Pet health and behavior monitoring method based on multi-sensor data

Abstract

The invention provides a pet health and behavior monitoring method based on multi-sensor data, and the method comprises the steps: carrying out the cleaning processing of a preliminary multi-dimensional data set, eliminating environment interference and sensor errors, and obtaining a standardized data set; hierarchical classification is carried out on the standardized data set, and key feature values of data of all dimensions are extracted; if the dynamic change trend exceeds a preset threshold range, marking the abnormal data points, and generating a marked abnormal data set; performing context correlation analysis on the marked abnormal data set to determine the correlation between the abnormal data and the health risk or behavior abnormality; performing comprehensive scoring on the whole health condition of the pet according to a correlation result, and obtaining a health risk level and a behavior abnormal probability; and generating a health and behavior report containing the trend chart and the risk prompt, and outputting the health and behavior report to the terminal equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of behavior recognition, and in particular, to a method for monitoring pet health and behavior based on multi-sensor data. Background Art

[0002] As an important innovative direction in the field of pet health management, the pet collar system undertakes the key mission of improving the quality of life of pets and the care ability of owners. With the increasing status of pets in families, how to pay attention to pet health and behavior in real time through technical means has become a research topic that cannot be ignored. The importance of this field lies in that it not only concerns the well-being of pets, but also directly affects the emotional investment and life experience of owners. However, most of the current pet management tools on the market have a single function, often only providing simple positioning or activity tracking, lacking comprehensive recording and in-depth analysis of the daily life data of pets. This limitation makes it difficult for owners to fully understand the health status of pets, let alone discover potential problems in time and miss the best intervention opportunity. In this context, the core challenges in this field have gradually emerged. The first and foremost is how to achieve accurate collection of multi-dimensional data of pets, including basic information such as activity level, sleep quality, diet, and excretion frequency, as well as specific behavior data such as barking and digging. Due to the variety and complexity of data types and sources, the comprehensiveness and accuracy of collection have become a major problem. And this problem further gives rise to another technical difficulty, that is, how to integrate and analyze the massive amount of collected data through effective algorithms to form an intuitive and instructive health and behavior trend report. If the depth of data analysis cannot be solved, simple data recording will lose its practical value and it is difficult to provide practical help for owners and veterinarians.

[0003] Therefore, how to design a pet collar system that can comprehensively and accurately record the multi-dimensional data of the daily life of pets and generate targeted health and behavior reports through intelligent analysis has become the key problem that this research urgently needs to overcome. Summary of the Invention

[0004] The present invention provides a method for monitoring pet health and behavior based on multi-sensor data, mainly including: The original data stream containing activity level, sleep status, and specific behaviors is obtained through a multi-sensor module deployed in a pet collar to generate a preliminary multi-dimensional data set; the preliminary multi-dimensional data set is cleaned to eliminate environmental interference and sensor errors, resulting in a standardized data group; by performing hierarchical classification on the standardized data group, the key feature values of each dimension of data are extracted; based on the key feature values, the dynamic change trends of each dimension of data are detected to judge the periodic changes in activity level, sleep quality, and behavior patterns; if the dynamic change trend exceeds the preset threshold range, the abnormal data points are marked to generate an annotated abnormal data set; through context correlation analysis of the annotated abnormal data set, the correlation between abnormal data and health risks or behavioral abnormalities is determined; based on the correlation results, a comprehensive score of the overall health status of the pet is obtained to acquire the health risk level and the probability of behavioral abnormalities; a health and behavior report containing trend charts and risk warnings is generated and output to the terminal device.

[0005] Furthermore, the original data stream containing activity level, sleep status, and specific behaviors is obtained through a multi-sensor module deployed in a pet collar to generate a preliminary multi-dimensional data set, including: setting independent acquisition frequencies and accuracy parameters for different data types to obtain the original data stream covering activity level, sleep status, and specific behaviors; using a preset filtering mechanism to denoise the original data stream and applying different smoothing strategies for various data types to obtain a processed clean data set; according to the change characteristics of the activity data in the clean data set, applying a classification algorithm to classify the data and judge the distribution of the pet's activity patterns at different time periods; if the activity pattern distribution shows abnormal fluctuations, in-depth feature extraction is performed on the original data of the abnormal time period to obtain potential behavior pattern identifiers; by matching and comparing the behavior pattern identifiers with preset specific behavior templates, the occurrence probability of specific behaviors is determined.

