demonstrates the technical effect of the system automatically executing a more stringent temperature screening range based on the deterioration of the fit.

[0086] The red dots in the figure mark the boundary oscillation points of the curved surface. It can be observed that when two normalized eigenvalues ​​simultaneously approach the high-interference boundary region of 0.9, the edge of the colored surface exhibits small fluctuations of approximately 0.05 degrees Celsius. These minor boundary fluctuations indicate that the currently employed nonlinear mapping algorithm triggers a rigid constraint mechanism in the underlying anti-aliasing filter when dealing with boundary conditions where multidimensional features are simultaneously highly coupled. This is not only a physical manifestation of the algorithm's forced truncation of interference signals in extreme non-steady-state regions, but also represents the optimal engineering trade-off between smoothness and absolute isolation against interference, ensuring filtering safety under extreme conditions.

[0087] Specifically, after completing the dynamic reconstruction of all parameters, the system sequentially performs physical validity filtering and interval filtering based on candidate interval thresholds on the original temperature values ​​of the infrared object within the current temperature measurement window, and combines the state machine to identify the validity.

[0088] The system constructs a target state machine that includes target loss state, target entry state, and target hold state. This state machine is essentially a finite state controller running in the main control memory, used to smoothly process discrete temperature measurement sequences.

[0089] In the target loss state, if the original infrared object temperature value of the current temperature measurement window passes through the strict candidate interval filter in the candidate interval threshold, and the number of consecutive passes through the strict candidate interval filter reaches the target entry count threshold in the target state machine count threshold, then the system state is switched to the target entry state. The strict candidate interval is used to ensure that the signal read when the pet collar initially contacts the skin has extremely high medical reliability.

[0090] In the target entry state, if the raw infrared object temperature value of subsequent temperature measurement windows continuously passes through the soft candidate interval in the candidate interval threshold, the system state is switched and maintained in the target holding state. The calibrated and corrected raw infrared object temperature value currently in the target holding state is marked as a valid pet body surface temperature sample. The use of a relatively wide soft candidate interval aims to introduce a hysteresis interval after the system has locked onto the target, increasing the algorithm's fault tolerance and preventing frequent oscillations and jumps in the state machine caused by the pet's slight breathing or swallowing movements.

[0091] In the target-held state, if the number of consecutive failures to pass the soft candidate interval filter reaches the target loss count threshold in the target state machine counting threshold, then the system switches back to the target loss state. This typically occurs when a pet violently shakes its body, causing the collar to completely detach from the effective temperature measurement area.

[0092] It should also be noted that the method includes a state machine-based effective value preservation mechanism to cope with signal loss in very short periods of time.

[0093] When in the target holding state, if the original infrared object temperature value of the current temperature measurement window does not pass the soft candidate interval filtering, the valid value holding operation is triggered, and the valid pet body surface temperature sample recorded in the previous temperature measurement window is used as the output sample of the current temperature measurement window.

[0094] The system counts the number of consecutive failures that fail to pass the soft candidate interval filter. If the number of failures does not exceed the number of valid value hold counts, the target hold state is maintained. If the number of failures reaches the number of valid value hold counts, the valid value hold operation is terminated and the system is forcibly switched to the target loss state. This zero-order hold mechanism can provide smooth data filling during the 1 to 2 seconds of momentary signal loss caused by a pet's brief scratching or shaking, avoiding abrupt changes and discontinuities in the temperature curve of the front-end application interface.

[0095] like Figure 4 The waveform diagram of the target state machine under dynamic disturbance is shown. The horizontal axis represents time in seconds, the left vertical axis represents temperature in degrees Celsius, and the right vertical axis represents the state machine level in dimensionless units.

[0096] Figure 4 It includes three graphical representations: gray scatter dots represent the original temperature scatter dots, red solid lines represent the final output temperature, and blue step lines represent the state machine level.

[0097] Within the stable range of 4 to 8 seconds, the gray original temperature scatter points are densely distributed around 38.5 degrees Celsius, and the blue step line is stable at the two-level position, indicating that the system is in the target holding state. At this time, the red final output temperature curve is smooth and continuous.

[0098] When a continuous and intense dynamic disturbance is introduced from the 8th to the 12th second, a large number of gray original temperature scatter points drop to around 36.5 degrees Celsius. In the 1.5 seconds before the disturbance occurs, the blue step line remains at level two, successfully triggering the effective value hold operation, so that the red final output temperature is maintained at the physiological normal level before the disturbance, verifying the system's ability to filter and hysteresis protect against short-term physical disturbances.

[0099] However, as the duration of the interference exceeded the maximum number of times the effective value could be maintained, the blue step line dropped to the zero level at 9.5 seconds, indicating a loss of target. The system then proactively stopped outputting unreliable temperature measurement data to prevent false alarms from the front end and to ensure the rigor of medical-grade monitoring.

[0100] Furthermore, after the original temperature scatter plot returns to normal at the 12-second mark, the blue step line needs to go through a target entry state at level one, and then climb back to level two and resume temperature output after a fixed decision period of approximately two seconds. This fixed time delay mechanism from target loss to target retention aims to completely filter out the aftershock noise in the early stages of interference termination through a strict recapture confidence confirmation period. This design reflects the state machine's high-reliability defense strategy in the face of complex dynamic interference, providing a solid data foundation for the smooth reconstruction of the subsequent effective temperature.

[0101] Optionally, after extracting the data cleaned by the state machine, the system performs the final core temperature correction step.

[0102] The system performs accuracy correction on the valid pet body surface temperature samples in conjunction with the ambient temperature value, and outputs the pet body surface temperature value and the corresponding fit level label. This accuracy correction step specifically includes the following filtering and nonlinear regression processes:

[0103] The system acquires the posterior temperature estimate output from the previous temperature measurement window and generates the prior estimate for the current temperature measurement window based on the posterior temperature estimate and a preset process noise variance. During this process, to suppress Gaussian white noise, the system performs prior deduction within a filter, calculating the prior estimate using the following formula:

[0104] ;

[0105] In the formula, Defined as the prior estimate of the current temperature measurement window; Defined as the posterior temperature estimate output from the previous temperature measurement window.

[0106] Since pet body temperature is a thermal inertial system, the prior prediction of the current state directly inherits the optimal posterior estimate from the previous moment. Subsequently, the system calculates the residuals and update weights based on the valid pet body surface temperature samples in the current temperature measurement window, and outputs a denoised initial screening temperature sequence through a prediction-correction filtering model.

[0107] Next, the system uses the synchronously acquired ambient temperature value as a feature input into a pre-established nonlinear regression model to calculate the ambient thermal radiation drift compensation. The temperature measurement principle of the infrared thermopile sensor relies on the temperature difference between the cold and hot ends.

[0108] When the ambient temperature fluctuates drastically, the reference cold junction temperature inside the sensor package shifts. A pre-trained nonlinear regression model can accurately calculate the amount of drift caused by this disruption of internal thermal equilibrium based on the current ambient temperature and its gradient.

[0109] Finally, the system superimposes and corrects the noise-reduced initial screening temperature sequence with the environmental thermal radiation drift compensation amount to generate and output the final pet body surface temperature value with medical reference significance.

[0110] For example, in order to achieve long battery life with the limited battery capacity of portable wearable devices, the method also includes a hierarchical power consumption scheduling step based on fit level, applied to systems containing multimodal sensors.

[0111] In normal monitoring mode, the system performs the steps of periodically collecting the multi-dimensional monitoring data and infrared temperature measurement filtering. At this time, most of the processor's peripherals are in deep sleep mode, maintaining basic temperature measurement operation with a microampere-level current.

[0112] When the pet's surface temperature values ​​output by a first preset number of temperature measurement windows deviate from the preset normal temperature range, and the current fitting level is "excellent," it indicates that a real body temperature abnormality has been detected even with excellent fitting. The system immediately generates a first wake-up command to activate the multimodal sensor and enter an abnormality enhancement mode to collect multidimensional vital sign data. For example, it activates the photoplethysmography (PPG) sensor and the blood oxygen sensor for cross-validation.

