Family blood pressure measurement and identity recognition system and method based on oscillatory wave characteristics

Abstract

The invention provides a family blood pressure measurement and identity recognition system and method based on oscillatory wave characteristics, and relates to the technical field of household medical equipment. The method comprises the following steps: firstly, acquiring original oscillatory wave signals during blood pressure measurement, judging whether the signals are available or not, further fusing time domain and frequency domain features to perform quality evaluation on available oscillatory wave signals, and screening high-quality signals; filtering and noise reduction preprocessing is carried out on the oscillatory wave signals with good quality; then respectively extracting time domain, frequency domain and time-frequency joint features from the preprocessed oscillatory wave signal to obtain multi-index features of the oscillatory wave signal; carrying out standardization processing and unsupervised feature fusion on the multi-index feature data; finally, non-invasive blood pressure measurement and identity recognition are completed. According to the method, the oscillation wave signals are subjected to fusion analysis according to the multi-index characteristics, non-invasive blood pressure measurement is achieved, the accuracy and calculation efficiency of identity recognition are enhanced, historical blood pressure measurement results of fixed members are conveniently stored, and a family blood pressure monitoring system is constructed.

Description

Technical Field

[0001] The present invention relates to the technical field of household medical devices, and particularly to a home blood pressure measurement and identity recognition system and method based on oscillatory wave characteristics. Background Art

[0002] In the field of home health monitoring, blood pressure measurement is an important means for chronic disease management. With the intelligent development of household medical devices, the scenario of multiple family members sharing a blood pressure monitor has become increasingly common. However, existing devices generally lack a reliable identity recognition function, resulting in the inability to match measurement data with specific family members one by one. The method of manually inputting identity information not only reduces the usage efficiency but also may cause data confusion due to user negligence, affecting the accuracy of long-term health trend analysis.

[0003] Current mainstream biometric technologies such as fingerprint recognition and face recognition require additional hardware support and are vulnerable to environmental interference (such as sensor contamination and light changes). In contrast, oscillatory wave signals, as accompanying data in blood pressure measurement, naturally contain individual physiological characteristic information. Research shows that there are significant differences in the amplitude distribution, frequency characteristics, and time-frequency dynamic changes of oscillatory wave waveforms among different individuals. These differences are closely related to vascular elasticity, hemodynamic characteristics, and individual health status, providing a biometric basis for identity recognition.

[0004] Chinese Patent CN118211182A performs identity recognition based on pulse wave signals. Its effective frequency band is narrow, feature extraction is limited, and it is vulnerable to environmental interference. The frequency range of oscillatory wave signals is wider, containing more high-frequency features, and can more accurately reflect individual differences, especially suitable for the scenario of multiple family members sharing devices. Moreover, due to different frequencies, the processing methods of the two waveforms are also different in terms of quality assessment, feature extraction, and feature fusion. In addition, the stability of oscillatory wave signals is higher. Combining blood pressure measurement to achieve identity recognition is highly innovative and solves the deficiencies of traditional pulse wave solutions in terms of practicality and recognition accuracy.

[0005] The differences in oscillatory wave characteristics among different family members stem from the natural differences in their physiological states. In the elderly, due to arteriosclerosis, the vascular elasticity decreases, and their oscillatory waves show characteristics such as reduced amplitude and posterior shift of the main peak; in middle-aged people, during the stable period of vascular function, the waveform presents a sharp main peak and the spectral energy is concentrated in the middle and low frequency bands; in children, due to high vascular elasticity and fast heart rate, the waveform shows high amplitude, narrow main peak, and an increased proportion of high-frequency components. When the system detects a waveform significantly different from the known template, it may indicate two situations: one is that a new member who has not been recorded is being measured, and the other is that there is a pathological change in an existing member (such as the disappearance of the dicrotic wave in arteriosclerosis patients and the spectral energy shift in diabetic patients). This difference based on physiological characteristics provides a biometric basis for multi-member identity recognition and also provides data basis for early warning of abnormal health status.

[0006] Through the multi-index fusion analysis of the oscillatory wave signal, the present invention realizes the automatic identification of the user's identity during measurement, and solves the problem of data confusion when multiple family members share devices. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to provide a home blood pressure measurement and identity recognition system and method based on oscillatory wave characteristics in view of the deficiencies of the above-mentioned prior art. Through the multi-index fusion analysis of the oscillatory wave signal, the present invention realizes the automatic identification of the user's identity during measurement, and solves the problem of data confusion when multiple family members share devices.

