Method and system for detecting wearing stability of smart watch and correcting motion data

Abstract

The application discloses a smart watch wearing stability detection and motion data correction method and system, including training a sweat mode prediction model, calling user physical sign data through a smart watch and setting motion attributes, inputting the sweat mode prediction model to determine user sweat mode and sweat prediction parameters, calculating predicted sweat rate and sweat interference index, and evaluating wearing stability, fitting historical correction coefficients of historical measured motion data and monitoring data of the smart watch with corresponding actual motion data to determine the historical correction coefficients, constructing a correction coefficient prediction model, inputting the predicted sweat rate and sweat interference index of the user into the correction coefficient prediction model to obtain a predicted correction coefficient, and correcting motion data. The method can not only improve the efficiency and accuracy of smart watch wearing stability detection and motion data correction, but also has good interpretability and can be directly applied to a smart watch wearing stability detection and motion data correction system.

Description

Technical Field

[0001] This invention relates to the field of smart wearable device technology, and in particular to a method and system for detecting the wearing stability and correcting motion data of a smartwatch. Background Technology

[0002] Smartwatches can monitor key physiological and activity indicators such as heart rate, steps, skin temperature, and ambient humidity in real time, providing users with value-added services such as exercise load assessment, health risk warnings, and training effect analysis. However, changes in the skin-device interface caused by sweat secretion during high-intensity exercise are a key bottleneck affecting data acquisition quality. This not only leads to device slippage, changes in electrode contact impedance, and sensor signal attenuation, but also results in significant deviations in the exercise data from existing smartwatches, making it difficult to meet the needs of precise medical-grade monitoring.

[0003] Existing technologies still have significant limitations: First, traditional solutions rely solely on macroscopic physical quantities such as temperature, humidity, or friction coefficient to characterize the wearing status, failing to delve into the electrophysiological essence of the skin-electrode interface, resulting in delayed stability assessment and a high misjudgment rate; second, existing algorithms lack physical modeling and advanced prediction capabilities for the dynamic process of sweat accumulation, failing to cope with the nonlinear data drift caused by rapid sweat bursts during high-intensity interval training, exhibiting insufficient real-time performance and significant correction delays; finally, existing technologies mostly employ static population classification based on gender and age or general filtering parameters, failing to fully consider the differentiated physiological responses of body fat percentage and exercise habits among individuals, leading to higher biases in exercise data for high BMI users or individuals with special physical conditions. Therefore, this invention proposes a method and system for detecting the wearing stability and correcting motion data of smartwatches. By integrating a skin contact quality assessment system based on multi-frequency bioimpedance spectroscopy, temperature and humidity, and friction sensing, a predictive model is constructed that integrates an unsteady diffusion equation and a long short-term memory network. This enables dynamic correction of sweat interference, breaking through the physical limits of traditional single-dimensional perception. Furthermore, the prediction-correction feedforward mechanism improves data accuracy, laying a technical foundation for the application of smart wearable devices in high-reliability scenarios such as competitive sports and clinical monitoring. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for detecting the wearing stability of smartwatches and correcting motion data.

[0005] To achieve the above objectives, the present invention is implemented according to the following technical solution: This invention includes the following steps: Collect vital sign data and corresponding exercise attributes and sweating pattern labels from different groups of people, train a sweating pattern prediction model, call the user's vital sign data and set exercise attributes through a smartwatch, and input the sweating pattern prediction model to determine the user's sweating pattern. The sweating prediction parameters are determined based on the user's sweating pattern, the predicted sweating rate is calculated, the sweat interference index is calculated based on the smartwatch monitoring data, and the wearing stability is evaluated. The historical correction coefficient is determined by fitting the historical measured and monitored exercise data of the smartwatch with the corresponding actual exercise data. A correction coefficient prediction model is constructed based on historical correction coefficients and corresponding predicted sweating rate and sweat interference index. The user's predicted sweating rate and sweat interference index are input into the correction coefficient prediction model to obtain the prediction correction coefficients and correct the exercise data. The vital signs data include height, weight, wrist size, and BMI. The motion attributes include motion type and motion intensity; The sweating prediction parameters include sweating rate prediction weight, sweating adjustment factor, and skin contact weight; The monitoring data includes humidity, temperature, friction index, and skin-electrode impedance modulus. The exercise data includes heart rate, steps, and speed.