[0113] Conversely, when the pet's surface temperature values ​​output by a first preset number of temperature measurement windows deviate from the preset normal temperature range, but the corresponding fit level is poor fit, the system will not directly wake up the high-power peripherals. Instead, it will increase the sampling frequency of the temperature measurement windows. Only after the fit level returns to good fit and the temperature is confirmed to deviate from the preset normal temperature range again will the first wake-up command be generated. This scheduling strategy avoids unnecessary power consumption caused by false alarms due to a loose collar.

[0114] Specifically, considering that pets are prone to encountering wireless signal blind spots when outdoors, the method also includes weak network fault tolerance and quality-graded upload steps.

[0115] The system assigns three levels of upload priority to the pet's body temperature value based on the fit level and whether the current pet's body temperature value is within the normal temperature range. The highest priority is given to the pet's body temperature value that is not within the normal temperature range and has excellent fit. The second highest priority is given to the pet's body temperature value that is within the normal temperature range and has excellent fit. The normal priority is given to the pet's body temperature value that has poor fit.

[0116] The system assigns priority to the data and writes it into a circular buffer queue of a preset fixed length for uploading. This queue is built in the microcontroller's static random access memory. When the pet is in an area without network coverage, such as a basement, and the circular buffer queue is determined to be full and receiving new data to be written, the queue management program will traverse the entire buffer space and prioritize overwriting the earliest written ordinary priority data in the queue, while retaining the highest priority data and the data at the head of the queue that is waiting for confirmation from the receiving end.

[0117] This dynamic priority packet loss strategy ensures that, even in extremely poor network communication environments, the very early fever warning data, which is crucial to the life safety of pets, will never be lost due to being squeezed out by regular heartbeat packet data.

[0118] Optionally, the physiological body temperature of each pet varies significantly due to differences in breed, coat thickness, and body fat percentage. Therefore, the method also includes an end-to-end cloud collaborative baseline update step.

[0119] The device packages the pet's surface temperature value and corresponding fit level label, and sends it to the cloud server via the Internet of Things (IoT) communication protocol. The cloud server executes a big data analysis program, receives the updated body temperature baseline data sent by the cloud server, and updates the local warning threshold; wherein, the updated body temperature baseline data is generated by the cloud server assigning baseline update weights to the pet's surface temperature value according to the fit level, and iteratively calculating based on a weighted moving average algorithm.

[0120] Specifically, the cloud algorithm highly trusts uploaded data marked with "excellent fit," assigning it a very high baseline iteration weight; while data marked with "poor fit" is assigned a very low weight or ignored entirely. Through this deep closed-loop iteration of end-to-end cloud collaboration, the collar will become increasingly adapted to the physiological characteristics of a specific pet during long-term wear, and its built-in health alarm thresholds will gradually evolve from an initial uniform definition to a personalized monitoring baseline based on the specific physiological characteristics of the individual.

[0121] In summary, existing technologies often rely on fixed-threshold temperature measurement schemes that struggle to distinguish between a pet's actual fever and sudden temperature changes caused by the device moving away from the skin due to running or scratching. These schemes are also prone to outputting irrelevant and inaccurate data when the pet is exercising vigorously or when the collar is loose.

[0122] The method provided in this embodiment calculates an objective comprehensive deviation score for fit using inertial sensor data, and uses this as the underlying logic engine to dynamically control the entire data link from interval filtering and calibration compensation to state machine operation. In practical applications, this technical solution can effectively suppress physical interference artifacts introduced by the hyperactivity of pets, reduce the frequency of false alarms from the device, and ensure that the pet's body temperature data presented to the user has high accuracy and clinical reference value, effectively improving the reliability of pet smart wearable devices in actual daily use.

[0123] Example 2:

[0124] In real-world applications, pets are rarely kept in a constant temperature environment for extended periods. For example, in winter, a pet may suddenly move from a heated indoor space (around 25 degrees Celsius) to the cold outdoors (where the temperature may be below 0 degrees Celsius); or in summer, a pet may move from an air-conditioned, low-temperature indoor space to a hot outdoor space. This rapid spatial movement across temperature zones exposes the infrared thermometer wearable system around the pet's neck to severe external thermal shock.

[0125] The underlying physical temperature measurement principle of infrared thermopile sensors is based on the Seebeck effect. The weak voltage signal output by the sensor depends not only on the amount of thermal radiation from the target object (pet skin), but also on the reference temperature of the cold junction inside the sensor.

[0126] Typically, a thermistor is encapsulated inside the sensor to measure the cold junction reference temperature in real time. However, when the external ambient temperature changes drastically, the external plastic housing, metal pins, and internal metal casing of the sensor each have their own independent physical thermal resistance and physical thermal capacity, which causes a significant time delay in the process of external temperature conduction into the sensor.

[0127] During this delay, a severe imbalance occurs between the actual hot-junction temperature and cold-junction temperature inside the sensor and the temperature measured by the thermistor, forming an internal transient temperature gradient. This transient temperature gradient generates additional parasitic thermoelectric potential, causing the temperature measurement system to output drift data that deviates significantly from physiological norms.

[0128] To address this technical problem, this embodiment introduces a transient compensation step for cross-temperature thermal shock before the accuracy correction step in Embodiment 1 or in the parallel process. The specific execution logic is as follows:

[0129] Specifically, the method described in this embodiment also includes a transient compensation step for cross-temperature zone thermal shock. This step first involves the microprocessor periodically calculating the rate of change of the ambient temperature value within a second preset number of temperature measurement windows.

[0130] In the underlying data processing architecture, the system maintains a dedicated first-in, first-out (FIFO) circular data buffer for storing ambient temperature values. With each temperature measurement window, the latest ambient temperature value is pushed into this buffer, while the oldest data is removed. The microprocessor extracts the ambient temperature value of the current measurement window from the historical ambient temperature values ​​corresponding to the second-to-last preset number of measurement windows, calculates the difference between the two, and divides it by the absolute physical time span between these two data points to obtain a high-precision rate of change in ambient temperature.

[0131] This rate of change is essentially the first derivative of the ambient temperature on the time axis. The sign of its value indicates whether the ambient temperature is in a process of rapid warming or rapid cooling, while its absolute value directly and objectively quantifies the severity of the external thermal shock.

[0132] It should be noted that relying solely on the rate of change of ambient temperature as a single indicator to determine cross-temperature zone migration carries a high risk of misjudgment.

[0133] In actual operation, if the pet is in a resting or sleeping state, and the cold air from an air conditioner or the hot air from a heating device happens to blow directly on the collar, it will also cause a drastic change in the ambient temperature.

[0134] However, in this scenario, the pet itself does not undergo spatial relocation, and the microclimate heat exchange between the sensor and the pet's skin remains relatively stable. Directly triggering thermal shock compensation would introduce a reverse calculation error. Therefore, the system incorporates a joint triggering logic.

[0135] The system determines that the thermal shock transient compensation mode is triggered only when the absolute value of the rate of change of the ambient temperature exceeds a preset thermal shock change threshold, and the motion intensity characteristic is continuously higher than a preset displacement judgment threshold. Here, the preset thermal shock change threshold is an empirical critical value calibrated through a large number of constant temperature chamber sudden change experiments; and the motion intensity characteristic is a value extracted based on triaxial acceleration in Example 1.

[0136] The system requires that the intensity of the movement must be continuously higher than the preset displacement judgment threshold. This means that when the ambient temperature changes abruptly, the pet must also exhibit continuous spatial physical movement behavior with a certain intensity (such as continuous walking or running).

[0137] Through cross-validation of this multimodal data, the system eliminates false abrupt changes caused by local turbulent winds, effectively improving the rigor and reliability of cross-temperature zone transfer determination.

