A comprehensive risk early warning monitoring method and system based on smart wearable devices

Abstract

The present invention discloses an all-round risk early warning monitoring method and system based on smart wearable devices, which relates to the field of health management technology, including collecting and compressing multimodal data through smart wearable devices, transmitting the compressed data to a central database and restoring the data; performing multimodal data analysis on the restored multimodal data and extracting comprehensive features, and constructing a logistic regression model to evaluate the user's health score. The present invention collects and compresses multimodal data through smart wearable devices, transmits the compressed data to a central database and restores the data; performing multimodal data analysis on the restored multimodal data and extracting comprehensive features, and constructing a logistic regression model to evaluate the user's health score; improving data processing and transmission efficiency, improving the accuracy and response speed of health monitoring, and enhancing the accuracy of data analysis.

Description

Technical Field

[0001] The present invention relates to the field of health management technology, and in particular to an all-round risk early warning monitoring method and system based on smart wearable devices. Background Art

[0002] With the rapid development of smart wearable devices, personalized health monitoring and management has become an important research direction. Traditional health monitoring methods rely on regular check-ups in hospitals or clinics and cannot capture users' dynamic physiological data in real time. The development of smart wearable devices not only makes real-time and comprehensive health monitoring possible, but also can collect multimodal data through sensors, including physiological parameters such as heart rate, blood oxygen, body temperature, and environmental factors such as air quality and sound. Through the Internet of Things technology, these data can be transmitted to the central database in real time for processing and analysis, thereby providing users with personalized health warning services. In the existing technology, multimodal data analysis technology has been widely used, but most technologies still have a lot of room for improvement in data processing accuracy, real-time and accuracy of personalized health assessment.

[0003] Although existing technologies have made certain progress in health monitoring and risk warning, there are still several obvious shortcomings. The data preprocessing and compression methods are not efficient enough, especially in the transmission and storage of multimodal data. There is a lot of data redundancy and the compression rate is not high, which can easily cause system performance degradation. Most of the existing health risk assessment models are based on only a single data source, ignoring the potential correlation of multimodal data, making it difficult to form a comprehensive and accurate health risk assessment model, as well as affecting users' daily movement and sensory disorders. Summary of the invention

[0004] In view of the problems existing in the above-mentioned existing omnidirectional risk early warning monitoring method and system based on smart wearable devices, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is that the data preprocessing and compression methods are not efficient enough, especially in the process of multimodal data transmission and storage, there is a lot of data redundancy and the compression rate is not high, which can easily cause the system performance to decline. Most of the existing health risk assessment models are based on only a single data source, ignoring the potential correlation of multimodal data, making it difficult to form a comprehensive and accurate health risk assessment model, as well as affecting the user's daily movement and sensory disorders.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a comprehensive risk early warning monitoring method based on smart wearable devices, which includes collecting and compressing multimodal data through smart wearable devices, transmitting the compressed data to a central database and restoring the data;

[0007] Perform multimodal data analysis on the restored multimodal data and extract comprehensive features, and build a logistic regression model to evaluate the user's health score;

[0008] Re-collect users’ multimodal data for monitoring, and build a user visualization interface to display the monitoring results in real time.

[0009] As a preferred solution of the all-round risk early warning monitoring method based on smart wearable devices of the present invention, wherein: the collecting and compressing of multimodal data through smart wearable devices refers to the user collecting and preprocessing multimodal data through smart wearable devices,

[0010] The smart wearable device includes optical heart rate, blood oxygen, body temperature, skin electricity, air quality and sound sensors;

[0011] The multimodal data includes heart rate, blood oxygen, body temperature, galvanic skin response, air quality and environmental sound data;

[0012] The preprocessing includes: using a dynamic time warping algorithm to time align the multimodal data, using a Z-score algorithm to identify the multimodal data and delete outliers, using an adaptive Kalman filter algorithm to denoise the multimodal data, and standardizing the denoised data;

[0013] Initialize the dictionary to a random matrix;

[0014] Use Lasso regression to solve the sparse coefficients;

[0015] Use singular value decomposition to update the dictionary, iteratively update the dictionary and sparse coefficients, and gradually optimize the dictionary;

[0016] Use the trained dictionary to sparsely represent the standardized multimodal data to obtain sparse data;

[0017] The Toeplitz matrix is ​​used to convert sparse data into compressed data for compressed sensing.

