Wearable device-based animal physiological signal telemetry method and system

Abstract

This invention relates to the field of signal sensing technology, and more particularly to a method and system for telemetry of animal physiological signals based on wearable devices. The method includes: acquiring photoelectric signals and inertial data from multiple acquisition nodes; using an adaptive filtering algorithm to obtain initially separated physiological fluctuations and artifact components; extracting motion pattern features of corresponding nodes, determining their dynamic correlation parameters with changes in the optical path, and when the dynamic correlation parameters exceed preset values, correcting the mapping model using a Kalman filter algorithm to obtain optimized artifact reference data; fusing multi-node information to determine the similarity of motion patterns between nodes; obtaining a shared artifact elimination template, processing the remaining mixed signals, and obtaining synchronously restored true physiological signals from multiple sites. This invention solves the problem that multi-node photoelectric physiological signals are easily interfered with by complex motion artifacts and are difficult to coordinate in a coordinated manner during free animal activity, and achieves high-precision synchronous telemetry of true physiological signals from multiple sites.

Description

Technical Field

[0001] This invention relates to the field of signal sensing technology, and in particular to a method and system for telemetry of animal physiological signals based on wearable devices. Background Technology

[0002] In livestock breeding, laboratory animal management, companion animal health monitoring, and field animal behavior research, the use of wearable devices for continuous, non-invasive, and remote acquisition of physiological signals from animals has become an important technological direction. Current methods typically collect photoelectric physiological signals such as heart rate and respiration by wearing sensor nodes on the animal's neck, trunk, and limbs, and then using wireless transmission for telemetry. However, animals in free-moving states generally exhibit complex movements such as running, jumping, head shaking, and twisting. Different body parts also produce asynchronous, multi-axis, and multi-frequency posture changes, resulting in a large number of motion artifacts in the acquired photoelectric signals caused by skin deformation, tissue displacement, changes in contact pressure, and optical path offset, severely affecting the accuracy of physiological parameter extraction.

[0003] Existing technologies mostly employ single-node filtering or simple threshold denoising to handle motion interference, making it difficult to accurately characterize the dynamic coupling relationship between inertial motion and optical path changes. Especially in multi-node network conditions, issues such as time misalignment, frequency differences, and inconsistent signal reliability exist between nodes in different locations, easily leading to insufficient artifact separation, inability to coordinate and utilize information between nodes, and low overall reconstruction accuracy. Therefore, there is an urgent need for a method and system for telemetry of animal physiological signals based on wearable devices, capable of dynamically modeling, coordinating, fusing, and grouping motion artifacts under multi-node conditions, to improve the accuracy and stability of telemetry of real animal physiological signals. Summary of the Invention

[0004] This invention provides a wearable device-based animal physiological signal telemetry system, which addresses the complex motion interference problems of animals in a free-moving state by achieving high-precision telemetry and reconstruction of multi-node physiological signals through collaborative processing.

[0005] In a first aspect, the present invention provides a method for telemetry of animal physiological signals based on wearable devices, the method comprising: Step S1: By collecting photoelectric signals and inertial data from each node, an adaptive filtering algorithm is used to process the initial mixed signal to obtain the preliminarily separated physiological fluctuations and artifact components; Step S2: Based on the initially separated artifact components, obtain the motion mode features of the corresponding nodes and determine their dynamic correlation parameters with the changes in the optical path; if the dynamic correlation parameters exceed the preset threshold, construct a mapping model that characterizes the mapping relationship between motion mode features and artifact components, and correct the mapping model through the Kalman filter algorithm to obtain optimized artifact reference data. Step S3: For the optimized artifact reference data, extract the shared features of multiple nodes, calculate the similarity of motion patterns between nodes, and verify the similarity based on both time and frequency domain judgments. Combine the node reliability index for weighted fusion to obtain the weighted similarity between nodes and obtain the relevant node groups and independent processing nodes. Step S4: Generate a shared artifact elimination template from the optimized artifact reference data of the relevant node group through principal component extraction, and determine the time offset parameter and amplitude scaling ratio parameter to form a unified filtering strategy; use the filtering strategy to perform adaptive cancellation filtering on the remaining mixed signal and combine it with Kalman filtering to obtain the real physiological signals of multiple parts synchronously restored.

[0006] Secondly, the present invention provides a wearable device-based animal physiological signal telemetry system for implementing the above-described method, the system comprising: The signal acquisition and processing unit is used to acquire photoelectric signals and inertial data from each node, and process the initial mixed signal using an adaptive filtering algorithm to obtain the initially separated physiological fluctuations and artifact components. The artifact modeling and optimization unit is used to obtain the motion mode features of the corresponding nodes based on the initially separated artifact components, and determine the dynamic correlation parameters between them and the changes in the optical path. If the dynamic correlation parameters exceed the preset threshold, a mapping model representing the mapping relationship between the motion mode features and the artifact components is constructed, and the mapping model is corrected by the Kalman filter algorithm to obtain optimized artifact reference data. The multi-node collaborative analysis unit is used to extract shared features of multiple nodes for optimized artifact reference data, calculate the similarity of motion patterns between nodes, verify the similarity based on both time and frequency domain judgments, and perform weighted fusion with node reliability indicators to obtain the weighted similarity between nodes and acquire relevant node groups and independent processing nodes. The filtering strategy generation unit is used to generate a shared artifact elimination template from the optimized artifact reference data of the relevant node group through principal component extraction, and to determine the time offset parameter and amplitude scaling ratio parameter to form a unified filtering strategy. The signal restoration output unit is used to perform adaptive cancellation filtering on the remaining mixed signal using the filtering strategy and combine it with Kalman filtering to obtain synchronously restored real physiological signals from multiple sites.

[0007] Thirdly, the present invention provides a telemetry device for animal physiological signals based on a wearable device, the device comprising: a memory and at least one processor, wherein the memory stores instructions; The at least one processor invokes the instructions in the memory to cause the wearable device-based animal physiological signal telemetry device to perform the above-described method.

[0008] Fourthly, the present invention also provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the above-described method.

