Animal motion state sampling data processing method and system

Abstract

The invention discloses an animal motion state sampling data processing method and system, and belongs to the technical field of information, and the method comprises the steps: obtaining animal motion state data; according to the animal motion state data, constructing a prediction model based on deep learning; according to the collected animal motion state data, the sampling mode of the monitoring data is judged by calculating energy; when the sampling mode is a high-frequency sampling mode, predicting an animal state signal of a next sampling point by using the prediction model; collecting an animal state signal of the next sampling point at the current sampling frequency; calculating an error E between the predicted animal state signal and the actually collected animal state signal; adjusting the sampling frequency according to the error E; when the sampling mode is a low-frequency sampling mode, sampling by using a preset frequency; and collecting real-time animal motion state data according to the sampling frequency. Aiming at high energy consumption caused by constant sampling frequency in the prior art, energy consumption in animal movement monitoring is reduced.

Description

Technical Field

[0001] The present application relates to the field of information technology, and more specifically, to a method and system for processing sampled data of animal motion states. Background Art

[0002] In recent years, with the rapid development of Internet of Things technology, sensors have been increasingly widely used in fields such as animal behavior monitoring, health management, and environmental monitoring. As a typical Internet of Things device, the electronic ear tag for animal state monitoring integrates sensor nodes to achieve real-time monitoring of animal motion, health, and behavior states, providing technical support for the intelligent management of the livestock industry. However, the high-frequency operation of sensors is usually accompanied by high energy consumption, which is an urgent problem to be solved for electronic ear tag devices relying on battery power.

[0003] Currently, there have been many research results on electronic ear tags for intelligent detection of animal states. For example, a Chinese patent with the publication number CN114928628A discloses an intelligent livestock animal sign collection system, including an active electronic ear tag, an active ear tag networking base station, and an active ear tag data processing platform. This system collects data such as the body temperature, heart rate, and motion of animals through the electronic ear tag and uploads it to the data processing platform via the networking base station. However, this system uses a fixed-frequency data collection method and fails to dynamically adjust the sampling frequency according to the animal state, resulting in high-frequency sampling still being maintained when the animal is in a stationary or low-activity state, causing waste of energy resources.

[0004] A patent with the publication number CN118901610A mentions a monitoring method and system for electronic ear tags for livestock. This system calculates the activity intensity in real time based on a triaxial accelerometer and dynamically adjusts the data sampling frequency to optimize the storage efficiency. Although this system considers the adjustment of the sampling frequency to a certain extent, its sampling strategy is mainly based on simple threshold judgment, lacking the ability to predict the future state of animals, and not fully considering the constraint conditions of the Nyquist sampling theorem, which may lead to insufficient sampling or data loss when the signal frequency changes rapidly. In addition, this system does not adopt an effective compression technology in the data transmission link, and the energy consumption during the transmission process is still high.

[0005] Li Weifeng also discloses a temperature measurement ear tag system in "Design and System Construction of Intelligent Temperature Measurement Ear Tags for Beef Cattle Based on LoRa". However, this system mainly focuses on temperature monitoring, lacks comprehensive monitoring of animal motion states, and also faces the problems of fixed sampling frequency and high energy consumption.

[0006] In the prior art, there are common problems such as a constant sampling frequency, lack of state prediction ability, and redundant data transmission. These problems lead to high energy consumption and short battery life, making it difficult to achieve low energy consumption while ensuring data quality. Therefore, there is an urgent need to develop an animal state monitoring method that can intelligently adjust the sampling frequency and optimize energy utilization to solve the above technical problems. Summary of the Invention

[0007] 1. Technical Problems to be Solved

[0008] In view of the high energy consumption caused by the constant sampling frequency in the prior art, this application provides a method for processing sampled data of animal motion states. By means of a dual-mode adaptive sampling mechanism based on the energy of animal motion states and a dynamic adjustment of the sampling frequency driven by a deep learning prediction model, etc., the energy consumption in animal motion monitoring is reduced.

[0009] 2. Technical Solutions

[0010] The objectives of this application are achieved through the following technical solutions.

