A health detection method based on weight scale

Abstract

The present application provides a health detection method based on a weight scale, comprising: collecting pressure distribution data exerted by the left and right feet of multiple users when they walk on the scale surface through a high-precision, high-sampling rate pressure sensor array arranged on the left and right sides of the weight scale, and analyzing the time series characteristics, frequency domain characteristics and spatial distribution characteristics of the pressure distribution data; if the evaluation result shows that the user has a balance ability deviation or there is a left-right difference in lower limb muscle strength, the corresponding exercise training program and rehabilitation guidance plan are pushed to the user; while the user is using the weight scale, the left and right foot pressure difference data is continuously collected, and daily monitoring is carried out by acquiring exercise habit data, and the target evaluation model of balance ability and muscle strength is adaptively optimized to improve the accuracy of the evaluation results.

Description

Technical Field

[0001] The present application relates to the field of information technology, and in particular to a health detection method based on a weight scale. Background Art

[0002] There are specialized gait analyzers on the market that can accurately measure the movement patterns and mechanical parameters of the human body during exercise. At the same time, weight scales, as the most common health monitoring equipment in daily life, have also been widely used.

[0003] Since the slight difference in pressure between the left and right feet of a user contains a wealth of health information, these differences may reflect the deviation of body balance ability and the imbalance of muscle strength distribution, which in turn affects the risk of sports injuries. Therefore, how to enable users to understand the health status of their gait while weighing themselves daily is a technical problem that needs to be solved urgently. Summary of the invention

[0004] The present application provides a health detection method based on a weight scale, which is used to solve the problem of how to enable users to understand their own health status through their own gait while weighing themselves daily.

[0005] The program includes:

[0006] The pressure distribution data of the left and right feet of multiple users walking on the scale are collected through a high-precision, high-sampling-rate pressure sensor array arranged on the left and right sides of the scale device, and the time series characteristics, frequency domain characteristics and spatial distribution characteristics of the pressure distribution data are analyzed and obtained;

[0007] The wavelet transform signal processing method is used to denoise and smooth the pressure distribution data, filter out the high-frequency interference introduced by the body's center of gravity shift and gait rhythm fluctuations, extract the statistical characteristics of the stable left and right foot pressure difference, and analyze its correlation with gait rhythm, body center of gravity shift and lower limb joint angle.

[0008] Based on the time series characteristics, frequency domain characteristics and spatial distribution characteristics of pressure distribution data, as well as the statistical characteristics of left and right foot pressure differences and the obtained correlation, a pressure feature database covering different ages, genders and physical conditions was constructed. The cluster analysis method was used to extract the left and right foot pressure difference patterns with significant statistical characteristics, and used as an evaluation indicator to measure body balance ability and lower limb muscle strength distribution.

[0009] For each pressure difference pattern, the user's exercise habit data is obtained and combined with medical diagnosis data for correlation analysis to determine the quantitative correspondence between the pressure difference pattern and body balance ability and lower limb muscle strength. A balance ability and muscle strength assessment model based on pressure characteristics is constructed, and its relationship with age, gender and exercise habits is analyzed;

[0010] The evaluation model is integrated into the embedded system of the weight scale device. During the weighing process, the pressure distribution data is collected in real time and the statistical characteristics and pressure difference patterns of the left and right feet are calculated and extracted. The pressure difference patterns are matched with the evaluation model to judge the user's body balance ability and lower limb muscle strength level, and give a quantitative evaluation result.

[0011] If the evaluation results show that the user has a balance deviation or a difference in lower limb muscle strength between the left and right sides, the corresponding exercise training and rehabilitation guidance suggestions will be pushed to the user through the interactive interface of the weight scale device;

[0012] When users use the scale, the pressure difference data between their left and right feet is continuously collected, and daily monitoring is carried out by obtaining exercise habit data. The balance ability and muscle strength assessment model is adaptively optimized to improve the accuracy of the assessment results.

[0013] The technical solution provided by the embodiments of the present application may have the following beneficial effects:

[0014] The present application discloses a health detection method based on a weight scale. The present application arranges a high-precision, high-sampling-rate pressure sensor array on the left and right sides of the weight scale to collect the pressure distribution data applied by the left and right feet of the user when walking on the scale surface, and performs denoising, smoothing and feature extraction on the data to construct a pressure feature database covering different ages, genders and physical conditions. Through cluster analysis, the left and right foot pressure difference patterns with significant statistical characteristics are extracted, and correlation analysis is performed in combination with medical diagnostic data and user exercise habit data to construct a balance ability and muscle strength target evaluation model based on pressure characteristics. This is beneficial for the user to use the weight scale for weighing in the future, and the user's body balance ability and lower limb muscle strength are evaluated in real time through the target evaluation model and the pressure distribution data of the user's left and right feet, so that the user can understand his or her own health status.

[0015] The weight scale of the present application is integrated with a target assessment model of balance ability and muscle strength based on pressure characteristics, so that when the user uses the weight scale, the user's body balance ability and lower limb muscle strength can be evaluated in real time, and by comparing with normal people and people of the same age and gender, it is judged whether there is a deviation in the user's physical function, and a quantitative assessment result is given. For users with problems, the present application can also push targeted sports training programs and rehabilitation guidance programs. By continuously collecting user data and performing incremental learning, the present application can continuously optimize the assessment model and improve the accuracy of the assessment results. The present application provides a convenient, non-invasive, and quantitative method for assessing body balance ability and lower limb muscle strength, which helps users to promptly discover and improve physical function problems, and is of great significance to health management and rehabilitation training. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A flow chart of a health detection method based on a weight scale provided in an embodiment of the present application;

[0017] Figure 2 A schematic diagram of a health detection method based on a weight scale provided in an embodiment of the present application;

[0018] Figure 3 Another schematic diagram of a health detection method based on a weight scale provided in an embodiment of the present application;

[0019] Figure 4 Another schematic diagram of a health detection method based on a weight scale provided in an embodiment of the present application. DETAILED DESCRIPTION

[0020] The technical solution of the present application will be described clearly and completely in conjunction with the embodiments below. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application.

[0021] When users walk on the scale, the slight difference in pressure between the left and right feet contains a wealth of health information. Using a scale for gait analysis can easily detect the state of the feet in daily life and provide users with valuable health tips. For example: when standing statically, the center of gravity of the body is obviously shifted, and one side bears significantly more weight than the other side, which may indicate scoliosis or unequal length of lower limbs, which in turn causes problems such as low back pain. When walking dynamically, phenomena such as different lengths of time the left and right feet touch the ground and asymmetric pressure distribution may be related to diseases such as poor gait, arthritis, and plantar fasciitis. Timely detection of these problems can help users take targeted intervention measures. However, how to accurately measure and quantify these tiny differences and establish a reliable association with the functional status of the body is a complex technical challenge.

[0022] Before introducing the health detection method based on a weight scale provided in the embodiment of the present application, it is first introduced how to obtain a balance ability and muscle strength evaluation model based on pressure characteristics.

[0023] like Figure 1 As shown, Figure 1 The construction process of the balance ability and muscle strength assessment model based on pressure characteristics is shown, and the method includes:

[0024] S101. By arranging a high-precision, high-sampling-rate pressure sensor array on the left and right sides of the weight scale device, the pressure distribution data exerted by the left and right feet of multiple users when they walk on the scale surface are collected, and the time series characteristics, frequency domain characteristics and spatial distribution characteristics of the pressure distribution data are analyzed and obtained.

[0025] like Figure 2 As shown in the figure, the pressure distribution data of the left and right feet of the user when walking is collected, and a two-dimensional pressure distribution matrix is ​​constructed according to the spatial position of the pressure sensor and the corresponding pressure value; the two-dimensional pressure distribution matrix is ​​subjected to bilinear interpolation smoothing to obtain a continuous pressure distribution surface; the pressure distribution surface is subjected to time series analysis to calculate the spectrum of the pressure signal; the main frequency components and their amplitudes are extracted according to the spectrum; the moments when the heel touches the ground and the toes leave the ground are identified through the main frequency components, and the gait cycle is divided; according to the spatial characteristics of the pressure distribution surface, the center of gravity position of the sole contact area at each moment is calculated; the center of gravity trajectory is tracked according to the center of gravity position to obtain the motion curve of the pressure center; polynomial regression fitting is performed on the motion curve to extract the center of gravity transfer speed, acceleration and curvature characteristic parameters; the gait feature vector is constructed using the gait cycle, spectrum characteristics and center of gravity trajectory characteristic parameters; the gait feature vector is normalized to obtain a normalized feature vector; the normalized feature vector is classified using the support vector machine algorithm; if the support vector machine classification result shows that the gait is abnormal, the abnormal gait characteristics are compared with the normal gait characteristics to quantify the difference in pressure distribution of the left and right feet.