[0009] The beneficial effects of this invention are as follows: This invention introduces the fusion of photoelectric signals and inertial data during the signal acquisition stage and achieves real-time decomposition of the initial mixed signal through adaptive filtering. While preserving physiological fluctuations such as heart rate and respiration, it effectively removes major motion-related artifacts, providing a reliable foundation for subsequent processing. Subsequently, by extracting motion pattern features from the artifacts and constructing dynamic correlation parameters between these features and changes in the optical path, the system can adaptively determine the degree of interference based on different motion intensities and states. In strongly coupled scenarios, Kalman filtering is further combined to perform online correction of the mapping model, thereby obtaining artifact reference data that more closely reflects the evolution of real interference and improving the accuracy and stability of artifact modeling. Based on this, a multi-node collaborative analysis mechanism is used to perform time alignment and similarity evaluation of the motion features of each node. Combining time-domain and frequency-domain dual judgment and node reliability weighting, accurate identification of the motion consistency of different parts is achieved, thus dividing the nodes... Dividing the data into related node groups and independent nodes ensures effective fusion of homogeneous motion information while avoiding interference from heterogeneous nodes in the overall processing. Furthermore, within related node groups, a shared artifact removal template is generated through principal component extraction, and a unified filtering strategy is constructed using time offset and amplitude scaling parameters, enabling artifact subtraction for each node within a unified reference frame. For nodes with significant differences, personalized templates are used for targeted processing, balancing overall consistency with local adaptability. Finally, through cascaded processing of adaptive destructive filtering and Kalman filtering, residual artifacts and random noise in the remaining mixed signal are eliminated in layers, and the temporal alignment of multi-node signals is completed, achieving synchronous restoration of real physiological signals from multiple body parts. Through the synergy of these technical solutions, the problem of difficult collaborative processing of multi-node signals under complex motion conditions is effectively solved, significantly improving the accuracy, synchronization, and stability of physiological signal extraction. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart of the animal physiological signal telemetry method based on wearable devices in the embodiment; Figure 2This is a schematic diagram illustrating the relationship between dynamic correlation parameters and time during animal movement in the embodiment.

[0012] Figure 3 This is a structural diagram of the animal physiological signal telemetry system based on wearable devices in the embodiment. Detailed Implementation

[0013] This invention provides a method and system for telemetry of animal physiological signals based on wearable devices. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0014] For ease of understanding, the specific process of the embodiments of the present invention will be described below, such as... Figure 1 As shown, an embodiment of the present invention provides a method for telemetry of animal physiological signals based on wearable devices, comprising: Step S1: By collecting photoelectric signals and inertial data from each node, an adaptive filtering algorithm is used to process the initial mixed signal to obtain the preliminarily separated physiological fluctuations and artifact components; specifically including: Photoelectric signals and inertial data are acquired from each node as initial mixed signals. An adaptive filtering algorithm is applied to the initial mixed signals for noise suppression and component decomposition to separate the initial physiological fluctuation components and artifact components. The filtering parameters are iteratively adjusted to minimize the residual error to obtain the initially separated physiological fluctuation and artifact components. The separation effect is verified by frequency domain analysis. If the spectral overlap is lower than the preset threshold, the initial separation is confirmed to be effective; otherwise, the adaptive filtering algorithm is repeated until the condition is met.

[0015] Specifically, wearable acquisition nodes are deployed at multiple key locations throughout the animal's body, and photoelectric signals and inertial data of the corresponding locations are collected synchronously according to the same sampling period. The aforementioned photoelectric signals are preferably the reflected light intensity change signals formed by the absorption, scattering, and reflection of light from the animal's body surface tissue after being irradiated by a light-emitting device, which can characterize the physiological fluctuations caused by the periodic changes in blood volume. The aforementioned inertial data are preferably output in real time by the triaxial accelerometer and triaxial gyroscope built into the nodes, used to characterize the three-dimensional spatial acceleration and angular velocity changes of the node's location. Each node aligns the photoelectric signals and inertial data at the same timestamp to form the initial mixed signal of the corresponding node. The above settings ensure that each initial mixed signal contains not only useful components corresponding to real physiological activities such as heart rate and respiration, but also motion artifacts and environmental noise introduced by the animal running, jumping, swinging, body twisting, relative displacement of the equipment, and local skin deformation.

[0016] After obtaining the initial mixed signals from each node, an adaptive filtering algorithm is applied to suppress noise and decompose the components, separating the preliminary physiological fluctuation components and artifact components. Specifically, inertial data is extracted from the initial mixed signals as a reference input signal, mainly reflecting motion information, and photoelectric signals are extracted as the desired response, containing both real physiological fluctuations and motion artifacts. An adaptive filter based on the normalized least mean square algorithm is constructed. This adaptive filter internally employs a finite-length unit impulse response filter structure, and its filter order can be set according to the motion complexity, tissue thickness, and device fit conditions at the node location. The filter order is rooted in... Based on the characteristics of the node's location, higher-order parameters (e.g., 64th order) are used for highly complex motion areas such as the limbs to capture intricate artifact dynamics, while lower-order parameters (e.g., 32nd order) are used for relatively simpler motion areas such as the torso. During processing, for each sampling point, the inertial data from the current and multiple past sampling times are used to construct a reference input vector. This vector is then multiplied by the weighting coefficient vector to obtain the estimated artifact at the current moment. This estimated artifact is subtracted from the photoelectric signal to obtain the preliminary physiological fluctuation component, i.e., the error signal. Subsequently, based on the error signal and the reference input vector, the weighting coefficients are updated using the NLMS algorithm. The normalization factor in the update formula is the energy of the reference input vector plus a small constant to prevent division by zero. Through this adaptive process, motion-related interference can be effectively removed while maintaining the integrity of the physiological rhythm structure.

[0017] To adapt the preliminary separation results to the complex conditions of rapid changes in movement and significant differences in body parts during free animal activity, the filter parameters are iteratively adjusted to minimize the residual error, thereby obtaining a more stable preliminary separation result. Specifically, the preliminary physiological fluctuation component is defined as the residual error, meaning that ideally, after sufficient estimation and subtraction of motion artifacts, the remaining signal should be as close as possible to the actual physiological fluctuations. Therefore, at each sampling point, the square of the Euclidean norm of the reference input signal is first calculated, and a very small positive constant is added as a normalization factor to prevent the denominator from being zero and to limit numerical divergence. Then, the current residual error, the reference input signal, and the preset step size factor are multiplied and divided by the normalization factor to obtain the gradient adjustment amount of the weight coefficients. This gradient adjustment amount is then superimposed on the current weight coefficient vector to update the weight coefficient vector for the next time step. By performing the above update process point by point, the mean square value of the residual error gradually decreases, thereby enabling the filter's estimation of motion-induced light path changes to gradually approach the actual artifact components. It should be noted that in animal telemetry scenarios, the tissue structure, blood perfusion conditions, and tightness of the fitting are not consistent in different parts of the body. When the same animal switches between different actions such as walking, running, and jumping, the coupling relationship between inertial data and artifact signals will exhibit obvious time-varying characteristics. The normalized least mean square algorithm enables the filter to track the changes in this coupling relationship in real time by adaptively updating the weight coefficient vector point by point. For example, when an animal suddenly switches from walking to jumping, the reflected light path in the body surface tissue will undergo a drastic and nonlinear change, and the residual error will increase rapidly in a short period of time. At this time, the algorithm uses the currently drastically changing reference input signal and the increased residual error to calculate a larger gradient adjustment amount, so as to correct the weight coefficient vector at a faster speed. This allows the estimated artifact components to keep up with the dynamic interference pattern changes caused by the sudden action in a timely manner, while the residual signal after subtraction retains the true fluctuation characteristics of physiological signals such as heartbeat and respiration in the time domain to the greatest extent.

