Intelligent pet health monitoring and intelligent evaluation system

Abstract

The invention discloses an intelligent pet health monitoring and intelligent evaluation system, which comprises a plurality of integrated sensors, and carries out real-time processing on multi-source data by adopting an improved unscented Kalman filtering and deep fusion algorithm. The system integrates a convolutional neural network, a long-short-term memory network and a Transform model at the same time, and realizes adaptive behavior recognition, health state evaluation and personalized health suggestion generation through online learning and incremental learning. The edge device is responsible for data acquisition and primary processing, including wavelet denoising, filtering and feature extraction, and performs data interaction and model updating with the cloud server through a standardized interface, such as MQTT or RESTful API. And the cloud server uses large-scale data to train a depth model, and continuously feeds back and optimizes an edge end algorithm. Through detailed parameter setting, interface design between modules and an early warning strategy, multi-dimensional data fusion, dynamic health assessment and multi-level early warning are realized, and an accurate, real-time and customized health management scheme is provided for pets.

Description

Technical Field

[0001] The present invention relates to a pet monitoring system, and in particular to an intelligent pet health monitoring and intelligent evaluation system. Background Art

[0002] Existing pet health monitoring systems mostly rely on single or a small number of sensors. For example, only an accelerometer is used to record the amount of exercise or a single heart rate sensor is used to collect physiological data, making it difficult to comprehensively reflect the overall health status of pets. In addition, these systems use simple statistical or rule algorithms for data processing, with low intelligence and lack of dynamic adaptation to individual differences of pets, resulting in one-sided monitoring results and an untimely warning mechanism.

[0003] In recent years, although research has introduced deep learning (such as CNN and LSTM) for behavior recognition, it is often limited to single behavior classification and cannot simultaneously take into account the comprehensive evaluation of physiology, behavior, and environment.

[0004] Patent WO2018000215A1 discloses a pet health monitoring system, which only uses an accelerometer and a heart rate sensor, the measured data is too single, and the data processing depends on rule algorithms and cannot dynamically adapt to individual differences.

[0005] Patent CN216254682U discloses an intelligent pet collar, which realizes basic data transmission but does not mention the collaborative optimization mechanism between edge computing and the cloud.

[0006] The Four Paws project of Carnegie Mellon University studied a pet behavior recognition model, but it only uses a single LSTM network and does not combine other network models, resulting in insufficient long-time series behavior modeling ability.

[0007] And the current PetPace collar products on the market usually only provide physiological index monitoring and lack the ability to model the spatio-temporal dependence of sensors based on GNN. Summary of the Invention

[0008] Object of the Invention: Aiming at the above problems, the object of the present invention is to provide an intelligent pet health monitoring and intelligent evaluation system to achieve comprehensive, accurate, and intelligent health management.

[0009] Technical solution: An intelligent pet health monitoring and intelligent evaluation system, including a multi-sensor fusion module, a behavior analysis module, a health evaluation and warning module, and an edge computing and cloud collaboration module. The multi-sensor fusion module includes a sensor module and a data fusion algorithm module. The sensor module includes an accelerometer, a gyroscope, a heart rate sensor, a body temperature sensor, a blood oxygen sensor, and an environmental sensor. The sensor module collects various data and performs data preprocessing through the data fusion algorithm module. The processed data is transmitted to the edge computing and cloud collaboration module. The edge computing and cloud collaboration module makes a determination based on the data priority, transmits the data to the behavior analysis module and / or uploads it to the cloud. The edge computing and cloud collaboration module updates the model through cloud data, optimizes the edge algorithm, and transmits the analysis result to the health evaluation and warning module to generate a warning signal or a suggestion.

[0010] Further, the signal is transmitted between the multi-sensor fusion module and the edge computing and cloud collaboration module through Bluetooth 5.0 or Wi-Fi. The environmental sensor includes a temperature and humidity sensor, an air quality sensor, and a noise sensor, and the sampling period of all is 1 - 1.5 min. The sampling frequency of the accelerometer is set to 50 - 100 Hz, and the sampling frequency of the heart rate sensor is 1 Hz.