[0008] To solve the above technical problem, the technical solution adopted by the present invention is as follows: On the one hand, the present invention provides a home blood pressure measurement and identity recognition system based on oscillatory wave characteristics, including an oscillatory wave signal acquisition and "usable" signal discrimination module, a preprocessing module, a multi-index feature extraction module, a multi-index feature standardization processing and unsupervised fusion module, and a non-invasive blood pressure measurement and identity recognition module; The oscillatory wave signal acquisition and "usable" signal discrimination module is used to continuously and real-time collect the original oscillatory wave signal through the cuff and the pressure sensor during blood pressure measurement, evaluate the influence of each interference on the signal quality according to the cuff fitting and arm movement conditions, and further fuse the time-domain and frequency-domain characteristics to evaluate the quality of the "usable" oscillatory wave signal, and screen out high-quality signals from the "usable" oscillatory wave signals; the "unusable" oscillatory wave signals are signals with distorted waveforms or completely disappeared signals; the signal quality of the "usable" oscillatory wave signals is not completely overwhelmed by noise or interference, and can be further distinguished between good and bad quality; The preprocessing module is used to preprocess the oscillatory wave signals with good quality, remove the noise in the oscillatory wave signals, and facilitate subsequent feature extraction; The multi-index feature extraction module extracts the time-domain characteristics of the oscillatory wave signal based on the Pan-Tompkins feature point detection algorithm, extracts the frequency-domain characteristics of the oscillatory wave signal based on the spectrum analysis method, and extracts the time-frequency characteristics of the oscillatory wave signal based on the short-time Fourier transform method; The multi-index feature standardization processing and unsupervised fusion module is responsible for standardizing the oscillatory wave signal feature data; first, perform a standardization operation on the oscillatory wave signal feature data to eliminate the differences in dimension and magnitude between different features; then use the PCA technology to perform dimensionality reduction processing on the standardized oscillatory wave signal feature data; then use the deep autoencoder to carry out further feature learning and fusion on the oscillatory wave feature data after PCA dimensionality reduction, and achieve the unsupervised feature fusion of the oscillatory wave signal; The non-invasive blood pressure measurement and identity recognition module is based on the fused features extracted by the deep autoencoder. On the one hand, it evaluates the importance of features based on the contribution of reconstruction error, that is, analyzes the change of the autoencoder reconstruction error after each fused feature is removed. The greater the contribution of the feature to the reconstruction error, the more important the feature is for the representation of the data. For the data after each feature removal, calculate the reconstruction error and compare it with the reconstruction error of the original model without removing any features. The increase in the reconstruction error after feature removal reflects the importance of the feature. Features with a significant increase in reconstruction error are considered more important for the model and data representation. On the other hand, according to the result of feature importance evaluation, weighted fusion of oscillatory wave features is carried out. First, feature weight allocation is performed. The weight size of the feature is positively correlated with the importance of the feature. Divide the importance score of each oscillatory wave feature by the sum of the importance scores of all features to obtain the weight of the feature. Apply the calculated weight to each oscillatory wave feature vector to obtain the weighted oscillatory wave feature vector. On this basis, use the pre-trained model for non-invasive blood pressure prediction to achieve non-invasive blood pressure measurement, and use the obtained weighted oscillatory wave feature vector as the feature template of the current detected object. Adopt the Euclidean distance metric algorithm to perform identity recognition in the pre-stored oscillatory wave feature database, that is, match it with the known identity template library, and finally output the result to complete the identity recognition.

[0009] Further, the deep autoencoder includes two parts: an encoder and a decoder. The function of the encoder is to further compress the input oscillatory wave feature data into a low-dimensional space, and the decoder attempts to reconstruct the original oscillatory wave features from this low-dimensional space. The specific process of unsupervised feature fusion is as follows: Use the oscillatory wave feature data after PCA dimensionality reduction processing as the input of the deep autoencoder, adopt the mean square error as the loss function, and train the model. The purpose is to minimize the difference between the input oscillatory wave features and the reconstructed features. After the model training is completed, the output of the decoder part, that is, the feature representation in the low-dimensional space, is used as the fused oscillatory wave feature. Map the input feature x2 to the low-dimensional latent space z: (12); Among them, and are the encoder weight matrices and are hidden layer parameters; and represent the encoder bias terms, σ is the activation function; Reconstruct the output from the latent space z : (13); Among them, and is the decoder weight matrix, and is the decoder bias top.

[0010] Furthermore, the implementation process of the non-invasive blood pressure measurement and identity recognition module for non-invasive blood pressure measurement is as follows: Based on the pre-stored oscillatory wave database, the extreme gradient boosting algorithm is used to optimize for individual differences; the objective function of XGBoost consists of a loss function and a regularization term, and the formula is as follows: (16); where y i is the true blood pressure value, is the blood pressure value predicted by the model, T k is the number of leaf nodes of the k-th tree, w k is the leaf weight of the k-th tree, γ and λ are the regularization coefficients for controlling the model complexity; Mix the oscillatory wave data of various categories of people to train the model, and control the category bias through cross-validation to complete the pre-training of the non-invasive blood pressure prediction model; use stratified K-fold cross-validation to ensure that the category distribution of each fold of data is consistent, and the model evaluation formula is: (17); where K is the number of cross-validation folds, is the k-th fold validation data set; At the same time, continuously use the newly collected oscillatory wave data to enable the non-invasive blood pressure prediction model to adapt to the individual characteristics of the test object, and iteratively update the model; when the new data arrives, the incremental training formula is: (18); where K is the number of trees in the existing model, M is the number of newly added trees, f K+m ( x i ) is the predicted value of the newly added m-th tree, is the predicted value of the original model, is the predicted value of the updated model; Input the weighted oscillatory wave feature vector as input data into the pre-trained model for non-invasive blood pressure prediction, perform the model prediction operation, and output two blood pressure values, systolic blood pressure and diastolic blood pressure, thus completing the non-invasive blood pressure measurement of the test subject.