[0138] Optionally, after the system has undergone the above-mentioned rigorous logical judgment and formally triggered and entered the thermal shock transient compensation mode, the microprocessor first executes the parameter call instruction.

[0139] In the thermal shock transient compensation mode, the pre-stored thermal resistance and thermal capacity parameters of the infrared temperature sensor package shell are acquired. During the hardware calibration stage before the smart collar leaves the factory, engineers will establish a database of the overall thermodynamic properties of the wearable device based on the thermal conductivity of the polymer material used in the collar shell, the geometric wall thickness of the shell, the material characteristics of the infrared filter of the infrared sensor, and the specific heat capacity of the metal shell.

[0140] Thermal resistance represents the resistance to heat conduction from the external environment to the core region of the sensor; thermal capacity represents the sensor's ability to absorb or release heat, causing a change in its own temperature. These two key physical parameters are stored in the microcontroller's read-only memory, serving as the fundamental constant inputs for subsequently building a dynamic thermodynamic model.

[0141] For example, after obtaining the above physical parameters, the system enters the core model calculation stage.

[0142] Based on the current ambient temperature, the estimated internal hot-end temperature of the previous temperature measurement window, and the thermal resistance and thermal capacity parameters, the system iteratively calculates and outputs the current real-time hot-end hysteresis temperature inside the sensor by constructing a first-order thermal equivalent model.

[0143] To achieve efficient computation in embedded microcontrollers with limited computing power, this embodiment employs the lumped parameter method, simplifying the complex partial differential heat conduction equation into a first-order resistive-capacitive equivalent network model. In this equivalent network, the thermal resistance parameter is equivalent to the resistance in the circuit, the thermal capacitance parameter is equivalent to the capacitance in the circuit, the ambient temperature is equivalent to the input voltage, and the real-time hot-junction hysteresis temperature inside the sensor is equivalent to the response voltage across the capacitor. The microprocessor uses discrete-time difference equations to perform periodic iterative calculations on this first-order thermal equivalent model. The specific calculation formula is given below:

[0144] ;

[0145] In the formula, Defined as the real-time hot-end hysteresis temperature output after the current temperature measurement window is calculated by the first-order thermal equivalent model; Defined as the estimated internal hot end temperature value calculated and stored in the register in the previous temperature measurement window; Defined as the physical time interval between two adjacent temperature measurement windows; Defined as the pre-stored thermal resistance parameter of the infrared temperature sensor package housing; Defined as the pre-stored thermal capacity parameter of the infrared temperature sensor package housing; Defined as the ambient temperature value obtained by the temperature measurement module at the current moment.

[0146] Since the internal temperature is consistent with the external temperature under steady-state conditions, when entering the thermal shock transient compensation mode for the first time, the initial estimated internal hot end temperature is assigned the ambient temperature value that was in thermal equilibrium before the trigger.

[0147] Through continuous iteration of this formula, the system can accurately simulate the slow rise or slow fall temperature response curve inside the sensor due to physical thermal inertia when thermal shock occurs, without increasing the hardware cost of additional internal temperature probes, using software algorithms.

[0148] like Figure 5 The graphs show the parameter convergence and internal hysteresis temperature prediction of the first-order thermal equivalent model under cross-temperature thermal shock. The horizontal axis represents time in seconds, and the vertical axis represents temperature in degrees Celsius.

[0149] Figure 5 It contains two core curves: the blue curve represents the external ambient temperature, and the red curve represents the sensor's internal hysteresis temperature, which is calculated iteratively based on a first-order thermodynamic model.

[0150] During the stable monitoring period from 0 to 100 seconds, the equipment was in a warm and constant temperature environment, with both the blue and red curves remaining stable at around 25 degrees Celsius.

[0151] When the time reaches 100 seconds, the blue external ambient temperature curve drops sharply and rapidly to 0 degrees Celsius, objectively reflecting the thermal shock effect caused by the sudden transfer of the pet from indoors to the cold outdoors.

[0152] Faced with drastic changes in external boundary temperature, the red internal hysteresis temperature curve did not follow suit with abrupt changes, but instead exhibited a smooth and slow exponential decreasing trend, gradually converging and approaching 0 degrees Celsius after approximately 400 seconds. This obvious waveform differentiation intuitively confirms that the thermal resistance and thermal capacity parameters used in this embodiment play an accurate physical inertial buffering role, indicating that the equivalent model can successfully reproduce the huge transient temperature gradient difference formed on the inner and outer sides of the sensor in the early stage of spatial transfer, providing a reliable calculation benchmark for the system to perform accurate transient thermal radiation compensation.

[0153] However, a closer look at the descent range after 100 seconds reveals that, influenced by the dynamic changes in the outdoor cold air velocity, the blue external ambient temperature curve exhibits dense high-frequency spikes. Meanwhile, due to the strong low-pass filtering physical characteristics of the fixed static parameters of the current first-order model, the red internal hysteresis temperature prediction curve filters out almost all external high-frequency disturbances, resulting in a completely smooth descent trajectory.

[0154] Furthermore, considering the complex external wind speed disturbances in practical applications, the current system intentionally filters out high-frequency noise from surface convection heat transfer caused by instantaneous strong convection through first-order low-pass physics characteristics. This allows the predicted cooling rate to smoothly filter out transient airflow interference. This first-order thermodynamic model design achieves stable fitting and accurate compensation convergence of external thermal shock trends on wearable devices with limited computing resources.

[0155] Specifically, after obtaining accurate internal temperature model data, the system proceeds to the drift calculation stage.

[0156] The system calculates the transient temperature gradient difference between the current ambient temperature and the real-time hot-end hysteresis temperature, and multiplies this transient temperature gradient difference by a preset transient drift compensation coefficient to calculate the transient thermal radiation compensation. The magnitude of the parasitic thermoelectric potential inside the sensor is highly positively correlated with the temperature gradient experienced by the hot and cold ends.

[0157] To quantify the impact of this parasitic error on the final body surface temperature measurement result, the system uses the following formula to calculate the transient compensation:

[0158] ;

[0159] In the formula, Defined as the transient compensation amount of thermal radiation calculated in the current prediction window; Defined as a preset transient drift compensation coefficient; and The definition is consistent with the one mentioned above.

[0160] The transient temperature gradient difference is obtained by subtracting the current ambient temperature from the real-time hot-end hysteresis temperature. The transient drift compensation coefficient is an engineering calibration constant characterizing the sensitivity of this specific type of infrared thermopile sensor to the internal thermal gradient.

[0161] When a pet moves from a warm indoor environment to a cold outdoor environment, the ambient temperature drops rapidly, while the internal lag temperature remains relatively high. The transient temperature gradient difference is negative, and the calculated transient thermal radiation compensation is also negative. Conversely, if a pet moves from a cold outdoor environment to a warm indoor environment, the transient thermal radiation compensation is positive. This directional gradient difference calculation accurately reflects the dynamic direction and potential energy of heat flow within the sensor.

[0162] It is important to note that after calculating the above compensation amount, it must be effectively integrated into the overall data filtering and correction process. During the accuracy correction step, the system superimposes the transient thermal radiation compensation amount onto the valid pet body surface temperature sample.

[0163] In Example 1, the system has already acquired valid pet surface temperature samples filtered by state machine and physical validity, and is preparing to perform steady-state nonlinear regression correction for ambient temperature. However, in this example, during the special period of thermal shock transient compensation mode, the system adds a linear compensation term to the end of the original accuracy correction formula. That is, the thermal radiation transient compensation amount calculated here is directly added to the final output value through addition.

[0164] If the sensor generates a false temperature rise signal due to cold or thermal shock, the calculated negative compensation value can just offset it; thus ensuring that the pet's body surface temperature value output by the device remains stable throughout the dynamic evolution of rapid temperature changes, shielding against false fluctuations caused by external thermal shock.