[0006] Furthermore, the preliminary multi-dimensional data set is cleaned to eliminate environmental interference and sensor errors, resulting in a standardized data group, including: performing an initial screening on the preliminary multi-dimensional data set to obtain a preliminarily filtered data group; applying a denoising algorithm to identify environmental interference in the preliminarily filtered data group, and if abnormal fluctuations are identified, smoothing the abnormal part to obtain a denoised data set; for the denoised data set, applying a calibration algorithm to correct sensor errors, and if it is detected that the data deviation exceeds the preset threshold, adjusting the deviated part to determine a calibrated data set; extracting key features from the calibrated data set, analyzing the data distribution using preset rules, and judging whether there is residual interference to obtain a data group after feature analysis; performing standardization processing on the data group after feature analysis to uniformly adjust the data range to obtain a standardized data set.

[0007] Furthermore, by classifying the standardized data set hierarchically and extracting the key characteristic values of each dimension of data, including: establishing characteristic extraction models for activity monitoring, sleep analysis, and behavior recording respectively, performing hierarchical processing on the standardized data set to obtain the first characteristic value of activity data, the second characteristic value of sleep state data, and the third characteristic value of specific behavior data; analyzing the distribution law of pet activity based on the first characteristic value; evaluating the persistence and depth of pet sleep state based on the second characteristic value; identifying the occurrence frequency and pattern of pet specific behaviors based on the third characteristic value; integrating the first characteristic value, the second characteristic value, and the third characteristic value into a key characteristic value set for subsequent dynamic change detection.

[0008] Furthermore, according to the key characteristic values, detect the dynamic change trends of each dimension of data, and judge the periodic changes of activity, sleep quality, and behavior patterns, including: using a preset time series analysis algorithm to process the key characteristic values to obtain the first change trend of activity data, the second change trend of sleep quality data, and the third change trend of behavior pattern data; judging the periodic fluctuations of pet activity based on the first change trend; analyzing the stability of pet sleep quality based on the second change trend; identifying the regular changes of pet behavior patterns based on the third change trend; if the first change trend, the second change trend, or the third change trend exceeds the preset threshold range, record the corresponding time period and data points to generate an abnormal change record.

[0009] Furthermore, if the dynamic change trend exceeds the preset threshold range, mark the abnormal data points to generate an annotated abnormal data set, including: identifying the data points whose dynamic change trend exceeds the preset threshold range through an anomaly detection module to obtain the distribution positions and time periods of the abnormal data points; classifying and marking according to the distribution positions and time periods of the abnormal data points in combination with the corresponding data types to generate a first abnormal data subset, a second abnormal data subset, and a third abnormal data subset; integrating the first abnormal data subset, the second abnormal data subset, and the third abnormal data subset to generate an annotated abnormal data set; storing the annotated abnormal data set in a preset database for subsequent context correlation analysis.

[0010] Furthermore, by performing context - correlation analysis on the labeled abnormal data set, the correlation between the abnormal data and health risks or behavioral abnormalities is determined, including: extracting the feature information of abnormal data points from the labeled abnormal data set, comparing and analyzing in combination with historical data records and current environmental information to obtain the first correlation value; judging the correlation between the abnormal data points and health risks according to the first correlation value to obtain the health - risk correlation result; judging the correlation between the abnormal data points and behavioral abnormalities according to the first correlation value to obtain the behavioral - abnormality correlation result; integrating the health - risk correlation result and the behavioral - abnormality correlation result to generate a comprehensive correlation analysis report; and determining the final correlation classification of the abnormal data according to the comprehensive correlation analysis report.

[0011] Furthermore, according to the correlation result, a comprehensive score of the overall health condition of the pet is obtained to get the health - risk level and the probability of behavioral abnormalities, including: inputting the correlation result into a preset health - status evaluation model to obtain a preliminary score value of the pet's health condition; adjusting the scoring weights in combination with the key feature values of activity level, sleep quality, and behavior pattern according to the preliminary score value to generate a final comprehensive score; determining the health - risk level according to the final comprehensive score; calculating the probability of behavioral abnormalities according to the final comprehensive score; storing the health - risk level and the probability of behavioral abnormalities in a preset database, and generating a corresponding evaluation record for subsequent report generation.