[0006] Furthermore, the method for determining the user's sweating pattern includes: Exercise attributes and corresponding physical characteristics data with sweating pattern labels were collected as a comprehensive sweating set, which was randomly divided into a sweating training set and a sweating test set in a 6:4 ratio. The sweating pattern prediction model was trained using the sweating training set, and the performance of the sweating pattern prediction model was tested using the sweating test set. The sweating pattern prediction model includes an input layer, an encoding layer, a clustering prediction layer, and an output layer. The encoding layer includes a static encoding channel for vital signs and a dynamic encoding channel for motion. The static encoding channel encodes the vital sign data to obtain static features, while the dynamic encoding channel captures the long-term dependencies of motion attributes to obtain dynamic features. The clustering prediction layer concatenates the outputs of the encoding layer and projects them onto a cluster-friendly unit hypersphere to obtain sweating fusion features. The sweating fusion features are then softly assigned and clustered to determine the sweating pattern. The sweating pattern prediction model improves the accuracy of its predictions through a multi-objective composite loss function, the expression of which is: ; ; ; in For multi-objective composite loss functions, For the sample size, For sample similarity, For joint embedding vectors, To and Embedding vectors belonging to the same sweating pattern For temperature coefficient, For the auxiliary distribution, the th The sample belongs to the first The target probability of each cluster, For the first The sample belongs to the first The soft assignment probability of each cluster. For physiological constraint loss, For physiological similarity constraint weights, For physiological differences, the constraint weights are... For the set of physiologically similar pairs, For physiologically distinct pairs, To reconstruct consistency loss, For the first A vector of vital sign data for each sample. For parameters Physiological feature decoder For the first The latent representation of a sample after being encoded by a physiological feature encoder For the first A vector of motion attributes of each sample. For parameters Motion feature decoder, For the first The latent representation of a sample after being encoded by a motion feature encoder; By accessing the user's vital signs data and setting exercise attributes through a smartwatch, the user's sweating pattern is determined by inputting a sweating pattern prediction model.

[0007] Furthermore, the method for predicting sweating rate includes: The predicted sweat rate is calculated by extracting the corresponding sweat rate prediction weights based on the user's sweating pattern, as expressed by: ; in Sweat mode for users The corresponding sweating rate, , , Sweat mode for users The corresponding sweating rate prediction weight, For wrist size adjustment factor, For exercise intensity, For sports type, For wrist size, Based on the standard wrist size.

[0008] Furthermore, the method for assessing wearing stability includes: The skin contact quality index is calculated based on data monitored by the smartwatch. The sweat interference index is then calculated segment by segment based on the skin contact quality index. The expression is: ; ; in The skin contact quality index. , For skin contact weight, The skin-electrode impedance modulus at 1 kHz. For individual reference impedance, , The standard deviation and mean of the skin-electrode impedance modulus in the frequency range of 10-100kHz are given. The sweat interference index, For real-time relative humidity, Humidity threshold At maximum humidity, For real-time skin temperature, As the reference temperature, This is the upper limit of temperature. For maximum friction, For real-time friction, It is a sweating regulator.

[0009] Furthermore, the method for correcting motion data includes: Obtain historical measured motion data, historical monitoring data, and corresponding historical actual motion data from the smartwatch, and perform linear fitting to obtain the historical correction coefficient, expressed as follows: ; in For actual motion data, For actual motion data, for Correction coefficients for motion-like data, for Correction factor for monitoring data, for Monitoring data deviation; The historical correction coefficients and the corresponding predicted sweating rate, sweat interference index, heart rate variability and acceleration sequence are combined to form a correction comprehensive set. The comprehensive set is randomly divided into a correction training set and a correction test set in a 6:4 ratio. The correction training set is used to train the correction coefficient prediction model, and the correction test set is used to test the performance of the correction coefficient prediction model. The correction coefficient prediction model includes an input layer, a physical constraint embedding layer, a temporal feature extraction layer, a correction coefficient prediction layer, and an output layer. The physical constraint embedding layer transforms the unsteady diffusion equation of sweat accumulation into a differentiable soft constraint and calculates the physical discrete residual; the expression for the unsteady diffusion equation of sweat accumulation is: ; in The sweat interference index, for The Laplace operator for the peristaltic perspiration interference index. Where is the diffusion coefficient. The skin evaporation coefficient, For real-time relative humidity, For real-time skin temperature, As the reference temperature, for Always sweating mode The corresponding sweating rate; The temporal feature extraction layer uses a two-layer bidirectional long short-term memory network to process sequence data, extract the temporal dependency relationship between motion, sweating, and disturbance, and fuse the physical discrete residuals with the temporal dependency relationship to obtain fused features; The correction coefficient prediction layer uses a temporal attention mechanism and context aggregation to predict physiological-physical features, and maps the physiological-physical features to the correction coefficient space through a two-layer fully connected network to obtain the predicted correction coefficients. The predicted sweating rate and sweat interference index of the user are input into the prediction model to obtain the prediction correction coefficient, which is then used to correct the exercise data.