[0165] Optionally, any dynamic compensation mechanism must have reasonable exit logic to prevent the introduction of continuous steady-state errors after the system re-establishes thermal equilibrium.

[0166] Therefore, the system continuously monitors the changes in the external environment. When the absolute value of the rate of change of the ambient temperature falls below the thermal shock change threshold and the duration reaches the preset thermal equilibrium stabilization time, the system exits the thermal shock transient compensation mode.

[0167] Specifically, when a pet completes a transfer across temperature zones and is in a new constant-temperature environment, such as after walking for 10 minutes in a cold outdoor environment, the ambient temperature no longer changes drastically, and the absolute value of its rate of change approaches 0 and is below the set thermal shock change threshold.

[0168] However, the compensation calculation cannot be stopped immediately because, according to thermodynamic principles, although the external ambient temperature is stable, the temperature inside the sensor is still slowly approaching the external temperature due to the thermal inertia of the encapsulation shell, and the gradient difference still exists.

[0169] Therefore, the system introduces a preset thermal equilibrium stabilization time as a time-dimensional hysteresis protection mechanism. Only when the ambient temperature change rate remains stable for a period exceeding this stabilization time can the system be physically certain that a complete isothermal thermal equilibrium has been re-established inside and outside the sensor. Once this composite exit condition is met, the microprocessor will stop calling the iterative calculation of the first-order thermal equivalent model, forcibly clear the transient thermal radiation compensation to zero, and seamlessly switch the system state machine back to the conventional steady-state accuracy correction mode.

[0170] like Figure 6The thermodynamic pseudo-color map of transient drift compensation as a function of ambient temperature gradient is shown. The horizontal axis represents the rate of change of ambient temperature in degrees Celsius per second, the vertical axis represents the transient temperature gradient difference in degrees Celsius, and the color legend on the right represents the calculated transient drift compensation in degrees Celsius.

[0171] Figure 6 The pseudo-color gradient from deep blue to deep red objectively reflects the dynamic output of transient thermal radiation compensation when the wearable system faces thermal shocks of different intensities across temperature zones.

[0172] When both the rate of change of ambient temperature and the difference in transient temperature gradient are in the central light green area near zero, it indicates that the equipment is in a constant temperature thermal equilibrium state, and the calculated transient drift compensation amount is close to 0 degrees Celsius.

[0173] When a pet is quickly moved from a warm indoor environment to a cold outdoor environment, as both the horizontal and vertical axes extend into the negative range, the color in the graph gradually transitions smoothly to the dark blue area. This demonstrates that the system accurately calculates and applies negative temperature compensation based on the gradient difference between the higher internal hysteresis temperature and the lower external temperature, effectively offsetting the positive temperature measurement drift error caused by the imbalance between the hot and cold ends of the sensor.

[0174] However, in the extreme regions near the upper and lower edges of the vertical axis, close to -20 degrees Celsius and +20 degrees Celsius, irregular mottled noise and color block distortion can be observed in the originally smooth color gradient. This color dispersion under high gradients is actually an anti-oversmoothing mechanism deliberately set by the underlying algorithm to realistically preserve the physical characteristics of abrupt changes in thermal radiation under extreme temperature differences. While ensuring absolute accuracy in compensation for the core physiological temperature zone, the system allows for a certain degree of numerical fluctuation in the extreme edge regions, thus providing original data for subsequent investigation of potential hardware-level anomalies. This reflects the optimal engineering trade-off between compensation accuracy and preservation of original features.

[0175] In general, most existing technologies for handling ambient temperature compensation rely on static lookup table methods or steady-state equations. These solutions are only effective when the ambient temperature changes extremely slowly. Once the device faces typical pet living scenarios such as switching between indoor and outdoor environments, transient thermal shocks can cause serious deviations in the compensation algorithms of existing technologies, resulting in temperature jumps of several degrees Celsius lasting for tens of minutes, leading to a large number of false fever alarms received by users.

[0176] The thermal shock transient compensation scheme disclosed in this embodiment uses the motion state provided by the inertial unit as a confidence level discrimination switch for spatial transfer, and simultaneously performs real-time calculations of the rigorous first-order differential equation of hardware thermodynamics in the embedded microcontroller. This scheme not only avoids modifying the original optical structure and hardware circuitry, thus avoiding increased hardware costs, but also achieves the reverse reconstruction and error cancellation of the physical-level heat conduction process through pure digital algorithms.

[0177] This overall solution effectively eliminates the dynamic interference of parasitic thermoelectric potential on infrared detection, enabling the wearable system to still output smooth, objective, and accurate animal body temperature data when dealing with extreme hot and cold environments, greatly expanding the all-weather application range of the smart collar.

[0178] Example 3:

[0179] In real-world pet ownership scenarios, a significant portion of a pet's life cycle (especially felines and long-haired dogs) is spent indoors resting or in deep sleep. When a pet enters a prolonged lying or curled-up sleeping position, the smart health collar worn around its neck will fit tightly against the pet's neck skin due to its own weight and the pressure of the pet's body.

[0180] During this stage, the inertial measurement module detects extremely low levels of motion. Following the basic logic of Example 1, the system accurately assesses the current state as "excellent fit." However, this high degree of fit, coupled with prolonged periods without relative physical movement, creates a closed microenvironment lacking air convection between the temperature measurement window inside the collar and the pet's neck skin. The physiological heat radiation continuously emitted by the pet's skin cannot be carried away by airflow, causing heat to accumulate in the temperature measurement window area, thus inducing a localized "heat island effect."

[0181] If relying solely on conventional steady-state temperature measurement models, the heat accumulation caused by external physical shielding can be misinterpreted by the system as a genuine increase in the pet's body surface temperature, leading to frequent false fever alarms without medical basis at night.

[0182] To address the aforementioned technical problems, this embodiment adds an independent nonlinear thermodynamic compensation mechanism to the temperature measurement and filtering framework of Embodiment 1. The specific execution steps are as follows:

[0183] The method described in this embodiment also includes a heat island effect temperature compensation step for long-term resting state, specifically including: when the fit level is determined to be excellent and the attitude stability feature is continuously lower than the preset resting stability threshold within the first preset observation time, extracting the mean vector of the triaxial acceleration in the three coordinate axis directions within the first preset observation time.

[0184] In the resource scheduling logic of the underlying embedded system, the microcontroller allocates a dedicated status monitoring buffer in static random access memory. When the main control program confirms that the current fit level is at the optimal level, it immediately activates the monitoring buffer and starts a long-cycle hardware timer to count the observation time.

[0185] The first preset observation period is typically set to a relatively long time span (e.g., 20 to 30 minutes) to filter out brief periods of rest or turning over during light sleep. During this extended observation period, the system examines newly generated posture stability features clockwise. Only when every data point is strictly below the resting stability threshold does the system determine that the pet may have entered deep sleep. Subsequently, the algorithm performs integral averaging on all raw triaxial acceleration sampling points within that period to extract a high-precision mean vector representing the current three-dimensional spatial orientation of the device.

[0186] It is important to note that in the theory of inertial navigation and sensor attitude calculation, when an object is in a state of absolute stillness or extremely low-frequency micro-motion, the multi-axis components sensed by its onboard accelerometer are actually the projection of the Earth's gravitational acceleration vector onto the three orthogonal coordinate axes of the sensor.

[0187] Therefore, the mean vector extracted by the system physically represents the exact three-dimensional tilt angle of the current collar device relative to gravity. Unlike simple motion intensity calculations, the mean vector retains directional attributes, which enables the system to perceive the pet's current absolute physical spatial posture (such as lying flat, on its side, or curled up), providing crucial spatial topological input for subsequent thermodynamic modeling.

[0188] Optionally, after extracting the high-precision mean vector, the system enters the attitude consistency verification stage.

