Remote rehabilitation monitoring auxiliary method and system based on artificial intelligence

Through an artificial intelligence-based approach, combining heart rate and behavioral data, identifying heart rate patterns and providing customized health assessments, the problem of single data analysis in the prior art is solved, and the analysis accuracy and adaptability of rehabilitation programs are improved.

CN118824548BActive Publication Date: 2024-12-06JILIN UNIV FIRST HOSPITAL
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411281793.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-13
Publication Date
2024-12-06
Estimated Expiration
2044-09-13

AI Technical Summary

Technical Problem

The existing telerehabilitation monitoring system is relatively single in data analysis, lacks a deep understanding of the complex relationship between heart rate and behavioral data, and cannot provide customized rehabilitation plans.

Method used

Using an artificial intelligence-based method, through timing feature extraction, clustering calculation, feature layer coding and change curve construction, combining user heart rate and behavioral data, heart rate patterns are identified and customized health assessments are provided to respond to heart rate abnormalities in real time.

Benefits of technology

It improves the accuracy of heart rate data analysis, enhances the accuracy of identifying user behavior, provides customized rehabilitation plans, and improves the safety and system intelligence of the rehabilitation process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118824548B_ABST
    Figure CN118824548B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of health management technology, and in particular to an artificial intelligence-based remote rehabilitation monitoring auxiliary method and system. The method comprises the following steps: obtaining user heart rate data and user behavior data; obtaining user heart rate time series feature data according to the user heart rate data, and performing clustering calculation according to the user behavior data and the user heart rate data to obtain user behavior heart rate clustering feature data; obtaining first user heart rate evaluation data according to the user heart rate time series feature data and preset healthy heart rate time series feature data; obtaining heart rate clustering feature coding data according to the user behavior heart rate clustering feature data; obtaining heart rate feature change curve data according to the heart rate clustering feature coding data; obtaining second user heart rate evaluation data according to the heart rate feature change curve data, so as to perform remote rehabilitation monitoring auxiliary operations. The present invention uses heart rate and behavior data for in-depth analysis to improve the evaluation accuracy of the remote rehabilitation monitoring system.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of health management technology, and in particular to an artificial intelligence-based remote rehabilitation monitoring auxiliary method and system. Background Art

[0002] With the rapid development of information technology and medical technology, telemedicine and rehabilitation monitoring have gradually become an important part of the modern medical system. Especially for patients with cardiovascular diseases, sports injuries or other patients who need long-term rehabilitation, remote monitoring technology can obtain patients' physiological data in real time, such as heart rate, blood pressure, blood oxygen, etc., so as to achieve real-time tracking and dynamic evaluation of patients' health status. It can not only effectively reduce the burden of patients going to the hospital frequently, but also help doctors to promptly detect abnormal conditions in the patient's rehabilitation process and formulate more customized rehabilitation plans.

[0003] In the existing technology, remote rehabilitation monitoring systems usually collect physiological data of patients through wearable devices, and generate simple health reports based on the patients' behavioral data (such as exercise, sleep, etc.). However, the monitoring and analysis methods of such systems are relatively simple, usually based on basic statistical analysis of heart rate or behavioral data, and lack a deep understanding of the complex relationships between data. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes an artificial intelligence-based remote rehabilitation monitoring auxiliary method and system to solve at least one of the above technical problems.

[0005] The present application provides a remote rehabilitation monitoring auxiliary method based on artificial intelligence, comprising the following steps:

[0006] Step S1: Obtain user heart rate data and user behavior data;

[0007] Step S2: extracting time series features based on the user's heart rate data to obtain time series feature data of the user's heart rate, and performing clustering calculation based on the user's behavior data and the user's heart rate data to obtain user behavior heart rate clustering feature data;

[0008] Step S3: fitting the user's heart rate time series characteristic data and preset healthy heart rate time series characteristic data to obtain first user heart rate evaluation data;

[0009] Step S4: performing feature layer encoding according to the user behavior heart rate clustering feature data to obtain heart rate clustering feature encoding data; constructing a change curve according to the heart rate clustering feature encoding data to obtain heart rate feature change curve data; performing health assessment according to the heart rate feature change curve data to obtain heart rate assessment data of the second user;

[0010] Step S5: performing remote rehabilitation monitoring auxiliary operations according to the first user's heart rate assessment data and the second user's heart rate assessment data.

[0011] In the present invention, by extracting time series features from user heart rate data, the method can capture the complex dynamic changes of heart rate and extract key time series features, thereby improving the analysis accuracy of heart rate data. By fitting the user heart rate data with the preset healthy heart rate data, the first user heart rate assessment data is generated, the difference between the user's heart rate and the standard healthy heart rate is identified, and a customized health assessment is provided. Through clustering calculation and feature layer encoding, different behavior-heart rate patterns can be identified based on user behavior data and heart rate data, and these patterns are encoded and analyzed. Change curve construction and trend analysis are introduced. By constructing the heart rate feature change curve, the method can capture the long-term change trend of the user's heart rate. The present invention can quickly respond to abnormal user heart rate conditions based on real-time data and evaluation results, provide immediate feedback and guidance, and improve safety during the rehabilitation process.

[0012] Preferably, step S1 specifically includes:

[0013] The user's heart rate is collected through a wearable device to obtain the user's heart rate data;

[0014] Obtaining user behavior table data and current time data, and generating user behavior based on the user behavior table data and current time data to obtain primary user behavior data;

[0015] Capturing user images through a camera device to obtain user image data, and capturing user posture data based on the user image data to obtain user posture data;

[0016] Generate user estimated posture according to primary user behavior data to obtain user estimated posture data;

[0017] Perform error evaluation based on the user's estimated posture data and the user's posture data to obtain user posture error data;

[0018] The primary user behavior data is adjusted according to the user posture error data to obtain the user behavior data.

[0019] The integration of multi-source data in the present invention ensures more accurate identification of user behavior and can effectively reduce the errors and uncertainties caused by a single data source. The real-time error evaluation and behavior data adjustment mechanism ensures that the system can promptly detect and correct the user's behavior deviations, and improves the system's real-time response capabilities to user behavior. By comparing the user's actual posture and estimated posture, this method can effectively identify the user's movement deviations in rehabilitation training, and make real-time corrections by adjusting the primary behavior data. The adjustment mechanism can be customized according to the specific circumstances of each user, thereby improving the adaptability and effectiveness of the rehabilitation program.

[0020] Preferably, the error assessment is specifically:

[0021] Generating posture components according to the user estimated posture data and the user posture data, respectively obtaining the user estimated posture component data and the user posture component data, wherein the user estimated posture component data includes the user estimated angle component data, the user estimated position component data and the user estimated speed component data, and the user posture component data includes the user angle component data, the user position component data and the user speed component data;

[0022] Calculate the error component according to the user's estimated posture component data and the user's posture component data to obtain the user's posture error component data;

[0023] The user posture error data is fused to obtain the user posture error data.

[0024] In the present invention, by decomposing the posture data into angle components, position components and speed components, and calculating the errors of these components respectively, the method can perform a more refined analysis of the user's posture to make deeper data adjustments. By generating and calculating posture component data of different dimensions such as angle, position and speed respectively, the user's posture can be evaluated in multiple dimensions. It ensures that the errors of each component can be effectively integrated, and the generated posture error data is more representative and accurate. Through in-depth analysis and synthesis of the error components, the method can generate detailed posture error information for precise posture adjustment.

[0025] Preferably, step S2 specifically includes:

[0026] Step S21: reconstructing the phase space according to the user's heart rate data to obtain the heart rate phase space data;

[0027] Step S22: constructing a phase space network for the heart rate phase space data to obtain the heart rate phase space network data;

[0028] Step S23: extracting network topology features from the heart rate phase space network data to obtain heart rate network topology feature data;

[0029] Step S24: performing entropy weight analysis based on the heart rate network topology feature data to obtain the user's heart rate time series feature data;

[0030] Step S25: Perform clustering calculation based on the user behavior data and the user heart rate data to obtain user behavior and heart rate clustering feature data.

[0031] In the present invention, through phase space reconstruction, the method can map the original one-dimensional heart rate time series data into a high-dimensional phase space, and the mapping can reveal the dynamic behavior and hidden patterns in the heart rate data, thereby enhancing the depth and breadth of data analysis. The heart rate phase space data is constructed as a network structure, which can capture the structural and correlation characteristics in the heart rate data, and the network analysis based on the heart rate data can reveal the nodes and connection patterns in the heart rate data. By extracting the topological features of the heart rate phase space network data, the method can identify the global and local structural characteristics of the heart rate data, such as node degree distribution, clustering coefficient, average path length, etc. Through entropy weight analysis, the method can quantify the complexity and information distribution of the topological features of the heart rate network. It can help identify the most informative features in the heart rate data, and generate the user's heart rate time series feature data based on these features. By combining user behavior data and heart rate data for clustering calculations, the method can identify the heart rate characteristics of users under different behavior patterns.