[0018] As a preferred solution of the all-round risk early warning monitoring method based on smart wearable devices of the present invention, wherein: the transmission of compressed data to the central database and data restoration refers to serializing the compressed data using a standard data packaging protocol and preparing to send it to the central database via a wireless network;

[0019] The central database receives the compressed data and unpacks it, restoring the multimodal data using the same compressed sensing matrix and sparse representation dictionary as used during compression.

[0020] As a preferred solution of the all-round risk early warning monitoring method based on smart wearable devices of the present invention, wherein: the multimodal data analysis of the restored multimodal data and extraction of comprehensive features refers to setting initial weights for the restored multimodal data according to the collected multimodal data;

[0021] Use the dynamic weight update formula to update the initial weight. The formula is:

[0022] ;

[0023] in is the current weight of the i-th restored multimodal data at time t, is the rate of change of the i-th restored multimodal data at time t, is the adjustment coefficient;

[0024] Use the Pearson correlation coefficient to calculate the correlation coefficient of the restored multimodal data at time t ;

[0025] Calculate the weighted Pearson correlation coefficient based on the current weights and correlation coefficients of the restored multimodal data ;

[0026] Based on the weighted Pearson correlation coefficient Absolute value, calculate the association weight between each pair of restored multimodal data ;

[0027] Based on the weighted Pearson correlation coefficient and associated weights And calculate the comprehensive characteristics at time t, the formula is:

[0028] ;

[0029] in is the comprehensive feature at time t, and N is the total number of restored multimodal data.

[0030] As a preferred solution of the all-round risk early warning monitoring method based on smart wearable devices described in the present invention, wherein: the construction of a logistic regression model to evaluate the health score of a user refers to collecting multimodal data with historical labels and preprocessing it;

[0031] Construct a logistic regression model to calculate the health risk score at time t ;

[0032] The weights of positive and negative classes are calculated based on the number of multimodal data category samples with historical labels. The formula is:

[0033] ; ;

[0034] Where M is the total number of class samples, is the number of positive samples, is the number of negative class samples;

[0035] Use the weighted cross entropy loss function to calculate the loss value L at time t , the formula is:

[0036] ;

[0037] Where T is the total time of multimodal data, is the weight of the positive class, is the weight of the negative class, is the true label, T is the time;

[0038] Calculate the partial derivative of the loss function with respect to the bias term ;

[0039] Using exponential decay to calculate the learning rate ;

[0040] Updating the bias term of a logistic regression model using the step-down formula ;

[0041] When the number of iterations reaches the set maximum number of iterations, the iteration stops.

[0042] As a preferred solution of the all-round risk early warning monitoring method based on smart wearable devices of the present invention, wherein: the re-collecting of the user's multimodal data for monitoring refers to re-collecting the user's multimodal data through the smart wearable device and pre-processing and extracting comprehensive features;

[0043] The extracted comprehensive features are input into the trained logistic regression model to calculate the health risk score, and the health risk score at time t is obtained;

[0044] The judgment threshold is set according to medical guidelines and doctor's advice. If the health risk score is greater than or equal to the judgment threshold, it is judged as having a health risk, and an early warning notification is sent through the smart device to remind the user to check the physical condition. If the health risk score is less than the judgment threshold, it is judged as risk-free, and the user's multimodal data is monitored persistently.

[0045] As a preferred solution of the all-round risk early warning monitoring method based on smart wearable devices described in the present invention, wherein: the construction of a user visualization interface to display the monitoring results in real time refers to using React.js to construct a user visualization interface to display the user's monitoring results in real time, and using D3.js to display the health risk score at time t;

[0046] Design an early warning panel at the top of the interface to display the currently triggered early warning information;

[0047] Customers who have passed real-name verification are allowed to view the information.

[0048] Another object of the present invention is to provide a comprehensive risk early warning monitoring system based on smart wearable devices, which includes:

[0049] The collection and transmission module is used to collect and compress multimodal data through smart wearable devices, transmit the compressed data to the central database and restore the data;

[0050] Extraction building module, used to perform multimodal data analysis on the restored multimodal data and extract comprehensive features, and build a logistic regression model to evaluate the user's health score;

[0051] The detection visualization module is used to re-collect the user's multimodal data for monitoring and build a user visualization interface to display the monitoring results in real time.

[0052] A computer device comprises: a memory and a processor; the memory stores a computer program, and the processor implements the steps of the above-mentioned all-round risk early warning monitoring method based on smart wearable devices when executing the computer program.

[0053] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned all-round risk early warning monitoring method based on smart wearable devices.