[0189] The system calculates the three-dimensional angle between the transient acceleration vector corresponding to each sampling point and the mean vector. When the variance of the three-dimensional angle is less than a preset variance threshold, the system determines that the target has entered a long-term resting prone posture.

[0190] In practical applications, even when a pet is asleep, its abdomen will rise and fall periodically with its breathing, occasionally accompanied by slight muscle twitches. These minute physiological movements do not disrupt the overall posture of deep sleep, but they will add weak high-frequency noise to the mean vector.

[0191] In order to accurately locate the pet's macroscopic sleeping posture while filtering out these physiological noises, the system uses a variance statistical model to quantitatively evaluate the dispersion of the three-dimensional angle.

[0192] For example, the system uses the following discrete statistics formula to calculate the variance of the three-dimensional included angle:

[0193] ;

[0194] In the formula, Defined as the variance of the calculated output three-dimensional included angle; Defined as the total number of sample points acquired for observation within the first preset observation period; Defined as the first The three-dimensional angle between the transient acceleration vector corresponding to each observation sample point and the mean vector; Defined as all The arithmetic mean of the three-dimensional angles corresponding to each sample point.

[0195] By calculating this variance, the system can objectively measure the deviation and dispersion of a pet's neck posture. When the calculated variance... When the value is less than the preset variance threshold calibrated through extensive clinical testing, it indicates that the pet's neck has maintained a nearly rigid, fixed angle throughout the entire long observation period, without any substantial rolling or posture change. At this point, the system sets the corresponding status flag register in the software state machine, officially determining that the target has entered a prolonged resting, lying posture.

[0196] like Figure 7 A scatter plot of the three-dimensional spatial distribution of transient acceleration vectors under long-term resting conditions is shown, where the three coordinate axes represent the first axial transient acceleration, the second axial transient acceleration, and the third axial transient acceleration, respectively, with the unit being meters per second squared.

[0197] Figure 7 It contains two different colored three-dimensional scatter clusters, where blue scatter points represent discrete sampling points when the pet is active, and red scatter points represent long-term resting cluster sampling points when the pet is within a first preset observation period.

[0198] It can be observed that the blue scatter points exhibit an extremely divergent and irregular distribution in three-dimensional space, with acceleration values ​​fluctuating wildly between -20 and +20; while the red scatter points are highly densely clustered in a specific region of coordinate space, forming a concentrated lattice cluster, with its value on the third axis stabilizing at around 9.8. This objectively reflects the Earth's gravitational acceleration component sensed by the device after the pet enters deep sleep.

[0199] The extremely small distribution dispersion of the red scatter points in this three-dimensional space can intuitively verify the effectiveness of the system's technical solution of using the three-dimensional variance of the angle between the transient acceleration vector and the mean vector corresponding to each sampling point to determine whether a pet has entered a long-term resting lying posture.

[0200] However, in the magnified area of ​​the red cluster lattice, it can be found that the lattice is not a standard three-dimensional equidistant sphere distribution, but rather exhibits a distinct ellipsoidal stretching shape along the second axis. This is mainly due to the periodic breathing and abdominal undulations of the pet during long sleep, which introduces low-frequency physiological vibration noise in a fixed direction into the system.

[0201] This stretched distribution pattern indicates that the current approach, relying solely on the global variance of a single three-dimensional angle for posture determination, not only accurately pinpoints the core posture during prolonged resting periods but also cleverly utilizes the natural tolerance of the variance envelope to automatically accommodate low-frequency physiological fluctuations caused by the breathing of large pets. This design avoids frequent false exits from prolonged resting states due to overly stringent posture determination, ensuring the continuity and stability of the heat island compensation mechanism throughout the entire deep sleep cycle.

[0202] Specifically, once it is confirmed that the pet has entered a special static posture that is highly prone to causing local heat accumulation, the system immediately activates the compensation mechanism.

[0203] If the patient is determined to have entered the prolonged resting prone posture, a heat accumulation timer is activated to record the dormancy duration. Based on a preset heat saturation time constant and a maximum heat accumulation temperature rise limit, a heat accumulation model that grows non-linearly with the dormancy duration is constructed to calculate the local heat island temperature rise compensation. This step is the core of the calculation for addressing complex microclimate changes in this embodiment.

[0204] When the status flag register is set, the low-power real-time clock (RTC) module inside the microcontroller is configured to second-level interrupt mode and starts working as an independent thermal accumulation timer. Its count value represents the duration of the pet's continuous sleep in a closed and warm state.

[0205] It is also important to note that the artificially high temperature caused by the local heat island effect does not increase linearly and infinitely over time. According to Newton's law of cooling and classical thermodynamics, in a closed system with a fixed heat source (pet's body temperature) and a fixed thermal resistance (collar material), the accumulated temperature will initially rise rapidly, then the rate of increase will gradually slow down, eventually approaching a thermal equilibrium extreme value.

[0206] Therefore, the system incorporates two key engineering calibration parameters. The first is the preset heat saturation time constant, which depends on the specific heat capacity and geometric thickness of the polycarbonate material in the collar's bottom shell, representing a time scale for the time required for heat to accumulate to saturation in the system. The second is the maximum heat accumulation temperature rise limit, which depends on the thermal insulation properties of the fur of a specific pet species, representing the highest spurious temperature rise that can be achieved in this enclosed microenvironment.

[0207] For example, the system utilizes the floating-point unit (FPU) of the microcontroller to construct a thermal accumulation model and calculate the compensation amount based on the following nonlinear exponential function equation:

[0208] ;

[0209] In the formula, Defined as the local heat island temperature rise compensation amount calculated currently; Defined as the maximum heat accumulation temperature rise limit pre-programmed into the controller flash memory; Defined as the base (constant) of the natural logarithm; Defined as the sleep duration currently output by the thermal accumulation timer; Defined as a preset thermal saturation time constant.

[0210] Using this nonlinear growth model, the microprocessor can accurately simulate the real-time temperature rise curve of the thin layer of air within the sensor's measurement area. In the initial stage of dormancy, due to... When the initial value is small, the exponential decay term is close to 1, and the overall compensation increases rapidly; however, as the pet's sleep time increases, after tens or even hundreds of minutes, the exponential term approaches 0, and the system's calculated... Approaching the setting infinitely. This closely matches the physical reality of thermodynamic equilibrium in a confined space.

[0211] Optionally, after obtaining the accurately fitted local heat island temperature rise compensation amount, the system must integrate it into the main link's data stream.

[0212] In the step of performing accuracy correction, the local heat island temperature rise compensation amount is subtracted from the effective pet body surface temperature sample. Specifically, in terms of software architecture implementation, after the main control program calls the environmental temperature nonlinear compensation model described in Embodiment 1 to output the noise reduction sequence, it adds an arithmetic logic subtraction operation node on the memory bus.

[0213] This node directly reads the real-time calculations from the thermal accumulation model. The variable is forcibly subtracted from the currently prepared valid pet body surface temperature sample. This reverse numerical cancellation operation, based on physical mechanisms, allows the system to provide the backend cloud server with a smooth and accurate baseline of body surface temperature that reflects the pet's core physiological metabolic level, even under harsh conditions such as when the pet is asleep and the local environment is severely overheated. This greatly reduces the probability of medical misdiagnosis caused by prolonged wear.

[0214] Specifically, as a complete closed-loop dynamic control system, the compensation mechanism is also designed with physical exit and residual heat dissipation paths.

[0215] When the variance of the three-dimensional angle is detected to be greater than the preset variance threshold, or the posture stability feature exceeds the resting stability threshold, it is determined that the long-term resting prone posture has been broken. The heat accumulation timer is stopped, and the local heat island temperature rise compensation amount is reduced to zero over time according to the preset physical heat dissipation attenuation rate, before exiting the heat island effect temperature compensation step. During the pet's sleep of several hours, the system's main loop continuously calculates the variance of the three-dimensional angle and the posture stability feature in parallel in the background.