[0010] Secondly, the system for detecting wearing stability and correcting motion data in smartwatches includes: Sweating pattern determination module: used to collect vital sign data and corresponding exercise attributes and sweating pattern labels of different groups of people, train the sweating pattern prediction model, call the user's vital sign data and set exercise attributes through the smartwatch, and input the sweating pattern prediction model to determine the user's sweating pattern. Wearing stability assessment module: used to determine sweat prediction parameters based on the user's sweating pattern, calculate the predicted sweating rate, calculate the sweat interference index based on the smartwatch monitoring data, and assess wearing stability; The sports data correction module is used to fit the historical measured sports data and monitoring data of the smartwatch with the corresponding actual sports data to determine the historical correction coefficient. Based on the historical correction coefficient and the corresponding predicted sweat rate and sweat interference index, a correction coefficient prediction model is constructed. The user's predicted sweat rate and sweat interference index are input into the correction coefficient prediction model to obtain the predicted correction coefficient to correct the sports data. Data Management Module: Used to present corrected data or alarm information, interact with the mobile app and cloud, and perform data synchronization and feedback.

[0011] The beneficial effects of this invention are: This invention relates to a method and system for detecting wearing stability and correcting motion data in smartwatches. Compared with existing technologies, this invention has the following technical advantages: This invention introduces multi-frequency bioimpedance spectroscopy technology and combines temperature, humidity and friction coefficient to construct a skin contact quality index, effectively overcoming the detection blind spot of loose watch straps, realizing multi-physical field collaborative sensing, and significantly improving the accuracy of interface state recognition. This invention constructs a physical information neural network that integrates unsteady diffusion equation constraints with bidirectional LSTM, and achieves predictive correction through a physical information fusion architecture, significantly improving the real-time performance and reliability of motion monitoring. This invention employs a dual-channel deep embedding clustering network and contrastive representation learning, which can accurately identify individual sweating patterns and configure personalized weight parameters based on user vital signs and exercise habits, thereby improving the adaptability of exercise monitoring. By combining prediction correction coefficients with wear assessment, this invention constructs a closed-loop system of perception-prediction-correction-feedback. This system not only provides high-precision motion data but also proactively intervenes in the device's wearing status, significantly reducing the decline in user trust caused by data drift. This provides key technical support for the promotion and application of smartwatches in the field of medical-grade health monitoring. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating the steps of the method for detecting the wearing stability and correcting motion data of a smartwatch according to the present invention. Figure 2 This is a schematic diagram of the structure of the smartwatch of the present invention; In the diagram: 1-Smartwatch dial; 2-Smartwatch strap; 3-Friction and impedance test box; 4-Temperature and humidity acquisition box; 5-Miniature sensor. Detailed Implementation

[0013] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0014] The method and system for detecting wearing stability and correcting motion data of a smartwatch according to the present invention include the following steps: like Figure 1 As shown, this embodiment includes the following steps: Collect vital sign data and corresponding exercise attributes and sweating pattern labels from different groups of people, train a sweating pattern prediction model, call the user's vital sign data and set exercise attributes through a smartwatch, and input the sweating pattern prediction model to determine the user's sweating pattern. The sweating prediction parameters are determined based on the user's sweating pattern, the predicted sweating rate is calculated, the sweat interference index is calculated based on the smartwatch monitoring data, and the wearing stability is evaluated. The historical correction coefficient is determined by fitting the historical measured and monitored exercise data of the smartwatch with the corresponding actual exercise data. A correction coefficient prediction model is constructed based on historical correction coefficients and corresponding predicted sweating rate and sweat interference index. The user's predicted sweating rate and sweat interference index are input into the correction coefficient prediction model to obtain the prediction correction coefficients and correct the exercise data. The vital signs data include height, weight, wrist size, and BMI. The motion attributes include motion type and motion intensity; The sweating prediction parameters include sweating rate prediction weight, sweating adjustment factor, and skin contact weight; The monitoring data includes humidity, temperature, friction index, and skin-electrode impedance modulus. The exercise data includes heart rate, steps, and speed.