[0032] Preferably, step S21 is specifically:

[0033] Determine the phase space embedding parameters according to the user's heart rate data to obtain the embedding delay time data;

[0034] Calculate the embedding dimension according to the user's heart rate data to obtain the embedding dimension data;

[0035] The heart rate phase space of the user's heart rate data is constructed according to the embedded delay time data and the embedded dimension data to obtain the heart rate phase space data.

[0036] In the present invention, by accurately calculating the embedding delay time and embedding dimension of the heart rate data, the method can reconstruct a phase space containing more information from the heart rate time series data. The phase space reconstruction method converts the original one-dimensional time series data into a high-dimensional space, which can better capture the dynamic characteristics and hidden patterns of the heart rate data, thereby enhancing the ability to analyze the heart rate behavior. After determining the embedding delay time and embedding dimension, the method can more accurately reconstruct the phase space and reveal the nonlinear characteristics existing in the heart rate data. By mapping the heart rate data to the high-dimensional phase space, the attractor structure and trajectory in the heart rate data can be more clearly displayed, helping to understand the behavior of the cardiac system as a complex dynamic system. The accurate calculation of the embedding delay time and embedding dimension ensures the effectiveness and accuracy of the phase space reconstruction.

[0037] Preferably, step S24 is specifically:

[0038] Perform conditional entropy calculation and joint entropy calculation according to the heart rate network topology feature data to obtain first heart rate entropy value feature data and second heart rate entropy value feature data respectively;

[0039] Performing entropy difference calculation based on the first heart rate entropy value characteristic data and the second heart rate entropy value characteristic data to obtain heart rate entropy difference data;

[0040] Performing differential weighting processing on the heart rate network topology feature data according to the entropy difference data to obtain differential weighted heart rate feature data;

[0041] Nonlinear feature mapping is performed based on the differentiated weighted heart rate feature data to obtain the user's heart rate time series feature data.

[0042] In the present invention, by respectively calculating the conditional entropy and joint entropy of the heart rate network topological feature data, the amount of information and complexity in the heart rate data can be analyzed at multiple levels. The conditional entropy can reveal the randomness and uncertainty of the heart rate characteristics under a specific state, while the joint entropy can quantify the total amount of information in the entire system. Through entropy difference calculation, the method can identify the difference between the conditional entropy and the joint entropy, reflecting the change in the amount of information of the heart rate network topological feature data under different conditions. By weighting these differences, the method can highlight those features that are more sensitive to heart rate changes. Through differential weighted processing, the heart rate feature data that show significant differences under different entropy conditions are extracted and emphasized. Subsequently, using nonlinear feature mapping technology, the method can map these weighted features to a high-dimensional space, thereby capturing the deep relationships and patterns hidden in the heart rate time series data. This nonlinear mapping not only improves the expressiveness of the features, but also enhances the ability to capture complex dynamic behaviors.

[0043] Preferably, step S3 is specifically:

[0044] Dynamically matching the user's heart rate time series feature data with preset healthy heart rate time series feature data to obtain heart rate feature matching data;

[0045] Perform residual calculation based on the heart rate feature matching data to obtain heart rate feature matching residual data;

[0046] Perform multivariate linear regression fitting on the residual data of heart rate feature matching to obtain the user's heart rate deviation data;

[0047] Perform residual curve analysis on the user's heart rate deviation data to obtain first user heart rate evaluation data.

[0048] The present invention dynamically matches the user's heart rate time series feature data with the preset healthy heart rate time series feature data, which can flexibly adapt to the changing trend of the individual heart rate. Dynamic matching can capture the slight changes in the user's heart rate data and compare it with the health standard in real time, ensuring the accuracy of the matching process, thereby providing high-quality basic data for subsequent residual analysis and evaluation. By performing residual calculation on the heart rate feature data after dynamic matching, the specific degree of deviation between the user's heart rate and the health standard is quantified. The residual data directly reflects the unconventional situation of the user's heart rate, providing users with more intuitive in-depth data. By performing multivariate linear regression fitting on the residual data, multi-dimensional heart rate feature information can be integrated, and the overall deviation of the user's heart rate can be comprehensively analyzed. The residual curve analysis of the user's heart rate deviation data can identify the trend and pattern of the heart rate deviation. Through the residual curve, this method can not only quantify the degree of heart rate deviation, but also evaluate the persistence and volatility of the deviation.

[0049] Preferably, step S4 is specifically:

[0050] Perform cluster feature encoding according to the user behavior heart rate cluster feature data to obtain heart rate cluster feature encoding data;

[0051] Embedding the encoded features of the heart rate clustering feature encoded data to obtain embedded heart rate feature data;

[0052] Constructing feature time series for embedded heart rate feature data to obtain heart rate clustering time series feature data;

[0053] Perform multi-scale curve analysis on the heart rate clustering time series feature data to obtain the heart rate feature change curve data;

[0054] Performing curve pattern recognition on the heart rate characteristic change curve data to obtain heart rate characteristic curve pattern data;

[0055] The heart rate characteristic curve pattern data is subjected to curve trend fitting to obtain the second user's heart rate evaluation data.

[0056] In the present invention, by encoding the user behavior heart rate clustering feature data, the method can effectively encode and represent the behavior and heart rate pattern, can structure the user's behavior and heart rate data, and reveal the heart rate response characteristics under the behavior pattern. After embedding the cluster feature encoding data, the method can map the high-dimensional features to the low-dimensional space, thereby compressing the dimension of the data, which not only reduces the computational complexity, but also retains the key information in the original feature data, so that the system can process complex data more efficiently. By constructing the embedded heart rate feature data in a time series, the method can capture the dynamic change characteristics of the user's heart rate data. The time series data can reflect the changing trend of the user's heart rate under different behavior patterns, help to identify the periodic fluctuations of the heart rate and the behavior-related change patterns, and further enhance the time sensitivity of the heart rate data analysis. By performing multi-scale curve analysis on the heart rate clustering time series feature data, the method can analyze the heart rate data on different time scales and capture the changes in the heart rate characteristics in the short and long term. By performing pattern recognition on the heart rate characteristic change curve data, this method can extract the typical heart rate change pattern of the user, which is helpful to analyze the heart rate change pattern of the user under different behavior patterns. After completing the curve pattern recognition, this method can predict the future change trend of the user's heart rate through trend fitting.

[0057] Preferably, step S5 is specifically:

[0058] Performing heart rate assessment difference analysis based on the heart rate assessment data of the first user and the heart rate assessment data of the second user to obtain heart rate assessment difference data;

[0059] The first user's heart rate assessment data, the second user's heart rate assessment data and the heart rate assessment difference data are used to generate a remote rehabilitation report to obtain remote rehabilitation report data for remote rehabilitation monitoring auxiliary operations.

[0060] In the present invention, by performing a difference analysis on the heart rate evaluation data of the first user and the heart rate evaluation data of the second user, the method can accurately identify the differences between different evaluation dimensions of the user's heart rate. Combining the first and second user heart rate evaluation data can perform a multi-dimensional comprehensive evaluation of the user's heart rate state. The first user's heart rate evaluation data focuses on the characteristics of the time series, while the second user's heart rate evaluation data covers the changing characteristics in the behavior pattern. By integrating the first and second user heart rate evaluation data and the difference analysis data to generate a remote rehabilitation report, customized, real-time, data-driven remote rehabilitation management is achieved, which significantly improves the intelligence and efficiency of the rehabilitation system.

[0061] Preferably, the present application also provides an artificial intelligence-based remote rehabilitation monitoring assistance system, which is used to execute the artificial intelligence-based remote rehabilitation monitoring assistance method as described above, and the artificial intelligence-based remote rehabilitation monitoring assistance system includes:

[0062] User heart rate basic data collection module, used to obtain user heart rate data and user behavior data;

[0063] A user heart rate feature extraction module is used to extract time series features based on user heart rate data to obtain user heart rate time series feature data, and to perform clustering calculation based on user behavior data and user heart rate data to obtain user behavior heart rate clustering feature data;

[0064] A first user heart rate evaluation module, used for fitting the user's heart rate time series characteristic data and preset healthy heart rate time series characteristic data to obtain first user heart rate evaluation data;

[0065] A second user heart rate assessment module is used to perform feature layer encoding according to the user behavior heart rate clustering feature data to obtain heart rate clustering feature encoding data; construct a change curve according to the heart rate clustering feature encoding data to obtain heart rate feature change curve data; perform health assessment according to the heart rate feature change curve data to obtain second user heart rate assessment data;

[0066] The remote rehabilitation monitoring auxiliary operation module is used to perform remote rehabilitation monitoring auxiliary operations according to the first user's heart rate evaluation data and the second user's heart rate evaluation data.