[0012] The technical solution provided by the embodiments of the present invention may include the following beneficial effects: The present invention discloses a multi - sensor - based pet health monitoring method. By deploying a multi - sensor module in a pet collar, raw data of activity level, sleep state, and specific behaviors are collected. After denoising, calibration, and feature extraction, activity - monitoring, sleep - analysis, and behavior - recording models are established. The present invention uses time - series analysis algorithms to detect the dynamic change trends of data in each dimension, mark abnormal data points and perform context - correlation analysis, and evaluate the health risks and the probability of behavioral abnormalities in combination with historical records. Finally, an intuitive health - behavior report including trend charts and risk warnings is generated and output to a terminal device. The present invention realizes the comprehensive monitoring of the pet's health condition and abnormal early warning, provides a scientific basis for pet owners' health management, helps to timely discover potential health problems, and improves the quality of life of pets. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a flowchart of a pet health and behavior monitoring method based on multi - sensor data of the present invention.

[0014] Figure 2 It is a schematic diagram of a pet health and behavior monitoring method based on multi - sensor data of the present invention.

[0015] Figure 3Another schematic diagram of a pet health and behavior monitoring method based on multi-sensor data according to the present invention.

[0016] Figure 4 Another schematic diagram of a pet health and behavior monitoring method based on multi-sensor data according to the present invention. Detailed implementation manners

[0017] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] Please refer to Figures 1 to 4 As shown, a pet health and behavior monitoring method based on multi-sensor data in this embodiment may specifically include: S101. Obtain the original data stream including activity amount, sleep state and specific behaviors through the multi-sensor module deployed in the pet collar, and generate a preliminary multi-dimensional data set.

[0019] Deployment and data acquisition of the multi-sensor module.

[0020] The accelerometer built in the pet collar collects three-dimensional axial motion data at a frequency of 50 Hz, and differentiates basic activity amounts such as stillness, walking, running, etc. through the dynamic threshold algorithm.

[0021] At the same time, the infrared sensor monitors the change of body surface temperature at a low frequency of 1 Hz, and combines the gyroscope attitude data to judge the sleep stage (light sleep, deep sleep, rapid eye movement).

[0022] The specific behavior monitoring adopts a customized strategy. For example, the barking frequency is captured by the sound sensor (sampling rate 8 kHz), and the scratching action is identified by cooperating with the high-frequency peak detection of the accelerometer (100 Hz). Synchronization and fusion of multi-source data streams; the hardware time stamp is used to align the data of each sensor. For example, the positioning data of the accelerometer and the GPS module are fused through the Kalman filter to correct the displacement error caused by signal delay. The temperature data and the activity amount data are aggregated according to the time window (10-second interval) to eliminate the clock drift between sensors.

[0023] The synchronization error is controlled within ±5 ms to ensure the timing correlation accuracy of behavior events (such as running immediately after barking). Generation logic of the multi-dimensional data set. After the original data stream is preprocessed by the edge computing node, it is classified and stored according to the behavior dimension: the basic activity amount data is compressed into the mean / variance per minute, the sleep data retains the original waveform characteristics (segmented every 30 seconds), and the specific behavior data retains the 10-second high-precision original records before and after the event trigger. The data set adopts a hierarchical structure, with the bottom layer being the millisecond-level original data, the middle layer being the second-level feature data, and the top layer being the minute-level statistical indicators. Dynamic adjustment mechanism for acquisition parameters.

[0024] The sampling rate is adaptively adjusted according to the ambient noise level. For example, the accelerometer frequency is reduced to 20Hz at night to reduce power consumption, and it automatically switches to 100Hz mode when intense movement is detected (variance exceeds 0.5g). The sound sensor is turned off in a quiet environment and is only activated after identifying ambient sound above 40dB, balancing data quality and energy consumption. Data quality control and exception handling.

[0025] Design redundancy strategies for sensor failures, such as automatically switching to the backup magnetometer attitude estimation when the gyroscope output exceeds the ±2000° / s range for three consecutive times; Linear interpolation of front and back windows (window size 5 seconds) was used for missing data, and continuous missing segments exceeding 10 seconds were marked as invalid segments.