[0054] The beneficial effects of the present invention are as follows: the present invention collects and compresses multimodal data through smart wearable devices, transmits the compressed data to a central database and performs data restoration; performs multimodal data analysis on the restored multimodal data and extracts comprehensive features, and constructs a logistic regression model to evaluate the user's health score; improves data processing and transmission efficiency, improves the accuracy and response speed of health monitoring, and enhances the accuracy of data analysis, thereby improving the safety of users' daily operations and the comprehensive assessment ability of sensory disorders. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0056] Figure 1 It is a flowchart of a comprehensive risk early warning monitoring method based on smart wearable devices;

[0057] Figure 2 This is a structural diagram of the comprehensive risk early warning monitoring system based on smart wearable devices. DETAILED DESCRIPTION

[0058] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0059] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0060] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0061] Example 1, reference Figure 1 , which is the first embodiment of the present invention, and which provides a comprehensive risk early warning monitoring method based on a smart wearable device, and the comprehensive risk early warning monitoring method based on a smart wearable device includes:

[0062] S1. Collect and compress multimodal data through smart wearable devices, transmit the compressed data to the central database and restore the data;

[0063] Specifically, collecting and compressing multimodal data through smart wearable devices means that users collect and pre-process multimodal data through smart wearable devices.

[0064] The smart wearable device includes optical heart rate, blood oxygen, body temperature, skin electricity, air quality and sound sensors;

[0065] The multimodal data includes heart rate, blood oxygen, body temperature, galvanic skin response, air quality and environmental sound data;

[0066] The preprocessing includes: using a dynamic time warping algorithm to time align the multimodal data, using a Z-score algorithm to identify the multimodal data and delete outliers, using an adaptive Kalman filter algorithm to denoise the multimodal data, and standardizing the denoised data;

[0067] Initialize the dictionary to a random matrix;

[0068] Using Lasso regression to solve sparse coefficients , the sparse solution is performed by minimizing the following objective function, the formula is:

[0069] ;

[0070] in For the dictionary, is the i-th multimodal data at time t, is the regularization coefficient, which is used to adjust the balance between sparsity and approximation accuracy. is the L1 norm, used to control sparsity;

[0071] Use singular value decomposition to update the dictionary, iteratively update the dictionary and sparse coefficients, and gradually optimize the dictionary;

[0072] Use the trained dictionary to sparsely represent the standardized multimodal data to obtain sparse data;

[0073] The Toeplitz matrix is ​​used to convert sparse data into compressed data for compressed sensing.

[0074] The advantage of multimodal data is that it can provide all-round health status information, so that the system can not only monitor the individual's physiological state, but also evaluate the environmental factors in which it is located. The data is time-aligned through the dynamic time warping algorithm to keep the data collected by different sensors in time synchronization. The preprocessing step provides a reliable basis for subsequent data analysis and ensures the accuracy and feasibility of the final evaluation results. Lasso regression is used to solve the sparse coefficients, and the singular value decomposition SVD is used to update the dictionary. The iterative update process of SVD can gradually optimize the dictionary to better adapt to the characteristics of the data. Through sparse representation, the dimension of the data is effectively compressed, which greatly reduces the bandwidth required for transmission. At the same time, the key feature information in the data is retained to ensure the accuracy of the data after compression. The Toeplitz matrix has the advantages of fast calculation and low storage requirements. In the process of compressed sensing, it can significantly improve the efficiency and compression ratio of data processing. In health monitoring applications, data compression not only reduces sensor power consumption, but also reduces the pressure of wireless network transmission, allowing the system to run for a long time and uninterruptedly. This is particularly important for the practical application of remote health monitoring systems, especially in resource-constrained environments.

[0075] Further, transmitting the compressed data to the central database and restoring the data means serializing the compressed data using a standard data packaging protocol and preparing to transmit the compressed data to the central database via a wireless network;

[0076] The central database receives the compressed data and unpacks it, restoring the multimodal data using the same compressed sensing matrix and sparse representation dictionary as used during compression.