[0067] The beneficial effect of the present invention is that by combining user heart rate data and behavior data, the present invention realizes the deep fusion of multi-source data. The user heart rate data reflects the individual's physiological response, while the behavior data provides the environment and activity context. The combination of data makes the analysis more comprehensive and insightful. By extracting the time series features of the user's heart rate data (step S2), the present method can accurately capture the time series dynamic features in the heart rate signal, including heart rate variability, fluctuation rules, etc. By combining the user's heart rate and behavior data for clustering calculation (step S2), the present method can automatically identify the heart rate change rules of the user under different behavior modes, which not only enhances the data classification ability, but also can accurately capture the association between specific behaviors and heart rate. Through the encoding and embedding of clustering features, the present invention can uniformly represent and process heart rate data under different modes (step S4). By fitting the user's heart rate time series feature data with the health standard (step S3), the present invention can automatically evaluate the user's heart rate deviation and quantitatively evaluate the gap between the user and the health status through residual analysis. By integrating the two types of evaluation data (step S5), the system can conduct a more in-depth and differentiated evaluation of the user's heart rate health, realize customized, real-time, data-driven remote rehabilitation management, and significantly improve the intelligence and efficiency of the rehabilitation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting implementations made with reference to the following drawings:

[0069] Figure 1 A flowchart showing a method for assisting remote rehabilitation monitoring based on artificial intelligence according to an embodiment of the present invention is provided;

[0070] Figure 2 A flowchart showing a method for assisting remote rehabilitation monitoring based on artificial intelligence according to an embodiment of the present invention is provided;

[0071] Figure 3 A flowchart showing a method for assisting remote rehabilitation monitoring based on artificial intelligence according to an embodiment of the present invention is provided;

[0072] Figure 4 A flowchart showing a method for assisting remote rehabilitation monitoring based on artificial intelligence according to an embodiment of the present invention is provided;

[0073] Figure 5 A flowchart of the steps of an artificial intelligence-based remote rehabilitation monitoring assistance method according to an embodiment is shown. DETAILED DESCRIPTION

[0074] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0075] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0076] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0077] See also Figures 1 to 5 , the present application provides a remote rehabilitation monitoring auxiliary method based on artificial intelligence, comprising the following steps:

[0078] Step S1: Obtain user heart rate data and user behavior data;

[0079] Specifically, wearable devices are used to collect the user's heart rate signals in real time, record changes in heart rate over time, and generate raw heart rate data. Through smart sensors or mobile devices, user behavior data is collected, including activity type, number of steps, exercise duration, etc. The heart rate data and behavior data are synchronized on the timeline to ensure that they match within the same time range to form the user's behavior-heart rate data set. Alternatively, user behavior data (rest, work, diet, or other custom behavior words) can be input through the data input interface to collect user behavior data.

[0080] Step S2: extracting time series features based on the user's heart rate data to obtain time series feature data of the user's heart rate, and performing clustering calculation based on the user's behavior data and the user's heart rate data to obtain user behavior heart rate clustering feature data;

[0081] Specifically, the heart rate data is segmented, and trend analysis is performed on each segment of heart rate data according to a preset time window (e.g., 5 minutes). Basic statistical features such as mean, volatility, maximum, and minimum values ​​within the time window are extracted. Dynamic change features over time, such as the rate of heart rate rise and fall, heart rate recovery time, etc., are further extracted to form a heart rate time series feature vector.

[0082] Based on the user's behavior data (such as exercise, stillness, walking, etc.) and heart rate data, the correlation between the behavior data and the heart rate data is calculated to construct a high-dimensional feature space. The feature space is scaled and normalized to ensure data scale consistency. The feature space is grouped to preliminarily divide the user's behavior pattern and the corresponding heart rate change pattern to obtain the behavior heart rate clustering feature data. At this time, the preliminary clustering center and classification label of the user's behavior and heart rate data are generated.

[0083] Step S3: fitting the user's heart rate time series characteristic data and preset healthy heart rate time series characteristic data to obtain first user heart rate evaluation data;

[0084] Specifically, the preset time series characteristic data of heart rate of healthy people is obtained to construct a healthy heart rate reference curve. The user's heart rate time series characteristic data is compared with the healthy heart rate reference curve, and fitting is performed on the time axis, and linear or nonlinear fitting is performed on the user's heart rate time series characteristics. According to the fitting error and similarity, the difference between the user's current heart rate and the healthy heart rate is calculated to generate fitting evaluation data.

[0085] Step S4: performing feature layer encoding according to the user behavior heart rate clustering feature data to obtain heart rate clustering feature encoding data; constructing a change curve according to the heart rate clustering feature encoding data to obtain heart rate feature change curve data; performing health assessment according to the heart rate feature change curve data to obtain heart rate assessment data of the second user;

[0086] Specifically, the clustered behavioral heart rate feature data is multi-dimensionally encoded, and different behavioral patterns (such as exercise, rest, and running) are encoded with corresponding heart rate features to generate heart rate cluster feature encoding data. During the encoding process, the heart rate change amplitude, change frequency, peak value and other characteristics are comprehensively considered and converted into digital feature codes.

[0087] The time variation curve of the heart rate characteristics is constructed based on the feature coding data to track the changing trend of the heart rate along with the behavior pattern. The trend of periodic, sudden or gradual changes in the heart rate can be identified through the variation curve.

[0088] According to the heart rate change curve, the fluctuation degree and change range of the heart rate are analyzed to detect abnormal heart rate fluctuations (such as changes that are too fast or too slow). The second user's heart rate evaluation data is generated by comprehensively considering indicators such as the stability of the curve, mutation points and recovery time to determine the user's heart rate health status.

[0089] Step S5: performing remote rehabilitation monitoring auxiliary operations according to the first user's heart rate assessment data and the second user's heart rate assessment data.

[0090] Specifically, the heart rate assessment data of the first user and the heart rate assessment data of the second user are integrated to form a comprehensive heart rate health report of the user. The consistency and difference in different assessment results are analyzed and weighted to generate the final user heart rate health assessment data, so as to generate suggestions through preset parameters and obtain rehabilitation monitoring suggestions. The rehabilitation monitoring suggestions are sent to the user through remote transmission technology, and the user's feedback data is recorded for further adjustment of the monitoring and evaluation plan.

[0091] In the present invention, by extracting time series features from user heart rate data, the method can capture the complex dynamic changes of heart rate and extract key time series features, thereby improving the analysis accuracy of heart rate data. By fitting the user heart rate data with the preset healthy heart rate data, the first user heart rate assessment data is generated, the difference between the user's heart rate and the standard healthy heart rate is identified, and a customized health assessment is provided. Through clustering calculation and feature layer encoding, different behavior-heart rate patterns can be identified based on user behavior data and heart rate data, and these patterns are encoded and analyzed. Change curve construction and trend analysis are introduced. By constructing the heart rate feature change curve, the method can capture the long-term change trend of the user's heart rate. The present invention can quickly respond to abnormal user heart rate conditions based on real-time data and evaluation results, provide immediate feedback and guidance, and improve safety during the rehabilitation process.

[0092] Preferably, step S1 specifically includes:

[0093] Step S11: collecting the user's heart rate through a wearable device to obtain the user's heart rate data;

[0094] Specifically, the user wears a wearable device, such as a smart bracelet or a heart rate monitoring device, and the device obtains the user's pulse signal through an optical sensor. The device converts the pulse signal into heart rate data and records it in units of beats per minute (BPM). The collected data stream is regularly transmitted to the background system for real-time recording and storage to form the user's heart rate time series data.

[0095] Step S12: obtaining user behavior table data and current time data, and generating user behavior according to the user behavior table data and current time data to obtain primary user behavior data;

[0096] Specifically, the user's daily activity information is read from the user behavior table, such as schedule, exercise habits, daily routine, etc. At the same time, the current timestamp data is obtained to determine which time period the user is currently in.