[0216] It's also important to note that the exit condition here is a logical "OR" relationship, which covers two different pet awakening physical scenarios:

[0217] The first scenario is that the variance of the three-dimensional angle is greater than the preset variance threshold. This means that although the pet's movement is extremely slight and slow, its overall sleeping posture has changed (for example, from curled up and lying down to lying on its back). At this time, the temperature measuring window of the collar, which was originally compressed and sealed, may be exposed to the air again.

[0218] The second scenario is that the posture stability characteristics exceed the resting stability threshold, which means that the pet may suddenly wake up and be accompanied by a large amount of shaking or getting up. The high-frequency angular velocity signal instantly breaks through the resting threshold defense line.

[0219] Whether it's a slow turning over or a sudden awakening, as long as either of the above conditions is met, the system will determine at the software level that the sealed microenvironment that caused the artificially high temperature has been broken.

[0220] like Figure 8 The nonlinear exponential saturation curve of the artificial temperature rise of the local heat island in a closed microenvironment is shown. The horizontal axis represents the continuous observation time in minutes, and the vertical axis represents the artificial temperature rise in degrees Celsius.

[0221] Figure 8 It contains two curves, where the gray curve represents the actual local heat island temperature rise, and the red curve represents the compensation amount calculated by the system model.

[0222] During the long resting and lying phase from 0 to 120 minutes, the actual local heat island temperature rise curve in the gray area shows a nonlinear growth trend of rapid initial rise followed by a gradual slowdown in the rate of rise, and gradually approaches the maximum heat accumulation temperature rise limit of 1.0 degrees Celsius around 120 minutes.

[0223] The compensation curve calculated by the red system model smoothly follows this exponential saturation physical process, indicating that the local heat island temperature rise compensation calculated by the system using the nonlinear growth model can objectively restore and offset the positive temperature drift caused by the collar-shaded microenvironment.

[0224] When the pet's posture is broken after 120 minutes, the heat accumulation timer stops, and the compensation curve calculated by the red system model rapidly decreases towards zero degrees Celsius according to the preset physical heat dissipation decay rate.

[0225] However, comparing the data from 150 to 180 minutes, it can be observed that the compensation curve calculated by the red system model has smoothly dropped to 0 degrees Celsius, but the actual local heat island temperature rise curve in the gray area still remains in the positive range of about 0.1 degrees Celsius, showing a long physical heat dissipation tail phenomenon.

[0226] This actual cooling lag indicates that the dense fur of dogs and cats inherently possesses a delayed release physical characteristic due to deep heat storage. The current system's single decay rate mechanism achieves a reasonable balance between ensuring algorithm real-time performance, conserving embedded computing power, and avoiding overcompensation. This heat island effect temperature compensation step, by accurately intercepting artificially high data during the core temperature rise phase, has essentially achieved effective offsetting of heat accumulation in the sleep microenvironment, sufficient to provide highly reliable baseline support for subsequent health warnings.

[0227] Optionally, upon confirming that the posture has been broken, the system immediately sends a command to the low-power real-time clock to stop the heat accumulation timer from accumulating. However, according to classical thermodynamic cooling laws, the heat accumulated inside the collar's plastic shell and in parts of the pet's fur will not disappear instantly within a fraction of a second of the pet getting up. The convective dissipation of heat to the surrounding cool air is also a time-consuming, non-linear decay process.

[0228] If the local heat island temperature rise compensation is directly reset to zero without buffering at this time, it will cause a large gradient step change in the final output temperature curve of the system. This not only violates the common sense of physics, but also seriously interferes with the subsequent health data analysis.

[0229] Therefore, the system retrieves the preset physical heat dissipation attenuation rate, causing the compensation variable originally residing in memory to enter a decreasing loop. In each temperature measurement cycle, this variable is multiplied by an attenuation coefficient less than 1 to simulate the cooling curve of convection heat dissipation. Only when the compensation variable numerically decays close to absolute zero does the system reset the relevant flag register in the state machine, officially exiting the heat island effect temperature compensation step and restoring to the basic temperature measurement filtering mode defined in Example 1.

[0230] In summary, most current wearable pet temperature monitoring technologies focus solely on filtering out discrete, transient noise caused by movement, neglecting the slow temperature rise and drift caused by physical contact and insulation effects during prolonged resting periods. When existing devices are used in conjunction with pets sleeping for up to eight hours overnight, the false temperature rise exceeding one degree Celsius in the enclosed microenvironment can easily trigger false alarms from conventional backend algorithms, thereby reducing the reference value of the device's monitoring data.

[0231] The solution described in this embodiment cleverly utilizes inexpensive microelectromechanical inertial sensors, not only as a tool for judging motion intensity but also as a detector for sensing three-dimensional absolute spatial attitude. By deeply integrating long-term attitude monitoring with nonlinear thermal accumulation equations at the pure software level, the system achieves high-precision simulation and inverse cancellation of complex local microclimates without incurring high hardware costs such as additional micro environmental probes.

[0232] The overall solution is logically consistent and has sufficient physical basis, which significantly enhances the robustness of the infrared temperature measurement filtering method in extreme resting environments and ensures the objectivity of the equipment's all-weather body temperature monitoring data and its clinical auxiliary diagnostic value.

[0233] Example 4:

[0234] In practical scenarios of home pet ownership or animal medical monitoring, the following embodiment addresses the technical problem mentioned in the background art: pets' daily activities (such as frequent running, jumping, splashing water, or scratching) can cause varying degrees of displacement, flipping, or loosening of the collar, leading to the infrared sensor deviating from the effective temperature measurement area and introducing interference from ambient cold air or fur. This embodiment provides a system device combining specific physical hardware and underlying control software. This system can physically solidify and engineer the data processing methods described in the above embodiment. Through the design of the underlying hardware architecture and the scheduling of microcontroller tasks, it effectively distinguishes between genuine body temperature changes and invalid detection data caused by device displacement.

[0235] like Figure 9 As shown in the figure, this embodiment discloses an infrared temperature measurement wearable system for pet liveness detection, including: an infrared temperature acquisition module for acquiring the original temperature value of the infrared object and the ambient temperature value; an inertial measurement module for acquiring triaxial acceleration and triaxial angular velocity data; and a main control processing module, which is communicatively connected to the infrared temperature acquisition module and the inertial measurement module respectively.

[0236] Specifically, the infrared temperature acquisition module typically employs a digital infrared thermopile sensor assembly packaged with high-precision medical-grade features. This module not only includes a thermopile array element sensitive to specific wavelengths of thermal radiation, but also integrates a high-precision NTC (negative temperature coefficient) thermistor on the same silicon chip for synchronously outputting the ambient temperature value of the sensor's internal cold junction.

[0237] To ensure the effectiveness of optical detection, the module has a physical temperature measurement window with an infrared transmission filter on the side of the wearable system facing the pet's neck, and stray light is isolated by a flexible silicone light-shielding sleeve. The inertial measurement module uses a six-axis microelectromechanical system (MEMS) chip, and its physical patch position on the printed circuit board (PCB) is designed to be rigidly coupled to the infrared temperature acquisition module, with the relative physical distance between the two being less than a preset tolerance (e.g., within 5 mm).

[0238] This tight, rigid coaxial layout ensures that the triaxial acceleration and triaxial angular velocity data collected by the inertial measurement module can accurately and without delay reflect the real-time motion intensity and three-dimensional attitude tilt angle of the infrared temperature measurement window on the pet's neck in physical space.

[0239] It is important to note that the main control processing module is the central hub for data throughput and algorithm scheduling of the entire wearable system, and typically employs a low-power 32-bit microcontroller (MCU) based on the ARM Cortex-M series architecture. This microcontroller integrates a floating-point unit (FPU), a direct memory access controller (DMA), and multi-channel timers. The main control processing module establishes high-speed data communication connections with the infrared temperature acquisition module and the inertial measurement module at the physical layer via an I2C or SPI bus.