[0011] One aspect of this application provides a method for processing sampled data of animal motion states, including: obtaining animal motion state data; constructing a prediction model based on deep learning according to the animal motion state data, where the prediction model is used to predict the animal state signal value at a future moment; judging the sampling mode of the monitoring data by calculating energy according to the collected animal motion state data, and the sampling mode includes a high-frequency sampling mode and a low-frequency sampling mode; when the sampling mode is the high-frequency sampling mode, using the prediction model to predict the animal state signal at the next sampling point; collecting the animal state signal at the next sampling point at the current sampling frequency; calculating the error E between the predicted and the actually collected animal state signals; adjusting the sampling frequency according to the error E; when the sampling mode is the low-frequency sampling mode, sampling at a preset frequency; collecting real-time animal motion state data according to the sampling frequency.

[0012] Further, the animal motion state data includes position, acceleration, and temperature; the high-frequency sampling mode indicates that the animal is in a state of vigorous motion; the low-frequency sampling mode indicates that the animal is in a state of stable motion.

[0013] Further, constructing a prediction model based on deep learning according to the animal motion state data, where the prediction model is used to predict the animal state signal value at a future moment, includes: performing normalization preprocessing on the animal motion state data; extracting features using feature engineering according to the preprocessed animal motion state data; performing time series modeling using a deep learning algorithm according to the extracted features to obtain an initial prediction model; optimizing the hyperparameters of the initial prediction model using the Bayesian algorithm to obtain a prediction model based on deep learning.

[0014] Further, according to the collected animal motion state data, the sampling mode of the monitoring data is judged by calculating the energy. The sampling mode includes a high-frequency sampling mode and a low-frequency sampling mode, and it includes: collecting the animal motion state data X in a time window t,i ; calculating the energy value of the animal motion state data X t,i ; comparing the calculated energy value with a preset high-frequency threshold and a low-frequency threshold: when the energy value is greater than or equal to the high-frequency threshold, judging that the current sampling mode is the high-frequency sampling mode; when the energy value is less than or equal to the low-frequency threshold, judging that the current sampling mode is the low-frequency sampling mode; when the energy value is between the high-frequency threshold and the low-frequency threshold, maintaining the current sampling mode.

[0015] Further, to calculate the energy value of the animal motion state data X t,i , the following formula is adopted: where n represents the number of dimensions, and X t,i represents the value of the i-th dimension at time t.

[0016] Further, adjusting the sampling frequency according to the error E includes: at the current time t, predicting the animal state signal value at time t + 1 using a prediction model; collecting the actual animal state signal value at time t + 1 at the current sampling frequency f current ; calculating the error E between the predicted and the collected animal state signal values at time t + 1; calculating the adjusted sampling frequency f new according to the error E; judging whether the error E is greater than or equal to a first threshold. If so, incrementally update the prediction model using the actual animal state signal value at time t + 1; if not, maintain the current prediction model; specifically, incrementally update the prediction model using the actually collected signal value. This technical feature enables the system to continuously learn new motion patterns and behavioral characteristics of animals, improves the accuracy of the prediction model and the long-term stability of the system, and reduces the frequent sampling mode switching and energy waste caused by inaccurate prediction.

[0017] Calculate the change amplitude between the predicted value at time t + 1 and the actual value at time t. If the change amplitude exceeds a second set threshold, trigger the sampling mode judgment:

[0018] Compare the motion condition data of the next sampling point with the preset high-frequency and low-frequency sampling states:

[0019]

[0020] where: f t : the current sampling frequency, f L : the low-frequency sampling frequency

[0021] If the motion condition is lower than the low-frequency threshold, enter the low-frequency sampling mode; otherwise, maintain the current sampling frequency; where the second threshold is greater than the first threshold.

[0022] Further, according to the error E, calculate the adjusted sampling frequency f new , including: f new = max(2·f max , f current +α·E + β·R), where f max is the maximum signal frequency, f current is the current sampling frequency, E is the prediction error E, R is the signal change rate, and α and β are adjustment coefficients.

[0023] Further, the adjustment coefficients α and β include: constructing an objective function: f(α,β;Z) = w1·PA(α,β;Z) - w2·EC, where w1, w2 are the importance weights for weighing prediction accuracy and energy consumption; Z = {z1,z2,z3,…,z m} represents the current signal characteristic vector, including characteristics such as the mean of the frequency distribution, variance, amplitude change rate, etc.; using the Bayesian optimization algorithm, calculate the adjustment coefficients α and β under the current animal motion state by minimizing the objective function.