[0097] Match the user behavior table with the current time data to determine the possible user behavior corresponding to the current time period, such as work, exercise, rest, etc. According to the historical behavior records in the behavior table, calculate the most likely behavior in the current time period and generate primary user behavior data. Primary user behavior data includes information such as behavior type, behavior duration, and behavior intensity.

[0098] Step S13: collecting user images through a camera device to obtain user image data, and collecting user posture data based on the user image data to obtain user posture data;

[0099] Specifically, a camera device installed indoors or worn by a user is used for image acquisition, and the camera captures the user's current posture image. The frame rate of the image data is optimized to ensure the continuity of image acquisition and form a continuous user image data stream.

[0100] The human body contour is extracted from the collected user image data to detect the human body position and posture information in the image. According to the key point detection method, the user's joint position, limb direction and other posture features are extracted to generate user posture data. The posture data includes the user's current body angle, hand and foot position, torso inclination and other specific position data of joints and body parts.

[0101] Step S14: generating user estimated posture according to the primary user behavior data to obtain user estimated posture data;

[0102] Specifically, the user's current behavior type (such as walking, sitting, running, etc.) is inferred based on the primary user behavior data, and the user's posture is inferred based on the intensity and duration of the behavior. The user's estimated posture data is generated based on the behavior characteristics, and the possible angles and positions of the user's joints and body parts are inferred. The generated estimated posture data includes the position of hands and feet, joint angles, and overall posture characteristics.

[0103] Step S15: performing error evaluation based on the user estimated posture data and the user posture data to obtain user posture error data;

[0104] Specifically, the user's estimated posture data is compared with the actual user posture data to calculate the difference in each joint position, angle and direction. The difference is calculated for each posture feature (such as arm angle, leg posture, etc.) to obtain the posture error data. The total error value between the user's current posture and the estimated posture is evaluated through weighted calculation of the overall difference.

[0105] Based on the size of the error value, the deviation between the user's current behavior and the estimated behavior is judged to identify behavioral changes. If the error value is large, it may indicate that the user's behavior is inconsistent with the estimate and further adjustment of the behavior judgment is required.

[0106] Step S16: adjusting the primary user behavior data according to the user posture error data to obtain user behavior data.

[0107] Specifically, the primary user behavior data is adjusted based on the posture error data. For larger errors, the user behavior type needs to be re-evaluated. If the user's actual posture data is significantly different from the primary user behavior data, the behavior label may need to be changed. For example, if the user is inferred to be "resting", but the actual posture indicates that the user is standing or moving, the behavior type needs to be adjusted to "standing" or "walking".

[0108] The integration of multi-source data in the present invention ensures more accurate identification of user behavior and can effectively reduce the errors and uncertainties caused by a single data source. The real-time error evaluation and behavior data adjustment mechanism ensures that the system can promptly detect and correct the user's behavior deviations, and improves the system's real-time response capabilities to user behavior. By comparing the user's actual posture and estimated posture, this method can effectively identify the user's movement deviations in rehabilitation training, and make real-time corrections by adjusting the primary behavior data. The adjustment mechanism can be customized according to the specific circumstances of each user, thereby improving the adaptability and effectiveness of the rehabilitation program.

[0109] Preferably, the error assessment is specifically:

[0110] Generating posture components according to the user estimated posture data and the user posture data, respectively obtaining the user estimated posture component data and the user posture component data, wherein the user estimated posture component data includes the user estimated angle component data, the user estimated position component data and the user estimated speed component data, and the user posture component data includes the user angle component data, the user position component data and the user speed component data;

[0111] Specifically, the rotation angles of joints are extracted from the user's estimated posture data, mainly including the three-dimensional rotation angles of joints such as arms, legs, and torso. By decomposing the joint position and direction of the posture data, the rotation angle of each joint is calculated to form the user's estimated angle component data. By calculating the three-dimensional position coordinates of each key point (such as wrist, knee, head, etc.) in the user's estimated posture, the specific position of the user's body parts in space is obtained to generate the user's estimated position component data. According to the time series data of the estimated posture, the displacement and speed of each key point at different time points are calculated to obtain the user's estimated speed component data.

[0112] Extract the rotation angle of the user's joints from the user's actual posture data, calculate its actual angle distribution in space, and generate the user's angle component data. Compare the position information of the user's estimated posture, extract the actual three-dimensional spatial position of each key point from the user's image posture data, and form the user's position component data. According to the time series of the user's posture image, calculate the actual speed of each key point at different time points to obtain the user's speed component data.

[0113] Calculate the error component according to the user's estimated posture component data and the user's posture component data to obtain the user's posture error component data;

[0114] Specifically, the user's estimated angle component data is compared with the user's actual angle component data to calculate the rotation angle difference of each joint. For the three-dimensional rotation angle, it can be decomposed into components on the X, Y, and Z axes, and the difference in each direction is calculated one by one. The angle error of each joint is weighted and the overall angle error component data is generated based on the importance of the joint (for example, the torso posture may be more important than the arm posture).

[0115] Compare the user's estimated position component data with the actual position component data in three-dimensional space, calculate the spatial distance difference of each key point, and generate the position offset. Use Euclidean distance or other distance measurement methods to calculate the spatial distance difference between the actual position and the estimated position of each key point. According to the offset of each key point, the position error is comprehensively evaluated to form the user position error component data.

[0116] Compare the user's estimated speed component data with the user's actual speed component data, and calculate the speed change difference at each key point. Consider the speed changes of each key point at different time points and calculate the speed error component. Combine the magnitude and direction of the speed change to evaluate the speed error as a whole and generate the user speed error component data.

[0117] The user posture error data is fused to obtain the user posture error data.

[0118] Specifically, based on the user posture error component data (angle error, position error, speed error), each component is weighted and fused. Different components have different weights in the overall posture assessment (for example, position error is more important than speed error), so the weight parameters need to be set according to actual needs. The weighted angle, position, and speed error component data are superimposed to obtain the comprehensive user posture error data. The fusion method can consider different calculation models to ensure the balance between angle, position, and speed.

[0119] Generate an overall posture error report based on the fused error data to evaluate the accuracy and deviation of the user's current posture.

[0120] In the present invention, by decomposing the posture data into angle components, position components and speed components, and calculating the errors of these components respectively, the method can perform a more refined analysis of the user's posture to make deeper data adjustments. By generating and calculating posture component data of different dimensions such as angle, position and speed respectively, the user's posture can be evaluated in multiple dimensions. It ensures that the errors of each component can be effectively integrated, and the generated posture error data is more representative and accurate. Through in-depth analysis and synthesis of the error components, the method can generate detailed posture error information for precise posture adjustment.

[0121] Preferably, step S2 specifically includes:

[0122] Step S21: reconstructing the phase space according to the user's heart rate data to obtain the heart rate phase space data;

[0123] Specifically, according to the user's heart rate data, a suitable time delay parameter is first selected. By calculating the time delay parameter, the optimal delay of the heart rate signal is determined. The heart rate time series is reconstructed according to the selected delay to generate time series with multiple embedding dimensions.

[0124] According to the time delay of the heart rate data, the number of embedding dimensions is determined, that is, the coordinate dimensions of the phase space are constructed. The phase space data of the user's heart rate is constructed by mapping the time series points into high-dimensional space points in the phase space.

[0125] Using the selected time delay and embedding dimension, the heart rate time series is mapped into the phase space to generate the heart rate phase space trajectory. The phase space data includes the coordinate points of multiple time series, which represent the dynamic change trajectory of the user's heart rate over time.

[0126] Step S22: constructing a phase space network for the heart rate phase space data to obtain the heart rate phase space network data;

[0127] Specifically, the mutual distances of the time series points in the heart rate phase space data are calculated to determine the Euclidean distance or other distances between each point. According to the set distance threshold, the adjacency matrix of the phase space is constructed, and each element of the matrix represents the connectivity (0 or 1) between two time points in the phase space.

[0128] Each time series point in the phase space is regarded as a node in the network to form a node set. According to the result of the adjacency matrix, the edges between the nodes are determined to generate the user heart rate phase space network structure.

[0129] Analyze the connectivity in the heart rate phase space network, calculate the node degree distribution in the network, and identify the most dense or sparse node connections in the network. The heart rate phase space network data contains the node set and edge set of the network, which describes the mutual correlation of each time point in the heart rate time series.

[0130] Step S23: extracting network topology features from the heart rate phase space network data to obtain heart rate network topology feature data;

[0131] Specifically, for each node in the heart rate phase space network, its node degree, that is, the number of edges connected to the node, is calculated to obtain the node degree distribution characteristics. By analyzing the degree distribution of different nodes, a set of nodes with significant characteristics in the network is extracted.