[0026] Specifically, the pressure distribution data of the left and right feet of multiple users when walking are collected from the high-precision pressure sensor array arranged on the left and right sides of the scale. A two-dimensional pressure distribution matrix is ​​constructed according to the spatial position of the pressure sensor and the corresponding pressure value. The matrix is ​​smoothed by the bilinear interpolation algorithm to obtain a continuous pressure distribution surface. The pressure distribution of the left and right feet is compared, and the pressure peak ratio and the center of gravity position offset of the left and right feet are calculated. The pressure distribution data is analyzed in time series, and the frequency spectrum of the pressure signal is calculated by the fast Fourier transform algorithm to extract the main frequency components and their amplitudes. The peak detection method is used to identify the moment when the heel touches the ground and the toe leaves the ground, and the gait cycle is divided to obtain the step frequency and step length information when the user walks. The gait cycle characteristics of the left and right feet are compared to identify the symmetry of the gait. According to the spatial characteristics of the pressure distribution, the center of gravity position of the sole contact area at each moment is calculated, and the center of gravity trajectory is tracked to obtain the motion curve of the pressure center. The polynomial regression method is used to fit the center of gravity trajectory, and the characteristic parameters such as the center of gravity transfer speed, acceleration and curvature are extracted. The deviation of the center of gravity trajectory of the left and right feet is calculated to judge the balance ability and stability of the user when walking. The time series features, frequency domain features, and spatial distribution features obtained in the previous steps are used to construct gait feature vectors. The feature data is normalized and the feature vectors are classified using the support vector machine algorithm. The optimal parameters are selected using the cross-validation method and the support vector machine model is trained. The trained model is used to identify the user's gait pattern and determine whether there is an abnormal gait. The features of the abnormal gait are compared with the normal gait to quantify the difference in pressure distribution between the left and right feet, providing data support for subsequent gait correction and rehabilitation training.

[0027] For example, the pressure sensor array is arranged on the left and right sides of the scale, with one pressure sensor array on each side, for example (the pressure sensor array includes 8x8 sensors), and the sampling frequency is 100Hz. The pressure data of the user when walking is collected to construct a 64x2 two-dimensional pressure distribution matrix. The matrix is ​​expanded to a 256x64 high-resolution pressure distribution map using a bilinear interpolation algorithm to achieve smoothing. The ratio of the left and right foot pressure peaks is calculated. If the peak value of the left foot is 500N and the peak value of the right foot is 450N, the ratio is 1.11, indicating that the left foot bears a slightly larger load. The center of gravity position offset is obtained by calculating the lateral distance between the pressure centers of the left and right feet. If the offset value is 2cm, it indicates that the center of gravity is slightly biased to the left. Fast Fourier transform is performed on the pressure data to obtain a spectrum. The main frequency component is 2Hz, corresponding to the step frequency. The peak detection method identifies the moment when the heel touches the ground and the toe leaves the ground, and the threshold is set to 10% of the body weight. When the pressure value rises from below the threshold to above the threshold, it is marked as the heel touches the ground; when the pressure value drops from above the threshold to below the threshold, it is marked as the toe leaves the ground. Calculate the time interval between two heel strikes and get a gait cycle of 0.5 seconds. According to the sampling frequency and footprint position, calculate the step length to be 60 cm. Compare the gait cycles of the left and right feet. If the difference is less than 5%, it is considered symmetrical. Track the trajectory of the center of gravity and obtain the motion curve of the pressure center. Use cubic polynomial regression to fit the trajectory of the center of gravity and get the function y=ax³+bx²+cx+d, where a, b, c, and d are fitting coefficients. a is the coefficient of the cubic term of the polynomial, b is the coefficient of the quadratic term of the polynomial, c is the coefficient of the linear term of the polynomial, and d is the constant term of the polynomial. The speed was obtained by calculating the slope of the tangent line of the curve at each point, the acceleration was obtained by the rate of change of the slope, and the curvature was calculated by the formula k=(y″) / (1+y′²)^(3 / 2). The center of gravity trajectories of the left and right feet were compared. If the maximum deviation exceeded 1 cm, it was judged to be unstable. A gait feature vector was constructed, including parameters such as step frequency, step length, pressure peak ratio, center of gravity offset, speed, acceleration and curvature. The feature data was normalized to the minimum and maximum, and the values ​​were mapped to the range of 0-1. The support vector machine with radial basis function kernel was used for classification, and the optimal parameters C=10 and γ=0.1 were selected by 5-fold cross validation. The model was trained to identify normal and abnormal gaits. If the predicted probability was greater than 0.8, it was judged to be abnormal gait. The feature differences between abnormal gait and normal gait were compared, such as a 10% decrease in step length and a 50% increase in center of gravity offset, to provide quantitative indicators for gait correction.

[0028] S102. Using wavelet transform signal processing method to denoise and smooth the pressure distribution data, extract the statistical characteristics of the stable left and right foot pressure difference, and analyze its correlation with gait rhythm, body center of gravity transfer and lower limb joint angle.

[0029] It can be understood that the use of wavelet transform signal processing method to denoise and smooth the pressure distribution data can filter out high-frequency interference introduced by the body's center of gravity transfer and gait rhythm fluctuations.

[0030] The multi-scale wavelet transform method was used to perform frequency decomposition on the pressure distribution data. According to the frequency decomposition results, the soft threshold was selected to perform noise reduction on the high-frequency wavelet coefficients. The pressure distribution data after noise reduction were smoothed by a sliding average filter. The statistical characteristics of the pressure difference between the left and right feet were extracted from the smoothed pressure distribution data, including the mean, standard deviation, skewness, and kurtosis. The displacement of the pressure center was calculated by the relative change rate of the left and right foot pressures. The gait rhythm was extracted according to the periodic changes of the pressure data, and the body center of gravity transfer was analyzed using the pressure center trajectory. The association model between the feature vector and the gait rhythm, body center of gravity transfer, and lower limb joint angle was established by the multivariate linear regression method. The pressure difference feature was selected as the independent variable, and the gait rhythm, center of gravity transfer, and joint angle were selected as the dependent variables.