[0077] The use of standard data packaging protocols ensures that the data collected by different sensors and devices can be uniformly packaged to adapt to various complex network environments and effectively avoid data loss or format errors during transmission. Serialization converts these data into byte streams so that they can be quickly transmitted over the network with minimal bandwidth usage. This not only improves the efficiency of data transmission, but also reduces the consumption of network bandwidth, effectively prolongs the battery life of smart wearable devices, and increases the portability and practicality of the devices. During wireless transmission, the integrity and reliability of the data are guaranteed, ensuring that the central database can receive correct compressed data for subsequent unpacking and data restoration. The Toeplitz matrix can not only compress data, but also can effectively reduce the cost of network bandwidth consumption, effectively prolong the battery life of smart wearable devices, and increase the portability and practicality of the devices. The integrity of the data structure is maintained, and the signal can be reconstructed through inverse operations during the restoration process, so that the system can maintain high-quality data reconstruction while reducing the sampling rate, greatly improving the efficiency and quality of data transmission. The compressed sensing technology reduces the occupation of sensors and network resources, so that the present invention achieves the optimal utilization of resources while ensuring data accuracy. After the central database receives the compressed data, the signal is restored through the same dictionary as the compression stage, ensuring the restoration accuracy of the data. The use of the sparse representation dictionary ensures the high efficiency of the data in compressed transmission and can reduce the consumption of computing resources, so that the system can still achieve real-time data processing and analysis under limited hardware conditions.

[0078] S2, perform multimodal data analysis on the restored multimodal data and extract comprehensive features, and build a logistic regression model to evaluate the user's health score;

[0079] Specifically, performing multimodal data analysis on the restored multimodal data and extracting comprehensive features refers to setting initial weights for the restored multimodal data based on the collected multimodal data;

[0080] Use the dynamic weight update formula to update the initial weight. The formula is:

[0081] ;

[0082] in is the current weight of the i-th restored multimodal data at time t, is the rate of change of the i-th restored multimodal data at time t, To adjust the coefficient, control the sensitivity of weight adjustment;

[0083] In health monitoring of multimodal data, the time-varying and volatility of data are key factors. The traditional static weight allocation method cannot dynamically respond to real-time changes in data, while the dynamic weight update formula can adjust the weight of each feature at each moment, so that the system can adaptively adjust according to the latest data to ensure that health risk assessment is more accurate. Compared with the fixed weight model, the dynamic formula provides higher real-time and flexibility. Multimodal data in health monitoring changes rapidly. The use of dynamic formulas can reflect the latest change information in real time and adapt to the health status of users. By introducing adjustment coefficients, weights can be adjusted according to the size of the data change rate, and flexibly respond to drastic fluctuations or small changes in data. This is crucial in dynamic health monitoring. The system can more keenly perceive the drastic changes in health status, thereby providing a more accurate budget. In the prior art, many systems use fixed or static weight allocation methods, which means that the weight of each data feature is constant throughout the analysis process. The dynamic weight update method can adjust the weight according to the data change rate at each moment, ensuring that the system reflects the latest health status in real time and improves the accuracy of health risk assessment. Compared with the existing technology that lacks flexibility or is only based on fixed models, the calculation method improves the adaptability and versatility of the system.

[0084] Calculate the rate of change of the i-th restored multimodal data at time t, the formula is:

[0085] ;

[0086] in is the i-th restored multimodal data at time t;

[0087] Use the Pearson correlation coefficient to calculate the correlation coefficient of the restored multimodal data at time t , the formula is:

[0088] ;

[0089] in and are the means of the i-th and j-th restored multimodal data at time t, and They are the i-th and j-th restored multimodal data at time t respectively;

[0090] Calculate the weighted Pearson correlation coefficient based on the current weights and correlation coefficients of the restored multimodal data , the formula is:

[0091] ;

[0092] Based on the weighted Pearson correlation coefficient Absolute value, calculate the association weight between each pair of restored multimodal data , the formula is:

[0093] ;

[0094] Where N is the total number of restored multimodal data;

[0095] Based on the weighted Pearson correlation coefficient and associated weights And calculate the comprehensive characteristics at time t, the formula is:

[0096] ;

[0097] in is the comprehensive feature at time t.

[0098] The core of this formula is to use the Pearson correlation coefficient and association weight to quantify the relationship between multimodal data. In practical applications, the relationship between different modal data may have an important impact on the user's health status. Only by using this weighted method based on correlation and weight can the complex interaction between different modal data be accurately captured. Other simple weighted or direct averaging methods cannot achieve such complex data association analysis. By dynamically adjusting the correlation coefficient and association weight at each moment, it is ensured that the system can reflect the changes of multimodal data in real time. The traditional static weight model cannot capture these dynamic changes in time. Therefore, the dynamic adjustment mechanism in this formula is irreplaceable. The data features in health monitoring often change over time. Only through this dynamic weighting method can the user's health status at each moment be effectively reflected. The comprehensive features are calculated by the weighted correlation of all multimodal data. This weighted-based personalized analysis is indispensable in practical applications. It can be personalized according to the data features of each user, thereby improving the accuracy of the prediction. In the prior art, the processing of multimodal data usually relies only on simple Simple weighted average or unweighted correlation analysis cannot reflect the relative importance and dynamic changes between different data. By introducing the weighted Pearson correlation coefficient, not only the correlation between different data is considered, but also the influence of different data at different time points is reflected through the associated weights, so that the system can more accurately evaluate the complex relationship between multimodal data, thereby improving the accuracy of health risk assessment. Traditional methods often use fixed weight allocations and ignore the changes in data at different time points. The introduced dynamic associated weights can adjust the weights in real time according to time and data characteristics to ensure that the system can respond to data changes in a timely manner. The dynamic nature makes the system more adaptable and can capture changes in health status in real time, rather than static analysis based on historical data or fixed weights. The dynamic weight method greatly improves the real-time and accuracy of the system. By conducting a global analysis of the correlation and weights of all multimodal data, it can more comprehensively reflect the global characteristics of health status. Through this global weighting and association analysis, the system can effectively identify potential health risks and provide more accurate and detailed health assessment reports.