[0132] The local clustering coefficient of each node in the heart rate phase space network is calculated to measure the degree of connection between the node and its neighboring nodes. The network clustering coefficient describes the local topological structure of the network and reflects the dynamic correlation between certain time points in the heart rate phase space.

[0133] The shortest path length between any two nodes in the heart rate phase space network is calculated to obtain the average path length of the network. The average path length reflects the overall connectivity between each time point in the heart rate time series and can describe the overall trend of heart rate changes.

[0134] Calculate the centrality characteristics of nodes in the network and identify key nodes in the network, such as degree centrality, closeness centrality, and betweenness centrality. These centrality characteristics reflect which nodes in the network play an important role in the heart rate change process and constitute the heart rate network topology feature data.

[0135] Step S24: performing entropy weight analysis based on the heart rate network topology feature data to obtain the user's heart rate time series feature data;

[0136] Specifically, the extracted heart rate network topology feature data is normalized to unify the data scales of different features. For features such as node degree, clustering coefficient, path length, and centrality, the maximum and minimum values ​​are normalized respectively for subsequent entropy weight analysis.

[0137] For each normalized network topology feature, calculate its information entropy value. Information entropy is used to measure the uniformity of feature distribution. The larger the entropy value, the more dispersed the feature is in the network, and the smaller the entropy value, the more concentrated the feature is on a few nodes.

[0138] According to the entropy value of each topological feature, its weight is calculated. Features with small information entropy will get higher weights, and features with large information entropy will get lower weights. According to the feature weights, the network topological feature data is weighted and fused to generate weighted feature values.

[0139] The weighted network topology feature values ​​are used as the time series features of the user's heart rate, reflecting the important features of the user's heart rate change process. The user's heart rate time series feature data describes the complex dynamic changes of the heart rate over time and is used for subsequent health assessment.

[0140] Step S25: Perform clustering calculation based on the user behavior data and the user heart rate data to obtain user behavior and heart rate clustering feature data.

[0141] Specifically, user behavior data (such as motion, stillness, and other behavior labels) are matched with heart rate time series feature data to construct a user behavior-heart rate feature matrix. Ensure that the behavior data and heart rate data are synchronized on the time axis so that the heart rate feature at each time point corresponds to a user behavior label.

[0142] After integrating the user's behavior data and heart rate features, a high-dimensional feature space is constructed, which includes behavior features and heart rate features. The data in the feature space is normalized to ensure that the data scales of different dimensions are consistent.

[0143] In the high-dimensional feature space, data clustering is performed based on the similarity of heart rate and behavior characteristics. By classifying the patterns between the user's heart rate and behavior, different heart rate-behavior patterns of the user are identified. The clustering results assign different heart rate-behavior patterns to different clusters, forming user behavior heart rate clustering feature data.

[0144] The clustering results are analyzed to identify the cluster centers representing different heart rates and behavior patterns. By analyzing the distribution and characteristics of different clusters, user behavior heart rate clustering feature data is generated for subsequent rehabilitation monitoring and auxiliary evaluation.

[0145] In the present invention, through phase space reconstruction, the method can map the original one-dimensional heart rate time series data into a high-dimensional phase space, and the mapping can reveal the dynamic behavior and hidden patterns in the heart rate data, thereby enhancing the depth and breadth of data analysis. The heart rate phase space data is constructed as a network structure, which can capture the structural and correlation characteristics in the heart rate data, and the network analysis based on the heart rate data can reveal the nodes and connection patterns in the heart rate data. By extracting the topological features of the heart rate phase space network data, the method can identify the global and local structural characteristics of the heart rate data, such as node degree distribution, clustering coefficient, average path length, etc. Through entropy weight analysis, the method can quantify the complexity and information distribution of the topological features of the heart rate network. It can help identify the most informative features in the heart rate data, and generate the user's heart rate time series feature data based on these features. By combining user behavior data and heart rate data for clustering calculations, the method can identify the heart rate characteristics of users under different behavior patterns.

[0146] Preferably, step S21 is specifically:

[0147] Determine the phase space embedding parameters according to the user's heart rate data to obtain the embedding delay time data;

[0148] Specifically, a continuous time series is extracted from the user's heart rate data to ensure the integrity and consistency of the heart rate data. The heart rate data is denoised to filter out abnormal heart rate values ​​and abnormally fluctuating noise signals to ensure the smoothness of the data.

[0149] The heart rate time series is segmented, the autocorrelation function value in each time period is calculated, and the autocorrelation changes under different time delays are found. By analyzing the autocorrelation function, the time delay point where the autocorrelation function value first drops to the minimum is selected as the embedded delay time. The embedded delay time data represents the temporal repetitive characteristics of the heart rate signal.

[0150] Calculate the embedding dimension according to the user's heart rate data to obtain the embedding dimension data;

[0151] Specifically, the user's heart rate time series is decomposed using the embedded delay time to generate delayed data series at multiple time points. According to the changing trend of the delayed time series, the range of the embedded dimension is preliminarily estimated. Generally, the higher the dimension, the stronger the ability to describe the dynamic changes of the system.

[0152] By calculating the average mutual information value of the heart rate data, we can find the trend of the average mutual information change under different embedding dimensions. When the mutual information value tends to be stable after increasing the dimension, the embedding dimension at this time is selected as the optimal embedding dimension. The embedding dimension data represents the complexity of the heart rate system, that is, how many dimensions are needed to fully describe the dynamic behavior of heart rate changes.

[0153] The heart rate phase space of the user's heart rate data is constructed according to the embedded delay time data and the embedded dimension data to obtain the heart rate phase space data.

[0154] Specifically, the user's heart rate time series is mapped into the phase space using the embedding delay time and embedding dimension to generate a set of points in the phase space. Each phase space point represents the user's heart rate state at a certain point in time, and the positions of these points in the phase space reflect the dynamic changes of the heart rate signal.

[0155] All phase space points are connected in time order to generate phase space trajectories, which depict the dynamic evolution of the heart rate signal. The phase space trajectory can reflect the periodicity, stability and possible chaotic behavior of heart rate changes.

[0156] The point set and trajectory information of the heart rate phase space are saved as heart rate phase space data as the basic data for subsequent analysis. The heart rate phase space data contains embedded multi-dimensional features, which describes the complex dynamic changes of heart rate over time.

[0157] In the present invention, by accurately calculating the embedding delay time and embedding dimension of the heart rate data, the method can reconstruct a phase space containing more information from the heart rate time series data. The phase space reconstruction method converts the original one-dimensional time series data into a high-dimensional space, which can better capture the dynamic characteristics and hidden patterns of the heart rate data, thereby enhancing the ability to analyze the heart rate behavior. After determining the embedding delay time and embedding dimension, the method can more accurately reconstruct the phase space and reveal the nonlinear characteristics existing in the heart rate data. By mapping the heart rate data to the high-dimensional phase space, the attractor structure and trajectory in the heart rate data can be more clearly displayed, helping to understand the behavior of the cardiac system as a complex dynamic system. The accurate calculation of the embedding delay time and embedding dimension ensures the effectiveness and accuracy of the phase space reconstruction.

[0158] Preferably, step S24 is specifically:

[0159] Perform conditional entropy calculation and joint entropy calculation according to the heart rate network topology feature data to obtain first heart rate entropy value feature data and second heart rate entropy value feature data respectively;

[0160] Specifically, the distribution analysis of the heart rate network topology feature data is performed to determine the connection mode of each node and its neighboring nodes and the local structure of the network. The uncertainty of the topological feature of a node given the topological feature of its neighboring nodes is calculated to obtain the conditional entropy of the node. By calculating the conditional entropy of all nodes in the network, the first heart rate entropy value feature data of the overall network is obtained.

[0161] The joint distribution of nodes and neighbor nodes in the heart rate network topology feature data is counted to calculate the common information between multiple nodes. According to the joint distribution of nodes in the network, the joint entropy value is calculated to reflect the global uncertainty in the entire network system, and the second heart rate entropy value feature data is obtained.

[0162] Performing entropy difference calculation based on the first heart rate entropy value characteristic data and the second heart rate entropy value characteristic data to obtain heart rate entropy difference data;

[0163] Specifically, more importantly, the heart rate entropy difference data includes the first heart rate entropy difference data and the second heart rate entropy difference data, and the entropy difference calculation is specifically as follows:

[0164] Calculate the node entropy difference according to the first heart rate entropy value characteristic data and the second heart rate entropy value characteristic data to obtain the first heart rate entropy difference data;

[0165] A global entropy difference calculation is performed based on the first heart rate entropy value characteristic data and the second heart rate entropy value characteristic data to obtain second heart rate entropy difference data.