[0026] When the temperature data exceeds the physiological range of 35-42℃, the recalibration process is triggered. The technical value of multi-dimensional data collection.

[0027] The accuracy of behavior recognition is improved through cross-validation of high-frequency activity data and low-frequency physiological data (for example, a temperature rise of 0.5°C / min during running is the normal threshold). The sleep stage judgment combines the number of body movements (less than 5 times per hour is deep sleep) and the temperature drop slope (0.1°C / 10min), which is 23% more accurate than a single sensor. The specific behavior template matching uses a dynamic time warping algorithm, and the distinction between similar actions (such as scratching and eating) reaches 89%.

[0028] S102, cleaning the preliminary multi-dimensional data set to eliminate environmental interference and sensor errors, and obtain a standardized data group; by stratifying and classifying the standardized data group, extracting key feature values ​​of each dimensional data.

[0029] When cleaning the preliminary multi-dimensional data set, it is first necessary to identify typical patterns of environmental interference and sensor errors.

[0030] For example, in the heart rate data collected by wearable devices, environmental interference may manifest as instantaneous high-frequency noise caused by exercise, and its numerical range usually exceeds the normal heart rate range (such as a sudden jump to more than 200 beats / minute). At this time, a sliding window mean filtering algorithm can be used, setting the window size to 5 data points, and replacing the abnormal values ​​in the window with the weighted average of adjacent normal values.

[0031] For sensor errors, if baseline drift is detected in the accelerometer data (such as a continuous offset of more than 0.2g on the Z axis), the zero-bias calibration algorithm is used to perform linear compensation on the dynamic data with reference to the baseline value in the static state. The hierarchical classification of the standardized data group needs to be combined with the design of feature extraction logic in combination with specific application scenarios.

[0032] Taking sleep analysis as an example, the key features extracted from triaxial acceleration data include: exercise intensity (the number of times the magnitude of the acceleration vector exceeds 0.5g per minute), body position switching frequency (the number of times the angle changes by more than 30 degrees per hour). For the dimension of behavior records, gait features need to be extracted, such as step frequency (a stable gait when the standard deviation of the number of steps per minute is less than 3), and swing amplitude (a normal walking when the hip joint angle change range is between 15 degrees and 25 degrees).

[0033] The activity monitoring layer needs to distinguish between high-intensity and low-intensity exercises, and achieve double verification through the coefficient of variation of heart rate (HRV standard deviation greater than 50ms for high intensity) combined with the acceleration peak value (exceeding 2g for more than 10 seconds). The core of eliminating environmental interference lies in distinguishing the frequency domain characteristics of signals and noises.

[0034] For example, the environmental light interference of a photoelectric heart rate sensor is manifested as 50Hz power frequency noise, which can be attenuated in the 48 - 52Hz frequency band through a band-stop filter. For the periodic fluctuations of the temperature sensor caused by ventilation conditions, the moving average algorithm (with a period set to 30 seconds) can effectively smooth short-term fluctuations and retain the true body temperature trend (with the error controlled within ±0.1℃).

[0035] Sensor error calibration requires establishing an error model. For example, the non-linear error of a blood oxygen sensor can be corrected by piecewise linear interpolation. When the original reading is in the 90% - 94% range, a pre-calibrated compensation coefficient of 1.03 is used for correction.

[0036] The extraction of key feature values depends on the statistical characteristics of data distribution.

[0037] In sleep stage classification, when the power ratio of δ waves (0.5 - 4Hz) exceeds 20% of the total power, it is marked as the deep sleep stage; the rapid eye movement stage is jointly determined by the electrooculogram signal amplitude (greater than 50μV) and the electromyogram signal entropy value (less than 0.8).

[0038] The recognition of sitting postures in behavior records requires integrating pressure sensor data (the symmetry degree of hip pressure distribution is greater than 85%) and the gyroscope pitch angle (-10 degrees to +10 degrees).

[0039] When performing hierarchical classification, the temporal correlation of each dimension feature needs to be quantified. For example, in activity monitoring, only when three consecutive high heart rate intervals (>120 beats per minute) are accompanied by high acceleration (>1.5g) can it be determined as a valid exercise event. Standardization processing needs to solve the problem of unifying the dimensions of multiple sensors. For example, normalizing the amplitude of the photoplethysmogram (PPG) signal to the 0 - 1 range for collaborative analysis with the data of the ambient light sensor (0 - 1000lux).