[0024] Further, at the current moment t, use the prediction model to predict the animal state signal value at t + 1, including: the calculation formula of the encoder: where: X is the input sequence, Y is the historical target sequence of the decoder, Z encoder is the output of the decoder, W is the weight matrix, M is the sparse mask matrix, and PosEnc(*) represents the position encoding.

[0025] The calculation formula of the decoder: where: Y is the input sequence of the decoder, is the weight matrix used in the decoder, H is the number of attention heads, W o is the final linear transformation matrix. Save the predicted signal value sequence Y Λ ; at the same time, save the true signal value Y of the next sampling point transmitted from the sensor end.

[0026] Further, when the sampling mode is the low-frequency sampling mode, sample using the preset frequency, including: setting a fixed sampling frequency f L : f L = 2f max +Δf, where Δf is the safety margin; f max represents the maximum frequency of the animal state signal; collect the animal state signal E at the current moment t according to the fixed sampling frequency f L t; Compare the value E at the previous moment t-1 with the true value E at this moment t as follows: If the change amplitude exceeds the third set threshold, trigger the sampling mode judgment: Compare the motion status data at this moment with the preset high-frequency and low-frequency sampling states: where: f t-1 : the sampling frequency at the previous moment, f H : the high-frequency sampling frequency. Among them, by comparing the signal change amplitude with different thresholds, trigger the switching of the sampling mode and the update of the prediction model. It balances the response speed and stability of the system, avoids data loss or signal distortion caused by sudden behavior changes of animals, and at the same time prevents the additional energy consumption caused by frequent mode switching. If the motion status is higher than the high-frequency threshold, enter the high-frequency sampling mode; otherwise, maintain the sampling frequency at the previous moment.

[0027] Another aspect of the present application also provides a processing system for sampling data of animal motion states, including: an acquisition module, which acquires monitoring data of animal motion states; a model construction module, which constructs a prediction model based on deep learning according to the monitoring data of animal motion states, and the prediction model is used to predict the animal state signal at the next moment; a mode judgment module, which judges the sampling mode of the monitoring data by calculating the energy value according to the monitoring data of animal motion states, and the sampling mode includes a high-frequency sampling mode and a low-frequency sampling mode; a processing module, when the sampling mode is the high-frequency sampling mode, uses the prediction model to predict the animal state signal at the next sampling point, calculates the error E between the predicted animal state signal and the actually acquired animal state signal, and adjusts the sampling frequency according to the error E; when the sampling mode is the low-frequency sampling mode, samples at a preset frequency; a switching module, which calculates the change amplitude of the error E and switches between the high-frequency sampling mode and the low-frequency sampling mode according to the change amplitude.

[0028] 3. Beneficial effects

[0029] Compared with the prior art, the advantages of the present application are:

[0030] The present application uses a deep learning prediction model to predict the animal state signal value at the next sampling point, and realizes the adaptive adjustment of the sampling frequency by calculating the error E between the predicted value and the actually acquired value. The system can increase the sampling frequency when the signal changes violently and decrease the sampling frequency when the signal changes gently, realizing the dynamic balance between data acquisition quality and energy consumption. The system always ensures that the sampling frequency is not lower than the minimum requirement to meet the Nyquist sampling theorem, fundamentally preventing signal distortion and aliasing phenomena. Description of the drawings

[0031] Figure 1This is a schematic diagram of the overall structure of the adaptive data collection and prediction system of this application;

[0032] Figure 2 This is a schematic diagram of the high-frequency mode data acquisition structure of this application;

[0033] Figure 3 This is a schematic diagram of the low-frequency mode data acquisition structure of this application;

[0034] Figure 4 This is a schematic diagram of the structure of the anomaly detection module of this application. DETAILED DESCRIPTION