[0166] Specifically, the first heart rate entropy feature data and the second heart rate entropy feature data are compared to calculate the difference between the conditional entropy and the joint entropy of each node. The entropy difference of each node in the heart rate network is calculated using a difference function (such as the absolute value of the entropy difference) to form node-level entropy difference data.

[0167] Specifically, the entropy differences of all nodes are normalized to ensure that the entropy difference data of different nodes are compared at the same scale. The global entropy difference value of the network is calculated to reflect the overall entropy difference of the heart rate network and obtain the heart rate entropy difference data.

[0168] Performing differential weighting processing on the heart rate network topology feature data according to the entropy difference data to obtain differential weighted heart rate feature data;

[0169] Specifically, different weights are assigned to each node in the heart rate network topology feature data according to the entropy difference data of each node. The greater the entropy difference of a node, the higher its importance in the network, and the weight increases accordingly. The heart rate network topology feature data is weighted so that nodes with larger entropy differences have a greater impact on the overall heart rate characteristics.

[0170] The weighted processing results of all nodes are summarized to generate differentiated weighted heart rate feature data. This data reflects the importance of different nodes in the heart rate network and the dynamic changes of the network at the global and local levels.

[0171] Nonlinear feature mapping is performed based on the differentiated weighted heart rate feature data to obtain the user's heart rate time series feature data.

[0172] Specifically, using differentiated weighted heart rate feature data as input, a nonlinear mapping model is constructed to capture the complex nonlinear relationship in the topological feature data of the heart rate network. By adjusting the parameters of the model, it is ensured that the mapping process can reflect the potential nonlinear changes in the heart rate network.

[0173] The differentiated weighted heart rate feature data is mapped through a nonlinear mapping model to generate a new feature space. This process can reveal the nonlinear association and dynamic evolution of the heart rate features. Feature extraction is performed on the mapped data to generate user heart rate time series feature data, reflecting the complex dynamic characteristics of the user's heart rate changing over time.

[0174] A multi-layer neural network in deep learning (such as a multi-layer perceptron or a graph-based deep learning model, such as a graph neural network) is used, and differentially weighted heart rate feature data is used as input. By adjusting the parameters of the model, the complex dynamic features are captured. Differentially weighted heart rate feature data, which includes feature weights of nodes and edges extracted from the topological features of the heart rate network, or weight data after entropy weight analysis. Assuming that the input feature dimension is n, the input data is a set of n-dimensional vectors, representing the weighted data of the heart rate topological features. Each data point contains information about topological features such as nodes, edges, local clustering coefficients, and path lengths. The middle layer consists of three hidden layers, each of which includes input, number of nodes, weight matrix, bias term, and activation function parameter settings. It is iteratively trained through the loss function (mean square error calculation formula), and finally output through the output layer (fully connected layer).

[0175] Specifically, a similar nonlinear mapping calculation method is used for mapping, such as ReLU or Leaky ReLU.

[0176] In the present invention, by respectively calculating the conditional entropy and joint entropy of the heart rate network topological feature data, the amount of information and complexity in the heart rate data can be analyzed at multiple levels. The conditional entropy can reveal the randomness and uncertainty of the heart rate characteristics under a specific state, while the joint entropy can quantify the total amount of information in the entire system. Through entropy difference calculation, the method can identify the difference between the conditional entropy and the joint entropy, reflecting the change in the amount of information of the heart rate network topological feature data under different conditions. By weighting these differences, the method can highlight those features that are more sensitive to heart rate changes. Through differential weighted processing, the heart rate feature data that show significant differences under different entropy conditions are extracted and emphasized. Subsequently, using nonlinear feature mapping technology, the method can map these weighted features to a high-dimensional space, thereby capturing the deep relationships and patterns hidden in the heart rate time series data. This nonlinear mapping not only improves the expressiveness of the features, but also enhances the ability to capture complex dynamic behaviors.

[0177] Preferably, step S3 is specifically:

[0178] Step S31: dynamically matching the user's heart rate time series characteristic data and preset healthy heart rate time series characteristic data to obtain heart rate characteristic matching data;

[0179] Specifically, the user's heart rate time series feature data is aligned with the preset healthy heart rate time series feature data on the time axis to ensure that the time nodes of the two data sets correspond. During the alignment process, if there is a time difference, the interpolation method is used to supplement or correct the time point to ensure that the two sets of feature data have a consistent time interval.

[0180] Specifically, based on the alignment results of the user's heart rate time series features and the healthy heart rate feature data, dynamic matching analysis is performed. The feature differences between the two at different time points are compared to generate matching data. The feature differences at each time point are calculated to form matching data. The matching data includes the similarity or difference of the heart rate features at each time point.

[0181] Step S32: performing residual calculation according to the heart rate feature matching data to obtain heart rate feature matching residual data;

[0182] Specifically, the difference between the user's heart rate feature data and the healthy heart rate feature data at each time point is calculated to obtain the residual at each time point. The residual represents the deviation between the user's heart rate feature and the healthy heart rate feature. The residual data is the difference between the user's actual heart rate state and the expected health state, reflecting the abnormality of the user's heart rate. The residual data at each time point is aggregated to form a residual data sequence for heart rate feature matching.

[0183] Step S33: performing multivariate linear regression fitting on the heart rate feature matching residual data to obtain the user's heart rate deviation data;

[0184] Specifically, a multivariate linear regression model is constructed using the residual data of heart rate feature matching as input. The goal of the model is to analyze the deviation trend of heart rate features by linearly fitting the residual data. The independent variables in the regression model are time points or other features related to heart rate, and the dependent variables are residual values.

[0185] The residual data is fitted through a multivariate linear regression model, and the regression coefficient is calculated to obtain the heart rate deviation trend. The fitting result indicates the overall trend and specific degree of deviation of the heart rate characteristics from the healthy state, forming the user's heart rate deviation data.

[0186] Step S34: performing residual curve analysis on the user's heart rate deviation data to obtain first user heart rate evaluation data.

[0187] Specifically, a residual curve is generated based on the user's heart rate deviation data. The residual curve shows the dynamic changes of the user's heart rate deviation from the healthy state. The fluctuation amplitude and trend change of the curve can reflect the severity and duration of the heart rate deviation.

[0188] Analyze the main features of the residual curve (such as peaks, troughs, slope, etc.) to identify the key periods and change patterns of heart rate deviation. By analyzing the trend of the residual curve, the user's heart rate deviation is judged through the preset threshold to determine whether it is stabilizing, deteriorating or improving, and generate heart rate health assessment data.

[0189] The results of the residual curve analysis are combined to obtain a comprehensive evaluation conclusion of the user's heart rate deviation from the healthy state, forming the first user heart rate evaluation data. The evaluation data includes the degree of deviation and deviation trend of the user's heart rate, which serves as the basis for subsequent rehabilitation monitoring.

[0190] The present invention dynamically matches the user's heart rate time series feature data with the preset healthy heart rate time series feature data, which can flexibly adapt to the changing trend of the individual heart rate. Dynamic matching can capture the slight changes in the user's heart rate data and compare it with the health standard in real time, ensuring the accuracy of the matching process, thereby providing high-quality basic data for subsequent residual analysis and evaluation. By performing residual calculation on the heart rate feature data after dynamic matching, the specific degree of deviation between the user's heart rate and the health standard is quantified. The residual data directly reflects the unconventional situation of the user's heart rate, providing users with more intuitive in-depth data. By performing multivariate linear regression fitting on the residual data, multi-dimensional heart rate feature information can be integrated, and the overall deviation of the user's heart rate can be comprehensively analyzed. The residual curve analysis of the user's heart rate deviation data can identify the trend and pattern of the heart rate deviation. Through the residual curve, this method can not only quantify the degree of heart rate deviation, but also evaluate the persistence and volatility of the deviation.

[0191] Preferably, step S4 is specifically:

[0192] Step S41: performing cluster feature coding according to the user behavior heart rate cluster feature data to obtain heart rate cluster feature coding data;

[0193] Specifically, each cluster in the user behavior heart rate cluster feature data is parsed to extract the cluster center and the distance between each data point and the cluster center, distribution characteristics, etc. Based on these cluster features, a coding rule is defined to convert the features of each cluster into a set of digital codes.