[0011] Furthermore, the process of the non-invasive blood pressure measurement and identity recognition module for identity recognition is as follows: Calculate the weighted Euclidean distance between the sample to be identified and each known identity sample; Using the weighted oscillatory wave feature vector, calculate the weighted Euclidean distance between the sample to be identified and each known identity sample; For the sample to be identified Y and the known identity sample R, the weighted Euclidean distance is expressed as: (19); where D(Y, R) is the weighted Euclidean distance between sample Y and sample R, x3(z, Y) and x3(z, R) are the values of sample Y and sample R on the z-th feature respectively, and w(z) is the weight of the z-th feature; Based on the weighted Euclidean distances between the sample to be identified and each known identity sample, calculate multiple identity recognition scores; The identity recognition score is defined as the reciprocal of the weighted Euclidean distance between the sample to be identified Y and the known identity sample R, as shown in the following formula: (20); where Score(Y) represents the identity recognition score of the sample to be identified Y, and the higher the score, the higher the similarity between the samples; By comparing the identity recognition scores Score(Y) between the sample to be identified Y and all known identity samples, the higher the score, the closer the sample Y is to a certain known identity sample, and thus the higher the credibility of the identity recognition. Select the known identity sample with the highest identity recognition score as the final identity recognition result of the sample to be identified; At the same time, continuously update the newly collected oscillatory wave data to the known identity samples of this member, so that the identity recognition process can adapt to the characteristics of the test object under different physiological states and iteratively update the existing identity samples.

[0012] On the other hand, the present invention also provides a home blood pressure measurement and identity recognition method based on oscillatory wave features, which is implemented by the above-mentioned home blood pressure measurement and identity recognition system based on oscillatory wave features, and includes the following steps: Step 1: During the blood pressure measurement process, continuously collect the original oscillatory wave signal of the measured person through the cuff and the pressure sensor; the system first detects the cuff fitting state and the arm movement condition, synchronously evaluates the influence degree of each interference factor on the signal quality, and discriminates the "usable" and "unusable" signals; for the "usable" signals that meet the basic acquisition requirements, the system will further carry out multi-dimensional quality grading evaluation; Step 2: Preprocess the oscillatory wave signal with good quality, including noise removal, filtering and downsampling operations to remove the noise in the oscillatory wave signal; Step 3: Use the Pan-Tompkins feature point detection algorithm to extract the time-domain features of the oscillatory wave signal from the preprocessed oscillatory wave signal, use the spectral analysis method to extract the frequency-domain features of the oscillatory wave signal, and use the short-time Fourier transform to extract the time-frequency features of the oscillatory wave signal, so as to obtain the multi-index features of the oscillatory wave signal; Step 4: Perform standardization processing and dimensionality reduction on the time-domain, frequency-domain, and time-frequency features of the oscillatory wave signal obtained in Step 3. Use a deep autoencoder to perform more in-depth feature learning and fusion on the oscillatory wave feature data x2 after dimensionality reduction processing, achieving unsupervised feature fusion of the oscillatory wave signal, and obtaining the fused oscillatory wave feature x3; Step 5: Based on the fused oscillatory wave feature x3 extracted from the deep autoencoder in Step 4, perform feature weighted fusion, and then realize non-invasive blood pressure measurement and identity score recognition; Step 5.1: Evaluate the importance of the oscillatory wave features based on the reconstruction error contribution, and clarify the role of each feature in the whole; Step 5.2: Perform weighted fusion on the oscillatory wave features according to the results of the feature importance evaluation; Step 5.2.1: Feature weight assignment; The weight size of the feature is positively correlated with the importance of the feature. Divide the importance score of each oscillatory wave feature by the sum of the importance scores of all features to obtain the weight of this feature; the fused feature vector is x3 = [x3(1), x3(2), …, x3(k)], and its corresponding weight is w = [w(1), w(2), …, w(k)]; Step 5.2.2: Perform feature weighted fusion based on the weight assignment; Apply the calculated weight to each oscillatory wave feature vector to obtain the weighted oscillatory wave feature vector X weighted = [x3(1)w(1), x3(2)w(2), …, x3(k)w(k)]; Step 5.3: Realize non-invasive blood pressure measurement based on multi-index feature weighted fusion; Based on the weighted oscillatory wave feature vector X obtained in Step 5.2 weighted , input it into the pre-trained model for non-invasive blood pressure prediction to realize non-invasive blood pressure measurement of the test subject; Step 5.4: Perform identity score recognition based on multi-index feature weighted fusion; Use the weighted oscillatory wave feature vector obtained in Step 5.2 as the feature template of the current detected object; adopt the Euclidean distance metric algorithm to perform identity recognition in the pre-stored oscillatory wave feature database and finally output the result.