[0011] Further, the data fusion algorithm module uses UKF to perform non-linear state estimation on the data of each sensor. Among them, the state transition matrix and the noise covariance are calibrated through pre-collected data. The state vector parameters include the displacement and velocity parameters of the accelerometer, and the measurement vector covers the data of each sensor;

[0012] The data fusion algorithm module also constructs a spatio-temporal dependence graph between sensors through GNN. GNN constructs edges based on mutual information or correlation coefficient weighting and uses supervised learning for training to complete real-time data fusion.

[0013] Model the spatio-temporal dependence relationship between sensors using a deep fusion algorithm combined with a graph neural network. GNN constructs nodes with sensors, edges based on mutual information or correlation coefficient weighting, and uses supervised learning for training to optimize the real-time fusion effect of multi-source data.

[0014] Further, the behavior analysis module is equipped with a multi-layer deep learning model integrating CNN, LSTM, and Transformer. The input data first extracts spatial features through CNN, then is concatenated with physiological and environmental data and input into LSTM to capture temporal dependencies, then models long-distance dependencies through the Transformer encoder, and finally performs behavior classification through the fully connected layer.

[0015] The CNN is responsible for extracting the spatial features of sensor data such as accelerometers and gyroscopes; the LSTM captures the temporal changes in physiological data such as heart rate and body temperature; the Transformer models long-distance dependencies.

[0016] Preferably, the multi-layer deep learning model is supervised and trained with cloud data, and the model is updated in real time using online learning and incremental learning algorithms. The parameters of the CNN, LSTM, and Transformer models are dynamically adjusted according to the behavioral characteristics of different pets, and continuously adjusted by the sliding window method.

[0017] Adopt online learning and incremental learning methods to update the model parameters in real time based on the pre-constructed labeled sample data set of multiple breeds of pets. Each model hyperparameter (such as learning rate 0.001 - 0.01, batch size 32 - 128, number of LSTM units 64 - 128) can be dynamically adjusted according to the statistical characteristics of the collected data. The system continuously updates using the sliding window method to ensure adaptation to the behavioral patterns of different pets.

[0018] Preferably, the health assessment and warning module calculates the health index and identifies potential risks through a preset multi-dimensional health assessment formula, sets low, medium, and high-level warning thresholds according to each health indicator, generates warning signals in sequence when the data exceeds the normal range, and immediately pushes them through the mobile APP or SMS.

[0019] Combining the pet's behavioral patterns, physiological parameters (such as heart rate, body temperature, blood oxygen, etc.) and environmental data, calculate the health index through a multi-dimensional assessment algorithm, and evaluate the visceral functions (such as cardiovascular health), musculoskeletal system, emotional state, and environmental adaptability. The health index is calculated by weighted summation, and the weights of each dimension are determined based on expert knowledge or machine learning techniques (for example, heart rate and blood oxygen have higher weights in the cardiovascular health score). Set the abnormal detection threshold (such as the normal range of heart rate, blood oxygen, and activity level determined based on the statistical distribution of historical data or veterinary guidelines). When the detected data deviates from the normal range, the system automatically generates low, medium, and high-level warning messages and immediately notifies the pet owner and veterinarian through the mobile application or SMS. For example, a high-level warning is triggered when the heart rate exceeds the 95th percentile.

[0020] Preferably, the system also includes a personalized recommendation module. The personalized recommendation module uses a hybrid recommendation system based on the output results of the health assessment and warning module to generate a customized health management plan based on the pet's individual information and health assessment results. Among them, the hybrid recommendation system combines content filtering and collaborative filtering.

[0021] Preferably, the edge computing and cloud collaboration module includes edge devices and cloud servers. The edge devices are designed with low power consumption to complete data preprocessing and real-time analysis, and the cloud servers use standard communication protocols to achieve data transmission, model training, and parameter update.