[0240] Furthermore, the main control processing module is configured to perform the following key steps: acquiring multi-dimensional monitoring data of the pet's neck with a preset temperature measurement window as the cycle. The multi-dimensional monitoring data includes the original temperature value of the infrared object, the ambient temperature value, and the triaxial acceleration and triaxial angular velocity data of the object, which are synchronously collected through a unified time base and contain multiple sampling points.

[0241] In actual operation, in order to strictly ensure the above-mentioned "unified time base", the main control processing module uses an internal general-purpose timer to generate a hardware trigger signal with a fixed frequency.

[0242] The trigger signal synchronously drives the I2C bus controller, which, through the DMA channel, transfers the raw register data of the infrared sensor and the inertial sensor in parallel to the pre-allocated sliding data buffer in the static random access memory (SRAM) without occupying the computing power of the main control CPU.

[0243] This underlying hardware synchronization mechanism eliminates the time phase error caused by asynchronous sampling of multiple sensors, so that each temperature sampling point contained in the preset temperature measurement window (such as a data segment with a length of 1 second or 2 seconds) has a unique and tightly aligned physical vector of acceleration and angular velocity on the time axis.

[0244] Optionally, after completing hardware-level data synchronization acquisition, the main control processing module is configured to continue executing the following steps using its floating-point arithmetic unit: based on the original values ​​of the triaxial acceleration, triaxial angular velocity, and infrared object temperature, extract motion intensity features, attitude stability features, and temperature fluctuation features; and calculate a comprehensive fit deviation score based on the motion intensity features, attitude stability features, and temperature fluctuation features to classify the fit level. After reading the data in the SRAM buffer, the main control processing module runs the corresponding feature extraction statistical algorithm to quantize and decode the intensity of physical motion and the stability of the infrared signal. The calculated fit level is stored in the microcontroller's global variable register as input for subsequent dynamic adjustment of hardware behavior.

[0245] Specifically, the main control processing module is configured to dynamically adjust the temperature measurement parameters based on the current fit level. These parameters include candidate interval thresholds, calibration offsets, target state machine count thresholds, and the number of valid value hold times. A multi-level parameter configuration matrix corresponding to the actual usage scenario is pre-programmed into the microcontroller's internal flash memory. The main control program dynamically switches to the corresponding parameter configuration area via an address pointer based on the fit level in the global variable register, thereby modifying the boundary conditions of the temperature measurement algorithm running in main memory in real time. This allows the wearable system to quickly tighten the data access threshold at the physical level when it detects that the device has become loose due to scratching by a pet, and to restore the normal configuration when the pet is lying quietly.

[0246] For example, the main control processing module is then configured to: use the dynamically adjusted temperature measurement parameters to sequentially perform physical validity filtering, interval filtering based on candidate interval thresholds, calibration correction based on calibration offsets, and target state machine jump control based on target state machine count thresholds on the raw temperature values ​​of the infrared object within the current temperature measurement window, thereby identifying valid pet body surface temperature samples.

[0247] This step is essentially controlled by a finite state machine subroutine running inside the main control processing module. Based on the modified threshold parameters, the state machine program performs point-by-point logical judgments on the timing input data stream, discarding or blocking interfering data that deviates significantly from the physiological threshold due to poor contact at the register level, and finally selecting valid body temperature sample points with physiological reference significance.

[0248] Finally, the main control processing module is configured to: perform accuracy correction on the valid pet surface temperature samples combined with the ambient temperature values, and output the pet surface temperature value and the corresponding fit level label. The main control processing module calls a nonlinear regression algorithm library to fit the selected valid samples with the ambient temperature values ​​obtained from the internal cold end, compensating for ambient thermal radiation drift. The finally generated pet surface temperature value and fit level label will be encapsulated in a standard data frame format and sent via serial port (UART) to the wearable system's wireless communication module (such as Bluetooth BLE or NB-IoT module), and then transmitted to the user terminal or cloud server.

[0249] Specifically, in practical use, the system, through the collaboration between the main control processing module and the underlying hardware of the dual-modal sensor, constructs a complete closed-loop device from synchronous signal acquisition and hardware status assessment to dynamic parameter feedback control. When the pet's vigorous activity causes the device to loosen or shift, the system can adaptively shrink the temperature measurement filter standard within a microsecond response time, relying on hardware computing power to intercept physical interference data; while ensuring continuous temperature output when the fit is stable. This deep integration of hardware and control logic significantly reduces the superimposed impact of physical movement displacement and external environmental fluctuations on the infrared temperature measurement components, ensuring that the system's final output of body surface temperature can more objectively and accurately reflect the pet's actual physiological state, thus improving the monitoring usability of the device in complex activity scenarios.

[0250] Example 5:

[0251] Corresponding to the above embodiments, the present invention also proposes an electronic device.

[0252] like Figure 10 The diagram shows a structural schematic of an electronic device according to the present invention. The electronic device 100 includes a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, for example, via a bus 102. Optionally, the electronic device 100 may further include a transceiver 104. It should be noted that in practical applications, the transceiver 104 is not limited to one unit, and the structure of this electronic device 100 does not constitute a limitation on the embodiments of the present invention.

[0253] Processor 101 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 101 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0254] Bus 102 may include a pathway for transmitting information between the aforementioned components. Bus 102 may be a PCI bus or an EISA bus, etc. Bus 102 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 10 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0255] The memory 103 stores a computer program corresponding to the infrared thermometry filtering method for pet liveness detection according to the above embodiments of the present invention. This computer program is controlled and executed by the processor 101. The processor 101 executes the computer program stored in the memory 103 to implement the content shown in the aforementioned method embodiments.

[0256] Among them, electronic devices 100 include, but are not limited to: mobile terminals such as laptops and PADs (tablet computers) and fixed terminals such as desktop computers. Figure 10 The electronic device 100 shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0257] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. An infrared thermometry filtering method for pet liveness detection, characterized in that, include: With a preset temperature measurement window as the cycle, multi-dimensional monitoring data of the pet's neck is acquired. The multi-dimensional monitoring data includes the original infrared object temperature value, ambient temperature value, and triaxial acceleration and triaxial angular velocity data, which are collected synchronously with a unified time base and contain multiple sampling points. Based on the original values ​​of triaxial acceleration, triaxial angular velocity, and infrared object temperature, motion intensity features, posture stability features, and temperature fluctuation features are extracted. Based on the motion intensity features, posture stability features, and temperature fluctuation features, a comprehensive deviation score of fit is calculated, and fit levels are classified. Based on the current fit level, the temperature measurement parameters are dynamically adjusted, including candidate interval threshold, calibration offset, target state machine count threshold, and number of valid value holding times. Using the dynamically adjusted temperature measurement parameters, the original temperature values ​​of the infrared objects within the current temperature measurement window are sequentially subjected to physical validity filtering, interval filtering based on candidate interval thresholds, calibration correction based on calibration offset, and target state machine jump control based on target state machine count thresholds to identify valid pet body surface temperature samples. The effective pet body surface temperature sample is combined with the ambient temperature value to perform accuracy correction, and the pet body surface temperature value and the corresponding fit level label are output.

2. The method according to claim 1, characterized in that, The extraction of motion intensity features, attitude stability features, and temperature fluctuation features based on the triaxial acceleration, triaxial angular velocity, and raw infrared object temperature includes: Calculate the standard deviation of the sum of triaxial acceleration values ​​at all sampling points within a single temperature measurement window, and use it as the motion intensity feature; The mean value of the triaxial angular velocity summation of all sampling points within a single temperature measurement window is calculated and used as the attitude stability feature. Calculate the variance of the original infrared object temperature values ​​within multiple consecutive temperature measurement windows, and use it as the temperature fluctuation characteristic; After normalizing the extreme values ​​of the motion intensity characteristics, posture stability characteristics, and temperature fluctuation characteristics, the results are weighted and summed using preset weighting coefficients to obtain the comprehensive deviation score of the fit. The overall deviation score of the fit is compared with a preset first grading threshold and a second grading threshold, wherein the first grading threshold is less than the second grading threshold, and the fit level is output as excellent, good, or poor.