[0013] Furthermore, the specific method of Step 1 is: Step 1.1: Adopt a multi-parameter joint error detection mechanism to detect errors in the collected oscillatory wave signals, monitor the signal integrity in real time, and determine whether the oscillatory wave signals collected by the cuff and the sensor are "usable"; When abnormal situations such as the cuff being too tight or not firmly attached, interference caused by the user's arm movement or exertion during the measurement, or severe baseline drift are detected, it is determined that this section of the signal is "unusable" and a re-acquisition instruction is triggered; for signals that do not trigger the error detection mechanism, the system marks them as "usable" and transfers them to the next evaluation link; Step 1.2: For the "usable" oscillatory wave signals, a quality assessment model based on the fusion of time-frequency domain features is constructed, and indicators in both the time domain and the frequency domain are comprehensively considered to evaluate the signal quality; the specific method is as follows: Step 1.2.1: Conduct time-domain analysis on the oscillatory wave signals, and use the amplitude change degree index, baseline drift degree index, and pulse rate consistency degree index as the signal quality assessment indicators in the time domain; Perform variational mode decomposition and period segmentation on the oscillatory wave signal pulse respectively to obtain the reconstructed signal bPulse and single-period signal Pulse that characterize the baseline drift of the signal i ; The amplitude change degree index is obtained through the amplitude difference of the signal waveforms in multiple cardiac cycles; the amplitude change degree index a(i) in the i-th cycle of the oscillatory wave signal is as follows: a(i)=|max(Pulse i )-min(Pulse i ) | (1); Among them, max(Pulse i ) and min(Pulse i ) are respectively the maximum and minimum values of the waveform amplitudes of the i-th cycle of the oscillatory wave signal after filtering processing; The baseline drift degree index is obtained by windowing the reconstructed signal; Perform windowing on the reconstructed signal bPulse to obtain the baseline drift degree index b(i) in the i-th cycle of the oscillatory wave signal, as follows: b(i)= max( |bPulse i |)-b (2); Among them, bPulse i is the reconstructed signal of the i-th cycle of the oscillatory wave signal, and b is the baseline standard value of the oscillatory wave signal in the stationary state; The pulse rate consistency degree index is calculated by statistically analyzing the time lengths of the single-period signals after period segmentation and calculating the consistency of the signal time lengths in multiple cycles; Set the period length of the oscillatory wave signal as length(Pulse i ) to obtain the pulse rate consistency index r(i) within the i-th period of the oscillatory wave signal, as shown in the following formula: (3); where SD(length(Pulse i )) is the standard deviation of the calculated i-th signal period length, and Mean(length(Pulse i )) is the mean of the calculated i-th signal period length; Step 1.2.2: Conduct frequency-domain analysis on the oscillatory wave signal. Select the effective frequency band of the oscillatory wave signal through fast Fourier transform, calculate the power spectral density as the frequency component index, and evaluate the quality of the oscillatory wave signal; The frequency component index within the i-th period of the oscillatory wave signal f ([[]] i ) is: (4); where P is the power of the oscillatory wave signal and f represents frequency; Step 1.2.3: Based on the signal quality evaluation indicators in the time domain and frequency domain, use the information entropy theory to calculate the weight coefficients of each index in the time-frequency domain, perform multi-index weighted fusion of the oscillatory wave signal in the time-frequency domain, and obtain the fusion index SQI(i), as shown in the following formula: (5); where w a (i), w b (i), w r (i), w f (i) are the entropy weight values of the evaluation indicators a(i), b(i), r(i), f(i) respectively; Step 1.2.4: Conduct threshold discrimination based on the fusion index, and screen out high-quality signals from the "usable" oscillatory wave signals; When the fusion index SQI(i) exceeds the preset threshold, it is determined as a "high-quality signal" and enters the subsequent processing flow. The signals that do not meet the standard will be marked as "sub-optimal signals", and the system will prompt for re-measurement.

[0014] Furthermore, the specific method of the said Step 2 is: Step 2.1: Adopt cascaded digital filtering technology. First, filter out high-frequency noise through a low-pass filter, and then apply a non-linear median filtering algorithm to eliminate burst pulse interference to form a preliminarily noise-reduced signal sequence; Step 2.2: Use sliding window mean filtering combined with a high-pass filter to filter, smooth the signal, and remove baseline drift and low-frequency noise; Step 2.3: Perform selective downsampling processing. First, restore the continuity of the original signal through spline interpolation technology, and then adopt a multi-stage decimation algorithm to retain key feature information.