[0099] By calculating the rate of change of each data point at time t, the system can adaptively adjust the data and reflect the health changes of the user in real time. The dynamic adjustment method greatly improves the real-time and accuracy of the system, enabling it to better cope with complex health data environments. By analyzing these correlations, the system can identify potential signals of health risks, thereby providing more accurate individualized analysis in health monitoring. The introduction of correlation analysis enables the system to not only rely on changes in a single physiological data, but also provide a more comprehensive assessment of the user's health through the interaction between different signals. By introducing weighting coefficients, the system can adjust according to the actual importance of the data, so that the assessment results are more in line with the user's health status. The comprehensive features calculated using the association weights and weighted Pearson correlation coefficients can better reflect the relative importance of each modal data, providing a more valuable reference indicator for health assessment. The final comprehensive feature calculation formula integrates the current weights and correlation coefficients of multimodal data to obtain a characteristic value with strong representativeness and large amount of information, which has significant advantages for personalized health management and can achieve early intervention in potential risks such as chronic diseases and cardiovascular diseases. The accurate calculation and reasonable use of comprehensive features are the core of the present invention in the field of health monitoring to achieve efficient and accurate risk assessment.

[0100] Furthermore, constructing a logistic regression model to evaluate the user's health score refers to collecting multimodal data with historical labels and preprocessing them;

[0101] Construct a logistic regression model to calculate the health risk score at time t , the formula is:

[0102] ;

[0103] in is the bias term;

[0104] The weights of positive and negative classes are calculated based on the number of multimodal data category samples with historical labels. The formula is:

[0105] ; ;

[0106] Where M is the total number of class samples, is the number of positive samples, is the number of negative class samples;

[0107] In health risk assessment, the data distribution of positive and negative classes is often unbalanced. It is common that healthy samples far outnumber risk samples. This imbalance will cause the model to tend to the majority class when predicting and reduce sensitivity to the minority class. Using weights to adjust the importance of positive and negative samples is a necessary means to deal with the problem of class imbalance. Other simple weighting methods cannot dynamically adjust weights according to the proportion of class samples, which may cause model imbalance. The data-driven approach ensures that weight allocation can dynamically adapt to the class distribution in different data sets, making the weight allocation more accurate and enhancing the adaptability of the model. Dynamic weight allocation is necessary because the data distribution in each health monitoring system may be different, and fixed weights cannot be used to handle all situations. During model training, the imbalance in the number of positive and negative samples will affect the optimization of the loss function, making the model easy to ignore the minority class. By using this formula to calculate the weights, the model can give different "penalties" to positive and negative samples, making the model more sensitive to the prediction of minority classes. This method is necessary for health risk prediction models because missed detection of high-risk samples can lead to serious consequences, while simple unweighted models may not be able to effectively identify these minority samples. In the prior art, many methods often use fixed weights to balance positive and negative samples, or ignore the impact of weights on class imbalance problems. The weights are automatically calculated through the statistics of historical samples, and personalized weight adjustments can be made for the positive and negative class distributions of different data sets. This dynamic adjustment makes the system more flexible and able to adapt to different health monitoring scenarios. Compared with the prior art, this targeted enhanced design enables the health monitoring system to more sensitively identify potential health risks and provide more accurate early warnings.