[0015] In this embodiment, the method for determining the user's sweating pattern includes: Exercise attributes and corresponding physical characteristics data with sweating pattern labels were collected as a comprehensive sweating set, which was randomly divided into a sweating training set and a sweating test set in a 6:4 ratio. The sweating pattern prediction model was trained using the sweating training set, and the performance of the sweating pattern prediction model was tested using the sweating test set. The sweating pattern prediction model includes an input layer, an encoding layer, a clustering prediction layer, and an output layer. The encoding layer includes a static encoding channel for vital signs and a dynamic encoding channel for motion. The static encoding channel encodes the vital sign data to obtain static features, while the dynamic encoding channel captures the long-term dependencies of motion attributes to obtain dynamic features. The clustering prediction layer concatenates the outputs of the encoding layer and projects them onto a cluster-friendly unit hypersphere to obtain sweating fusion features. The sweating fusion features are then softly assigned and clustered to determine the sweating pattern. The sweating pattern prediction model improves the accuracy of its predictions through a multi-objective composite loss function, the expression of which is: ; ; ; in For multi-objective composite loss functions, For the sample size, For sample similarity, For joint embedding vectors, To and Embedding vectors belonging to the same sweating pattern For temperature coefficient, For the auxiliary distribution, the th The sample belongs to the first The target probability of each cluster, For the first The sample belongs to the first The soft assignment probability of each cluster. For physiological constraint loss, For physiological similarity constraint weights, For physiological differences, the constraint weights are... For the set of physiologically similar pairs, For physiologically distinct pairs, To reconstruct consistency loss, For the first A vector of vital sign data for each sample. For parameters Physiological feature decoder For the first The latent representation of a sample after being encoded by a physiological feature encoder For the first A vector of motion attributes of each sample. For parameters Motion feature decoder, For the first The latent representation of a sample after being encoded by a motion feature encoder; By accessing the user's vital signs data and setting exercise attributes through a smartwatch, the user's sweating pattern is determined by inputting a sweating pattern prediction model. In the actual evaluation, exercise sweating data (cumulative sweat volume, sweating initiation delay) of multiple subjects (covering different BMI gradients: [18.5, 24], [24, 28], [28, +∞) were simultaneously collected in different controlled environments (with temperature and humidity as variables) using a transdermal water loss meter and capacitive sweat patches. An adaptive Gaussian mixture model was used to pre-classify the exercise sweating data at different exercise stages to determine the sweating pattern: profuse sweating (sweating initiation delay less than 60s, cumulative sweat volume greater than 20μL / cm). 2 / min), progressive type (sweating initiation delay between [60, 120] s, cumulative sweat volume between [10, 20] μL / cm 2 / min), sustained-release (sweating initiation delay greater than 120s, cumulative sweat volume less than 10μL / cm), 2 The sweating patterns are categorized into three types: localized (identified by thermal infrared imaging with a sweat distribution heterogeneity index greater than 0.6) and sensitive (where small changes in exercise intensity result in a cumulative sweat volume variation coefficient greater than 0.3). The labels are then semantically calibrated to ultimately create sweating pattern labels with physiological semantic constraints. In the sweating pattern prediction model, the encoding rules for the static encoding channel of vital signs and the dynamic encoding of movement are as follows:

[0016]

[0017] in For the first The latent representation of a sample after being encoded by a physiological feature encoder For the first The latent representation of a sample after being encoded by a motion feature encoder This is the first layer weight matrix. Second layer weight matrix, For the first Original vital sign data vectors of individual users This is the first layer bias vector. This is the second layer bias vector. This is a one-dimensional convolution operation. It is a sequence of motion intensity; The clustering prediction layer concatenates the outputs of the encoding layer and maps them to a cluster-friendly unit hypersphere through a metric learning projection head to obtain sweating fusion features. A Student t-distribution kernel is introduced to calculate the sample... Belonging to the cluster center Soft-assignment clustering determines the sweating pattern.