[0035] The present application is described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0036] like Figure 1 As shown, first obtain the animal motion state data of a certain time window as the initial data set. Calculate the current motion state of the collected data, and judge whether to enter the high-frequency or low-frequency sampling mode according to the motion state: when the motion state exceeds the high-frequency threshold, enter the refined sampling mode, and when the motion state is lower than the low-frequency threshold, use energy-saving fixed frequency sampling; in the high-frequency mode, the system realizes multi-level intelligent decision-making: use the prediction model to predict the signal value at the next moment; calculate the error based on the actual collected value and the predicted value; when the error is greater than the set threshold, trigger the prediction model update; evaluate the degree of motion change and determine whether the sampling mode needs to be adjusted. Monitor the state of the low-frequency mode, use a fixed lower sampling frequency to collect data; calculate the degree of motion change, set the trigger condition; when the change exceeds the threshold, re-evaluate whether it is necessary to switch to the high-frequency mode. This application uses a dual-mode adaptive sampling mechanism (high-frequency / low-frequency mode) based on the energy of the animal's motion state to automatically adjust the sampling strategy according to the energy characteristics of the animal's real-time motion state. When the animal is in a stable or stationary state, switch to the low-frequency sampling mode, use a fixed lower sampling frequency, reduce unnecessary high-frequency sampling, reduce the operating frequency and data processing of the sensor circuit, and directly reduce the system energy consumption.

[0037] Specifically, historical data preprocessing and feature extraction, through the sensors installed on the animals to collect historical data of animal movement status, including multi-dimensional features such as position, acceleration, temperature, etc. Data cleaning algorithm is used to remove abnormal data and improve data quality. A moving median filter (window size is 5) is used to filter out high-frequency noise in acceleration data; the 3σ principle is applied to remove outliers, that is, data outside the range of (μ-3σ,μ+3σ) is marked as outliers, where μ is the data mean and σ is the standard deviation; linear interpolation is used to repair missing data to ensure data continuity.

[0038] Use feature engineering methods to refine key data features and improve model training efficiency. Key features include but are not limited to: calculating statistical features such as the mean, standard deviation, kurtosis, and skewness of each dimension of data; performing fast Fourier transform (FFT) on acceleration data to extract power spectrum density features; preferably, after feature extraction, use principal component analysis (PCA) to reduce dimensionality and retain the principal components that explain 95% of the variance to reduce the amount of subsequent model calculations.

[0039] A deep learning model is used for time series modeling to improve the ability to predict changes in animal status. Bayesian optimization is combined to adjust hyperparameters to balance prediction accuracy and computational cost with the optimal model. After training, the model parameters are distributed to each sensor node to achieve local reasoning capabilities.

[0040] During operation, the sensor nodes continuously collect real-time data and compare it with the model prediction value. If the error exceeds the set threshold, the model fine-tuning mechanism is triggered to perform online learning based on incremental data to improve adaptability. Using the federated learning mechanism, each sensor node regularly reports updated information to the central server to optimize the global model while avoiding high-energy data transmission.

[0041] The model predicts the energy consumption under different sampling modes and dynamically adjusts the sampling strategy to achieve a balance between data accuracy and power consumption. Figure 4 As shown, combined with the anomaly detection module, the sampling frequency is automatically adjusted during strenuous exercise or special circumstances to ensure that key data is not lost. Specifically, the sampling decision in this application is not only based on the current state, but also predicts future state changes through a deep learning model. The sampling strategy can be adjusted in advance before the state change occurs to achieve forward-looking sampling. The sampling frequency is no longer limited to a simple "high / low" binary choice, but the optimal sampling frequency is calculated based on the prediction error E.

[0042] Specifically, for data collection and motion state calculation, the sensor node collects a sequence of data X at the initial set sampling frequency. t,i , according to the energy calculation formula, evaluate the current motion status and save the data. The specific calculation formula is as follows: Among them, X t,i It represents the value of the i-th dimension at time t, and n is the number of dimensions.

[0043] Sampling mode interpretation, compare the saved motion data with the preset high-frequency and low-frequency sampling:

[0044]

[0045] Where: f H : High frequency sampling frequency, f L : low-frequency sampling frequency, S t: Current sampling frequency

[0046] If the motion condition exceeds the high-frequency threshold, enter the high-frequency sampling mode; if it is lower than the low-frequency threshold, enter the low-frequency sampling mode; otherwise, maintain the current sampling frequency. The value range of the high-frequency threshold is 0.7 - 0.8; the value range of the low-frequency threshold is 0.2 - 0.3.

[0047] Maintain or switch the sampling mode. If the sensor maintains the current sampling frequency. Otherwise, enter the high-frequency or low-frequency sampling mode according to the judgment.