[0018] To avoid misclassifying some physiological components into artifacts based solely on time-domain residual convergence, frequency-domain analysis was used to verify the separation effect. Based on the verification results, closed-loop adjustment of filter parameters was performed. Specifically, Fast Fourier Transform was applied to the preliminary physiological fluctuation components and artifact components obtained after iterative processing, converting the time-domain signals into corresponding physiological and artifact spectrum amplitude sequences. In animal physiological signal telemetry, the physiological frequency band can be set according to the type, age, body size, and monitoring indicators of the monitored object. For example, in heart rate monitoring, the target physiological frequency band can be set to a preset range. Subsequently, within the preset physiological frequency band, the intersection area of ​​the physiological spectrum amplitude sequence and the artifact spectrum amplitude sequence is calculated, and this intersection area is defined as the spectral overlap portion. Then, the spectral overlap portion is divided by the physiological frequency band. The total area of ​​the spectral amplitude sequence within the frequency band is used to obtain the overlap ratio. If the overlap ratio is lower than a preset threshold, it indicates that the real physiological components mixed in with the artifact components are relatively few, and the current separation result can be considered valid. The current artifact components and the corresponding preliminary physiological fluctuation components are then output. If the overlap ratio is not lower than the preset threshold, it indicates that the current filter is over-tracking or underfitting the artifacts, causing some physiological frequency band energy to be incorrectly allocated to the artifact components. In this case, the preset step size factor is reduced, and the process returns to the adaptive filtering stage to re-execute artifact estimation, residual update, and weight iteration. Through the above technical solution, the output artifact reference data is decoupled from the real physiological frequency bands in a spectral sense as much as possible, thus providing a basis for subsequent motion pattern correlation analysis and artifact coordination processing between multiple nodes.

[0019] Step S2: Based on the initially separated artifact components, obtain the motion mode features of the corresponding nodes and determine their dynamic correlation parameters with the changes in the optical path; if the dynamic correlation parameters exceed the preset threshold, construct a mapping model that characterizes the mapping relationship between motion mode features and artifact components, and correct the mapping model through the Kalman filter algorithm to obtain optimized artifact reference data. In step S2, the dynamic correlation parameters between motion mode characteristics and optical path changes are determined, including: The initially separated artifact components are segmented using a sliding window method. Temporal and amplitude features are extracted for each segment and used as local motion pattern features. Correlation analysis is performed with the corresponding photoelectric signal change data to obtain segment-level dynamic correlation parameters. The segment-level dynamic correlation parameters are then fused by weighted averaging to obtain the overall dynamic correlation parameters. The local deviation between the segment-level dynamic correlation parameters and the overall dynamic correlation parameters is calculated. If the local deviation exceeds a preset difference, the window size is adjusted and the motion pattern features are extracted again.

[0020] Specifically, since animals are usually in a constantly changing dynamic behavioral state during telemetry, the artifact morphology and optical path perturbation relationship at the same node are not constant at different action stages. Therefore, a sliding window method is used to segment the artifact components. Specifically, the time step of the window moving forward along the time axis is set, and the sliding window slides along the time axis of the artifact components segment by segment with the above step. For each segment, temporal features and amplitude features are extracted. The temporal features are used to characterize the rhythm of the artifact components on the time axis, which can be obtained by calculating the time interval between adjacent peaks in the artifact components and the frequency of the signal crossing the zero level. The former reflects the periodicity of motion perturbation, and the latter reflects the speed of perturbation change. The amplitude features are used to characterize the energy fluctuations of the artifact components, which can be obtained by calculating the root mean square amplitude of the artifact components within a set time period to reflect the strength of interference caused by local tissue deformation, relative equipment displacement, and microscopic scattering changes. The above-mentioned time interval between adjacent peaks, zero level crossing frequency, and root mean square amplitude are combined as the above-mentioned motion pattern features.

[0021] After obtaining the motion pattern features, correlation analysis is performed between them and the photoelectric signal change data acquired synchronously within the corresponding time period to obtain the segment-level dynamic correlation parameters corresponding to that segment. The aforementioned photoelectric signal change data is preferably the gradient change value of the reflected intensity formed after the light beam propagates within the animal's body tissue. The correlation analysis can be performed using the Pearson correlation coefficient calculation method. The time series of any feature in the local motion pattern features is taken as the first variable series, and the synchronous optical path change data is taken as the second variable series. The covariance of the two series is calculated, and the standard deviation of the two series is calculated separately. Then, the covariance is divided by the product of the two standard deviations to obtain the Pearson correlation coefficient. The absolute value of the correlation coefficient is taken as the correlation strength. The mean of the correlation strength between all features in the aforementioned motion pattern features and the aforementioned optical path change data is taken as the aforementioned segment-level dynamic correlation parameter. The value of this parameter ranges from 0 to 1. The larger the value, the stronger the synchronicity and coupling between the current node's motion pattern change and optical path change, that is, the more significant the interference of animal movement on the photoelectric signal. The above process is repeated after the window slides to characterize the change in the degree of transient interference at different time periods during the running process.