[0031] Specifically, multi-scale wavelet transform was used to perform frequency decomposition on the pressure distribution data, and Daubechies wavelet (db4) was selected as the basis function for 5-layer decomposition. The soft threshold was selected by Stein unbiased risk estimation (SURE) method to reduce the noise of high-frequency wavelet coefficients. The signal was reconstructed to obtain the denoised pressure distribution data, while retaining the periodic characteristics of gait rhythm. The 11-point sliding average filter was applied to the denoised pressure distribution data for smoothing. The symmetric extension method was used to deal with the window boundary problem, and the mean of the data in the window was calculated as the smoothed data point. The high-frequency interference introduced by the body center of gravity transfer and gait rhythm fluctuation was filtered out, and the main change trend of pressure distribution was retained. The statistical features of the pressure difference between the left and right feet were extracted from the smoothed pressure distribution data, including mean, standard deviation, skewness and kurtosis. The relative change rate of the left and right foot pressure was calculated, which was obtained by (right foot pressure-left foot pressure) / total pressure. The displacement of the pressure center was calculated, which was obtained by the Euclidean distance of the left and right foot pressure centers. The pressure difference feature vector was constructed, which contained the above features and was used for subsequent correlation analysis. The gait rhythm is extracted through the periodic changes of pressure data, the body center of gravity transfer is analyzed by the pressure center trajectory, and the changes in the lower limb joint angle are inferred according to the changes in pressure distribution. According to the extracted pressure difference feature vector, the association model between the feature vector and the gait rhythm, body center of gravity transfer and lower limb joint angle is established by the multivariate linear regression method. The pressure difference feature is selected as the independent variable, and the gait rhythm, center of gravity transfer and joint angle are selected as the dependent variables. The stepwise regression method is used to deal with the multicollinearity problem and select the most relevant features. The correlation coefficient matrix between the variables is calculated to determine the correlation between the pressure difference feature and the gait parameters, and to quantify the impact of the pressure distribution difference between the left and right feet on the gait. The collected pressure distribution data is subjected to multi-scale wavelet transform, and the db4 wavelet basis function is selected for 5-layer decomposition. The original data sampling frequency is 100Hz, and after decomposition, sub-band signals with frequency bands of 0-3.125Hz, 3.125-6.25Hz, 6.25-12.5Hz, 12.5-25Hz and 25-50Hz are obtained. The SURE method is used to calculate the optimal threshold for each subband, such as the threshold of the first layer high-frequency subband is 0.05, the second layer is 0.08, and so on. The high-frequency subband coefficients are soft-thresholded, and the coefficients less than the threshold are set to zero, and the coefficients greater than the threshold are subtracted from the threshold. The reconstructed signal is used to obtain the denoised pressure distribution data, retaining the 2-3Hz gait cycle characteristics. An 11-point sliding average filter is applied to the denoised data with a window size of 0.11 seconds. The boundary is processed by symmetric extension, such as the data point [1,2,3,4,5] is extended to [3,2,1,2,3,4,5,4,3,2,1]. The mean in the window is calculated as the smoothed data point to obtain the smoothed pressure distribution data.Extract the pressure difference features of the left and right feet from the smoothed data, and calculate the mean (such as 300N for the left foot and 320N for the right foot), standard deviation (20N for the left foot and 25N for the right foot), skewness (0.1 for the left foot and -0.2 for the right foot), and kurtosis (2.8 for the left foot and 3.1 for the right foot). Calculate the relative rate of change (320-300) / (320+300)=3.23%. Calculate the Euclidean distance through the pressure center coordinates and get the pressure center displacement such as 2cm. Analyze the periodicity of the pressure data and get the gait rhythm such as 1.2 steps / second. Track the trajectory of the pressure center and calculate the center of gravity transfer speed such as 10cm / s. According to the change in pressure distribution, estimate the range of knee joint angle change such as 0-60 degrees. Construct a feature vector containing the above 11 features. Use multivariate linear regression to establish an association model, with the 11 pressure difference features as independent variables and the gait rhythm, center of gravity transfer speed and knee joint angle as dependent variables. Stepwise regression was used, and the F-statistic threshold for entering the model was set to 4.0, and the threshold for leaving the model was set to 3.9, and the seven most relevant features were screened out. The correlation coefficient matrix was calculated, and the correlation coefficient between the mean pressure difference and the gait rhythm was 0.85, the correlation coefficient with the center of gravity transfer speed was 0.72, and the correlation coefficient with the knee joint angle was 0.63, which quantified the influence of the difference in pressure distribution between the left and right feet on gait.

[0032] S103. Based on the time series characteristics, frequency domain characteristics and spatial distribution characteristics of the pressure distribution data, as well as the statistical characteristics of the left and right foot pressure differences and the obtained correlation, a pressure characteristic database covering different ages, genders and physical conditions is constructed. The cluster analysis method is used to extract the left and right foot pressure difference patterns with significant statistical characteristics, and use them as evaluation indicators to measure body balance ability and lower limb muscle strength distribution.

[0033] The gait cycle, pressure peak, spectrum energy distribution and pressure center trajectory indicators are obtained to construct a multidimensional feature vector; the pressure feature data are collected by stratified sampling method according to the multidimensional feature vector to obtain a pressure feature database; the data in the pressure feature database are standardized to obtain standardized pressure feature data; the standardized pressure feature data are clustered, and the optimal number of clusters is determined by calculating the silhouette coefficient; the standardized pressure feature data are divided into different pressure difference pattern categories according to the optimal number of clusters, and the center point and discrete degree of each category are calculated; for each pressure difference pattern category, significant statistical features are extracted, and the significant statistical features include the mean, standard deviation, skewness and kurtosis of the pressure difference between the left and right feet; a scoring standard is established according to the significant statistical features, and the scoring standard maps each category to a preset scoring range; the body balance ability and lower limb muscle strength distribution are quantified through the scoring standard to obtain the final evaluation score.

[0034] Specifically, a multidimensional feature vector was constructed based on the time series characteristics, frequency domain characteristics and spatial distribution characteristics of the pressure distribution data, as well as the statistical characteristics and correlation of the pressure difference between the left and right feet. The feature vector contains indicators such as gait cycle, pressure peak, spectral energy distribution and pressure center trajectory. The power spectral density was calculated using fast Fourier transform to obtain the spectral energy distribution. The pressure center trajectory was extracted by calculating the weighted average position of the pressure at each time point. At the same time, the age, gender and physical condition information of the subjects were recorded as auxiliary features. A stratified sampling method was used to collect pressure feature data from people of different age groups, genders and physical conditions to build a comprehensive pressure feature database. The collected data were Z-score standardized to convert each feature into a distribution with a mean of 0 and a standard deviation of 1 to eliminate the dimension effect. The balance ability was evaluated by the stability of the pressure center trajectory, and the root mean square error of the trajectory was calculated. The muscle strength distribution was evaluated by the pressure difference between the left and right feet, and the mean and standard deviation of the pressure difference were calculated. The K-means clustering algorithm was applied to the standardized pressure feature data, and the optimal number of clusters was determined by calculating the silhouette coefficient. The method for calculating the silhouette coefficient is as follows: for each sample, calculate its average distance a(i) from other samples of the same type and its average distance b(i) from the nearest samples of other types. The silhouette coefficient s(i)=(b(i)-a(i)) / max(a(i),b(i)), where max(a(i),b(i)) represents the maximum value of the two distance values ​​a(i) and b(i) of sample i. The cluster number with the largest average silhouette coefficient is selected as the optimal value. The data is divided into different pressure difference pattern categories, and the center point and dispersion degree of each category are calculated. Based on the clustering results, the significant statistical features of each category are extracted, including the mean, standard deviation, skewness and kurtosis of the pressure difference between the left and right feet. These features are used as evaluation indicators to establish a scoring standard. The scoring standard is established by mapping each category to a scoring range of 0-100 based on the clustering results, considering the distance between the category center and the ideal state, as well as the relative position of the sample in the category. The body balance ability and the distribution of lower limb muscle strength are quantified to obtain the final evaluation score. The stability of clustering results and the reliability of scoring criteria were evaluated using cross-validation methods to ensure the accuracy and consistency of the evaluation results. Multidimensional feature vectors were extracted from the pressure distribution data, including gait cycle (e.g., 1.2 seconds), pressure peak (300N for the left foot and 320N for the right foot), spectral energy distribution, and pressure center trajectory. The power spectral density was calculated using a 1024-point fast Fourier transform to obtain the spectral energy distribution in the range of 0-50Hz. The pressure center trajectory was extracted by calculating the weighted average position of the pressure at each time point, such as {(x1,y1),(x2,y2),...}. Subject information was recorded, such as age 35 years, male, and BMI index 22. A stratified sampling method was used to collect data from 1,000 subjects with different backgrounds to construct a pressure feature database.The data was Z-score standardized to convert the features into a distribution with a mean of 0 and a standard deviation of 1. The root mean square error of the pressure center trajectory was calculated, such as 0.5cm, to evaluate balance ability. The mean of the pressure difference between the left and right feet was calculated to be 20N and the standard deviation was 5N to evaluate muscle strength distribution. The K-means clustering algorithm was applied, and 2-10 clusters were tried, and the average silhouette coefficient was calculated for each case. Assuming that the average silhouette coefficient is the highest when there are 5 clusters, which is 0.68, this is selected as the optimal number of clusters. The data was divided into 5 pressure difference pattern categories, and the center point and dispersion of each category were calculated. The significant statistical features of each category were extracted, such as the mean of the pressure difference between the left and right feet of category 1 is 15N, the standard deviation is 4N, the skewness is 0.1, and the kurtosis is 2.8. A scoring standard of 0-100 was established, and the category center was mapped to a specific score, such as the center of category 1 is mapped to 85 points. Consider the relative position of the sample in the category, such as full marks within 0.5 standard deviations from the center, 5 points deducted within 1 standard deviation, and so on. The final evaluation score, such as 92 points, quantifies the body balance ability and lower limb muscle strength distribution. The clustering stability is evaluated using 5-fold cross validation. For example, the adjusted Rand index of the 5-fold clustering results is 0.85, which proves that the clustering results are stable. The reliability of the scoring criteria is verified by the test and retest method. For example, the correlation coefficient of the two test results is 0.92, indicating that the scoring criteria are highly reliable.