[0108] Use the weighted cross entropy loss function to calculate the loss value L at time t , the formula is:

[0109] ;

[0110] Where T is the total time of multimodal data, is the weight of the positive class, is the weight of the negative class, is the true label, T is the time, For health risks, To be risk-free;

[0111] By maximizing the log-likelihood, the cross-entropy loss can most effectively drive the parameter update of the model, so that the probability value predicted by the model is closer to the actual label. Compared with other loss functions such as mean square error, cross entropy is more efficient in dealing with classification problems and has better numerical stability. Therefore, it is a reasonable choice of loss function in classification problems. The model parameters are updated by calculating the partial derivative of the bias term. When processing multimodal data, this precise partial derivative calculation helps the model to adjust parameters according to the importance of different signals, so as to better adapt to complex health data characteristics. The imbalance problem of positive and negative classes is usually ignored, resulting in poor prediction of high-risk categories. Through weighted cross-entropy loss, the present invention can more accurately identify high-risk users in health risk assessment, avoiding the biased prediction of the model due to uneven data distribution. In multimodal data, data from different signal sources contribute differently to health status. , and weighted cross entropy can provide a more suitable optimization path for multimodal data by distinguishing and processing different categories of data. Especially in health monitoring, various physiological signals have different indicative effects on health status, and the importance distribution of positive and negative data is also different. Therefore, the weighting mechanism is particularly necessary. Many existing technologies only use cross entropy loss for optimization, and usually only update the weights, ignoring the influence of bias terms. The present invention not only considers the optimization of weights, but also adjusts the bias terms by calculating partial derivatives, so that the model can fit complex health data more flexibly. In actual health monitoring, different physiological signals have different influences, and the positive class is often more important. The weighting mechanism of the loss function can make the model better adapt to these scenarios and be more sensitive when processing abnormal data. Compared with the existing generalized model optimization method, the present invention provides a more customized optimization path for health monitoring.

[0112] Calculate the partial derivative of the loss function with respect to the bias term , the formula is:

[0113] ;

[0114] The learning rate is calculated using the exponential decay method, and the formula is:

[0115] ;

[0116] in is the initial learning rate, randomly set the initial learning rate, is the learning rate at the kth iteration, k is the number of iterations, is the decay coefficient, which controls the speed at which the learning rate decreases;

[0117] Updating the bias term of a logistic regression model using the step-down formula , the formula is:

[0118] ;

[0119] The maximum number of iterations is set according to expert experience, and the iteration is stopped when the number of iterations reaches the set maximum number of iterations.

[0120] The results based on comprehensive features avoid the user's daily movement and sensory disorders, thereby improving the safety of the user's daily operation and the comprehensive assessment ability of sensory disorders. The advantage of multimodal data is that it can provide a full range of health status information, making the assessment results more comprehensive and accurate. By using historical label data, the system can conduct a more personalized analysis of the current user's health status, and adjust the model's learning process by calculating the weights of category samples. The introduction of weights strengthens the influence of negative samples, ensuring that the model can more sensitively identify health risks. Through flexible adjustment of weights, the system has higher robustness and can adapt to different health status distributions. Through weighted cross entropy, the assessment of health risks is more accurate. The setting of weights enables the model to learn its characteristics more effectively when facing high-risk samples, so that it can better predict potential health problems in practical applications. It not only improves the sensitivity of health risk assessment, but also provides users with more accurate health monitoring services. The exponential decay method dynamically adjusts the learning rate. As the number of iterations increases, the learning rate gradually decreases, thereby ensuring that the model can learn quickly in the early stage and fine-tune more stably in the later stage. The model can learn complex health data features more efficiently and avoid the risk of overfitting. This method is particularly suitable for health risk assessment models because it can ensure the accuracy of the calculation of health risk scores while ensuring the stability of the model. After each iteration, the system will adjust the model parameters according to the current learning situation until the predetermined maximum number of iterations is reached. This process ensures that the model can fully learn the data features and ultimately form a highly accurate logistic regression model. The system can provide real-time and accurate health scores based on its historical and current multimodal data, and effectively identify potential health risks.

[0121] S3, re-collect the user's multimodal data for monitoring, and build a user visualization interface to display the monitoring results in real time;

[0122] Specifically, re-collecting the user's multimodal data for monitoring means re-collecting the user's multimodal data through the smart wearable device and pre-processing and extracting comprehensive features;

[0123] The extracted comprehensive features are input into the trained logistic regression model to calculate the health risk score, and the health risk score at time t is obtained;

[0124] The judgment threshold is set according to medical guidelines and doctor's advice. If the health risk score is greater than or equal to the judgment threshold, it is judged as having a health risk, and an early warning notification is sent through the smart device to remind the user to check the physical condition. If the health risk score is less than the judgment threshold, it is judged as risk-free, and the user's multimodal data is monitored persistently.