[0022] After obtaining multiple paragraph-level dynamic correlation parameters, to reduce the impact of occasional mutations or local anomalies on the overall judgment, a weighted average is used to fuse the paragraph-level dynamic correlation parameters to obtain the overall dynamic correlation parameters. Specifically, the variance of the artifact components within each paragraph is calculated. The smaller the variance, the more stable the signal fluctuation and the higher the data reliability within that paragraph. Therefore, the reciprocal of the variance of each paragraph is used as the weight of the corresponding paragraph-level dynamic correlation parameter. All paragraph-level dynamic correlation parameters are weighted and summed, and then divided by the total weight to obtain the overall dynamic correlation parameters of the current node. To enable the dynamic correlation parameters to adapt to the time-varying interference characteristics of different parts and different motion states, it is also necessary to calculate the local deviation between each paragraph-level dynamic correlation parameter and the overall dynamic correlation parameters. That is, the absolute value of the difference between the two is taken and compared with a preset difference value. When the local deviation... When the difference exceeds a preset value, it indicates that the current sliding window length cannot accurately adapt to the dynamic evolution of the optical path changes under the current motion state of the node. In this case, the window size is adjusted according to the zero-level crossing frequency in the corresponding local motion mode features. If the zero-level crossing frequency is greater than the preset frequency reference value, it indicates that the node is in a high-frequency, violent motion state, and the sliding window length should be reduced to enhance the ability to capture rapid optical path changes. If the zero-level crossing frequency is less than or equal to the frequency reference value, it indicates that the node is in a low-frequency, gentle motion state, and the sliding window length should be increased to cover a more complete motion cycle and avoid insufficient feature extraction. After adjusting the window size, the artifact components are re-segmented, features are extracted, and correlation analysis is performed until dynamic correlation parameters that can stably characterize the node's motion and optical path coupling relationship are obtained, such as... Figure 2 As shown in the figure, this diagram illustrates the changes in dynamic correlation parameters over time during animal movement. The dark solid line represents the overall correlation parameters after weighted fusion, while the gray dashed line represents the segment-level correlation parameters calculated by the sliding window. When the local deviation between the segment-level parameters and the overall parameters exceeds a preset threshold (see the shaded area), the sliding window length is automatically adjusted to adapt to changes in movement state, ensuring the quantification accuracy of the coupling relationship between movement patterns and optical path disturbances. Through this technical solution, a quantitative mapping relationship between artifact features and optical path disturbances can be adaptively established for the differentiated movement patterns of different body parts of the animal, which helps improve the accuracy of artifact analysis and the consistency of multi-node reconstruction during animal physiological signal telemetry.

[0023] Further, in step S2, optimized artifact reference data is obtained, including: If the dynamic correlation parameter is less than the preset value, the initially separated artifact components are directly used as artifact reference data. If the dynamic correlation parameter is greater than or equal to the preset value, a mapping model is constructed. This mapping model characterizes the mapping relationship between motion pattern features and artifact components. Three-dimensional motion features are extracted from the three-dimensional motion data collected by the inertial sensors at the corresponding nodes and input into the mapping model, outputting predicted artifact component values. The predicted artifact component values ​​are used as prior model data, and the initially separated artifact components are used as observation data. The Kalman filter algorithm is used to fuse the prior model data and the observation data, iteratively updating the state vector to correct the mapping model. The correction process is performed independently for each node, and the filter gain is adjusted based on the covariance matrix to obtain optimized artifact reference data. After the correction is completed, the fit between the optimized artifact reference data and the original signal is verified.

[0024] Specifically, when the aforementioned dynamic correlation parameters are less than the preset values, it indicates that the current motion posture has little impact on the optical path change, and the relationship between the motion mode and artifacts in the current node is relatively stable. In this case, the preliminary artifact components obtained from the initial separation can be directly used as the artifact reference data output for that node. When the aforementioned dynamic correlation parameters are greater than or equal to the preset values, it is determined that the current node is in a state of strong coupling interference, and the artifact components obtained from the initial separation are difficult to fully reflect the evolution law of the real artifacts. By establishing a mapping model describing the transformation relationship from motion mode features to artifact components, its essence is to establish a quantitative mapping between inertial motion state and photoelectric artifact response. The Kalman filter algorithm is used to estimate and predict the state of this mapping model. Specifically, the three-dimensional motion features most relevant to the current artifact generation are extracted from the three-dimensional motion data collected by the node's inertial sensor, including at least three-axis acceleration and three-axis angular velocity. Alternatively, the components that contribute more to the artifacts can be selected according to the main motion direction of the node's location, such as the neck node. The model emphasizes the angular velocity component, while the back or limb nodes emphasize the vertical acceleration component. These three-dimensional motion features are input into a mapping model, preferably constructed as a polynomial regression-based mathematical expression. This model quantifies the influence of changes in spatial acceleration and angular velocity on the propagation, scattering, and reflection paths of the light beam within the animal's tissues, and outputs predicted artifact components at corresponding moments. Simultaneously, preliminary artifact components obtained from the mixed signals actually acquired by the photoelectric sensor are used as observation data input into the Kalman filter algorithm. Subsequently, the predicted artifact components output by the mapping model are used as prior model data. The residual between the observation data and the prior model data is calculated, and this residual is used to update the state vector in the Kalman filter algorithm. This state vector contains the polynomial coefficients of various orders in the mapping model. By continuously iterating and updating the state vector, real-time correction of the mapping model parameters is achieved, enabling the model output to more closely reflect the actual changes in artifacts under the animal's current movement.

[0025] To make this correction process applicable to the heterogeneous interference characteristics of different body parts in a multi-node animal telemetry network, the correction process is further applied independently to each node, and a covariance matrix is ​​introduced to adjust the filter gain. Specifically, in the multi-node physiological monitoring network, each acquisition node deployed at different body parts such as the animal's neck, back, and limbs is assigned an independent computing thread, allowing each node to execute its own Kalman estimation process. Within each computing thread, the node's specific covariance matrix is ​​initialized based on the motion characteristics, skin deformation complexity, and artifact fluctuation characteristics of the corresponding body part. This covariance matrix includes at least the process noise. Covariance and measurement noise covariance are used to quantify the uncertainty of state vector prediction and the uncertainty of observation data, respectively. During the iteration of Kalman filtering, the Kalman gain of the corresponding node is calculated using the above covariance matrix, and this Kalman gain is applied to the residual to adjust the update step size of the state vector. When the noise of the node observation data is large, the Kalman gain can be reduced by increasing the measurement noise covariance, thereby reducing the sensitivity of the state update to the current observation value and avoiding over-correction of the model due to instantaneous abnormal disturbances. When the node observation data is relatively stable, the observation weight can be increased relatively, so that the model can track the changes in artifacts more quickly.