3. The method according to claim 1, characterized in that, The target state machine counting threshold includes a target entry counting threshold and a target loss counting threshold; the dynamic adjustment of the temperature measurement parameters based on the current fit level specifically involves: Establish a mapping relationship between the fit level and the temperature measurement parameters, query the mapping relationship based on the currently acquired fit level, and extract the corresponding temperature measurement parameters; The fit level is divided into excellent fit, good fit, and poor fit. As the fit level transitions from excellent to poor, the range of the candidate interval threshold in the mapping relationship decreases, the corresponding calibration offset decreases, the corresponding target entry count threshold increases, the corresponding target loss count threshold decreases, and the corresponding number of valid value retentions increases.

4. The method according to claim 3, characterized in that, The candidate interval threshold includes a strict candidate interval for target entry determination and a soft candidate interval for target retention determination; the specific steps of dynamically adjusting the candidate interval threshold in the temperature measurement parameters are as follows: Obtain the preset lower and upper limits of the baseline, and map the excellent fit, good fit, and poor fit to the first, second, and third level codes with decreasing numerical values, respectively; The interval parameter corresponding to the first level code is used as the benchmark; when the current fit level is not excellent, the level difference between the level code corresponding to the current fit level and the first level code is calculated. The current upper limit after dynamic adjustment is obtained by subtracting the product of the preset adjustment step size and the level difference from the benchmark upper limit. The lower limit of the baseline is then added to the product of the preset adjustment step size and the level difference, which is used as the dynamically adjusted current lower limit to generate the strict candidate interval and soft candidate interval of the current temperature measurement window.

5. The method according to claim 1, characterized in that, The process of sequentially performing physical validity filtering, interval filtering based on candidate interval thresholds, calibration correction based on calibration offset, and target state machine jump control based on target state machine count thresholds on the raw temperature values ​​of the infrared objects within the current temperature measurement window, to identify valid pet body surface temperature samples, includes: Construct a target state machine that includes the target loss state, the target entry state, and the target hold state; In the target loss state, if the original value of the infrared object temperature in the current temperature measurement window passes through the strict candidate interval filter in the candidate interval threshold, and the number of times it passes through the strict candidate interval filter consecutively reaches the target entry count threshold in the target state machine count threshold, then the system state is switched to the target entry state. In the target entry state, if the raw infrared object temperature value of the subsequent temperature measurement window continues to pass through the soft candidate interval in the candidate interval threshold, the system state will be switched and maintained in the target holding state, and the raw infrared object temperature value that has been calibrated and corrected in the target holding state will be marked as a valid pet body surface temperature sample. In the target holding state, if the number of consecutive failures to pass the soft candidate interval filter reaches the target loss count threshold in the target state machine counting threshold, then switch back to the target loss state.

6. The method according to claim 5, characterized in that, The method also includes a state machine-based effective value preservation mechanism: When in the target holding state, if the original value of the infrared object temperature in the current temperature measurement window does not pass the soft candidate interval filtering, the valid value holding operation is triggered, and the valid pet body surface temperature sample recorded in the previous temperature measurement window is used as the output sample of the current temperature measurement window. The number of consecutive failures that fail to pass the soft candidate interval filter is counted. If the number of failures does not exceed the number of valid value retentions, the target retention state is maintained. If the number of failures reaches the number of valid value retentions, the valid value retention operation is terminated and the target loss state is forcibly switched.

7. The method according to claim 1, characterized in that, The process of performing accuracy correction on the valid pet body surface temperature sample in conjunction with the ambient temperature value, and outputting the pet body surface temperature value, includes: Obtain the posterior temperature estimate output from the previous temperature measurement window, and generate the prior estimate of the current temperature measurement window based on the posterior temperature estimate and the preset process noise variance. The residuals and update weights are calculated by combining the effective pet body surface temperature samples in the current temperature measurement window, and the noise-reduced initial screening temperature sequence is output through the prediction-correction filtering model. The synchronously collected ambient temperature values ​​are used as features input into a pre-established nonlinear regression model to calculate the ambient thermal radiation drift compensation amount. The noise-reduced initial screening temperature sequence is superimposed and corrected with the environmental thermal radiation drift compensation amount to generate and output the pet's body surface temperature value.

8. The method according to claim 1, characterized in that, The method further includes a hierarchical power scheduling step based on fit level, applicable to systems containing multimodal sensors, specifically including: In normal monitoring mode, the steps of periodically collecting the multi-dimensional monitoring data and infrared temperature measurement filtering are performed. When the pet's body surface temperature value output by the first preset number of temperature measurement windows deviates from the preset normal temperature range, and the current fitting level is excellent, a first wake-up command is generated to wake up the multimodal sensor to enter the abnormal enhancement mode to collect multidimensional vital sign data. When the pet's body surface temperature value output by the first preset number of temperature measurement windows deviates from the preset normal temperature range, and the corresponding fit level is poor fit, the sampling frequency of the temperature measurement window is increased. After the fit level recovers to good fit and the temperature is confirmed to deviate from the preset normal temperature range again, the first wake-up command is generated.

9. The method according to claim 1, characterized in that, The method also includes a weak network fault tolerance and quality-graded upload step: Based on the fit level and whether the current pet's body surface temperature is within the normal temperature range, three levels of upload priority are assigned to the pet's body surface temperature value. The highest priority is given to the temperature that is not within the normal temperature range and has excellent fit, the second highest priority is given to the temperature that is within the normal temperature range and has excellent fit, and the normal priority is given to the data with poor fit. The data, after being assigned a priority, is written into a circular buffer queue with a preset fixed length and uploaded in sequence. When the circular buffer queue is determined to be full and new data to be written is received, the queue is traversed and the earliest written ordinary priority data in the queue is overwritten first, while the highest priority data and the data at the head of the queue that is waiting for confirmation from the receiving end are retained.

10. A wearable infrared temperature measurement system for pet liveness detection, characterized in that, include: The infrared temperature acquisition module is used to acquire the raw temperature value of the infrared object and the ambient temperature value; Inertial measurement module, used to acquire triaxial acceleration and triaxial angular velocity data; The main control processing module is communicatively connected to both the infrared temperature acquisition module and the inertial measurement module, and is configured to perform the following steps: With a preset temperature measurement window as the cycle, multi-dimensional monitoring data of the pet's neck is acquired. The multi-dimensional monitoring data includes the original temperature value of the infrared object, the ambient temperature value, and the triaxial acceleration and triaxial angular velocity data, which are collected synchronously with a unified time base and contain multiple sampling points. Based on the original values ​​of triaxial acceleration, triaxial angular velocity, and infrared object temperature, motion intensity features, posture stability features, and temperature fluctuation features are extracted. Based on the motion intensity features, posture stability features, and temperature fluctuation features, a comprehensive deviation score of fit is calculated, and fit levels are classified. Based on the current fit level, the temperature measurement parameters are dynamically adjusted, including candidate interval threshold, calibration offset, target state machine count threshold, and number of valid value holding times. Using the dynamically adjusted temperature measurement parameters, the original temperature values ​​of the infrared objects within the current temperature measurement window are sequentially subjected to physical validity filtering, interval filtering based on candidate interval thresholds, calibration correction based on calibration offset, and target state machine jump control based on target state machine count thresholds to identify valid pet body surface temperature samples. The effective pet body surface temperature sample is combined with the ambient temperature value to perform accuracy correction, and the pet body surface temperature value and the corresponding fit level label are output.

Images