[0018] In this embodiment, the method for predicting sweating rate includes: The predicted sweat rate is calculated by extracting the corresponding sweat rate prediction weights based on the user's sweating pattern, as expressed by: ; in Sweat mode for users The corresponding sweating rate, , , Sweat mode for users The corresponding sweating rate prediction weight, For wrist size adjustment factor, For exercise intensity, For sports type, For wrist size, Based on the standard wrist size.

[0019] In this embodiment, the method for evaluating wearing stability includes: The skin contact quality index is calculated based on data monitored by the smartwatch. The sweat interference index is then calculated segment by segment based on the skin contact quality index. The expression is: ; ; in The skin contact quality index. , For skin contact weight, The skin-electrode impedance modulus at 1 kHz. For individual reference impedance, , The standard deviation and mean of the skin-electrode impedance modulus in the frequency range of 10-100kHz are given. The sweat interference index, For real-time relative humidity, Humidity threshold At maximum humidity, For real-time skin temperature, As the reference temperature, This is the upper limit of temperature. For maximum friction, For real-time friction, It is a sweating regulator.

[0020] In this embodiment, the method for correcting motion data includes: Obtain historical measured motion data, historical monitoring data, and corresponding historical actual motion data from the smartwatch, and perform linear fitting to obtain the historical correction coefficient, expressed as follows: ; in For actual motion data, For actual motion data, for Correction coefficients for motion-like data, for Correction factor for monitoring data, for Monitoring data deviation; The historical correction coefficients and the corresponding predicted sweating rate, sweat interference index, heart rate variability and acceleration sequence are combined to form a correction comprehensive set. The comprehensive set is randomly divided into a correction training set and a correction test set in a 6:4 ratio. The correction training set is used to train the correction coefficient prediction model, and the correction test set is used to test the performance of the correction coefficient prediction model. The correction coefficient prediction model includes an input layer, a physical constraint embedding layer, a temporal feature extraction layer, a correction coefficient prediction layer, and an output layer. The physical constraint embedding layer transforms the unsteady diffusion equation of sweat accumulation into a differentiable soft constraint and calculates the physical discrete residual; the expression for the unsteady diffusion equation of sweat accumulation is: ; in The sweat interference index, for The Laplace operator for the peristaltic perspiration interference index. Where is the diffusion coefficient. The skin evaporation coefficient, For real-time relative humidity, For real-time skin temperature, As the reference temperature, for Always sweating mode The corresponding sweating rate; The temporal feature extraction layer uses a two-layer bidirectional long short-term memory network to process sequence data, extract the temporal dependency relationship between motion, sweating, and disturbance, and fuse the physical discrete residuals with the temporal dependency relationship to obtain fused features; The correction coefficient prediction layer uses a temporal attention mechanism and context aggregation to predict physiological-physical features, and maps the physiological-physical features to the correction coefficient space through a two-layer fully connected network to obtain the predicted correction coefficients. The user's predicted sweating rate and sweat interference index are input into the correction coefficient prediction model to obtain the prediction correction coefficient, which is then used to correct the exercise data. In practical evaluation, within the physical constraint embedding layer, the physical discrete residual, as a physical loss term, constrains the evolution of the LSTM hidden state during backpropagation, avoiding predictions that violate thermodynamics (such as sweat backflow or instantaneous evaporation). The formula for calculating the physical discrete residual is: ; in for The physical discrete residual at time step, For time step, For spatial step size, Evaporation rate; In the temporal feature extraction layer, the two-layer bidirectional long short-term memory network includes a forward LSTM and a backward LSTM. The forward LSTM captures the forward evolution from the past to the present (such as the delayed sweating effect caused by the gradual increase in exercise intensity), and the backward LSTM captures the backward constraints of future information on the current state (such as adjusting the correction strategy in advance when it is known that a high-interference state is about to be entered). The latent features of the forward LSTM and the backward LSTM are concatenated and fused. The physical residual is mapped to a physical guiding vector through a fully connected layer. The physical guiding vector and the fused latent features are multiplied element by element through a gating mechanism to obtain the fused features. In the correction coefficient prediction layer, attention weights are calculated and normalized using a temporal attention mechanism. These attention weights are then combined with contextualized aggregation of physiological and physical features from the past 60 seconds. The aggregated features are then mapped to the correction coefficient space through a two-layer fully connected network, as expressed in the following expression: ; ; ; in for