[0040] The data consistency check adopts the dynamic threshold method. If the difference in blood oxygen saturation between adjacent data points exceeds 5%, then linear interpolation based on time series is triggered.

[0041] Before storing in the database, ensure the integrity of each batch of data through hash verification (such as SHA-256). An abnormal data packet (verification failure rate > 0.1%) triggers an automatic retransmission mechanism.

[0042] S103. Detect the dynamic change trends of data in each dimension according to the key feature values, and judge the periodic changes of activity level, sleep quality and behavior patterns.

[0043] S104. If the dynamic change trend exceeds the preset threshold range, mark the abnormal data points and generate an annotated abnormal data set.

[0044] S105. Determine the correlation between the abnormal data and health risks or behavioral abnormalities by performing context correlation analysis on the annotated abnormal data set.

[0045] S106. Comprehensively score the overall health status of the pet according to the correlation results, obtain the health risk level and the probability of behavioral abnormalities; generate a health and behavior report containing trend charts and risk warnings, and output it to the terminal device.

[0046] S107. Obtain the original data stream containing activity level, sleep status and specific behaviors through the multi-sensor module deployed in the pet collar, and generate a preliminary multi-dimensional data set, including: setting independent acquisition frequencies and precision parameters for different data types to obtain the original data stream covering activity level, sleep status and specific behaviors; using a preset filtering mechanism to denoise the original data stream, applying different smoothing strategies for various data types to obtain a processed clean data set; according to the change characteristics of the activity data in the clean data set, applying a classification algorithm to classify the data and judge the activity pattern distribution of the pet at different time periods; if the activity pattern distribution shows abnormal fluctuations, perform in-depth feature extraction on the original data of the abnormal time period to obtain potential behavior pattern identifiers; match and compare the behavior pattern identifiers with the preset specific behavior templates to determine the occurrence probability of specific behaviors.

[0047] S108. Clean the preliminary multi-dimensional data set to eliminate environmental interference and sensor errors, and obtain a standardized data set, including: performing an initial screening on the preliminary multi-dimensional data set to obtain a preliminarily filtered data set; applying a denoising algorithm to identify environmental interference in the preliminarily filtered data set. If abnormal fluctuations are identified, smooth the abnormal part to obtain a denoised data set; for the denoised data set, apply a calibration algorithm to correct sensor errors. If it is detected that the data deviation exceeds a preset threshold, adjust the deviated part to determine a calibrated data set; extract key features from the calibrated data set, analyze the data distribution using preset rules, and determine whether there is residual interference to obtain a data set after feature analysis; perform standardization processing on the data set after feature analysis, uniformly adjust the data range, and obtain a standardized data set.

[0048] S109. Extract the key feature values of each dimension data by performing hierarchical classification on the standardized data set, including: establishing feature extraction models for activity monitoring, sleep analysis, and behavior recording respectively, performing hierarchical processing on the standardized data set to obtain the first feature value of activity data, the second feature value of sleep state data, and the third feature value of specific behavior data; analyzing the distribution law of the pet's activity level according to the first feature value; evaluating the persistence and depth of the pet's sleep state according to the second feature value; identifying the occurrence frequency and pattern of the pet's specific behavior according to the third feature value; integrating the first feature value, the second feature value, and the third feature value into a key feature value set for subsequent dynamic change detection.

[0049] S1010. Detect the dynamic change trends of each dimension data according to the key feature values, and judge the periodic changes of activity level, sleep quality, and behavior pattern, including: using a preset time series analysis algorithm to process the key feature values to obtain the first change trend of activity data, the second change trend of sleep quality data, and the third change trend of behavior pattern data; judging the periodic fluctuations of the pet's activity level according to the first change trend; analyzing the stability of the pet's sleep quality according to the second change trend; identifying the regular changes of the pet's behavior pattern according to the third change trend; if the first change trend, the second change trend, or the third change trend exceeds the preset threshold range, record the corresponding time period and data points to generate an abnormal change record.