[0015] Furthermore, the specific method of step 3 is as follows: Step 3.1: Use the improved Pan-Tompkins algorithm for cycle identification. Locate the waveform starting point through adaptive threshold detection technology to achieve per-cycle segmentation processing; For a single-cycle waveform, extract the following contour feature parameters: the vertical distance between the main peak vertex and the baseline, the main peak amplitude H, the vertical distance between the late peak vertex and the baseline, the main peak amplitude H1, the maximum slope value of the waveform rising edge, the maximum slope K, the vertical distance between the dicrotic notch point and the baseline, the notch depth D, time parameters: the duration from the main peak to the starting point T1, the duration from the main peak to the dicrotic notch point T2, the duration from the dicrotic notch point to the late peak T3, the duration from the late peak to the end point T4, the complete cycle duration T, area parameters: the waveform area A1 in the T1 period, the waveform area A2 in the T2 period, the waveform area A3 in the T3 period, the waveform area A4 in the T4 period, the number of the complete cycle area A; Step 3.2: Perform multi-resolution spectrum analysis based on the spectrum analysis method, and use the spectral energy ratio of different sub-band frequencies as the frequency domain feature of the oscillatory wave signal; First, use the sub-band division method to equally divide the concentrated frequency band of the effective information into multiple sub-band intervals, and calculate the power spectral density of each sub-band by combining downsampling and Fourier transform; the power spectral energy E calculation formula in the c-d frequency band range is: (6); Among them, P(Pulse) is the power spectral density of the oscillatory wave signal in the c-d frequency band, as shown in the following formula: (7); Among them, FFT(Pulse) is the spectrum obtained by the oscillatory wave signal using FFT, and fs is the sampling frequency of the oscillatory wave signal; Combine the main information distribution characteristics of the oscillatory wave signal to perform multiple analyses on the multi-spectrum energy of the oscillatory wave signal, and generate the power spectral energy ratio feature parameter through normalization processing: PSER m =E m / E all (8); Among them, m represents the number of sub-band frequencies, E m is the power spectral energy value of the m-th sub-band frequency, and E all is the total energy of all frequency bands; Step 3.3: Extract the time-frequency diagram features of the oscillatory wave signal based on the short-time Fourier transform; Convert the one-dimensional time series into a time-frequency matrix through short-time Fourier transform, and use dynamic threshold segmentation technology to extract the core area with an energy ratio of more than 90%; obtain the frequency axis range through the contour tracking algorithm, calculate the contour height, and locate the local maximum points in the time-frequency diagram based on the energy gradient detection algorithm. Select the N extreme points with the largest amplitude to form a time-frequency feature vector, where each coordinate point corresponds to the significant feature position of the energy distribution in the time-frequency diagram; Perform short-time Fourier transform on the oscillatory wave signal Pulse(t) using a sliding window w(t) to obtain the time-frequency distribution of the oscillatory wave signal Figure X (t, f), where t represents time; Detect the main energy distribution region on the time-frequency distribution Figure X (t, f), extract the edge contour of this region, and represent it with a closed curve L(t, f); Calculate the maximum value f max and the minimum value f min on the frequency f axis of the closed curve L(t, f), and obtain the contour height range feature h of the oscillatory wave signal: h = f max - f min (9); Based on the contour height of the oscillatory wave signal, locate the local maximum points in the time-frequency distribution Figure X (t, f) using the energy gradient detection algorithm, select the N extreme values with the largest amplitude, and form a feature sequence from the time and frequency coordinates of these points to form the time-frequency local maximum point feature M of the oscillatory wave signal: (10); where, S(t n , f n ) is the local maximum point, n = 1, 2,..., N.

[0016] Furthermore, the specific method of step 4 is as follows: Step 4.1: Standardize the characteristic data included in the oscillatory wave signal to eliminate the differences in dimension and magnitude between different characteristics; the standardization formula is as follows: (11); where, x is the original characteristic value of the oscillatory wave signal, μ is the average value of the characteristic, σ1 is the standard deviation of the characteristic, and x1 is the standardized characteristic value; Step 4.2: Use the principal component analysis method to perform dimensionality reduction on the standardized oscillatory wave signal characteristic data; the specific method is as follows: Step 4.2.1: Calculate the covariance matrix: For the standardized characteristic data of the blood pressure oscillatory wave signal, calculate its covariance matrix and analyze the correlation between each characteristic; Step 4.2.2: Eigenvalue Decomposition: Perform eigenvalue decomposition on the obtained covariance matrix to obtain a series of eigenvalues and their corresponding eigenvectors, providing a basis for subsequent principal component selection; Step 4.2.3: Select Principal Components: According to the magnitudes of the eigenvalues, retain the eigenvectors corresponding to Q eigenvalues with a contribution exceeding 90%, and determine them as the principal components to focus on the key information; Step 4.2.4: Data Projection: Project the original oscillatory wave signal feature data onto the selected principal components to obtain the oscillatory wave signal feature data x2 after dimensionality reduction processing; Step 4.3: Use a deep autoencoder to perform more in-depth feature learning and fusion on the oscillatory wave feature data after dimensionality reduction to achieve unsupervised feature fusion of the oscillatory wave signal; the specific method is as follows: Step 4.3.1: Construct a deep autoencoder, including an encoder and a decoder; the role of the encoder is to further compress the input oscillatory wave features into a low-dimensional space, and the decoder attempts to reconstruct the original oscillatory wave features from this low-dimensional space; Map the input feature x2 to the low-dimensional latent space z: (12); where, and are the encoder weight matrices and are the hidden layer parameters; and represent the encoder bias terms, σ is the activation function; Reconstruct the output from the latent space z : (13); where, and are the decoder weight matrices, and are the decoder bias terms; Step 4.3.2: Model Training: Use the oscillatory wave feature data x2 after PCA dimensionality reduction as the input of the deep autoencoder, and use the Mean-Square Error (MSE) as the loss function to train the model, aiming to minimize the difference between the input oscillatory wave features and the reconstructed features; the MSE calculation formula is: (14); where, n is the number of samples, represents the i-th original oscillatory wave feature data after PCA dimensionality reduction, is the i-th oscillatory wave feature data reconstructed by the deep autoencoder; Step 4.3.3: Feature fusion representation: After the model training is completed, the output of the decoder part is the feature representation in the low-dimensional space, which serves as the fused oscillatory wave feature x3.