Attention weight at any moment For attention context vectors, The hidden state transformation weight matrix is... for The fusion characteristics of moments For context transformation weight matrix, The context vector from the previous time step. For the final context vector, The attention weights are normalized. The aggregation window duration. To provide an intermediate representation for the prediction correction coefficient, , Here are the weight matrix and bias vector of the first fully connected layer. , The weight matrix and bias vector of the second fully connected layer; Taking user Zhang San (male, 35 years old) as an example of using a smartwatch for exercise monitoring for the first time, basic physical characteristics data (height 175cm, weight 85kg, wrist size 18cm, BMI 27.8) are collected through the input interface of the watch. The user selects and confirms the exercise type through the crown (exercise type is aerobic jogging, exercise intensity is 0.75, and exercise intensity is calculated according to the heart rate reserve method). The above data is input into the sweating pattern prediction model to determine that the user's sweating pattern is progressive high sweating group (confidence of progressive high sweating group is 0.88, confidence of profuse sweating is 0.45, and confidence of slow-release sweating is 0.21), and the corresponding sweating prediction parameters are retrieved. The predicted weights for sweating rate were set to 0.08 / 1.2 / 0.5, with a wrist size adjustment factor of 0.02 and a baseline wrist size of 16cm. The calculated sweating rate was 3.664ml / min. The skin-electrode impedance was collected in real time by the watch's sensor array, and the skin contact quality index was calculated to be 0.35 (the skin contact weight was determined to be 0.5 / 0.3 based on BMI). Environmental parameters were collected simultaneously (real-time relative humidity 65%, humidity threshold 50%, maximum humidity 100%; real-time skin temperature 35℃, reference temperature 25℃, upper temperature limit 40℃; real-time friction index 0.4, maximum friction 1.0; sweat adjustment factor for the progressive high-sweating group was 1.1). First, it was determined that the real-time relative humidity of 65% was greater than the humidity threshold of 50%. The sweat interference index was further calculated to be 0.169, and the wearing stability was evaluated. At this point, the wearing status was considered excellent contact, but continuous monitoring was required. The wearing stability evaluation criteria were as follows: a sweat interference index less than 0.3 was considered excellent contact (good wearing status); a sweat interference index between [0.3, 0.6] was considered ordinary contact (average wearing status, with slight interference from sweat on exercise data); and a sweat interference index greater than 0.6 was considered poor contact (poor wearing status, with discontinuous monitoring data and significant interference from sweat on exercise data). Based on the monitoring data sequence, the predicted sweat rate sequence, sweat interference index sequence, and heart rate variability sequence are calculated, and the corresponding acceleration sequence is extracted and input into the correction coefficient prediction model. Taking heart rate correction as an example, the heart rate correction coefficient is 0.73, and the correction coefficients for the monitoring data are (humidity / temperature / friction index / skin-electrode impedance modulus) 0.28 / 0.25 / 0.12 / 0.15, which are used to correct the heart rate data.

[0021] Secondly, the system for detecting wearing stability and correcting motion data in smartwatches includes: Sweating pattern determination module: used to collect vital sign data and corresponding exercise attributes and sweating pattern labels of different groups of people, train the sweating pattern prediction model, call the user's vital sign data and set exercise attributes through the smartwatch, and input the sweating pattern prediction model to determine the user's sweating pattern. Wearing stability assessment module: used to determine sweat prediction parameters based on the user's sweating pattern, calculate the predicted sweating rate, calculate the sweat interference index based on the smartwatch monitoring data, and assess wearing stability; The sports data correction module is used to fit the historical measured sports data and monitoring data of the smartwatch with the corresponding actual sports data to determine the historical correction coefficient. Based on the historical correction coefficient and the corresponding predicted sweat rate and sweat interference index, a correction coefficient prediction model is constructed. The user's predicted sweat rate and sweat interference index are input into the correction coefficient prediction model to obtain the predicted correction coefficient to correct the sports data. Data management module: used to present corrected data or alarm information, interact with the mobile app and cloud, and perform data synchronization and feedback; The structure of the smartwatch of this invention is as follows: Figure 2As shown, the back of the smartwatch face 1 integrates a friction and impedance testing box 3 and a temperature and humidity acquisition box 4. Miniature sensors are evenly distributed in the middle of the smartwatch strap 2. Temperature and humidity data are collected through the temperature and humidity acquisition box 4, friction index and skin-electrode impedance modulus are collected through the friction and impedance testing box 3, wrist size is determined by the contact between the miniature sensor 5 and the skin, and heart rate is collected by the miniature sensor 5 and other sensors on the back of the smartwatch face 1.