[0035] S104. For each pressure difference pattern, the exercise habit data of the collected user is obtained, and correlation analysis is performed in combination with the medical diagnosis data to determine the quantitative correspondence between the pressure difference pattern and the body balance ability and lower limb muscle strength, build a balance ability and muscle strength assessment model based on pressure characteristics, and analyze its relationship with age, gender and exercise habits.

[0036] The user's exercise habit data and medical diagnosis data are obtained, the exercise habit data includes exercise type, frequency, intensity and duration, and the medical diagnosis data includes balance test results and muscle strength test data; a mapping relationship with the pressure difference pattern is established based on the user's exercise habit data and medical diagnosis data; a multiple linear regression method is used to establish a quantitative correspondence model between pressure characteristics and body balance ability and lower limb muscle strength, with pressure characteristics as independent variables and balance ability and muscle strength test results as dependent variables; a neural network model is constructed based on the quantitative correspondence model, the neural network model uses pressure characteristic parameters as the input layer, the hidden layer uses the ReLU activation function, and the output layer corresponds to the balance ability and muscle strength score; age, gender and exercise habit data are added as additional input features to the neural network model to obtain an extended neural network model.

[0037] Specifically, for each pressure difference pattern, the exercise habit data of the tested users, including exercise type, frequency, intensity and duration, were collected through questionnaires and wearable devices. The exercise frequency was converted into the number of exercises per week, and the exercise intensity was expressed in metabolic equivalents. At the same time, the user's medical diagnosis data, such as balance test results and muscle strength test data, were collected. A mapping relationship between user ID and pressure difference pattern was established, and the collected data was associated with the corresponding pressure difference pattern. The multivariate linear regression method was used to establish a quantitative correspondence model between pressure characteristics and body balance ability and lower limb muscle strength, with the pressure difference pattern characteristics as independent variables and the balance ability and muscle strength test results as dependent variables. The variance inflation factor was calculated. If the variance inflation factor was greater than 10, the relevant variables were deleted or merged to deal with the multicollinearity problem. The regression coefficient and determination coefficient were calculated to evaluate the goodness of fit of the model. ANOVA was used to test whether there were significant differences in the pressure difference patterns between different age groups, gender groups and exercise habit groups. Based on the established quantitative correspondence, a neural network model was constructed. A three-layer structure was used, with the number of neurons in the input layer equal to the number of features, 64 neurons in the hidden layer, and 2 neurons in the output layer corresponding to the balance ability and muscle strength scores respectively. The input layer included stress feature parameters, and the hidden layer used the ReLU activation function. The model parameters were optimized by the back propagation algorithm to minimize the mean square error between the predicted value and the actual test result. The generalization ability of the model was evaluated by k-fold cross validation, and the mean absolute error and root mean square error were calculated to quantify the model performance. Age, gender, and exercise habit data were used as additional input features to expand the neural network model, retrain, and compare the model performance. The importance score of each input feature was calculated using the permutation feature importance method, and the feature importance was evaluated by randomly shuffling the value of each feature and observing the change in model performance. The influence of stress features, demographic characteristics, and exercise habits on the balance ability and muscle strength assessment was analyzed. Based on the analysis results, the feature selection was optimized to improve the prediction accuracy and explanatory power of the model. User data was collected through smart questionnaires and wearable devices. For example, the exercise habit data of user ID U001 showed that the frequency of exercise was 3 times per week, the average intensity was 6METs, and the duration was 45 minutes. Medical diagnostic data included a balance test score of 85 points out of 100 and a lower limb strength test result of 280N. U001 was associated with the pressure difference pattern M2. Data was collected from 1,000 users to establish a complete user-pressure pattern mapping dataset. Multiple linear regression was used with 10 pressure feature parameters as independent variables and balance ability and muscle strength scores as dependent variables. The variance inflation factor was calculated and the VIF values ​​of features x3 and x7 were found to be 12.3 and 15.6, respectively. These two features were merged into a new feature through principal component analysis. The coefficient of determination R² obtained from regression analysis was 0.78, indicating a good fit of the model.The results of variance analysis showed that there were significant differences in the pressure difference patterns between different age groups, F=15.3, p<0.001, and exercise habit groups, F=10.2, p<0.01. A 3-layer neural network was constructed with 12 neurons in the input layer, including 10 pressure features and age and gender, 64 neurons in the hidden layer, and 2 neurons in the output layer. The Adam optimizer was used with a learning rate of 0.001, a batch size of 32, and 50 cycles of training. The results of 5-fold cross validation showed a mean absolute error of 3.2 points and a root mean square error of 4.1 points. Through permutation feature importance analysis, it was found that the pressure center trajectory (importance score 0.25) and the pressure difference between the left and right feet (score 0.22) were the two most important features, while age (score 0.15) and exercise frequency (score 0.12) also showed high importance. Based on these results, 8 features with importance scores greater than 0.1 were retained and the model was retrained, and the prediction accuracy of the final model on the test set was improved to 91.

[0038] The following will describe how to use a weight scale to give a user a quantitative assessment result, such as Figure 2 As shown, Figure 3 A health detection method based on a weight scale is provided in an embodiment of the present application, and the method comprises:

[0039] S201. Integrate the evaluation model into the embedded system of the weight scale. During the user weighing process, collect the pressure distribution data in real time and calculate and extract the statistical characteristics of the pressure difference between the left and right feet and the pressure difference pattern.

[0040] like Figure 4 As shown, the evaluation model is obtained and converted into a lightweight version for accessing the embedded system. The lightweight version of the evaluation model is loaded into the embedded processor of the weight scale. The pressure distribution data of the user during weighing is collected in real time through the pressure sensor array. The statistical characteristics of the pressure difference between the left and right feet are calculated based on the pressure distribution data, and the statistical characteristics include mean, standard deviation, skewness and kurtosis. The statistical characteristics of the pressure difference are normalized to obtain the characteristic value that matches the input of the evaluation model.

[0041] The method provided in the embodiment of the present application can be executed by an electronic device, which can be a server, a mobile phone, a tablet, a wearable device or a weight scale. As an example, the weight scale in the embodiment of the present application can also be a body fat scale.

[0042] S202. Match the statistical characteristics of the pressure difference between the left and right feet and the pressure difference pattern with the evaluation model to determine the user's body balance ability and lower limb muscle strength level, and provide a quantitative evaluation result.

[0043] For example, the eigenvalues ​​are input into the evaluation model in the embedded processor, and the quantitative evaluation results of the user's body balance ability and lower limb muscle strength level are obtained through forward propagation calculation. The level of the quantitative evaluation results is determined according to the preset scoring criteria.