[0017] Further, the specific method of step 5.1 is as follows: Step 5.1.1: Feature removal: In the input oscillatory wave data, each fused feature is removed in sequence, and then the remaining features are used to reconstruct through the trained deep autoencoder; Step 5.1.2: For the oscillatory wave data obtained after each feature removal, calculate its reconstruction error and compare it with the reconstruction error of the original model; the reconstruction error formula is as follows: (15); where, represents the reconstruction output of the decoder for the i-th sample after removing the z-th feature; Step 5.1.3: Importance evaluation: According to the increase in the reconstruction error after feature removal, reflect the importance of this feature in the oscillatory wave data. The greater the increase, the more important this feature is; The specific method of step 5.3 is as follows: Step 5.3.1: Based on the pre-stored oscillatory wave database, use the extreme gradient boosting algorithm to optimize for individual differences; mix the oscillatory wave data of various population categories to train the model, and control the category bias through cross-validation to complete the pre-training of the non-invasive blood pressure prediction model; at the same time, continuously use the newly collected oscillatory wave data to enable the non-invasive blood pressure prediction model to adapt to the individual characteristics of the test object and iteratively update the model; The objective function of XGBoost consists of a loss function (mean squared error is used for regression tasks) and a regularization term, and the formula is as follows: (16); where, y i is the true blood pressure value, is the blood pressure value predicted by the model, T k is the number of leaf nodes of the k-th tree, w k is the leaf weight of the k-th tree, γ and λ are the regularization coefficients for controlling the model complexity; Use stratified K-fold cross-validation to ensure that the class distribution of each fold of data is consistent. The model evaluation formula is: (17); where, K is the number of cross-validation folds, is the k-th fold of the validation dataset; When new data When arriving, the incremental training formula is as follows: (18); Among them, K is the number of trees in the existing model, M is the number of newly added trees, f K+m ( x i ) is the predicted value of the m-th newly added tree, is the predicted value of the original model, is the predicted value of the updated model; Step 5.3.2: Non-invasive blood pressure measurement of the test subject; Take the weighted oscillatory wave feature vector X weighted as input data and input it into the pre-trained model for non-invasive blood pressure prediction, perform the model prediction operation, and output two blood pressure values of systolic blood pressure and diastolic blood pressure, thereby completing the non-invasive blood pressure measurement of the test subject; The specific method of the said Step 5.4 is as follows: Step 5.4.1: Calculate the weighted Euclidean distance between the sample to be identified and each known identity sample; Use the weighted feature vector X weighted , and calculate the weighted Euclidean distance between the sample to be identified and each known identity sample; For the sample to be identified Y and the known identity sample R, its weighted Euclidean distance is expressed as: (19); Among them, D(Y, R) is the weighted Euclidean distance between sample Y and sample R, x3(z, Y) and x3(z, R) are the values of sample Y and sample R on the z-th feature respectively, and w(z) is the weight of the z-th feature; Step 5.4.2: Calculate multiple identity recognition scores based on the weighted Euclidean distance between the sample to be identified and each known identity sample; The identity recognition score is defined as the reciprocal of the weighted Euclidean distance between the sample to be identified Y and the known identity sample R: (20); Among them, Score(Y) represents the identity recognition score of the sample to be identified Y. The higher the score, the higher the similarity between the samples; By comparing the identity recognition scores Score(Y) between the sample to be identified Y and all known identity samples, the higher the score, the closer the sample Y is to a certain known identity sample, and thus the higher the credibility of the identity recognition. Select the known identity sample with the highest identity recognition score as the final identity recognition result of the sample to be identified; At the same time, the newly collected oscillation wave data is continuously updated to the known identity samples of the member, so that the identity recognition process can adapt to the characteristics of the test subject under different physiological states and iteratively update the existing identity samples.

[0018] The beneficial effect of adopting the above technical scheme is that the family blood pressure measurement and identity recognition system and method based on oscillation wave characteristics provided by the present invention significantly improves the accuracy and practicality of family health monitoring through multi-dimensional signal analysis and intelligent feature fusion technology. The system uses a high-sensitivity pressure sensor to collect oscillation wave signals in real time, and dynamically screens high-quality data in combination with a time-frequency joint quality assessment mechanism, effectively overcoming the signal distortion problem caused by environmental noise or improper wearing of traditional equipment. In the preprocessing stage, cascade filtering and dynamic baseline correction technology are used to optimize signal quality and retain key physiological characteristics. The improved Pan-Tompkins algorithm, multi-resolution spectrum analysis and short-time Fourier transform technology extract composite feature sets including main peak amplitude, power spectrum energy ratio, etc. from the time domain, frequency domain and time-frequency domain to comprehensively characterize cardiovascular dynamic characteristics and individual differences. The core technology uses principal component analysis (PCA) dimensionality reduction and deep autoencoder unsupervised feature fusion to mine the potential structure of data and generate low-dimensional high-density features, combined with dynamic evaluation of reconstruction error and entropy weighting mechanism, so that the accuracy of identity recognition is improved by more than 30% compared with traditional solutions. The system's built-in XGBoost pre-trained model supports iterative updates and can adapt to individual differences and changes in physiological status of users to ensure long-term stability. By integrating the biometric attributes of oscillation waves, the system can automatically distinguish family members without external sensors, solving the pain point of data confusion in shared devices. The cloud storage function provides a complete data chain for chronic disease health trend analysis. Its efficient architecture adapts to the resource limitations of home devices, achieving seamless integration of medical-grade blood pressure measurement and accurate identity recognition while reducing hardware costs, providing an intelligent, integrated solution for family cardiovascular health management. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A structural block diagram of a family blood pressure measurement and identity recognition system based on oscillation wave characteristics provided by an embodiment of the present invention; Figure 2 A flow chart of a method for measuring blood pressure at home and identifying an individual based on oscillation wave characteristics provided by an embodiment of the present invention; Figure 3 A flow chart of signal acquisition and signal quality determination provided by an embodiment of the present invention; Figure 4 A flowchart for obtaining multi-index features provided by an embodiment of the present invention; Figure 5 A schematic diagram of the characteristics of an oscillation wave signal provided by an embodiment of the present invention; Figure 6Flow chart of multi - class feature data standardization processing and unsupervised feature fusion for oscillatory wave signals provided by the embodiments of the present invention; Figure 7 Flow chart of non - invasive blood pressure measurement and identity score recognition provided by the embodiments of the present invention; Figure 8 Feature map of oscillatory wave signals of a 70 - year - old male provided by the embodiments of the present invention; Figure 9 Feature map of oscillatory wave signals of a 40 - year - old female provided by the embodiments of the present invention; Figure 10 Feature map of oscillatory wave signals with elevated blood pressure in the urine - retention state of a certain subject A provided by the embodiments of the present invention; Figure 11 Feature map of oscillatory wave signals in the normal state of a certain subject A provided by the embodiments of the present invention. Specific embodiments