[0022] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting wearing stability and correcting motion data in a smartwatch, characterized in that, Includes the following steps: S1. Collect vital sign data and corresponding exercise attributes and sweating pattern labels of different groups of people, train the sweating pattern prediction model, call the user's vital sign data and set the exercise attributes through the smartwatch, and input the sweating pattern prediction model to determine the user's sweating pattern. S2. Determine the sweating prediction parameters based on the user's sweating pattern, calculate the predicted sweating rate, calculate the sweat interference index based on the smartwatch monitoring data, and evaluate the wearing stability. S3. Fit the historical measured motion data and monitoring data of the smartwatch with the corresponding actual motion data to determine the historical correction coefficient; S4. Construct a correction coefficient prediction model based on historical correction coefficients and corresponding predicted sweating rate and sweat interference index. Input the user's predicted sweating rate and sweat interference index into the correction coefficient prediction model to obtain the prediction correction coefficient and correct the exercise data. The vital signs data include height, weight, wrist size, and BMI. The motion attributes include motion type and motion intensity; The sweating prediction parameters include sweating rate prediction weight, sweating adjustment factor, and skin contact weight; The monitoring data includes humidity, temperature, friction index, and skin-electrode impedance modulus. The exercise data includes heart rate, steps, and speed.

2. The method for detecting wearing stability and correcting motion data of a smartwatch according to claim 1, characterized in that, The method for determining a user's sweating pattern includes: Exercise attributes and corresponding physical characteristics data with sweating pattern labels were collected as a comprehensive sweating set, which was randomly divided into a sweating training set and a sweating test set in a 6:4 ratio. The sweating pattern prediction model was trained using the sweating training set, and the performance of the sweating pattern prediction model was tested using the sweating test set. The sweating pattern prediction model includes an input layer, an encoding layer, a clustering prediction layer, and an output layer. The encoding layer includes a static encoding channel for vital signs and a dynamic encoding channel for motion. The static encoding channel encodes the vital sign data to obtain static features, while the dynamic encoding channel captures the long-term dependencies of motion attributes to obtain dynamic features. The clustering prediction layer concatenates the outputs of the encoding layer and projects them onto a cluster-friendly unit hypersphere to obtain sweating fusion features. The sweating fusion features are then softly assigned and clustered to determine the sweating pattern. The sweating pattern prediction model improves the accuracy of its predictions through a multi-objective composite loss function, the expression of which is: ； ； ； in For multi-objective composite loss functions, For the sample size, For sample similarity, For joint embedding vectors, To and Embedding vectors belonging to the same sweating pattern For temperature coefficient, For the auxiliary distribution, the th The sample belongs to the first The target probability of each cluster, For the first The sample belongs to the first The soft assignment probability of each cluster. For physiological constraint loss, For physiological similarity constraint weights, For physiological differences, the constraint weights are... For a set of physiologically similar pairs, For physiologically distinct pairs, To reconstruct consistency loss, For the first A vector of vital sign data for each sample. For parameters Physiological feature decoder For the first The latent representation of a sample after being encoded by a physiological feature encoder For the first A vector of motion attributes of each sample. For parameters Motion feature decoder, For the first The latent representation of a sample after being encoded by a motion feature encoder; By accessing the user's vital signs data and setting exercise attributes through a smartwatch, the user's sweating pattern is determined by inputting a sweating pattern prediction model.