[0026] After completing the above corrections, the effectiveness of the obtained optimized artifact reference data needs to be verified. Specifically, this can be done by calculating the correlation coefficient between the optimized artifact reference data and the original photoelectric signal. The higher the correlation coefficient, the stronger the ability of the optimized artifact reference data to interpret the interference components in the original mixed signal. The above technical solution achieves high-precision modeling and optimized estimation of complex motion artifacts in the process of animal physiological signal telemetry. It retains the processing efficiency of directly using the preliminary artifact components in weak interference scenarios, and significantly improves the accuracy and stability of artifact reference data in strong coupling interference scenarios.

[0027] Step S3: For the optimized artifact reference data, extract multi-node shared features, calculate the similarity of motion patterns between nodes, and verify the similarity based on both time and frequency domain judgments. Combine this with node reliability indicators for weighted fusion to obtain the weighted similarity between nodes and acquire relevant node groups and independently processed nodes; specifically including: Multi-node shared features are extracted from the optimized artifact reference data of each node. A reference node is selected, and a signal coordination mechanism is used to synchronize and align the shared features of each node. Based on the aligned shared features, the preliminary similarity of motion patterns between each node is calculated. The preliminary similarity is judged in both the time domain and the frequency domain. If the time domain similarity and the frequency domain similarity simultaneously meet the preset conditions, the similarity is confirmed to be valid; otherwise, the preliminary similarity is attenuated. The confirmed similarity is weighted and fused according to the reliability index of each node to obtain the weighted similarity between each node. Then, a clustering algorithm is applied to the fused multi-node information to group similar motion patterns. If the weighted similarity between nodes in a group is higher than a preset threshold, it is marked as a related node group; otherwise, it is marked as an independent processing node.

[0028] Specifically, multi-node shared features are extracted from optimized artifact reference data. For the 3D spatial acceleration and angular velocity sequences synchronously acquired by each node, the composite vector amplitude of the 3D spatial acceleration is calculated within each time window to obtain the motion envelope of the corresponding node. The motion envelope is used to characterize the overall motion intensity change of the node within the time window. Furthermore, a fast Fourier transform is performed on the 3D spatial acceleration sequences of each node to extract the energy distribution value within the main frequency band, reflecting the frequency structure and energy concentration range of the current node's motion. The motion envelope is combined with the energy distribution value within the main frequency band to form multi-node shared features that can be used for mutual comparison. The motion envelope mainly reflects the change in motion amplitude in the time domain, while the energy distribution value in the main frequency band mainly reflects the motion rhythm characteristics in the frequency domain. Together, they characterize the external manifestation and internal structure of the node's motion pattern.

[0029] After obtaining the shared features of multiple nodes, a signal coordination mechanism is further employed to synchronize and weightedly fuse the information of multiple nodes to calculate the similarity of motion patterns among nodes. The trunk node, which can represent the overall posture change trend, is selected as the reference node. The cross-correlation function between the motion envelope of each other node and the motion envelope of the reference node is calculated. Since the movements of different body parts are not strictly synchronized when animals run or jump, for example, the acceleration of the forelimbs usually precedes the overall displacement of the trunk. If the unaligned data is directly compared, it is easy to misjudge essentially related movements as unrelated ones. Therefore, the time delay corresponding to the maximum value of the cross-correlation function is found, and the motion envelope of the corresponding node is shifted on the time axis according to the time delay, thereby completing the synchronization and alignment of the information of multiple nodes. After completing the time alignment, the Pearson correlation coefficient between the energy distribution values ​​in the main frequency band of each node is calculated and used as the preliminary similarity of motion patterns among nodes, thereby avoiding the distortion of similarity calculation caused by phase misalignment.

[0030] To ensure the accuracy of similarity judgment, the preliminary similarity is further verified through dual judgment in the time domain and frequency domain. Specifically, in the time domain, the Euclidean distance between the motion envelopes of each node after synchronization and alignment is calculated. If the Euclidean distance is less than a first distance threshold, the related nodes are determined to have similar amplitude change trends in the time domain. In the frequency domain, the power spectral density of the three-dimensional spatial acceleration sequence of each node is calculated, and the cross-entropy between the power spectral densities of different nodes is further calculated. The cross-entropy is used to measure the consistency between two spectral distributions. If the cross-entropy is less than a first entropy threshold, the related nodes are determined to have similar energy distribution structures in the frequency domain. The preliminary similarity is confirmed to be valid only when both time domain similarity and frequency domain similarity are satisfied. Otherwise, the preliminary similarity is multiplied by a preset attenuation coefficient for penalty downgrading. Through this dual judgment mechanism, node combinations that appear similar only in amplitude change but have significantly different actual motion mechanisms can be effectively eliminated, so that the final similarity can more accurately reflect the real coupling relationship between motion artifacts and optical path changes.

[0031] Based on this, a weighting factor is introduced during the fusion process to adjust the contribution based on node reliability. Specifically, the ratio of the effective value of the AC component to the effective value of the DC component of the photoelectric signal at each node within the current time window is calculated to obtain the local perfusion index, which is used to characterize the detectability of local tissue blood flow fluctuations and the effectiveness of photoelectric measurement at that node. A higher local perfusion index generally indicates that the effective physiological components in the photoelectric signal at that node are relatively clearer. Simultaneously, the variance of the corresponding three-dimensional spatial acceleration sequence is calculated to characterize the intensity of the current motion disturbance at that node. The ratio of the local perfusion index to the variance is defined as the node reliability index, and the reliability indices of all nodes are normalized to obtain the weighting factor for each node. Subsequently, the confirmed effective preliminary similarity is multiplied by the corresponding weighting factor to obtain the weighted similarity between nodes. After obtaining the weighted similarity, further... The next step involves applying a clustering algorithm to group similar motion patterns based on the fused multi-node information. Specifically, an undirected graph is constructed with each node as a vertex and the weighted similarity between nodes as edge weights. The adjacency matrix of this undirected graph is then input into the K-means clustering algorithm. After setting the initial number of cluster centers, all nodes are divided into multiple clusters through iterative calculation. Subsequently, the minimum weighted similarity between nodes within each cluster is extracted. If the minimum weighted similarity is higher than a set threshold, the cluster is marked as a related node group, and the nodes within the group share subsequent artifact removal reference data. If the minimum weighted similarity is not higher than the set threshold, the cluster is split, and the nodes within it are marked as independent processing nodes. Computational resources are allocated to them separately for artifact separation and subsequent signal restoration. This approach can identify node sets driven by the same source motion while avoiding interference from isolated nodes with special motion patterns in the overall processing.