[0125] By continuously acquiring new data, the system can always grasp the user's latest health status, discover potential health risks in a timely manner, and thus improve the reliability of health risk prediction. By extracting comprehensive features, the system can identify those key indicators that are most relevant to health status, thereby reducing data redundancy and reducing computational complexity. This not only improves the computational efficiency of the model, but also ensures that the data input into the logistic regression model is highly representative, thereby improving the accuracy and stability of the health risk score. The introduction of the comprehensive feature extraction step effectively enhances the system's ability to analyze complex health data, laying the foundation for more accurate health risk prediction. The logistic regression model can identify the nonlinear relationship between multimodal data and health risks by learning from historical data. , and outputs a score that reflects the current health status of the user. The advantage of the logistic regression model is that it can respond quickly and accurately to the input multimodal data, provide users with timely health feedback, and has good interpretability, ensuring that users can discover health abnormalities at an early stage, so that they have the opportunity to intervene in time and reduce the deterioration of health risks. The early warning mechanism provides users with a convenient and efficient health management tool, which enhances the initiative of individual health monitoring. The health status of users changes dynamically. Through real-time updated data, the system can continuously track user health trends and ensure timely intervention when risks increase. This continuous monitoring mechanism provides users with higher security and also provides an important data basis for long-term health management.

[0126] Further, building a user visualization interface to display monitoring results in real time means using React.js to build a user visualization interface to display the user's monitoring results in real time, and using D3.js to display the health risk score at time t;

[0127] Design an early warning panel at the top of the interface to display the currently triggered early warning information;

[0128] Customers who have passed real-name verification are allowed to view the information.

[0129] The construction of the user visualization interface not only improves the user's interactive experience, but also enables the system to continuously monitor the user's health status in the background and present the monitoring results to the user in real time in an intuitive and friendly way. This real-time display design can help users keep abreast of changes in health risks and enhance their ability to manage their health status. The visualization not only makes the changes in health risk scores more intuitive, but also can timely remind users to pay attention to fluctuations in health status through changes in charts. The design of the early warning panel is simple and intuitive, and can clearly display the current risk level and specific information, allowing users to quickly understand their own health status and make corresponding decisions. Users can get timely reminders through the early warning panel to avoid potential health crises, which not only improves users' trust in health monitoring, but also provides users with effective health management tools. Through real-name verification, the system can provide customized health management services for each user and ensure the high confidentiality of data. This protection measure greatly enhances their trust in the system and ensures that personal privacy will not be violated. Users can view their health data and trend changes through the interface and respond when necessary. Real-time interaction not only enhances users' understanding of health status, but also helps users better manage health risks in daily life. The introduction of continuous monitoring functions enables health risks to be discovered in the early stages, providing strong support for users' long-term health management.

[0130] Example 2, reference Figure 2 , which is the second embodiment of the present invention, and which is different from the previous embodiment, provides a comprehensive risk early warning monitoring system based on smart wearable devices, including:

[0131] The collection and transmission module is used to collect and compress multimodal data through smart wearable devices, transmit the compressed data to the central database and restore the data;

[0132] Extraction building module, used to perform multimodal data analysis on the restored multimodal data and extract comprehensive features, and build a logistic regression model to evaluate the user's health score;

[0133] The detection visualization module is used to re-collect the user's multimodal data for monitoring and build a user visualization interface to display the monitoring results in real time.

[0134] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0135] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0136] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0137] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit with a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit with a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

Claims

1. A comprehensive risk early warning monitoring method based on smart wearable devices, characterized by: include, Collect and compress multimodal data through smart wearable devices, transmit the compressed data to the central database and restore the data; Perform multimodal data analysis on the restored multimodal data and extract comprehensive features, and build a logistic regression model to evaluate the user's health score; Re-collect users’ multimodal data for monitoring, and build a user visualization interface to display monitoring results in real time; The collecting and compressing of multimodal data through the smart wearable device refers to the user collecting and preprocessing the multimodal data through the smart wearable device. The smart wearable device includes optical heart rate, blood oxygen, body temperature, skin electricity, air quality and sound sensors; The multimodal data includes heart rate, blood oxygen, body temperature, galvanic skin response, air quality and environmental sound data; The preprocessing includes: using a dynamic time warping algorithm to time align the multimodal data, using a Z-score algorithm to identify the multimodal data and delete outliers, using an adaptive Kalman filter algorithm to denoise the multimodal data, and standardizing the denoised data; Initialize the dictionary to a random matrix; Use Lasso regression to solve the sparse coefficients; Use singular value decomposition to update the dictionary, iteratively update the dictionary and sparse coefficients, and gradually optimize the dictionary; Use the trained dictionary to sparsely represent the standardized multimodal data to obtain sparse data; The Toeplitz matrix is ​​used to convert sparse data into compressed data for compressed sensing.