[0048] Such as Figure 2 As shown, in the high-frequency sampling mode, for data processing and model prediction, in the high-frequency sampling mode, the collected data sequence is encoded and decoded through a prediction model based on the Transformer architecture to achieve automatic update of the sampling frequency.

[0049] Calculation formula of the encoder:

[0050]

[0051] Where: X is the input sequence, Y is the historical target sequence of the decoder, Z encoder is the output of the decoder, W is the weight matrix, M is the sparse mask matrix, represents the positional encoding.

[0052] Calculation formula of the decoder:

[0053]

[0054] Where: Y is the input sequence of the decoder, is the weight matrix used in the decoder, H is the number of attention heads, W o is the final linear transformation matrix.

[0055] Save the predicted signal value sequence Y Λ ; At the same time, save the true signal value Y of the next sampling point transmitted from the sensor side.

[0056] Adaptive sampling rate adjustment, dynamically adjust the sampling frequency according to the adaptive algorithm to reduce energy consumption and simultaneously meet the Nyquist sampling theorem:

[0057]

[0058] Among them, 2f max : Ensure that the sampling frequency always meets the Nyquist sampling theorem to prevent aliasing. f sampling(t): represents the current sampling frequency. α: prediction error adjustment coefficient, which controls the impact of prediction error on the sampling rate. β: change rate adjustment coefficient, which controls the impact of signal change rate on the sampling frequency. When the signal changes rapidly, the sampling frequency is increased in advance to adapt to the upcoming more complex signals. When the signal changes slowly, the sampling frequency can be reduced to save energy.

[0059] Specifically, the lower limit of the sampling frequency is set by max(2·f_max,...). This setting ensures that under any circumstances, the sampling frequency is not lower than twice the highest frequency of the signal, further preventing signal aliasing and distortion, and ensuring the integrity and accuracy of the collected data. In addition, the sampling frequency is dynamically adjusted to avoid energy waste caused by oversampling. When the signal is stable, the sampling frequency is automatically reduced to save energy, and the sampling frequency is only increased when necessary (when the signal changes rapidly or the prediction is inaccurate), achieving precise control of energy usage.

[0060] Considering that fixed weights cannot be fully applicable under different motion states, this solution uses Bayesian optimization to dynamically adjust the α and β coefficients. f(α,β;Z) = w1·PA(α,β;Z) - w2·EC, where w1 and w2 are the importance weights that balance prediction accuracy and energy consumption. Z = {z1,z2,z3,…,z m}: the current signal characteristic vector, including characteristics such as the mean of frequency distribution, variance, and amplitude change rate. This application uses the Bayesian optimization algorithm to dynamically adjust the α and β coefficients. By constructing an objective function, the system can automatically optimize and adjust parameters according to the motion characteristics of different animal individuals, improving the adaptability and generality of the system, reducing the workload of manual parameter adjustment, and achieving the optimal balance between energy consumption and accuracy under different motion modes.

[0061] Meanwhile, if |Y(t) - Y ∧ (t) ≥ ω|, that is, the error value between the predicted value and the true value is larger than the set value, then the true value is passed into the prediction model for training to update the prediction model.

[0062] Specifically, the traditional fixed-parameter method cannot adapt to the diverse motion patterns of animals. For example, the signal characteristics of the same cow are significantly different during rest, foraging, and running. If the fixed α and β parameters are optimized for the rest state, there will be insufficient sampling during the running state; conversely, there will be energy waste during the rest state. Bayesian optimization fundamentally solves the technical limitation of the "one-size-fits-all" parameter setting by intelligently exploring the parameter space to automatically find the optimal parameter configuration for different motion patterns.

[0063] In addition, by introducing the signal characteristic vector Z = {z1,z2,z3,…,z m} (including frequency distribution mean, variance, amplitude change rate and other characteristics), the system can perceive the key characteristics of the animal's current movement state. This makes the parameter optimization process no longer a blind adjustment, but based on rich signal context information, so that the algorithm can select the most matching parameters according to the specific signal pattern. For example, when the signal presents high frequency and high variance characteristics, the algorithm will automatically increase the β value to enhance the sensitivity to changes.