[0032] Step S4: Generate a shared artifact elimination template from the optimized artifact reference data of the relevant node group through principal component extraction, and determine the time offset parameter and amplitude scaling ratio parameter to form a unified filtering strategy; use the filtering strategy to perform adaptive cancellation filtering on the remaining mixed signal and combine it with Kalman filtering to obtain the real physiological signals of multiple parts synchronously restored.

[0033] In step S4, determining the unified filtering strategy includes: From the optimized artifact reference data of each node in the relevant node group, a shared artifact removal template is generated through principal component extraction. A template matching algorithm is used to determine the time offset parameter and amplitude scaling parameter by calculating the normalized cross-correlation coefficient between the shared artifact removal template and the optimized artifact reference data of each node. Combined with an adaptive cancellation filter type, a unified filtering strategy is formed. For nodes in the relevant node group, if their cross-correlation coefficient with the shared template is not lower than a preset similarity threshold, the unified filtering strategy is applied for artifact subtraction. If their cross-correlation coefficient is lower than the preset similarity threshold, they are determined to be independent nodes, and their own optimized artifact reference data is used as a personalized template for artifact subtraction.

[0034] Specifically, after completing the division of relevant node groups, a shared artifact elimination template is generated from the optimized artifact reference data of the relevant node groups. Optimized artifact reference data from each acquisition node within the relevant node groups is extracted. Principal component extraction is performed on the extracted multi-path optimized artifact reference data to obtain the first principal component data sequence. The covariance matrix of the optimized artifact reference data is calculated, and then the eigenvalues ​​and eigenvectors of this covariance matrix are solved. The eigenvalues ​​are then arranged in descending order, and the direction represented by the eigenvector corresponding to the largest eigenvalue is selected as the dominant change direction, forming the first principal component data sequence. Since the direction corresponding to the largest eigenvalue represents the common change pattern with the largest variance contribution in the original multi-path data, the first principal component data sequence can retain the common artifact morphology within the relevant node groups to the greatest extent, while suppressing random noise and local abnormal fluctuations in individual nodes. Finally, this first principal component data sequence is used as a shared artifact elimination template. This template not only reduces the computational workload of repetitive modeling node by node but also more centrally characterizes the core artifact morphology of the relevant node group, thereby improving the consistency and representativeness of subsequent filtering strategies.

[0035] After obtaining the shared artifact removal template, the template parameters are further optimized using a template matching algorithm to determine a unified filtering strategy. This unified filtering strategy includes two parts: filter type and parameter settings. Specifically, normalized cross-correlation calculation is used to calculate the cross-correlation coefficient sequence between the shared artifact removal template and the optimized artifact reference data corresponding to each node in the relevant node group. The normalized cross-correlation calculation first performs zero-mean processing on the two sequences, then calculates their covariance and divides it by the product of the standard deviations of the two sequences, thereby eliminating amplitude differences while measuring waveform morphology similarity. Furthermore, by finding the peak position of the cross-correlation coefficient sequence, the optimal alignment time offset of the shared template relative to the optimized artifact reference data of each node is determined. This offset is the time offset parameter, which describes the specific amount the template should be shifted on the time axis to adapt to different... The time delay differences between nodes are caused by variations in motion transmission paths and local tissue response speeds. Simultaneously, the amplitude scaling parameter of the shared template is determined based on the peak value of the cross-correlation coefficient to reflect the proportional relationship between the node artifact amplitude and the shared template amplitude, thus adapting to differences in attachment tightness, skin thickness, muscle deformation amplitude, and optical path stability among different nodes. After obtaining the time offset parameter and amplitude scaling parameter, adaptive cancellation filtering is set as the filtering type in the unified filtering strategy, and the aforementioned time offset parameter and amplitude scaling parameter are used as corresponding parameter settings to form a unified filtering strategy suitable for the relevant node group.

[0036] For nodes with high similarity, the same template is applied for artifact subtraction. Specifically, it is determined whether the cross-correlation coefficient between each node in the relevant node group and the shared artifact elimination template is lower than a preset similarity threshold. If the cross-correlation coefficient is lower than the preset similarity threshold, it indicates that although the node has a certain similarity with other nodes in the group in terms of overall motion pattern, there are still significant differences in its local skin deformation, relative device displacement, or microscopic sight path scattering process. The shared template cannot fully cover the details of its artifact changes. In this case, the node is determined to be an independent node, and its optimized artifact reference data is extracted as a personalized template. The personalized template is then input into an adaptive destructive filter for artifact subtraction. Conversely, if the cross-correlation coefficient is greater than or equal to the preset similarity threshold, it indicates that the node and the shared template are similar. When the matching degree is high and the shared template can accurately represent the artifact-dominant mode, the shared artifact elimination template is integrated into the unified filtering strategy. Using the shared template with time offset parameters and amplitude scaling parameters, the original mixed signal of the node is degraded by adaptive cancellation filtering. For example, the limb nodes generally swing periodically when running, but there may be slight differences in the local skin stretching direction, force changes and optical path disturbance patterns of the forelimbs and hindlimbs at the moment of ground contact. When the cross-correlation coefficient between a certain forelimb node and the shared template is lower than the preset similarity threshold, it means that the shared template cannot completely represent its local complex artifacts. In this case, using a personalized template can avoid the distortion of physiological signals caused by template mismatch and ensure that the noise reduction process of each node fits its real artifact structure as closely as possible.

[0037] Furthermore, in step S4, the real physiological signals of multiple sites are obtained synchronously, including: The remaining mixed signal of each node is obtained. The remaining mixed signal is the initial mixed signal minus the initially separated artifact components. According to the unified filtering strategy, combined with the shared artifact elimination template, time offset parameter and amplitude scaling parameter, the remaining mixed signal is subjected to adaptive cancellation filtering to remove residual artifacts. Then, the filtered output is used as the observation value to input the Kalman filter for state update and optimal estimation, filter out random noise, and perform time alignment to obtain the synchronously restored real physiological signals of multiple parts.