[0022] Integrate a low-power edge computing module in the pet-wearing device to complete data collection and preliminary processing (such as wavelet denoising, high-pass filtering, time-domain and frequency-domain feature extraction), and upload the data using standard communication protocols such as Bluetooth 5.0, Wi-Fi, and MQTT protocols. The cloud server is responsible for large-scale data storage, batch deep model training and optimization, and regularly pushes the updated model to the edge device through a secure encrypted channel to achieve continuous improvement of the overall system performance. The data interface, transmission protocol, and model update mechanism all follow international communication standards to ensure data security and stable transmission.

[0023] Preferably, this system is installed on a pet-wearing device, and the pet-wearing device supports wireless charging or replaceable batteries.

[0024] Beneficial effects: Compared with the prior art, the advantages of the present invention are:

[0025] The present invention is a novel pet health monitoring system that combines multi-sensor data fusion, advanced filtering and deep learning algorithms, and edge-cloud collaborative processing. It can detect various health indicators of pets, adapt to the individual behavior patterns of pets, ensure the accuracy of behavior recognition and health index evaluation, and provide accurate, real-time, and customized health management solutions for pets. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a flow chart of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0027] The present invention will be further clarified below in conjunction with the drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention.

[0028] An intelligent pet health monitoring and intelligent evaluation system, as Figure 1 shown, includes a multi-sensor fusion module, a behavior analysis module, a health evaluation and warning module, a personalized recommendation module, and an edge computing and cloud collaboration module. The sensor fusion module includes a sensor module and a data fusion algorithm module.

[0029] First, continue data collection. The sensor module includes an accelerometer, a gyroscope, a heart rate sensor, a body temperature sensor, and a blood oxygen sensor embedded in the pet collar. At the same time, an environmental sensor is configured, and the environmental sensor includes temperature and humidity, air quality, and noise sensors. Each sensor meets the requirements of low power consumption and high sensitivity. The device supports wireless charging or replaceable batteries to ensure continuous operation. The sampling frequency of the accelerometer is set to 100 Hz; the sampling frequency of the heart rate sensor is 1 Hz; the sampling period of the environmental sensor is set to 60 seconds. Using the Bluetooth 5.0 or Wi-Fi or MQTT (Message Queuing Telemetry Transport) module, the collected raw data is sent to the edge computing module in the form of data packets in real time, and an encryption protocol is used during the transmission process to ensure data security. MQTT is a lightweight communication protocol for low-power, high-reliability data transmission between edge devices and cloud servers, and supports real-time warning message pushing.

[0030] The collected data is processed and fused. The edge device preprocesses the raw data: uses wavelet transform for denoising, smooths the data using a high-pass filter, and extracts time-domain (mean, variance) and frequency-domain (power spectral density) features. The data fusion algorithm module uses UKF (Unscented Kalman Filter) for non-linear state estimation of multi-sensor data. Among them, the state transition matrix and noise covariance are calibrated through pre-collected data and support online adaptive update; UKF is a non-linear state estimation algorithm used to handle noise and uncertainty in multi-sensor data fusion, approximates the non-linear system distribution through preset Sigma points, and improves the estimation accuracy of the pet's motion state (such as displacement, speed). Combine the deep fusion algorithm to further integrate different sensor features; use GNN (Graph Neural Network) to construct a spatio-temporal dependence graph between sensors. GNN constructs edges weighted based on mutual information or correlation coefficient and is trained using supervised learning to achieve high-precision real-time data fusion. GNN is a deep learning model that models the spatio-temporal dependence relationship of sensor data by constructing a topological relationship graph (such as edges weighted based on mutual information or correlation coefficient) between sensor nodes and optimizes the multi-source data fusion effect. The processed data is transmitted to the cloud server through standard interfaces such as RESTful API (Representational State Transfer Application Programming Interface) to ensure unified data format.