[0044] Specifically, the evaluation model is converted into a lightweight version suitable for embedded systems, and the model size is reduced and the computational efficiency is optimized through weight pruning and quantization techniques. 32-bit floating-point numbers are converted to 8-bit fixed-point numbers to adapt to the computing power of embedded systems. The compressed model parameters and structure are loaded into the embedded processor of the weight scale device to ensure that all calculations are completed within about 10 seconds during the user's standing period. During the user's weighing process, the pressure distribution data is collected in real time through a high-precision pressure sensor array, and the original data is segmented using a 1-second sliding window method, sliding once every 0.5 seconds. The window boundary data is processed using a zero-filling method. For each data segment, the statistical characteristics of the pressure difference between the left and right feet are calculated, including the mean, standard deviation, skewness, and kurtosis. A data caching mechanism is set to continuously update the evaluation results during the user's weighing process to improve the accuracy and stability of the evaluation. The calculated statistical characteristics of the pressure difference are normalized, and the eigenvalues ​​are mapped to a range that matches the input of the evaluation model. The principal component analysis method is used to reduce the feature dimension, and the cumulative contribution rate threshold is set to 95% to determine the number of principal components to be retained, and the main pressure difference patterns are extracted. An abnormal detection mechanism is added to identify and handle situations such as unstable user posture or repeated weighing to ensure the reliability of the evaluation results. The processed feature data is input into the evaluation model in the embedded system, and the quantitative evaluation results of the user's body balance ability and lower limb muscle strength level are obtained through forward propagation calculation. According to the preset scoring criteria, the numerical results are converted into easy-to-understand grade ratings. The evaluation results are divided into 5 levels: excellent (90-100 points), good (80-89 points), general (70-79 points), poor (60-69 points), and poor (below 60 points). The evaluation results are continuously updated until the user weighing process is completed, and the final evaluation level is output. In the embedding of the weight scale, the original evaluation model is compressed by 50% through weight pruning, and the 32-bit floating point number is quantized to 8-bit fixed point number, and the model size is reduced from 10MB to 2.3MB. The compressed model runs on the ARMCortex-M4 processor, and the single inference time does not exceed 100 milliseconds. The pressure sensor array collects data at a sampling rate of 100Hz, and uses a sliding window of 1 second (100 data points) and slides every 0.5 seconds (50 data points) for feature extraction. For each window, the mean, standard deviation, skewness and kurtosis of the pressure difference between the left and right feet are calculated to obtain 8 eigenvalues. A 256KB circular buffer is set to store the feature data of the most recent 25 windows. The features are Z-score normalized, and then the principal component analysis is used to reduce the dimension, retaining the principal components with a cumulative contribution rate of 95%, and usually reducing the 8-dimensional features to 5 dimensions. Anomaly detection uses the 3-times standard deviation principle. If any eigenvalue exceeds the mean ±3-times standard deviation range, it is considered abnormal and the window data is discarded. The evaluation model updates the results every 0.5 seconds and outputs a continuous score of 0-100.The scores are mapped to five levels: 90-100 is excellent, 80-89 is good, 70-79 is fair, 60-69 is poor, and below 60 is poor. It takes an average of 7.5 seconds from the time the user stands on the scale to the output of a stable evaluation result, meeting the requirement of completing the evaluation within 10 seconds. The system achieved an evaluation accuracy rate of 95% among 100 test users, and the correlation coefficient with the evaluation results of professional medical equipment reached 0.92.

[0045] By analyzing the difference between the user's pressure difference pattern and the normal population, it is determined whether the user's balance ability is deviated, and a quantitative index of the degree of deviation is given. By analyzing the statistical characteristics of the pressure difference between the user's left and right feet, it is determined whether the user's lower limb muscle strength is different between the left and right sides, and a quantitative index of the degree of difference is given.

[0046] Obtain the pressure difference pattern data of healthy adults and select key pressure difference features. Calculate the mean vector and covariance matrix based on the pressure difference pattern data, and construct a multivariate Gaussian distribution model as a normal population benchmark model. Use the Mahalanobis distance to calculate the degree of difference between the user's pressure difference pattern and the normal population benchmark model, and obtain a quantitative indicator of the degree of deviation in the user's balance ability. Calculate the statistical characteristics of the pressure difference between the user's left and right feet, and determine whether there is a left-right difference in the user's lower limb muscle strength by training a support vector machine. If there is a left-right difference, calculate the distance from the user's feature vector to the classification hyperplane to determine the quantitative indicator of the degree of difference in the user's lower limb muscle strength.

[0047] Specifically, the pressure difference pattern data of 1,000 healthy adults without balance disorders were extracted from the database, and 10 key pressure difference features were selected. The data were detected for outliers, and samples that were beyond the range of ±3 times the standard deviation of the mean were removed. The mean vector and covariance matrix of the remaining samples were calculated, and a multivariate Gaussian distribution model was constructed as a baseline model for the normal population. This baseline model was regularly verified and updated with newly collected data to ensure its representativeness. The Mahalanobis distance was used to calculate the degree of difference between the user's pressure difference pattern and the baseline model of the normal population. The Mahalanobis distance calculation formula is √[(x-μ)^TΣ^(-1)(x-μ)], where x is the user feature vector, μ is the mean vector of the normal population, Σ is the covariance matrix, and (x−μ)^T is the transpose of the vector (x−μ). The calculated Mahalanobis distance value is used as a quantitative indicator of the degree of deviation of the user's balance ability. The threshold is set to divide the degree of deviation into mild 0-2, moderate 2-4, and severe>4. Explain the quantitative indicators, such as "Your balance ability deviation is 3.2, which is a moderate deviation." Calculate the statistical features of the difference in pressure between the left and right feet of the user, such as the mean, variance, skewness, and kurtosis. Use 1000 labeled normal samples and 1000 abnormal samples to train the support vector machine, using the radial basis kernel function. Standardize the input data to ensure that the feature scale is consistent. Use grid search and cross-validation to optimize the hyperparameters of the support vector machine. Input the user's statistical features into the trained support vector machine to determine whether there is a left-right difference in the user's lower limb muscle strength. Based on the classification results of the support vector machine, calculate the distance from the user's feature vector to the classification hyperplane. Use the output value of the decision function as a distance metric and normalize it to a range of 0-100. The normalized distance value is used as a quantitative indicator of the difference in the degree of lower limb muscle strength between the left and right sides. The larger the distance, the more significant the difference. Divide the quantitative indicators into five levels: no difference 0-20, slight difference 20-40, moderate difference 40-60, significant difference 60-80, and severe difference 80-100. Provide result explanations, such as "The left and right difference index of your lower limb muscle strength is 75, which is a significant difference level." In actual applications, the pressure difference pattern data of 1,000 healthy adults are extracted from the database, and 10 key features such as pressure center offset and left and right foot pressure ratio are selected. The 3σ principle is used for outlier detection, and 27 abnormal samples are eliminated. The mean vector μ and covariance matrix Σ are calculated for the remaining 973 samples to construct a multivariate Gaussian distribution model. The model is updated with 100 newly collected samples every month to maintain its timeliness. For user A, the system calculates the Mahalanobis distance between its feature vector x and the normal population model as 3.2. According to the preset threshold, the degree of deviation in the balance ability of user A is determined to be moderate. Then the statistical characteristics of the pressure difference between the left and right feet of user A are calculated, and the mean difference, variance ratio, skewness difference, and kurtosis difference are obtained, which are 0.15, 1.8, 0.4, and 0.7, respectively.These features are input into a pre-trained support vector machine (using RBF kernel, C=10, γ=0.1), which determines that user A has a difference in muscle strength between the left and right sides. The distance from the feature vector to the classification hyperplane is calculated, and the original output is 2.3, which is normalized to the range of 0-100 to get 75 points. Explanation: "The degree of deviation in your balance ability is 3.2, which is a moderate deviation; the difference index of the left and right sides of the lower limb muscle strength is 75, which is a significant difference level. Targeted balance training and muscle strength balance exercises are recommended." The system stores these results in the health record of user A for subsequent tracking and comparative analysis.

[0048] Compare the user's balance ability and lower limb muscle strength level with the average level of people of the same age and gender to determine whether the user's physical function is within the normal range and give quantitative comparison results.

[0049] According to the user's age and gender information, the corresponding population sample data is obtained, and the population samples are divided into intervals according to age; the mean and standard deviation of the balance ability index and lower limb muscle strength index of each group of samples after division are calculated; whether the index data of each group of samples conforms to the normal distribution; if the index data conforms to the normal distribution, a normal distribution model is constructed as a reference benchmark; if the index data does not conform to the normal distribution, the median and interquartile range are used as reference indicators; the user's balance ability index and lower limb muscle strength index are converted into Z scores, and the Z score is calculated by subtracting the group mean from the user's index value and dividing it by the group standard deviation; according to the Z score, the standard normal distribution table is searched to obtain the percentile corresponding to the user's index; the percentiles are divided into multiple levels according to the preset percentile range to determine whether the user's physical function is within the normal range; the difference percentage between the user's index and the group average level is calculated, and the difference percentage is equal to the user's index value minus the group mean divided by the group mean and then multiplied by the percentage coefficient; according to the difference percentage and percentile level, the corresponding evaluation conclusion is given.