[0038] Specifically, the residual mixed signal of each node is acquired. This residual mixed signal is the initial mixed signal of the corresponding node minus the initially separated artifact components. The centralized processing unit receives the initial mixed signals and corresponding macroscopic motion artifact components uploaded by multiple acquisition nodes, and performs a time-domain subtraction operation to obtain the residual mixed signal. After this processing, macroscopic artifacts caused by large-scale spatial acceleration, overall limb swaying, and trunk undulation are preferentially subtracted. The residual mixed signal mainly retains real physiological fluctuations and residual artifacts formed by the coupling of factors such as local skin deformation, microscopic scattering of light in tissues, and subtle changes in adhesion pressure. Subsequently, a unified filtering strategy is applied. Based on the shared template, and combining the peak position of the cross-correlation coefficient (i.e., the time offset) and signal length, a continuous reference artifact sequence of equal length and time alignment with the remaining mixed signal is reconstructed using the shared template, i.e., the principal component eigenvector of the reference artifact. Subsequently, this continuous reference artifact sequence is used as the input for the first-stage filtering, i.e., a minimum mean square error filter is applied to the remaining mixed signal for the first-stage filtering. During the filtering process, the filtering weights are continuously updated based on the error between the input and the target signal, so that residual artifact components highly correlated with local skin stretching, compression, or torsion are adaptively subtracted. After the first-stage filtering output, the preliminary denoised signal is used as the current observation value and input to the Kalman filter for the second-stage filtering. The Kalman filter uses a state transition matrix constructed based on a priori models of real animal physiological rhythms, such as the periodicity of heart rate, to predict the internal state of the current physiological signal, and combines this with the current observation value for state updates and optimal estimation, thereby further filtering out uncorrelated random noise caused by microscopic scattering and other processes. Through the above multi-stage processing, structural residual artifacts and random fragmented noise can be removed in layers, yielding real physiological signals from multiple locations.

[0039] The above scheme enables the hierarchical elimination of multi-level motion artifacts in multi-node photoelectric signals under conditions of free animal movement, and ensures the consistency of signals from multiple locations at both the temporal and physiological logic levels, thereby significantly improving the fidelity, stability, and engineering practicality of real physiological signal restoration.

[0040] This invention also provides a wearable device-based animal physiological signal telemetry system for implementing the above-described method, such as... Figure 3 As shown, the system includes: The signal acquisition and processing unit is used to acquire photoelectric signals and inertial data from each node, and process the initial mixed signal using an adaptive filtering algorithm to obtain the initially separated physiological fluctuations and artifact components. The artifact modeling and optimization unit is used to obtain the motion mode features of the corresponding nodes based on the initially separated artifact components, and determine the dynamic correlation parameters between them and the changes in the optical path. If the dynamic correlation parameters exceed the preset threshold, a mapping model representing the mapping relationship between the motion mode features and the artifact components is constructed, and the mapping model is corrected by the Kalman filter algorithm to obtain optimized artifact reference data. The multi-node collaborative analysis unit is used to extract shared features of multiple nodes for optimized artifact reference data, calculate the similarity of motion patterns between nodes, verify the similarity based on both time and frequency domain judgments, and perform weighted fusion with node reliability indicators to obtain the weighted similarity between nodes and acquire relevant node groups and independent processing nodes. The filtering strategy generation unit is used to generate a shared artifact elimination template from the optimized artifact reference data of the relevant node group through principal component extraction, and to determine the time offset parameter and amplitude scaling ratio parameter to form a unified filtering strategy. The signal restoration output unit is used to perform adaptive cancellation filtering on the remaining mixed signal using the filtering strategy and combine it with Kalman filtering to obtain synchronously restored real physiological signals from multiple sites.

[0041] The present invention also provides a telemetry device for animal physiological signals based on a wearable device, the device comprising: a memory and at least one processor, wherein the memory stores instructions; The at least one processor invokes the instructions in the memory to cause the wearable device-based animal physiological signal telemetry device to perform the above-described method.

[0042] The present invention also provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the above-described method.

[0043] In summary, this invention introduces the fusion of photoelectric signals and inertial data during the signal acquisition stage and achieves real-time decomposition of the initial mixed signal through adaptive filtering. While preserving physiological fluctuations such as heart rate and respiration, it effectively removes major motion-related artifacts, providing a reliable foundation for subsequent processing. Subsequently, by extracting motion pattern features from the artifacts and constructing dynamic correlation parameters between these features and changes in the optical path, the system can adaptively determine the degree of interference based on different motion intensities and states. In strongly coupled scenarios, Kalman filtering is further combined to perform online correction of the mapping model, thereby obtaining artifact reference data that more closely reflects the evolution of real interference and improving the accuracy and stability of artifact modeling. Based on this, a multi-node collaborative analysis mechanism is used to perform time alignment and similarity evaluation of the motion features of each node. Combining time-domain and frequency-domain dual judgment and node reliability weighting, accurate identification of motion consistency in different parts is achieved, thus... The nodes are divided into related node groups and independent nodes, which ensures the effective fusion of homogeneous motion information and avoids interference from heterogeneous nodes on the overall processing. Furthermore, within related node groups, a shared artifact removal template is generated through principal component extraction, and a unified filtering strategy is constructed by combining time offset and amplitude scaling parameters, enabling each node to perform artifact subtraction within a unified reference frame. For nodes with significant differences, personalized templates are used for targeted processing, thus balancing overall consistency and local adaptability. Finally, through cascaded processing of adaptive destructive filtering and Kalman filtering, residual artifacts and random noise in the remaining mixed signal are eliminated in layers, and the temporal alignment of multi-node signals is completed, achieving synchronous restoration of real physiological signals from multiple body parts. Through the synergy of the above technical solutions, the problem of difficult collaborative processing of multi-node signals under complex motion conditions is effectively solved, significantly improving the accuracy, synchronization, and stability of physiological signal extraction.

[0044] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0045] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0046] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for telemetry of animal physiological signals based on wearable devices, characterized in that, The method includes: Step S1: By collecting photoelectric signals and inertial data from each node, an adaptive filtering algorithm is used to process the initial mixed signal to obtain the preliminarily separated physiological fluctuations and artifact components; Step S2: Based on the initially separated artifact components, obtain the motion mode features of the corresponding nodes and determine their dynamic correlation parameters with the changes in the optical path; if the dynamic correlation parameters exceed the preset threshold, construct a mapping model that characterizes the mapping relationship between motion mode features and artifact components, and correct the mapping model through the Kalman filter algorithm to obtain optimized artifact reference data. Step S3: For the optimized artifact reference data, extract the shared features of multiple nodes, calculate the similarity of motion patterns between nodes, and verify the similarity based on both time and frequency domain judgments. Combine the node reliability index for weighted fusion to obtain the weighted similarity between nodes and obtain the relevant node groups and independent processing nodes. Step S4: Generate a shared artifact elimination template from the optimized artifact reference data of the relevant node group through principal component extraction, and determine the time offset parameter and amplitude scaling ratio parameter to form a unified filtering strategy; use the filtering strategy to perform adaptive cancellation filtering on the remaining mixed signal and combine it with Kalman filtering to obtain the real physiological signals of multiple parts synchronously restored.