[0194] Each cluster is encoded and its features are converted into cluster feature encoding data. The encoding method can encode different clusters in order according to features such as cluster center, cluster radius, point density, etc., to generate heart rate cluster feature encoding data, which represents the differences between each cluster and its characteristic pattern.

[0195] Step S42: embedding the encoded features of the heart rate clustering feature encoded data to obtain embedded heart rate feature data;

[0196] Specifically, the heart rate cluster feature encoding data is preprocessed to unify the length and dimension of the encoding data to ensure that different encoding features have the same data format. The encoding data is standardized to eliminate the scale differences between different encoding values.

[0197] A feature embedding model is constructed to map the cluster feature encoding data into a multidimensional space, so that different features can better reflect their positions and relationships in the cluster. Through the embedding model, embedded heart rate feature data is generated, and the embedded feature data can capture the deep associations between the encoded features.

[0198] Step S43: constructing a characteristic time series for the embedded heart rate characteristic data to obtain heart rate clustering time series characteristic data;

[0199] Specifically, according to the heart rate feature embedded data, the data points are sorted in chronological order to generate a time series structure. Each time point corresponds to an embedded heart rate feature, forming a complete heart rate clustering time series feature data.

[0200] Smoothing the time series features eliminates short-term fluctuations in the data and makes the changes in the series more stable and coherent. The smoothed time series can reflect the overall trend of heart rate changes over time.

[0201] Step S44: performing multi-scale curve analysis on the heart rate clustering time series feature data to obtain heart rate feature change curve data;

[0202] Specifically, the heart rate clustering time series feature data is decomposed into different time scales, and the change patterns of heart rate features are analyzed at different time scales. On the short-term scale, the fast-fluctuating heart rate changes are analyzed; on the long-term scale, the slowly changing trends are analyzed.

[0203] The heart rate characteristics at different scales are curve-fitted to generate multiple heart rate change curves. Each curve corresponds to a time scale, reflecting the dynamic changes of the heart rate at that scale. The curve data of all scales are aggregated to generate the heart rate characteristic change curve data, which comprehensively displays the multi-scale change characteristics of the heart rate.

[0204] Step S45: performing curve pattern recognition on the heart rate characteristic change curve data to obtain heart rate characteristic curve pattern data;

[0205] Specifically, according to the heart rate characteristic change curve data, the main patterns in the curve are identified, such as periodic fluctuations, mutation points, stable segments, etc. Pattern classification is performed on each change curve, and the curve data is marked according to the identified pattern to obtain the pattern data of the curve.

[0206] The characteristic parameters of each curve pattern are extracted, such as cycle length, fluctuation amplitude, frequency change, etc., to form heart rate characteristic curve pattern data, which can help identify typical characteristics in the user's heart rate changes, such as abnormal heart rate fluctuations or recovery trends.

[0207] Step S46: Perform curve trend fitting on the heart rate characteristic curve pattern data to obtain the second user's heart rate evaluation data.

[0208] Specifically, the curve pattern data is used to perform trend fitting on the changes in heart rate characteristics and analyze the overall change direction of the curve. During the fitting process, by calculating the key features such as the rising or falling trend and inflection point of each curve, the state trend of the heart rate value is judged based on the features. Linear or nonlinear fitting is performed on the curve data of multiple time periods to generate the trend data of the user's heart rate in each time period.

[0209] Combined with the fitted trend data, the overall change trend of the user's heart rate value is evaluated to determine whether the heart rate value is in a normal or abnormal state, and the second user's heart rate evaluation data is generated. The second user's heart rate evaluation data includes the deviation degree, trend, and abnormal fluctuation of the heart rate.

[0210] In the present invention, by encoding the user behavior heart rate clustering feature data, the method can effectively encode and represent the behavior and heart rate pattern, can structure the user's behavior and heart rate data, and reveal the heart rate response characteristics under the behavior pattern. After embedding the cluster feature encoding data, the method can map the high-dimensional features to the low-dimensional space, thereby compressing the dimension of the data, which not only reduces the computational complexity, but also retains the key information in the original feature data, so that the system can process complex data more efficiently. By constructing the embedded heart rate feature data in a time series, the method can capture the dynamic change characteristics of the user's heart rate data. The time series data can reflect the changing trend of the user's heart rate under different behavior patterns, help to identify the periodic fluctuations of the heart rate and the behavior-related change patterns, and further enhance the time sensitivity of the heart rate data analysis. By performing multi-scale curve analysis on the heart rate clustering time series feature data, the method can analyze the heart rate data on different time scales and capture the changes in the heart rate characteristics in the short and long term. By performing pattern recognition on the heart rate characteristic change curve data, this method can extract the typical heart rate change pattern of the user, which is helpful to analyze the heart rate change pattern of the user under different behavior patterns. After completing the curve pattern recognition, this method can predict the future change trend of the user's heart rate through trend fitting.

[0211] Preferably, step S5 is specifically:

[0212] Performing heart rate assessment difference analysis based on the heart rate assessment data of the first user and the heart rate assessment data of the second user to obtain heart rate assessment difference data;

[0213] Specifically, the heart rate assessment data of the first user and the heart rate assessment data of the second user are aligned according to the same time period. During the alignment process, for data with inconsistent time points, interpolation or smoothing is used to fill in the missing data to ensure the continuity of the time series.

[0214] The heart rate evaluation data of the first user and the heart rate evaluation data of the second user at each time point are compared point by point to calculate the difference between the two. The difference may include deviation of the heart rate value, trend difference, curve fluctuation difference, etc. The difference value may be expressed by calculating the absolute difference or relative difference (such as percentage difference) to obtain the heart rate evaluation difference data, which reflects the evaluation difference of the user's heart rate state under different analysis methods.

[0215] Statistical analysis of the heart rate assessment difference data was performed to identify time periods with large differences or time points with large abnormal fluctuations. By analyzing these difference points, the main differences between the assessment methods were determined, and possible abnormalities or deviations in the heart rate assessment trends were identified.

[0216] The first user's heart rate assessment data, the second user's heart rate assessment data and the heart rate assessment difference data are used to generate a remote rehabilitation report to obtain remote rehabilitation report data for remote rehabilitation monitoring auxiliary operations.

[0217] Specifically, the first user's heart rate assessment data, the second user's heart rate assessment data and the heart rate assessment difference data are integrated into a comprehensive assessment data set. The data set contains the multi-angle assessment results of the user's heart rate, as well as the difference information between the assessments. Based on the integrated assessment data, a remote rehabilitation report is automatically generated. The report content includes the user's heart rate status assessment results, heart rate change trends, abnormal fluctuation points, health deviation degree and difference analysis conclusions. The report should also include a visualized heart rate change curve chart, trend analysis chart, etc., so that users can intuitively understand their own heart rate health status.

[0218] Report generation can be fitted and mapped through the preset user heart rate health knowledge graph to obtain health suggestions, such as exercise adjustment, lifestyle improvement, etc., to help users conduct remote rehabilitation.

[0219] The generated remote rehabilitation report data is stored in the cloud system or remote monitoring platform to ensure data security and accessibility. Through remote network transmission technology, the rehabilitation report is sent to the user's device or rehabilitation team for remote rehabilitation monitoring assistance.

[0220] In the present invention, by performing a difference analysis on the heart rate evaluation data of the first user and the heart rate evaluation data of the second user, the method can accurately identify the differences between different evaluation dimensions of the user's heart rate. Combining the first and second user heart rate evaluation data can perform a multi-dimensional comprehensive evaluation of the user's heart rate state. The first user's heart rate evaluation data focuses on the characteristics of the time series, while the second user's heart rate evaluation data covers the changing characteristics in the behavior pattern. By integrating the first and second user heart rate evaluation data and the difference analysis data to generate a remote rehabilitation report, customized, real-time, data-driven remote rehabilitation management is achieved, which significantly improves the intelligence and efficiency of the rehabilitation system.

[0221] Preferably, the present application also provides an artificial intelligence-based remote rehabilitation monitoring assistance system, which is used to execute the artificial intelligence-based remote rehabilitation monitoring assistance method as described above, and the artificial intelligence-based remote rehabilitation monitoring assistance system includes:

[0222] User heart rate basic data collection module, used to obtain user heart rate data and user behavior data;

[0223] A user heart rate feature extraction module is used to extract time series features based on user heart rate data to obtain user heart rate time series feature data, and to perform clustering calculation based on user behavior data and user heart rate data to obtain user behavior heart rate clustering feature data;

[0224] A first user heart rate evaluation module, used for fitting the user's heart rate time series characteristic data and preset healthy heart rate time series characteristic data to obtain first user heart rate evaluation data;

[0225] A second user heart rate assessment module is used to perform feature layer encoding according to the user behavior heart rate clustering feature data to obtain heart rate clustering feature encoding data; construct a change curve according to the heart rate clustering feature encoding data to obtain heart rate feature change curve data; perform health assessment according to the heart rate feature change curve data to obtain second user heart rate assessment data;

[0226] The remote rehabilitation monitoring auxiliary operation module is used to perform remote rehabilitation monitoring auxiliary operations according to the first user's heart rate evaluation data and the second user's heart rate evaluation data.