3. The method for detecting wearing stability and correcting motion data of a smartwatch according to claim 1, characterized in that, The method for predicting sweating rate includes: The predicted sweat rate is calculated by extracting the corresponding sweat rate prediction weights based on the user's sweating pattern, as expressed by: ； in Sweat mode for users The corresponding sweating rate, , , Sweat mode for users The corresponding sweating rate prediction weight, For wrist size adjustment factor, For exercise intensity, For sports type, For wrist size, Based on the standard wrist size.

4. The method for detecting wearing stability and correcting motion data of a smartwatch according to claim 1, characterized in that, The method for assessing wearing stability includes: The skin contact quality index is calculated based on data monitored by the smartwatch. The sweat interference index is then calculated segment by segment based on the skin contact quality index. The expression is: ； ； in The skin contact quality index. , For skin contact weight, The skin-electrode impedance modulus at 1 kHz. For individual reference impedance, , The standard deviation and mean of the skin-electrode impedance modulus in the frequency range of 10-100kHz are given. The sweat interference index, For real-time relative humidity, Humidity threshold At maximum humidity, For real-time skin temperature, As the reference temperature, This is the upper limit of temperature. For maximum friction, For real-time friction, For sweating adjustment factors; Wearing stability is assessed based on the sweat interference index.

5. The method for detecting wearing stability and correcting motion data of a smartwatch according to claim 1, characterized in that, The method for correcting motion data includes: Obtain historical measured motion data, historical monitoring data, and corresponding historical actual motion data from the smartwatch, and perform linear fitting to obtain the historical correction coefficient, expressed as follows: ； in For actual motion data, For actual motion data, for Correction coefficients for motion-like data, for Correction factor for monitoring data, for Monitoring data deviation; The historical correction coefficients and the corresponding predicted sweating rate, sweat interference index, heart rate variability and acceleration sequence are combined to form a correction comprehensive set. The comprehensive set is randomly divided into a correction training set and a correction test set in a 6:4 ratio. The correction training set is used to train the correction coefficient prediction model, and the correction test set is used to test the performance of the correction coefficient prediction model. The correction coefficient prediction model includes an input layer, a physical constraint embedding layer, a temporal feature extraction layer, a correction coefficient prediction layer, and an output layer. The physical constraint embedding layer transforms the unsteady diffusion equation of sweat accumulation into a differentiable soft constraint and calculates the physical discrete residual; the expression for the unsteady diffusion equation of sweat accumulation is: ； in The sweat interference index, for The Laplace operator for the peristaltic perspiration interference index. Where is the diffusion coefficient. The skin evaporation coefficient, For real-time relative humidity, For real-time skin temperature, As the reference temperature, for Always sweating mode The corresponding sweating rate; The temporal feature extraction layer uses a two-layer bidirectional long short-term memory network to process sequence data, extract the temporal dependency relationship between motion, sweating, and disturbance, and fuse the physical discrete residuals with the temporal dependency relationship to obtain fused features; The correction coefficient prediction layer uses a temporal attention mechanism and context aggregation to predict physiological-physical features, and maps the physiological-physical features to the correction coefficient space through a two-layer fully connected network to obtain the predicted correction coefficients. The predicted sweating rate and sweat interference index of the user are input into the prediction model to obtain the prediction correction coefficient, which is then used to correct the exercise data.

6. A system for detecting the wearing stability and correcting motion data of a smartwatch, used to perform the method according to any one of claims 1-5, characterized in that, include: Sweating pattern determination module: used to collect vital sign data and corresponding exercise attributes and sweating pattern labels of different groups of people, train the sweating pattern prediction model, call the user's vital sign data and set exercise attributes through the smartwatch, and input the sweating pattern prediction model to determine the user's sweating pattern. Wearing stability assessment module: used to determine sweat prediction parameters based on the user's sweating pattern, calculate the predicted sweating rate, calculate the sweat interference index based on the smartwatch monitoring data, and assess wearing stability; The sports data correction module is used to fit the historical measured sports data and monitoring data of the smartwatch with the corresponding actual sports data to determine the historical correction coefficient. Based on the historical correction coefficient and the corresponding predicted sweat rate and sweat interference index, a correction coefficient prediction model is constructed. The user's predicted sweat rate and sweat interference index are input into the correction coefficient prediction model to obtain the predicted correction coefficient to correct the sports data. Data Management Module: Used to present corrected data or alarm information, interact with the mobile app and cloud, and perform data synchronization and feedback.

Images