2. The method according to claim 1, characterized in that, In step S1, the initially separated physiological fluctuations and artifact components are obtained, including: Photoelectric signals and inertial data are acquired from each node as initial mixed signals. An adaptive filtering algorithm is applied to the initial mixed signals for noise suppression and component decomposition to separate the initial physiological fluctuation components and artifact components. The filtering parameters are iteratively adjusted to minimize the residual error to obtain the initially separated physiological fluctuation and artifact components. The separation effect is verified by frequency domain analysis. If the spectral overlap is lower than the preset threshold, the initial separation is confirmed to be effective; otherwise, the adaptive filtering algorithm is repeated until the condition is met.

3. The method according to claim 1, characterized in that, In step S2, the dynamic correlation parameters between the optical path and the changes are determined, including: The initially separated artifact components are segmented using a sliding window method. Temporal and amplitude features are extracted for each segment and used as local motion pattern features. Correlation analysis is performed with the corresponding photoelectric signal change data to obtain segment-level dynamic correlation parameters. The segment-level dynamic correlation parameters are then fused by weighted averaging to obtain the overall dynamic correlation parameters. The local deviation between the segment-level dynamic correlation parameters and the overall dynamic correlation parameters is calculated. If the local deviation exceeds a preset threshold, the window size is adjusted and the motion pattern features are extracted again.

4. The method according to claim 1, characterized in that, In step S2, optimized artifact reference data is obtained, including: If the dynamic correlation parameter is less than the preset value, the initially separated artifact components are directly used as artifact reference data. If the dynamic correlation parameter is greater than or equal to the preset value, a mapping model is constructed. This mapping model characterizes the mapping relationship between motion pattern features and artifact components. Three-dimensional motion features are extracted from the three-dimensional motion data collected by the inertial sensors at the corresponding nodes and input into the mapping model, outputting predicted artifact component values. The predicted artifact component values ​​are used as prior model data, and the initially separated artifact components are used as observation data. The Kalman filter algorithm is used to fuse the prior model data and the observation data, iteratively updating the state vector to correct the mapping model. The correction process is performed independently for each node, and the filter gain is adjusted based on the covariance matrix to obtain optimized artifact reference data. After the correction is completed, the fit between the optimized artifact reference data and the original signal is verified.

5. The method according to claim 1, characterized in that, Step S3 includes: Multi-node shared features are extracted from the optimized artifact reference data of each node. A reference node is selected, and a signal coordination mechanism is used to synchronize and align the shared features of each node. Based on the aligned shared features, the preliminary similarity of motion patterns between nodes is calculated. The preliminary similarity is evaluated in both the time domain and the frequency domain. If both time domain similarity and frequency domain similarity meet the preset conditions, the similarity is confirmed as valid; otherwise, the preliminary similarity is attenuated. The confirmed similarity is weighted and fused according to the reliability index of each node to obtain the weighted similarity between nodes. Then, a clustering algorithm is applied to the fused multi-node information to group similar motion patterns. If the similarity between nodes in a group is higher than a preset threshold, it is marked as a related node group; otherwise, it is marked as an independent processing node.

6. The method according to claim 1, characterized in that, In step S4, a unified filtering strategy applicable to the relevant nodes is determined, including: From the optimized artifact reference data of each node in the relevant node group, a shared artifact removal template is generated through principal component extraction. A template matching algorithm is used to determine the time offset parameter and amplitude scaling parameter by calculating the normalized cross-correlation coefficient between the shared artifact removal template and the optimized artifact reference data of each node. Combined with an adaptive cancellation filter type, a unified filtering strategy is formed. For nodes in the relevant node group, if their cross-correlation coefficient with the shared template is not lower than a preset similarity threshold, the unified filtering strategy is applied for artifact subtraction. If their cross-correlation coefficient is lower than the preset similarity threshold, they are determined to be independent nodes, and their own optimized artifact reference data is used as a personalized template for artifact subtraction. The parameters of the unified filtering strategy are evaluated and adjusted through cross-validation.

7. The method according to claim 6, characterized in that, In step S4, the real physiological signals of multiple sites are obtained synchronously, including: The remaining mixed signal of each node is obtained. The remaining mixed signal is the initial mixed signal minus the initially separated artifact components. According to the unified filtering strategy, combined with the shared artifact elimination template, time offset parameter and amplitude scaling parameter, the remaining mixed signal is subjected to adaptive cancellation filtering to remove residual artifacts. Then, the filtered output is used as the observation value to input the Kalman filter for state update and optimal estimation, filter out random noise, and perform time alignment to obtain the synchronously restored real physiological signals of multiple parts.

8. A wearable device-based animal physiological signal telemetry system for implementing the method as described in any one of claims 1-7, characterized in that, The system includes: The signal acquisition and processing unit is used to acquire photoelectric signals and inertial data from each node, and process the initial mixed signal using an adaptive filtering algorithm to obtain the initially separated physiological fluctuations and artifact components. The artifact modeling and optimization unit is used to obtain the motion mode features of the corresponding nodes based on the initially separated artifact components, and determine the dynamic correlation parameters between them and the changes in the optical path. If the dynamic correlation parameters exceed the preset threshold, a mapping model representing the mapping relationship between the motion mode features and the artifact components is constructed, and the mapping model is corrected by the Kalman filter algorithm to obtain optimized artifact reference data. The multi-node collaborative analysis unit is used to extract shared features of multiple nodes for optimized artifact reference data, calculate the similarity of motion patterns between nodes, verify the similarity based on both time and frequency domain judgments, and perform weighted fusion with node reliability indicators to obtain the weighted similarity between nodes and acquire relevant node groups and independent processing nodes. The filtering strategy generation unit is used to generate a shared artifact elimination template from the optimized artifact reference data of the relevant node group through principal component extraction, and to determine the time offset parameter and amplitude scaling ratio parameter to form a unified filtering strategy. The signal restoration output unit is used to perform adaptive cancellation filtering on the remaining mixed signal using the filtering strategy and combine it with Kalman filtering to obtain synchronously restored real physiological signals of multiple parts.

9. A telemetry device for animal physiological signals based on wearable devices, characterized in that, The device includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor invokes the instructions in the memory to cause the wearable device-based animal physiological signal telemetry device to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the method as described in any one of claims 1-7.

Images