[0064] Real-time monitoring of the degree of drastic changes in movement, and the predicted value E of the next sampling point t and the true value E at this moment t-1 For comparison: If the change amplitude exceeds the set threshold ε, the sampling mode conversion is triggered:

[0065] like Figure 3 As shown in the low-frequency sampling mode, a fixed sampling frequency is set in the low-frequency mode to ensure that the Nyquist sampling theorem is met and aliasing is avoided: f sampling =f L , where: f L =2f max +Δf, Δf is a safety margin to prevent insufficient sampling due to critical signal changes.

[0066] Similarly, the anomaly detection module is set to take the predicted value E of the next sampling point t and the true value E at this moment t-1 For comparison: If the change amplitude exceeds the set threshold ε, the sampling mode conversion is triggered:

[0067]

[0068] like Figure 1 As shown, the compressed sensing transmission module performs sparse representation of the signal, performs sparse representation of the collected animal state data, and converts the signal into a discrete cosine transform (DCT) domain to obtain its sparse representation.

[0069] Compressed sampling uses observation matrices to compress sparse signals and obtain observation values that are less than the original data volume, thereby reducing the amount of transmitted data and reducing the energy consumption of sensor nodes. This application sparsely represents the collected animal status data, converts the signal to the frequency domain through methods such as discrete cosine transform (DCT), and uses observation matrices to compress and sample sparse signals to obtain observation values that are less than the original data volume. From the data transmission level, the amount of data that needs to be transmitted is significantly reduced, and the energy consumption of sensor nodes during wireless communication is reduced. At the same time, the original signal is restored at the receiving end through a compressed sensing reconstruction algorithm to ensure the integrity and accuracy of the data.

[0070] Signal reconstruction: At the receiving end, the reconstruction algorithm of compressive sensing is used, combined with the observation matrix and the received observation values, to reconstruct the original signal.

[0071] The present invention and its implementation manners are schematically described above. The description is not restrictive. Without departing from the spirit or basic features of the present application, the present application can be implemented in other specific forms. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. Any reference numeral in the claims should not limit the claimed claim. Therefore, if those of ordinary skill in the art are inspired by it and design similar structural manners and embodiments to the technical solution without creative efforts without departing from the purpose of the present invention, they shall fall within the protection scope of this patent. In addition, the term "comprising" does not exclude other elements or steps, and the term "a" before an element does not exclude including "a plurality of" such elements. The plurality of elements stated in the product claims can also be implemented by one element through software or hardware. The terms "first", "second", etc. are used to denote names and do not denote any particular order.

Claims

1. A method for processing sampled data of an animal's motion state, characterized in that, Including: Obtain animal motion state data; Construct a prediction model based on deep learning according to the animal motion state data, where the prediction model is used to predict the animal state signal value at a future moment; According to the collected animal motion state data, judge the sampling mode of the monitoring data by calculating energy, and the sampling mode includes a high-frequency sampling mode and a low-frequency sampling mode; When the sampling mode is the high-frequency sampling mode, use the prediction model to predict the animal state signal at the next sampling point; collect the animal state signal at the next sampling point at the current sampling frequency; calculate the error E between the predicted and the actually collected animal state signals; adjust the sampling frequency according to the error E; When the sampling mode is the low-frequency sampling mode, then sample at a preset frequency; Collect real-time animal motion state data according to the sampling frequency.

2. The processing method of animal motion state sampling data according to claim 1, characterized in that: The animal motion state data includes position, acceleration and temperature; The high-frequency sampling mode indicates that the animal is in a state of intense motion; The low-frequency sampling mode indicates that the animal is in a state of stable motion.

3. The processing method of animal motion state sampling data according to claim 2, characterized in that: Constructing a prediction model based on deep learning includes: Preprocess the animal motion state data; Extract features using feature engineering according to the preprocessed animal motion state data; Perform time series modeling using a deep learning algorithm according to the extracted features to obtain an initial prediction model; Optimize the hyperparameters of the initial prediction model using the Bayesian algorithm to obtain a prediction model based on deep learning.

4. The processing method of animal motion state sampling data according to claim 2, characterized in that: Judging the sampling mode of the monitoring data by calculating energy includes: Collect the animal motion state data X within a time window t,i , X t,i represents the value of the i-th dimension at time t; Calculate the animal motion state data X t,i of the energy value; Compare the calculated energy value with a preset high-frequency threshold and a low-frequency threshold: When the energy value is greater than or equal to the high-frequency threshold, judge that the current sampling mode is the high-frequency sampling mode; When the energy value is less than or equal to the low-frequency threshold, judge that the current sampling mode is the low-frequency sampling mode; When the energy value is between the high-frequency threshold and the low-frequency threshold, maintain the current sampling mode.