[0050] Specifically, according to the user's age and gender, the corresponding population samples are screened from the database and divided into groups of 5 years old to ensure that the sample size of each group is not less than 1,000 people. The mean and standard deviation of the balance ability and lower limb muscle strength indicators of the group are calculated. The Shapiro-Wilk test is used to determine whether the data distribution conforms to the normal distribution. If it conforms, a normal distribution model is constructed as a reference benchmark; if it does not conform, the median and interquartile range are used as reference indicators. The reference benchmark data is updated regularly to reflect the changing trend of the population's physical function. The user's balance ability and lower limb muscle strength indicators are converted into Z scores. The calculation method is (user indicator value-group mean) / group standard deviation, and the relative position of the user in the same age and gender population is obtained. The interquartile range method (IQR) is used to identify and process outliers, and data outside the range of Q1-1.5IQR and Q3+1.5IQR are regarded as outliers and replaced with corresponding boundary values. At the same time, combined with factors such as the user's BMI index and exercise habits, the Z score is weighted and adjusted to provide a more comprehensive physical function assessment. According to the Z score, the standard normal distribution table is searched to obtain the percentile corresponding to the user indicator. The percentiles are divided into multiple levels: 0-10% is extremely low, 10-30% is low, 30-70% is normal, 70-90% is high, and 90-100% is extremely high. Determine whether the user's physical function is within the normal range. The normal range is defined as the percentile between 30% and 70%. For non-normally distributed data, the percentile is calculated using the empirical distribution function, and the same level division method is used. Calculate the percentage difference between the user indicator and the average level of the group, the difference percentage = (user indicator value-group mean) / group mean × 100%, as a quantitative comparison result. Set different difference percentage thresholds: within ±5% is basically the same, ±5%-15% is a slight difference, ±15%-30% is a significant difference, and above ±30% is a great difference. According to the degree of difference and the percentile level, give the corresponding evaluation conclusion. For example, "Your balance ability is at the 80% level of people of the same age and gender, which is in the high range, 20% higher than the average level, and performs well." At the same time, combined with factors such as BMI and exercise habits, personalized health advice is provided. A sample of a 35-year-old male user of the same age and gender was screened from the database, totaling 1,256 people. The mean of the balance ability index of this group was calculated to be 78.5 points, with a standard deviation of 12.3 points; the mean of the lower limb muscle strength index was 320N, with a standard deviation of 45N. The Shapiro-Wilk test result showed that the p value was 0.072, which was greater than 0.05, so the data was considered to be in line with the normal distribution. The user's balance ability score was 85 points, and the Z score was calculated as (85-78.5) / 12.3=0.69. The lower limb muscle strength test value was 350N, and the Z score was (350-320) / 45=0.67.The interquartile range method was used to detect abnormal values. Q1 was 72 points, Q3 was 85 points, IQR was 13 points, the lower limit was 52.5 points, and the upper limit was 104.5 points. The user data was within the normal range. The standard normal distribution table was queried, and the percentile of balance ability was 70.19%, and the percentile of lower limb muscle strength was 74.86%. Both indicators were at the "high" level. The difference percentage was calculated, and the balance ability was 8.28% higher than the average level, which was a slight difference; the lower limb muscle strength was 9.38% higher than the average level, which was also a slight difference. Combined with the user's BMI index of 22.5 (normal range) and the information that the frequency of exercise is 3 times a week, the system generates an evaluation conclusion: "Your balance ability and lower limb muscle strength are both at the 70%-75% level of people of the same age and gender, which is on the high side and about 9% higher than the average level, which is a good performance. Combined with your BMI index and exercise habits, it is recommended that you maintain the current exercise intensity and appropriately increase the proportion of balance training to further improve your physical function." The system stores this evaluation result in the user's health file for subsequent tracking and comparison.

[0051] S203. If the evaluation result shows that the user has a balance deviation or there is a left-right difference in lower limb muscle strength, the corresponding exercise training program and rehabilitation guidance plan are pushed to the user through the interactive interface of the weight scale device.

[0052] In a possible implementation of the present application, if the evaluation result indicates that the user has a body balance deviation or a left-right difference in lower limb muscle strength, a corresponding exercise training and rehabilitation guidance program is output for the user, including:

[0053] The Z score is calculated based on the assessment results. The Z score is used to quantify the degree of balance deviation and left-right difference in lower limb muscle strength:

[0054]

[0055] Where Z represents the Z score, X represents the user's evaluation score, μ represents the mean of the evaluation score, and σ represents the standard deviation of the evaluation score.

[0056] According to the degree of deviation, user age, gender, BMI index and physical condition, suitable training programs and rehabilitation guidance plans are screened out from the pre-established sports training database.

[0057] According to the degree of deviation, the corresponding training difficulty and rehabilitation guidance plan are designed. Among them, low-intensity training is used for mild deviation, medium-intensity training is used for moderate deviation, and high-intensity training is used for severe deviation.

[0058] Integrate the selected training programs and rehabilitation guidance plans into personalized push content, including text descriptions, pictures and video instructions. Display personalized push content to users through the touch screen display of the scale device or the associated mobile application. Obtain users' execution and effect evaluation of training programs and rehabilitation guidance, process user feedback data, and recommend training programs that similar users have achieved good results.

[0059] Specifically, based on the user's evaluation results, the Z score is used to quantify the degree of balance ability deviation and left-right difference in lower limb muscle strength. |Z|<1 is defined as mild, 1≤|Z|<2 is defined as moderate, and |Z|≥2 is defined as severe. Based on factors such as the user's age, gender, BMI index, and physical condition, suitable training programs and rehabilitation guidance plans are screened from the sports training database. Safety checks are performed on the screening results to ensure that the recommended training programs will not cause potential harm to the user. According to different degrees of deviation and difference, corresponding training difficulty and rehabilitation guidance plans are designed. Mild deviations use low-intensity training, 3 times a week, 20 minutes each time; moderate deviations use moderate-intensity training, 4 times a week, 30 minutes each time; severe deviations use high-intensity training, 5 times a week, 40 minutes each time. The upper limit of training duration and frequency is set according to the user's age and BMI index. For example, the single training duration of users over 60 years old does not exceed 30 minutes, and the training frequency of users with a BMI greater than 30 does not exceed 4 times a week. The screened training programs and rehabilitation guidance plans are integrated into personalized push content, including text descriptions, pictures, and video descriptions. Show it to the user through the touch screen display of the scale device or the associated mobile application, set the daily push frequency and reminder function. Add user feedback collection function in the interactive interface, including training completion status, feeling score and difficulty evaluation. Regularly evaluate the user's progress, such as conducting balance ability and muscle strength tests every two weeks, and adjust the training plan accordingly. Design a user feedback mechanism to collect users' execution and effect evaluation of training programs and rehabilitation guidance. Use collaborative filtering algorithm to process feedback data and recommend training programs that similar users have achieved good results. Use decision tree algorithm to dynamically adjust training plans according to user characteristics and feedback history to optimize the recommendation system. Introduce social functions to allow users to anonymously share training experience and progress. The system evaluates the effectiveness of shared content based on the number of likes and comments, and incorporates high-quality user experience into the recommendation system. Continuously optimize the pertinence and effectiveness of training suggestions to improve user participation and training results. Analyze the evaluation results of a 45-year-old male user. The balance ability Z score is -1.5, and the left and right side difference in lower limb muscle strength Z score is 1.8. According to the preset standards, the balance ability is judged to be moderately biased and the muscle strength difference is moderate. Considering the user's BMI is 27, suitable training items are screened out from the database: standing on one leg, walking on tiptoes, and squats. Safety checks show that these items are suitable for this user. The system designs a moderate-intensity training program: 4 times a week, 25 minutes each time. Considering the slightly higher BMI, the standard duration is reduced by 5 minutes. Personalized training videos and graphic instructions are pushed through the user's mobile phone APP, and a daily reminder is set at 20:00. After the user completed the first week of training, feedback was given through the APP: completion rate 80%, difficulty score 7 / 10, and overall feeling 6 / 10. Retesting two weeks later showed that the balance ability Z score increased to -1.2, and the muscle strength difference decreased to 1.5.The system uses a collaborative filtering algorithm to find that users with similar conditions benefit from "yoga balance ball training" and adds it to the recommendation list. The decision tree algorithm adjusts the plan based on feedback, adjusting the duration of a single training session to 30 minutes and slightly reducing the difficulty. In the social function, the user's "stick to brushing teeth on one leg" skill received a lot of likes, and the system pushed it as a supplementary suggestion. After 4 weeks, the user's balance ability and muscle strength differences improved to a slight level, and the system adjusted to low-intensity maintenance training accordingly.

[0060] In a possible embodiment of the present application, the method provided in the embodiment of the present application may further include:

[0061] S204. When the user uses the weight scale, the pressure difference data between the left and right feet of the user is continuously collected, and daily monitoring is performed by obtaining exercise habit data, and the balance ability and muscle strength assessment model is adaptively optimized to improve the accuracy of the assessment results.