[0227] The beneficial effect of the present invention is that by combining user heart rate data and behavior data, the present invention realizes the deep fusion of multi-source data. The user heart rate data reflects the individual's physiological response, while the behavior data provides the environment and activity context. The combination of data makes the analysis more comprehensive and insightful. By extracting the time series features of the user's heart rate data (step S2), the present method can accurately capture the time series dynamic features in the heart rate signal, including heart rate variability, fluctuation rules, etc. By combining the user's heart rate and behavior data for clustering calculation (step S2), the present method can automatically identify the heart rate change rules of the user under different behavior modes, which not only enhances the data classification ability, but also can accurately capture the association between specific behaviors and heart rate. Through the encoding and embedding of clustering features, the present invention can uniformly represent and process heart rate data under different modes (step S4). By fitting the user's heart rate time series feature data with the health standard (step S3), the present invention can automatically evaluate the user's heart rate deviation and quantitatively evaluate the gap between the user and the health status through residual analysis. By integrating the two types of evaluation data (step S5), the system can conduct a more in-depth and differentiated evaluation of the user's heart rate health, realize customized, real-time, data-driven remote rehabilitation management, and significantly improve the intelligence and efficiency of the rehabilitation system.

[0228] Therefore, from any point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the attached application documents rather than the above description, and it is intended that all changes falling within the meaning and scope of equivalent elements of the application documents are included in the present invention.

[0229] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A remote rehabilitation monitoring auxiliary method based on artificial intelligence, characterized in that: The following steps are involved: Step S1: Obtain user heart rate data and user behavior data; Step S2: extracting time series features based on the user's heart rate data to obtain time series feature data of the user's heart rate, and performing clustering calculation based on the user's behavior data and the user's heart rate data to obtain user behavior heart rate clustering feature data; Step S3: fitting the user's heart rate time series characteristic data and preset healthy heart rate time series characteristic data to obtain first user heart rate evaluation data; Step S4: performing feature layer encoding according to the user behavior heart rate clustering feature data to obtain heart rate clustering feature encoding data; Constructing a change curve according to the heart rate clustering feature coding data to obtain heart rate feature change curve data; performing health assessment according to the heart rate feature change curve data to obtain heart rate assessment data of the second user; Step S5: performing remote rehabilitation monitoring auxiliary work according to the heart rate evaluation data of the first user and the heart rate evaluation data of the second user; Wherein step S2 is specifically: Step S21: reconstructing the phase space according to the user's heart rate data to obtain the heart rate phase space data; Step S22: constructing a phase space network for the heart rate phase space data to obtain the heart rate phase space network data; Step S23: extracting network topology features from the heart rate phase space network data to obtain heart rate network topology feature data; Step S24: performing entropy weight analysis based on the heart rate network topology feature data to obtain the user's heart rate time series feature data; Step S25: performing clustering calculation based on the user behavior data and the user heart rate data to obtain user behavior and heart rate clustering feature data; Wherein step S21 is specifically as follows: Determine the phase space embedding parameters according to the user's heart rate data to obtain the embedding delay time data; Calculate the embedding dimension according to the user's heart rate data to obtain the embedding dimension data; The heart rate phase space of the user's heart rate data is constructed according to the embedded delay time data and the embedded dimension data to obtain the heart rate phase space data.

2. The method according to claim 1, characterized in that Step S1 is specifically as follows: The user's heart rate is collected through a wearable device to obtain the user's heart rate data; Obtaining user behavior table data and current time data, and generating user behavior based on the user behavior table data and current time data to obtain primary user behavior data; Capturing user images through a camera device to obtain user image data, and capturing user posture data based on the user image data to obtain user posture data; Generate user estimated posture according to primary user behavior data to obtain user estimated posture data; Perform error evaluation based on the user's estimated posture data and the user's posture data to obtain user posture error data; The primary user behavior data is adjusted according to the user posture error data to obtain the user behavior data.

3. The method according to claim 2, characterized in that The error assessment is specifically as follows: Generating posture components according to the user estimated posture data and the user posture data, respectively obtaining the user estimated posture component data and the user posture component data, wherein the user estimated posture component data includes the user estimated angle component data, the user estimated position component data and the user estimated speed component data, and the user posture component data includes the user angle component data, the user position component data and the user speed component data; Calculate the error component according to the user's estimated posture component data and the user's posture component data to obtain the user's posture error component data; The user posture error data is fused to obtain the user posture error data.

4. The method according to claim 1, characterized in that Step S24 is specifically as follows: Perform conditional entropy calculation and joint entropy calculation according to the heart rate network topology feature data to obtain first heart rate entropy value feature data and second heart rate entropy value feature data respectively; Performing entropy difference calculation based on the first heart rate entropy value characteristic data and the second heart rate entropy value characteristic data to obtain heart rate entropy difference data; Performing differential weighting processing on the heart rate network topology feature data according to the entropy difference data to obtain differential weighted heart rate feature data; Nonlinear feature mapping is performed based on the differentiated weighted heart rate feature data to obtain the user's heart rate time series feature data.

5. The method according to claim 1, characterized in that Step S3 is specifically as follows: Dynamically matching the user's heart rate time series feature data with preset healthy heart rate time series feature data to obtain heart rate feature matching data; Perform residual calculation based on the heart rate feature matching data to obtain heart rate feature matching residual data; Perform multivariate linear regression fitting on the residual data of heart rate feature matching to obtain the user's heart rate deviation data; Perform residual curve analysis on the user's heart rate deviation data to obtain first user heart rate evaluation data.

6. The method according to claim 1, characterized in that Step S4 is specifically as follows: Perform cluster feature encoding according to the user behavior heart rate cluster feature data to obtain heart rate cluster feature encoding data; Embedding the encoded features of the heart rate clustering feature encoded data to obtain embedded heart rate feature data; Constructing feature time series for embedded heart rate feature data to obtain heart rate clustering time series feature data; Perform multi-scale curve analysis on the heart rate clustering time series feature data to obtain the heart rate feature change curve data; Performing curve pattern recognition on the heart rate characteristic change curve data to obtain heart rate characteristic curve pattern data; The heart rate characteristic curve pattern data is subjected to curve trend fitting to obtain the second user's heart rate evaluation data.

7. The method according to claim 1, characterized in that Step S5 is specifically as follows: Performing heart rate assessment difference analysis based on the heart rate assessment data of the first user and the heart rate assessment data of the second user to obtain heart rate assessment difference data; The first user's heart rate assessment data, the second user's heart rate assessment data and the heart rate assessment difference data are used to generate a remote rehabilitation report to obtain remote rehabilitation report data for remote rehabilitation monitoring auxiliary operations.

8. A remote rehabilitation monitoring auxiliary system based on artificial intelligence, characterized in that: For executing the artificial intelligence-based remote rehabilitation monitoring assistance method as claimed in claim 1, the artificial intelligence-based remote rehabilitation monitoring assistance system comprises: User heart rate basic data collection module, used to obtain user heart rate data and user behavior data; A user heart rate feature extraction module is used to extract time series features based on user heart rate data to obtain user heart rate time series feature data, and to perform clustering calculation based on user behavior data and user heart rate data to obtain user behavior heart rate clustering feature data; A first user heart rate evaluation module, used for fitting the user's heart rate time series characteristic data and preset healthy heart rate time series characteristic data to obtain first user heart rate evaluation data; A second user heart rate assessment module is used to perform feature layer encoding according to the user behavior heart rate clustering feature data to obtain heart rate clustering feature encoding data; construct a change curve according to the heart rate clustering feature encoding data to obtain heart rate feature change curve data; perform health assessment according to the heart rate feature change curve data to obtain second user heart rate assessment data; The remote rehabilitation monitoring auxiliary operation module is used to perform remote rehabilitation monitoring auxiliary operations according to the first user's heart rate evaluation data and the second user's heart rate evaluation data.

Citation Information

Patent Citations

  • Health data sharing method and system based on Internet of Things

    CN116913541A

  • Intelligent auxiliary system and method for rehabilitation nursing

    CN117556220A