[0020] The following further describes in detail the specific embodiments of the present invention in conjunction with the accompanying drawings and embodiments. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0021] In this embodiment, a home blood pressure measurement and identity recognition system based on oscillatory wave features, as Figure 1 shown, includes an oscillatory wave signal acquisition and "usable" signal discrimination module, a pre - processing module, a multi - index feature extraction module, a multi - index feature standardization processing and unsupervised fusion module, and a non - invasive blood pressure measurement and identity recognition module.

[0022] The oscillatory wave signal acquisition and "usable" signal discrimination module is used to continuously and real-time collect the original oscillatory wave signals through the cuff and the pressure sensor during blood pressure measurement. According to the cuff fitting and arm movement conditions, it evaluates the influence of noise and other interferences on the signal quality, and further fuses the time-domain and frequency-domain features to evaluate the quality of the "usable" oscillatory wave signals, and screens out high-quality signals from the "usable" oscillatory wave signals. The "unusable" oscillatory wave signals are signals with distorted waveforms or completely disappeared signals. The signal quality of the "usable" oscillatory wave signals is not completely overwhelmed by noise or interference, and can be further distinguished between good and poor quality. The preprocessing module is used to preprocess the oscillatory wave signals with good quality, remove the noise in the oscillatory wave signals, and facilitate subsequent feature extraction. The multi-index feature extraction module extracts the time-domain features of the oscillatory wave signals based on the Pan-Tompkins feature point detection algorithm, extracts the frequency-domain features of the oscillatory wave signals based on the spectrum analysis method, and extracts the time-frequency features of the oscillatory wave signals based on the short-time Fourier transform method. The multi-index feature standardization processing and unsupervised fusion module is responsible for the standardization task of the oscillatory wave signal feature data. First, perform standardization operations on the oscillatory wave signal feature data, and then use PCA technology to perform dimensionality reduction processing on the standardized data. Then, further feature learning and fusion are carried out on the features after PCA dimensionality reduction with the help of a deep autoencoder. The non-invasive blood pressure measurement and identity recognition module is based on the fused features extracted by the deep autoencoder. On the one hand, it evaluates the importance of features based on the contribution of reconstruction error; on the other hand, it performs weighted fusion on the multi-index features according to the feature weight allocation principle. On this basis, a pre-trained model for non-invasive blood pressure prediction is used to achieve non-invasive blood pressure measurement, and identity recognition is finally completed by matching with a known identity template library.

[0023] In this embodiment, the oscillatory wave signal acquisition and "usable" signal discrimination module discriminates the acquired original oscillatory wave signal as "usable" and "unusable" signals based on the cuff fitting and the movement of the arm. The oscillatory wave signal is acquired by means of a sphygmomanometer cuff and an internal pressure sensor. The "unusable" oscillatory wave signal is usually caused by interference generated by other reasons such as too tight or insecure cuff fitting, movement or exertion of the user's arm during the measurement process, resulting in serious distortion of the oscillatory wave waveform or even complete disappearance of the signal. Since the "unusable" signal has lost the effective information and is difficult to restore, it must be acquired again. The "usable" signal indicates that its signal quality has not been completely damaged by noise or interference, and further distinction between good and bad signal quality can be made subsequently. Time-frequency analysis is performed on the "usable" oscillatory wave signal to evaluate the signal quality. First, time-domain analysis is performed on the usable signal. The baseline drift characteristics are extracted by variational mode decomposition, and the amplitude fluctuation coefficient, baseline offset standard deviation and pulse rate consistency index are calculated in combination with the period segmentation technique. At the same time, in the frequency domain, the frequency band is equally divided into 6 sub-bands, and the proportion of the power spectral density of each sub-band is calculated to construct a frequency-domain feature vector. Based on the information entropy theory, the weight coefficients of the time-frequency domain indexes are determined, and a comprehensive quality index (SQI) is generated through weighted linear combination. When the SQI exceeds the preset threshold, it is determined as a high-quality signal and enters the subsequent processing, and the signal that does not meet the standard triggers a re-acquisition instruction. This scheme provides a reliable data basis for subsequent feature extraction and identity recognition by quantitatively analyzing the signal stability, noise level and physiological feature integrity.