[0062] For example, the electronic device can receive measurement information carrying left and right foot pressure difference data, which is sent when the measurement information is collected by the weight scale. The electronic device obtains the user's exercise habit data based on the measurement information, and the exercise habit data includes exercise type, frequency and intensity; stores the left and right foot pressure difference data in a time series database, which is indexed by timestamps and has left and right foot pressure values ​​as fields; preprocesses the left and right foot pressure difference data, including: using linear interpolation to process missing values ​​and remove outliers; calculates statistical features based on the preprocessed left and right foot pressure difference data, and the statistical features include mean, standard deviation, and kurtosis; performs incremental learning on the balance ability and muscle strength assessment model, and the incremental learning uses newly collected data as training samples; determines whether the performance of the balance ability and muscle strength assessment model reaches the expected level, including: obtaining the latest data as a validation set, and calculating the accuracy and mean square error of the balance ability and muscle strength assessment model on the validation set; if the performance of the balance ability and muscle strength assessment model reaches the expected level, the balance ability and muscle strength assessment model is deployed to the weight scale device for real-time assessment of the user's balance ability and muscle strength level.

[0063] Specifically, every time a user uses a weight scale device, the pressure difference data of the left and right feet is collected, and the user's exercise habit data, including information such as exercise type, frequency and intensity, is obtained through the associated mobile application. A time series database is used to store the pressure difference data, with timestamps as indexes and left and right foot pressure values ​​as fields. The exercise habit data is stored in JSON format, including date, exercise type, duration and intensity level. During the data collection process, the validity of the collected data is ensured by detecting the stability of the user's standing posture. The data are integrated and stored at daily intervals. The collected pressure difference data and exercise habit data are preprocessed and feature extracted. Linear interpolation is used to handle missing values ​​in pressure data, and the 3σ principle is used to detect and eliminate outliers. The statistical features of pressure difference are calculated, including mean, standard deviation, kurtosis, etc. The quantitative indicators of exercise habits include weekly exercise time and exercise intensity score, where the exercise intensity score is calculated based on METs (metabolic equivalents) values. The user's diet and sleep data are also included in the feature extraction process to form a multimodal feature vector. An online gradient descent algorithm is used to incrementally learn the existing balance ability and muscle strength assessment model. Use the newly collected data as training samples, gradually adjust the model parameters, and achieve adaptive optimization of the model. Set a learning rate decay mechanism, multiply the learning rate by 0.9 after processing every 1,000 samples. Use SHAP values ​​to analyze feature importance and dynamically adjust feature weights. During the training process, integrate stress data, exercise habits, diet, and sleep information to improve the comprehensiveness of the model. Evaluate the performance of the optimized model regularly, such as monthly. Use the data from the most recent month as the validation set to calculate indicators such as the accuracy and mean square error of the model on the validation set. If the performance improvement is lower than the preset threshold, such as the accuracy improvement is less than 1%, adjust the feature extraction method or add new features. Analyze the results based on the SHAP value method, identify and add new features with greater impact. Repeat the feature extraction and model update process until the model performance reaches the expected level or there is no significant improvement after three consecutive evaluations. Deploy the final optimized model to the weight scale for real-time evaluation of the user's balance ability and muscle strength level. When the user weighs, the smart weight scale collects pressure data of the left and right feet every 0.1 seconds and stores it in time series format. For example, 2024-03-15 08:00:00.1 records 300N for the left foot and 280N for the right foot. The user records a 30-minute jog on the same day through a mobile phone APP, with an intensity of 6METs. The system stores this data in a time series database, while also recording the user's diet (2000 calories) and sleep (7 hours) information. During the data preprocessing phase, it was found that the data at 08:00:00.3 was missing, and linear interpolation was used to fill it. The 3σ principle detected the outlier at 08:00:01.5 (1000N for the left foot) and removed it. The calculated pressure difference has a mean of 20N, a standard deviation of 5N, and a kurtosis of 2.8.The exercise intensity score is 180, including 30 minutes * 6METs. The feature vector contains these statistics as well as diet and sleep data. The online gradient descent algorithm uses these new data to update the model, with an initial learning rate of 0.01 and a decay of 10% every 1000 samples. SHAP value analysis shows that the standard deviation of pressure difference has the greatest impact on the model, with an importance of 0.3. One month later, the model was evaluated using the last 30 days of data as a validation set, and the accuracy increased from 85% to 86.5%. Since the improvement exceeded 1%, the system retained the new model and deployed it to the device for real-time evaluation of user balance and muscle strength, with an output score of 92.

[0064] What is disclosed above is only the preferred embodiment of the present application, and it certainly cannot be used to limit the scope of rights of the present application. A person skilled in the art can understand that all or part of the processes of the above embodiments and equivalent changes made according to the claims of the present application still fall within the scope of the invention.

Claims

1. A health detection method based on a weight scale, characterized in that: A target evaluation model of balance ability and muscle strength based on pressure characteristics is integrated in the weight scale, and pressure sensor arrays are distributed on both sides of the weight scale. The method includes: When the user uses the weight scale to weigh himself, the pressure sensor array collects the pressure distribution data of the left and right feet of the user in real time; Determine the statistical characteristics of the pressure difference between the left and right feet and the pressure difference mode according to the pressure distribution data of the left and right feet, wherein the pressure difference mode is used to determine whether there is a deviation in the body balance ability of the user; the statistical characteristics of the pressure difference between the left and right feet are used to determine whether there is a difference in the muscle strength of the user's lower limbs between the left and right sides; Matching the statistical characteristics of the left and right foot pressure difference and the pressure difference pattern with the target evaluation model to determine the user's body balance ability and lower limb muscle strength, and obtain a quantitative evaluation result; If the evaluation result indicates that the user has a body balance deviation and / or a left-right difference in lower limb muscle strength, outputting a sports training program and a rehabilitation guidance program for the user; During the process of the user using the weight scale, continuously collecting the pressure difference data of the left and right feet of the user and acquiring the exercise habit data of the user; The target evaluation model is adaptively optimized according to the left and right foot pressure difference data of the user and the exercise habit data of the user. When the user uses the weight scale, the left and right foot pressure difference data is continuously collected, and the exercise habit data is obtained for daily monitoring to adaptively optimize the balance ability and muscle strength evaluation model, including: receiving measurement information carrying left and right foot pressure difference data, the measurement information being collected by the weight scale; Acquire the user's exercise habit data according to the measurement information, wherein the exercise habit data includes exercise type, frequency and intensity; The pressure difference data of the left and right feet is stored in a time series database, wherein the time series database uses the timestamp as an index and the pressure values ​​of the left and right feet as fields; Preprocessing the left and right foot pressure difference data by using a linear interpolation method to process missing values ​​and abnormal values ​​in the left and right foot pressure difference data; Calculating statistical features based on the preprocessed left and right foot pressure difference data, wherein the statistical features include mean, standard deviation and kurtosis; Performing incremental learning on the target evaluation model, wherein the incremental learning uses newly collected data as training samples; When the performance of the target evaluation model reaches the expected level, the target evaluation model is deployed to the weight scale. The performance of the target evaluation model reaches the expected level including: taking the latest data as a validation set, and the accuracy and mean square error of the target evaluation model on the validation set are less than or equal to preset values.

2. The health detection method based on a weight scale according to claim 1, characterized in that: If the evaluation result indicates that the user has a body balance deviation and / or a left-right difference in lower limb muscle strength, outputting a sports training program and a rehabilitation guidance program for the user, including: Calculating a Z score according to the evaluation result, wherein the Z score is used to quantify the degree of deviation of the user's body balance ability and / or the degree of deviation of the difference between the left and right sides of the user's lower limb muscle strength; According to the degree of deviation and the individual parameters of the user, selecting a sports training program and a rehabilitation guidance program suitable for the user from a pre-established sports training database; The lower the deviation degree, the lower the intensity of the training program, and the individual parameters include at least one or more of the following: age, gender, BMI index and physical condition; Integrate the selected sports training programs and rehabilitation guidance programs into personalized push content, wherein the personalized push content includes text descriptions, pictures, and video descriptions; The personalized push content is displayed through the touch screen display of the weight scale or an associated mobile application.