[0050] S1011. If the dynamic change trend exceeds the preset threshold range, mark the abnormal data points to generate an annotated abnormal data set, including: identifying the data points whose dynamic change trend exceeds the preset threshold range through an anomaly detection module to obtain the distribution positions and time periods of the abnormal data points; classifying and marking according to the distribution positions and time periods of the abnormal data points in combination with the corresponding data types to generate a first abnormal data subset, a second abnormal data subset, and a third abnormal data subset; integrating the first abnormal data subset, the second abnormal data subset, and the third abnormal data subset to generate an annotated abnormal data set; storing the annotated abnormal data set in a preset database for subsequent context correlation analysis.

[0051] S1012. Determine the correlation between the abnormal data and health risks or behavioral abnormalities by performing context correlation analysis on the annotated abnormal data set, including: extracting the feature information of the abnormal data points from the annotated abnormal data set, comparing and analyzing it in combination with historical data records and current environmental information to obtain a first correlation value; judging the correlation between the abnormal data points and health risks according to the first correlation value to obtain a health risk correlation result; judging the correlation between the abnormal data points and behavioral abnormalities according to the first correlation value to obtain a behavioral abnormality correlation result; integrating the health risk correlation result and the behavioral abnormality correlation result to generate a comprehensive correlation analysis report; determining the final correlation classification of the abnormal data according to the comprehensive correlation analysis report.

[0052] S1013. Comprehensively score the overall health status of the pet according to the correlation result to obtain a health risk level and a probability of behavioral abnormality, including: inputting the correlation result into a preset health status evaluation model to obtain a preliminary score value of the pet's health status; adjusting the scoring weights according to the preliminary score value in combination with the key feature values of activity level, sleep quality, and behavior pattern to generate a final comprehensive score; determining the health risk level according to the final comprehensive score; calculating the probability of behavioral abnormality according to the final comprehensive score; storing the health risk level and the probability of behavioral abnormality in a preset database and generating corresponding evaluation records for subsequent report generation.

[0053] The above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A method for monitoring pet health and behavior based on multi-sensor data, characterized in that, Including: Obtain the original data stream containing activity level, sleep status, and specific behaviors through the multi-sensor module deployed in the pet collar, and generate a preliminary multi-dimensional data set; Clean the preliminary multi-dimensional data set to eliminate environmental interference and sensor errors, and obtain a standardized data group; Extract the key feature values of each dimension data by performing hierarchical classification on the standardized data group; Detect the dynamic change trends of each dimension data according to the key feature values, and judge the periodic changes of activity level, sleep quality, and behavior patterns; If the dynamic change trend exceeds the preset threshold range, mark the abnormal data points to generate an annotated abnormal data set; Determine the correlation between the abnormal data and health risks or behavioral abnormalities through context correlation analysis of the annotated abnormal data set; Comprehensively score the overall health status of the pet according to the correlation results to obtain the health risk level and the probability of behavioral abnormalities; Generate a health and behavior report containing trend charts and risk warnings, and output it to the terminal device.

2. The method for monitoring pet health and behavior based on multi-sensor data according to claim 1, wherein Obtain the original data stream containing activity level, sleep status, and specific behaviors through the multi-sensor module deployed in the pet collar, and generate a preliminary multi-dimensional data set, including: setting independent acquisition frequencies and accuracy parameters for different data types to obtain the original data stream covering activity level, sleep status, and specific behaviors; Denoise the original data stream using a preset filtering mechanism, and apply different smoothing strategies for various data types to obtain a processed clean data set; According to the change characteristics of the activity data in the clean data set, apply a classification algorithm to classify the data and judge the activity pattern distribution of the pet in different time periods; If the activity pattern distribution shows abnormal fluctuations, perform in-depth feature extraction on the original data of the abnormal time period to obtain potential behavior pattern identifiers; Match and compare the behavior pattern identifiers with the preset specific behavior templates to determine the occurrence probability of specific behaviors.