5. The processing method of animal motion state sampling data according to claim 4, characterized in that: Calculate the animal motion state data X t,i of the energy value, using the following formula: Where n represents the number of dimensions.

6. The processing method of animal motion state sampling data according to claim 2, characterized in that: Adjusting the sampling frequency according to the error E includes: At the current moment t, use the prediction model to predict the animal state signal value at the moment t + 1; With the current sampling frequency f current Collect the actual animal state signal value at time t + 1; Calculate the error E between the predicted and the collected animal state signal values at the moment t + 1; Calculate the adjusted sampling frequency f according to the error E new ; Judge whether the error E is greater than or equal to the first threshold. If so, incrementally update the prediction model using the actual animal state signal value at the moment t + 1; if not, maintain the current prediction model; Calculate the change amplitude between the predicted value at the moment t + 1 and the actual value at the moment t. If the change amplitude exceeds the second set threshold, trigger the sampling mode judgment: Compare the motion status data E(x t ) of the next sampling point with the preset high-frequency and low-frequency sampling states: If the motion condition data E(x t ) is lower than the low-frequency threshold θ L , then enter the low-frequency sampling mode; if the motion condition data E(x t ) is higher than or equal to the high-frequency threshold θ H , then maintain the current sampling frequency; where the second threshold is greater than the first threshold.

7. The processing method of animal motion state sampling data according to claim 6, characterized in that: Calculate the adjusted sampling frequency f new , using the following formula: f new = max(2·f max , f current + α·E + β·R) where f max is the maximum signal frequency, f current is the current sampling frequency, E is the prediction error E, R is the signal change rate, and α and β are adjustment coefficients.

8. The processing method of the animal motion state sampling data according to claim 7, characterized in that: Adjusting coefficients α and β includes: Constructing an objective function: f(α, β; Z) = w1·PA(α, β; Z) - w2·EC Among them, w1 and w2 are the importance weights for balancing prediction accuracy and energy consumption; Z = {z1, z2, z3, …, z m} represents the current signal feature vector, including the mean value of frequency distribution, variance, and amplitude change; Using the Bayesian optimization algorithm, by minimizing the objective function, calculate the adjustment coefficients α and β in the current animal motion state.

9. The processing method of the animal motion state sampling data according to claim 2, characterized in that: When the sampling mode is the low-frequency sampling mode, sampling is performed using a preset frequency, including: Set a fixed sampling frequency f L : f L = 2f max + Δf where Δf is the safety margin; f max represents the maximum frequency of the animal state signal; According to the fixed sampling frequency f L Collect the animal state signal E at the current moment t t ; Compare the value E at the previous moment t-1 with the true value E at this moment t as follows: If the change amplitude exceeds the third set threshold, trigger the sampling mode judgment: Compare the motion status data E(x t ) at this moment with the preset high-frequency and low-frequency sampling states: If the motion status data E(x t ) is higher than the high-frequency threshold, enter the high-frequency sampling mode; if the motion status data E(x t ) is lower than the high-frequency threshold, maintain the sampling frequency of the previous moment.

10. A processing system for sampling data of an animal's motion state, characterized in that, Including: An acquisition module that obtains the monitoring data of the animal motion state; A model construction module that constructs a prediction model based on deep learning according to the monitoring data of the animal motion state, and the prediction model is used to predict the animal state signal at a future moment; A mode judgment module that judges the sampling mode of the monitoring data by calculating the energy value according to the monitoring data of the animal motion state, and the sampling mode includes a high-frequency sampling mode and a low-frequency sampling mode; A processing module that, when the sampling mode is the high-frequency sampling mode, uses the prediction model to predict the animal state signal at the next sampling point, calculates the error E between the predicted animal state signal and the actually collected animal state signal, and adjusts the sampling frequency according to the error E; When the adopted mode is the low-frequency sampling mode, sampling is performed using a preset frequency; A switching module that calculates the change amplitude of the error E and switches between the high-frequency sampling mode and the low-frequency sampling mode according to the change amplitude.

Images