3. The health detection method based on a weight scale according to claim 1, characterized in that: The method further comprises: Acquiring user feedback data of the user regarding the sports training program and rehabilitation guidance program, wherein the user feedback data includes execution status and effect evaluation; Based on the user feedback data, a target sports training program is recommended to the user, wherein the target sports training program is a sports training program that can achieve good results for users with a similar degree of deviation in body balance ability to the user and / or a similar degree of deviation in left and right side difference in lower limb muscle strength to the user.

4. The health detection method based on a weight scale according to claim 1, characterized in that: The statistical characteristics of the pressure difference between the left and right feet and the pressure difference pattern are matched with the target evaluation model to determine the user's body balance ability and lower limb muscle strength, and obtain a quantitative evaluation result, including: Normalizing the pressure difference statistical characteristics to obtain characteristic values ​​that match the target assessment model input; The characteristic value is input into the target evaluation model, and a quantitative evaluation result of the user's body balance ability and lower limb muscle strength level is obtained through forward propagation calculation; Determine the level of the quantitative evaluation results according to the preset scoring criteria; Analyze the difference between the pressure difference pattern of the user and that of normal people, determine whether there is a deviation in the balance ability of the user, and provide a quantitative indicator of the degree of deviation; Analyze the statistical characteristics of the pressure difference between the left and right feet of the user, determine whether there is a difference in the muscle strength of the user's lower limbs between the left and right sides, and provide a quantitative indicator of the degree of difference; The user's body balance ability and lower limb muscle strength level are compared with the average level of people of the same age and gender to determine whether the user's body function is within the normal range and provide a quantitative comparison result.

5. The health detection method based on a weight scale according to claim 4, characterized in that: The analysis of the difference between the user's pressure difference pattern and that of normal people, judging whether the user's balance ability has deviation, and providing a quantitative index of the degree of deviation, analyzing the statistical characteristics of the pressure difference between the user's left and right feet, judging whether the user's lower limb muscle strength has left and right side differences, and providing a quantitative index of the degree of difference, includes: Obtain stress difference pattern data of healthy adults and select key stress difference features; The mean vector and covariance matrix were calculated based on the pressure difference pattern data, and a multivariate Gaussian distribution model was constructed as a benchmark model for normal people. The Mahalanobis distance is used to calculate the difference between the user's pressure difference pattern and the normal population benchmark model, and a quantitative index of the deviation degree of the user's balance ability is obtained; Calculate the statistical characteristics of the pressure difference between the user's left and right feet, and determine whether there is a left-right difference in the user's lower limb muscle strength by training a support vector machine; If there is a left-right difference, the distance from the user's feature vector to the classification hyperplane is calculated to determine a quantitative indicator of the degree of difference in the user's lower limb muscle strength between the left and right sides.

6. The health detection method based on a weight scale according to claim 4, Among them, it is characterized in that the user's balance ability and lower limb muscle strength level are compared with the average level of people of the same age and gender to determine whether the user's physical function is within the normal range, and a quantitative comparison result is given, including: According to the user's age and gender information, obtain the corresponding population sample data, and divide the population sample into intervals according to age; Calculate the mean and standard deviation of the balance ability index and lower limb muscle strength index of each group of samples after division; Determine whether the indicator data of each group of samples conforms to the normal distribution; If the indicator data conforms to the normal distribution, a normal distribution model is constructed as a reference benchmark; If the indicator data does not conform to the normal distribution, the median and interquartile range are used as reference indicators; Convert the user's balance ability index and lower limb muscle strength index into Z scores. The Z score is calculated by subtracting the group mean from the user's index value and dividing it by the group standard deviation. Look up the standard normal distribution table based on the Z score to get the percentile corresponding to the user indicator; Divide the percentiles into multiple levels according to the preset percentile range to determine whether the user's physical function is within the normal range; Calculate the percentage difference between the user's index and the group's average level. The percentage difference is equal to the user's index value minus the group's average value, divided by the group's average value, and then multiplied by the percentage coefficient. According to the difference percentage and percentile rank, the corresponding evaluation conclusion is given.

7. The health detection method based on a weight scale according to any one of claims 1 to 6, characterized in that: The method further comprises: The pressure sensor arrays arranged on the left and right sides of the weight scale are used to collect a plurality of pressure distribution data of the left and right feet of a plurality of different users when they are weighed on the weight scale; Analyze the time series characteristics, frequency domain characteristics and spatial distribution characteristics of the pressure distribution data of multiple different users; The pressure distribution data is denoised and smoothed using a wavelet transform signal processing method to extract statistical features of the pressure differences between the left and right feet of different users; Analyze the correlation between the statistical characteristics of left and right foot pressure differences of different users and their gait rhythm, body center of gravity transfer, and lower limb joint angles; Based on the time series characteristics, frequency domain characteristics and spatial distribution characteristics of pressure distribution data, as well as the statistical characteristics of left and right foot pressure differences and the obtained correlation, a pressure characteristic database covering different ages, genders and physical conditions is constructed; Cluster analysis method was used to extract the left and right foot pressure difference patterns with significant statistical characteristics from the pressure feature database, and used as evaluation indicators to measure body balance ability and lower limb muscle strength distribution; For each pressure difference pattern, the exercise habit data of the collected users is obtained, and correlation analysis is performed in combination with medical diagnosis data to determine the quantitative correspondence between the pressure difference pattern and the body's balance ability and lower limb muscle strength. A target assessment model of balance ability and muscle strength based on pressure characteristics is constructed, and its relationship with age, gender and exercise habits is analyzed.

8. The health detection method based on a weight scale according to claim 7, characterized in that: The method collects a plurality of pressure distribution data of the left and right feet of a plurality of different users when they are weighed on the weight scale by means of a pressure sensor array arranged on the left and right sides of the weight scale, including: Collect the pressure distribution data of the left and right feet of the user when walking, and construct a two-dimensional pressure distribution matrix based on the spatial position of the pressure sensor and the corresponding pressure value; Performing bilinear interpolation smoothing processing on the two-dimensional pressure distribution matrix to obtain a continuous pressure distribution surface; Performing time series analysis on the pressure distribution surface to calculate the frequency spectrum of the pressure signal; Extract the main frequency components and their amplitudes according to the spectrum; The gait cycle is divided by identifying the heel strike and toe lift-off moments through the main frequency components; According to the spatial characteristics of the pressure distribution surface, the center of gravity position of the sole contact area at each moment is calculated; According to the position of the center of gravity, the trajectory of the center of gravity is tracked to obtain the motion curve of the pressure center; Perform polynomial regression fitting on the motion curve to extract the characteristic parameters of center of gravity transfer speed, acceleration and curvature; The gait feature vector is constructed using the gait cycle, spectrum features and center of gravity trajectory feature parameters; Normalizing the gait feature vector to obtain a normalized feature vector; The support vector machine algorithm is used to classify the normalized feature vectors; If the SVM classification results show an abnormal gait, the abnormal gait characteristics are compared with the normal gait characteristics to quantify the difference in pressure distribution between the left and right feet.

9. The health detection method based on a weight scale according to claim 7, characterized in that: The wavelet transform signal processing method is used to perform denoising and smoothing on the pressure distribution data, and the statistical characteristics of the pressure difference between the left and right feet of different users are extracted; and the correlation between the statistical characteristics of the pressure difference between the left and right feet of different users and the gait rhythm, the body center of gravity transfer and the lower limb joint angle is analyzed, including: The multi-scale wavelet transform method is used to perform frequency decomposition on the pressure distribution data. According to the frequency decomposition results, a soft threshold is selected to perform noise reduction on the high-frequency wavelet coefficients. Applying a sliding average filter to smooth the pressure distribution data after noise reduction; Extract the statistical characteristics of the pressure difference between the left and right feet from the smoothed pressure distribution data, including mean, standard deviation, skewness and kurtosis; The displacement of the center of pressure is calculated by the relative rate of change of the pressure of the left and right feet; The gait rhythm is extracted based on the periodic changes of pressure data, and the pressure center trajectory is used to analyze the body's center of gravity transfer; A multivariate linear regression method was used to establish an association model between the feature vector and the gait rhythm, the body center of gravity transfer and the lower limb joint angle, wherein the pressure difference feature was the independent variable, and the gait rhythm, the body center of gravity transfer and the lower limb joint angle were the dependent variables.

Images