[0031] Subsequently, behavioral analysis and health assessment are carried out. The behavioral analysis module constructs a deep model integrating CNN (Convolutional Neural Network), LSTM (Long Short-Term Memory Network), and Transformer, and uses the labeled dataset for supervised training. Example settings for each layer's parameters: the convolutional kernel size is 3×3, the number of LSTM units is 128, and the number of Transformer heads is 8, which are determined after optimizing the experimental results. After deployment, the model parameters are updated in real time using online learning and incremental learning algorithms, and continuously adjusted through the sliding window method to adapt to the individual behavioral patterns of pets, ensuring accurate behavior recognition (rest, exercise, eating, abnormal stress) and health index assessment. The health assessment and warning module combines the output of the deep model with a preset multi-dimensional health assessment formula (weighted summation, and the weights of each dimension are determined based on expert knowledge or machine learning) to calculate the health index and identify potential risks (such as too fast heart rate, sudden drop in activity level). Three levels of warning thresholds (e.g., heart rate exceeding the 95th percentile triggers a high-level warning) are set according to various health indicators. When the data exceeds the normal range, warning signals are generated in sequence and immediately pushed through the mobile APP or SMS.

[0032] CNN is used to extract the spatial features of sensor data (such as accelerometers, gyroscopes), for example, the local pattern recognition of pet movements. LSTM is a variant of the recurrent neural network, which is good at capturing the long-term dependencies of time series data (such as heart rate, body temperature) and is used to analyze the dynamic changes of pet physiological parameters. Transformer is a deep learning model based on the self-attention mechanism, which is good at modeling long-distance dependencies and is used for the global feature extraction of pet behavior sequences (such as the correlation analysis of continuous movement and rest states).

[0033] After that, the personalized recommendation module uses a hybrid recommendation system to generate customized recommendations (such as adjusting the diet ratio, increasing the appropriate exercise time) based on the pet's historical data, breed, age, and health assessment results. For example, the daily exercise time is recommended according to the pet's activity level, and the diet ratio is adjusted according to the historical health data. The system regularly analyzes historical and real-time data and continuously updates the health management plan through incremental learning to ensure that the recommended plan is scientific and meets the actual needs.

[0034] The edge computing and cloud collaboration module includes edge devices and cloud servers. The edge devices complete data collection, preprocessing, and local model operation on the pet-wearing devices through a low-power edge computing module to ensure low-latency response. The cloud server is responsible for big data storage, batch model training, global optimization, and security management, and pushes the updated model to the edge devices through a secure encrypted channel to achieve closed-loop feedback. The updated model is regularly pushed to the edge devices through a secure encrypted channel to continuously improve the overall performance of the system. The data interface, transmission protocol, and model update mechanism all follow international communication standards to ensure data security and stable transmission.

[0035] The sensor module of the present invention collects various raw data of the pet and transmits it to the data fusion algorithm module through Bluetooth 5.0 / Wi-Fi. The data fusion module uses the UKF and GNN algorithms to complete preprocessing (denoising, feature extraction, spatio-temporal dependence modeling). The processed data is transmitted to the edge computing and cloud collaboration module through a standardized interface (such as MQTT / RESTful API). The edge computing and cloud collaboration module is the main control module of the present invention. The main control module decides whether to immediately trigger local behavior analysis (such as abnormal movement detection) or upload the data to the cloud according to the data priority. When judging abnormalities, the Z-Score (standard score) can be introduced to quantify the degree of data deviation from the mean. For example, it is judged whether the sudden change in heart rate exceeds the normal range (such as Z-Score > 3 is regarded as abnormal). The main control module also conducts information interaction with the cloud server, uploading key data (such as long-term health trends) from the main control module to the cloud for deep model training, and the cloud sends the updated model parameters to the main control module to optimize the edge-side algorithm. The main control module transmits the analysis results (such as behavior classification, health index) to the health assessment module to generate warning signals or suggestions. The present invention realizes multi-dimensional data fusion, dynamic health assessment, and multi-level early warning, providing a precise, real-time, and customized health management solution for pets.