2. The all-round risk early warning monitoring method based on smart wearable devices according to claim 1, characterized in that: The transmitting of the compressed data to the central database and data restoration refers to serializing the compressed data using a standard data packaging protocol and preparing to transmit the compressed data to the central database via a wireless network; The central database receives the compressed data and unpacks it, restoring the multimodal data using the same compressed sensing matrix and sparse representation dictionary as used during compression.

3. The all-round risk early warning monitoring method based on smart wearable devices as claimed in claim 2, characterized in that: The performing multimodal data analysis on the restored multimodal data and extracting comprehensive features refers to setting initial weights for the restored multimodal data according to the collected multimodal data; Use the dynamic weight update formula to update the initial weight. The formula is: ； in is the current weight of the i-th restored multimodal data at time t, is the rate of change of the i-th restored multimodal data at time t, is the adjustment coefficient; Use the Pearson correlation coefficient to calculate the correlation coefficient of the restored multimodal data at time t ; Calculate the weighted Pearson correlation coefficient based on the current weights and correlation coefficients of the restored multimodal data ; Based on the weighted Pearson correlation coefficient Absolute value, calculate the association weight between each pair of restored multimodal data ; Based on the weighted Pearson correlation coefficient and associated weights And calculate the comprehensive characteristics at time t, the formula is: ； in is the comprehensive feature at time t, and N is the total number of restored multimodal data.

4. The all-round risk early warning monitoring method based on smart wearable devices as claimed in claim 3, characterized in that: The said constructing a logistic regression model to evaluate the user's health score refers to collecting multimodal data with historical labels and performing preprocessing; Construct a logistic regression model to calculate the health risk score at time t ; The weights of positive and negative classes are calculated based on the number of multimodal data category samples with historical labels. The formula is: ； ； in M is the total number of class samples, is the number of positive samples, is the number of negative class samples; Use the weighted cross entropy loss function to calculate the loss value L at time t , the formula is: ； Where T is the total time of multimodal data, is the weight of the positive class, is the weight of the negative class, is the true label, T is the time; Calculate the partial derivative of the loss function with respect to the bias term ; Using exponential decay to calculate the learning rate ; Updating the bias term of a logistic regression model using the step-down formula ; When the number of iterations reaches the set maximum number of iterations, the iteration stops.

5. The all-round risk early warning monitoring method based on smart wearable devices as claimed in claim 4, characterized in that: The re-collecting of the user's multimodal data for monitoring refers to re-collecting the user's multimodal data through the smart wearable device and performing pre-processing and extracting comprehensive features; The extracted comprehensive features are input into the trained logistic regression model to calculate the health risk score, and the health risk score at time t is obtained; The judgment threshold is set according to medical guidelines and doctor's advice. If the health risk score is greater than or equal to the judgment threshold, it is judged as having a health risk, and an early warning notification is sent through the smart device to remind the user to check the physical condition. If the health risk score is less than the judgment threshold, it is judged as risk-free, and the user's multimodal data is monitored persistently.

6. The all-round risk early warning monitoring method based on smart wearable devices as claimed in claim 5, characterized in that: The construction of a user visualization interface to display the monitoring results in real time refers to using React.js to construct a user visualization interface to display the user's monitoring results in real time, and using D3.js to display the health risk score at time t; Design an early warning panel at the top of the interface to display the currently triggered early warning information; Customers who have passed real-name verification are allowed to view the information.

7. A comprehensive risk warning monitoring system based on a smart wearable device based on the comprehensive risk warning monitoring method based on a smart wearable device according to any one of claims 1 to 6, characterized in that: include, The collection and transmission module is used to collect and compress multimodal data through smart wearable devices, transmit the compressed data to the central database and restore the data; Extraction building module, used to perform multimodal data analysis on the restored multimodal data and extract comprehensive features, and build a logistic regression model to evaluate the user's health score; The detection visualization module is used to re-collect the user's multimodal data for monitoring and build a user visualization interface to display the monitoring results in real time.

8. A computer device comprising: Memory and processor; The memory stores a computer program, characterized in that: when the processor executes the computer program, the steps of the all-round risk warning monitoring method based on a smart wearable device according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the all-round risk early warning monitoring method based on a smart wearable device described in any one of claims 1 to 6 are implemented.

Images