3. The method for monitoring pet health and behavior based on multi-sensor data according to claim 1, characterized in that, Clean the preliminary multi-dimensional data set to eliminate environmental interference and sensor errors, and obtain a standardized data group, including: performing an initial screening on the preliminary multi-dimensional data set to obtain a preliminarily filtered data group; Identify environmental interference in the preliminarily filtered data group using a denoising algorithm. If abnormal fluctuations are identified, smooth the abnormal part to obtain a denoised data set; For the denoised data set, apply a calibration algorithm to correct sensor errors. If it is detected that the data deviation exceeds the preset threshold, adjust the offset part to determine the calibrated data set; Extract key features from the calibrated data set, analyze the data distribution using preset rules, and judge whether there is residual interference to obtain a data group after feature analysis; Perform standardization processing on the data group after feature analysis, uniformly adjust the data range, and obtain a standardized data set.

4. The method for monitoring pet health and behavior based on multi-sensor data according to claim 1, characterized in that By performing hierarchical classification on the standardized data set, key characteristic values of data in each dimension are extracted, including: establishing characteristic extraction models for activity monitoring, sleep analysis, and behavior recording respectively, performing hierarchical processing on the standardized data set to obtain the first characteristic value of activity data, the second characteristic value of sleep state data, and the third characteristic value of specific behavior data; Analyze the distribution law of the pet's activity level according to the first characteristic value; Evaluate the persistence and depth of the pet's sleep state according to the second characteristic value; Identify the occurrence frequency and pattern of the pet's specific behavior according to the third characteristic value; Integrate the first characteristic value, the second characteristic value, and the third characteristic value into a set of key characteristic values for subsequent dynamic change detection.

5. The method for monitoring pet health and behavior based on multi - sensor data according to claim 1, characterized in that, Detect the dynamic change trend of data in each dimension according to the key characteristic values, and judge the periodic changes in activity level, sleep quality, and behavior pattern, including: using a preset time series analysis algorithm to process the key characteristic values to obtain the first change trend of activity data, the second change trend of sleep quality data, and the third change trend of behavior pattern data; Judge the periodic fluctuations of the pet's activity level according to the first change trend; Analyze the stability of the pet's sleep quality according to the second change trend; Identify the regular changes in the pet's behavior pattern according to the third change trend; If the first change trend, the second change trend, or the third change trend exceeds the preset threshold range, record the corresponding time period and data points to generate an abnormal change record.

6. The method for monitoring pet health and behavior based on multi-sensor data according to claim 1, wherein If the dynamic change trend exceeds the preset threshold range, mark the abnormal data points to generate an annotated abnormal data set, including: identifying the distribution positions and time periods of abnormal data points through an anomaly detection module for data points where the dynamic change trend exceeds the preset threshold range; According to the distribution positions and time periods of the abnormal data points, perform classification marking in combination with the corresponding data types to generate a first abnormal data subset, a second abnormal data subset, and a third abnormal data subset; Integrate the first abnormal data subset, the second abnormal data subset, and the third abnormal data subset to generate an annotated abnormal data set; Store the annotated abnormal data set in a preset database for subsequent context correlation analysis.

7. The method for monitoring pet health and behavior based on multi-sensor data according to claim 1, wherein, Determine the correlation between abnormal data and health risks or behavioral abnormalities by performing context correlation analysis on the annotated abnormal data set, including: extracting the characteristic information of abnormal data points from the annotated abnormal data set, and performing comparison and analysis in combination with historical data records and current environmental information to obtain a first correlation value; Judge the correlation between abnormal data points and health risks according to the first correlation value to obtain a health risk correlation result; Judge the correlation between abnormal data points and behavioral abnormalities according to the first correlation value to obtain a behavioral abnormality correlation result; Integrate the health risk correlation result and the behavioral abnormality correlation result to generate a comprehensive correlation analysis report; Determine the final correlation classification of the abnormal data according to the comprehensive correlation analysis report.

8. The method for monitoring pet health and behavior based on multi-sensor data according to claim 1, characterized in that Based on the correlation results, a comprehensive score of the overall health status of the pet is obtained, and the health risk level and the probability of behavioral abnormalities are acquired, including: inputting the correlation results into a preset health status assessment model to obtain a preliminary score value of the pet's health status; According to the preliminary score value, combining the key characteristic values of activity level, sleep quality and behavior pattern, adjusting the scoring weights, and generating a final comprehensive score; Determining the health risk level according to the final comprehensive score; Calculating the probability of behavioral abnormalities according to the final comprehensive score; Storing the health risk level and the probability of behavioral abnormalities in a preset database, and generating corresponding evaluation records for subsequent report generation.

Images