Claims

1. An intelligent pet health monitoring and intelligent evaluation system, characterized in that: It includes a multi-sensor fusion module, a behavior analysis module, a health assessment and warning module, and an edge computing and cloud collaboration module. The multi-sensor fusion module includes a sensor module and a data fusion algorithm module. The sensor module includes an accelerometer, a gyroscope, a heart rate sensor, a body temperature sensor, a blood oxygen sensor, and an environmental sensor. The sensor module collects various data and performs data preprocessing through the data fusion algorithm module. The processed data is transmitted to the edge computing and cloud collaboration module. The edge computing and cloud collaboration module makes a determination based on the data priority, transmits the data to the behavior analysis module and / or uploads it to the cloud. The edge computing and cloud collaboration module updates the model through cloud data, optimizes the edge algorithm, and transmits the analysis result to the health assessment and warning module to generate a warning signal or a suggestion.

2. The intelligent pet health monitoring and intelligent evaluation system according to claim 1, wherein: The signal is transmitted between the multi-sensor fusion module and the edge computing and cloud collaboration module through Bluetooth 5.0 or Wi-Fi. The environmental sensor includes a temperature and humidity sensor, an air quality sensor, and a noise sensor, and the sampling period of all of them is 1 - 1.5 minutes. The sampling frequency of the accelerometer is set to 50 - 100 Hz, and the sampling frequency of the heart rate sensor is 1 Hz.

3. The intelligent pet health monitoring and intelligent evaluation system according to claim 1, wherein: The data fusion algorithm module uses UKF to perform non-linear state estimation on the data of each sensor. Among them, the state transition matrix and the noise covariance are calibrated through pre-collected data. The state vector parameters include the displacement and velocity parameters of the accelerometer, and the measurement vector covers the data of each sensor. The data fusion algorithm module also constructs a spatio-temporal dependence graph between sensors through GNN. GNN constructs edges weighted based on mutual information or correlation coefficient and uses supervised learning for training to complete real-time data fusion.

4. An intelligent pet health monitoring and intelligent evaluation system according to claim 1, characterized in that: The behavior analysis module is equipped with a multi-layer deep learning model that integrates CNN, LSTM, and Transformer. The input data first passes through CNN to extract spatial features, then is concatenated with physiological and environmental data and input into LSTM to capture temporal dependencies, then the long-distance dependencies are modeled through the Transformer encoder, and finally, the behavior classification is performed through the fully connected layer.

5. The intelligent pet health monitoring and intelligent evaluation system according to claim 4, wherein: The multi-layer deep learning model is supervised and trained through cloud data, and the model is updated in real time using online learning and incremental learning algorithms. The parameters of the CNN, LSTM, and Transformer models are dynamically adjusted according to the behavior characteristics of different pets, and are continuously adjusted through the sliding window method.

6. The intelligent pet health monitoring and intelligent evaluation system according to claim 1, characterized in that: The health assessment and warning module calculates the health index and identifies potential risks through a preset multi-dimensional health assessment formula, sets low, medium, and high-level warning thresholds according to each health indicator. When the data exceeds the normal range, warning signals are generated in sequence and are immediately pushed through the mobile APP or SMS.

7. An intelligent pet health monitoring and intelligent evaluation system according to claim 1, characterized in that: It also includes a personalized suggestion module. The personalized suggestion module uses a hybrid recommendation system based on the output result of the health assessment and warning module to generate a customized health management plan based on the pet's individual information and health assessment result. Among them, the hybrid recommendation system combines content filtering and collaborative filtering.

8. An intelligent pet health monitoring and intelligent evaluation system according to claim 1, characterized in that: The edge computing and cloud collaboration module includes edge devices and cloud servers. The edge devices are designed with low power consumption to complete data preprocessing and real-time analysis, and the cloud servers use standard communication protocols to achieve data transmission, model training, and parameter update.

9. An intelligent pet health monitoring and intelligent evaluation system according to claim 1, characterized in that: This system is installed on a pet-wearing device, which supports wireless charging or replaceable batteries.

Images