[0024] In this embodiment, the preprocessing module performs filtering and denoising preprocessing on the original oscillatory wave signal with good quality to eliminate the noise in the oscillatory wave signal and improve the signal quality. First, digital filtering technology is applied. High-frequency noise is filtered out by a low-pass filter and burst interference is eliminated in combination with median filtering. Subsequently, a moving average filter is used in combination with a high-pass filter to construct a dynamic baseline correction system, effectively eliminating the baseline drift and low-frequency noise caused by respiratory movement. Finally, selective downsampling is implemented through a multi-stage decimation algorithm to reduce the data volume while retaining the main feature information. This preprocessing scheme effectively improves the signal quality and system processing efficiency while ensuring the integrity of the signal features through time-frequency domain joint filtering, baseline correction and intelligent downsampling techniques.

[0025] In this embodiment, the multi-index feature extraction module extracts time-domain, frequency-domain, and time-frequency features from the preprocessed oscillatory wave signal respectively to obtain multi-class index features of the oscillatory wave signal. First, the improved Pan-Tompkins algorithm is used to detect the starting point of the waveform period, and the preprocessed oscillatory wave signal is separated period by period to extract 14 time-domain profile features including the main peak amplitude, notch depth, rising slope, etc. Subsequently, through the multi-resolution spectrum analysis technology, the effective frequency band is equally divided into 6 sub-bands, the ratio of the power spectral density of each sub-band to the total spectral energy is calculated, and a frequency-domain feature vector is constructed. Finally, the short-time Fourier transform is used to generate a time-frequency matrix, the core region with an energy ratio of more than 90% is located by dynamic threshold segmentation, the frequency axis range feature is extracted, and 8 local extreme points with the largest amplitudes in the time-frequency diagram are located based on the energy gradient detection algorithm to form a time-frequency joint feature vector. This feature extraction system comprehensively characterizes the morphological features, frequency distribution, and time-varying characteristics of the signal through multi-dimensional parameter combinations, providing a rich data basis for subsequent processing.

[0026] In this embodiment, the multi-index feature standardization and unsupervised fusion module is responsible for standardizing the multi-index feature data of the oscillatory wave signal and performing unsupervised feature fusion. For the standardization of the multi-index features of the oscillatory wave signal, the core objective is to transform all features into the same scale range. The significance of this is to ensure that the contribution degrees of each feature to the objective function are roughly equivalent, prevent the situation where some features dominate during the parameter update process, so that the model can treat each feature more fairly. To reduce the data complexity while retaining as much original information as possible, principal component analysis (PCA) is used to perform dimensionality reduction on the standardized oscillatory wave signal feature data. The specific steps of PCA dimensionality reduction are as follows: First, based on the standardized feature data, calculate the covariance matrix, which can reflect the correlation between features. Then, perform eigenvalue decomposition on the covariance matrix to obtain a series of eigenvalues and their corresponding eigenvectors. Next, screen according to the magnitudes of the eigenvalues, and retain the eigenvectors corresponding to N eigenvalues whose contributions exceed 90%, and determine them as the principal components. These principal components can represent the main variation information of the original data. Finally, project the original data onto the selected principal components to obtain the dimensionality-reduced feature data. After completing the PCA dimensionality reduction, a deep autoencoder is used to perform more in-depth feature learning and fusion on the dimensionality-reduced features. The deep autoencoder consists of two parts: an encoder and a decoder. The role of the encoder is to further compress the input features into a low-dimensional space to extract the core features of the data; the decoder then attempts to reconstruct the original features from this low-dimensional space to verify the effectiveness of the features extracted by the encoder. Take the oscillatory wave feature data after PCA dimensionality reduction as the input of the deep autoencoder, and use the mean square error as the loss function to train the model, aiming to minimize the difference between the input oscillatory wave features and the reconstructed features; after completing the model training, the output of the decoder part finally, that is, the feature representation in the low-dimensional space, is used as the fused feature to provide more representative features for subsequent analysis and applications.

[0027] In this embodiment, the non-invasive blood pressure measurement and identity recognition module performs feature weighted fusion and identity score recognition. Based on the fusion features extracted from the deep autoencoder, the feature importance is evaluated based on the contribution of the reconstruction error, that is, analyzing the change in the autoencoder reconstruction error after each fusion feature is removed. The greater the contribution of the feature to the reconstruction error, the more important the feature is for the representation of the data. For the data after each feature removal, calculate the reconstruction error and compare it with the reconstruction error of the original model (when no feature is removed). The increase in the reconstruction error after feature removal reflects the importance of the feature. Features with a significant increase in the reconstruction error are considered more important for the model and data representation. Then, divide the importance score of each feature by the sum of the importance scores of all features to obtain the weight of the feature. Based on the weighted feature matrix, input it into the pre-trained blood pressure prediction model obtained by applying the pre-stored oscillatory wave database to output the blood pressure value of the test subject. Apply the calculated weight to each feature vector, and the weighted feature vector serves as the feature template of the currently detected object. Use the Euclidean distance metric algorithm for identity recognition in the pre-stored oscillatory wave feature database and finally output. By comparing the identity recognition scores between the sample to be recognized and all known identity samples, effectively utilize the information in the weighted feature vector, and select the known identity sample with the highest score as the recognition result of the sample to be recognized.

[0028] In this embodiment, a method for home blood pressure measurement and identity recognition based on oscillatory wave features is implemented through the above-mentioned home blood pressure measurement and identity recognition system based on oscillatory wave features, as Figure 2 shown, and includes the following steps: Step 1: Obtain the original oscillatory wave signal of the sample to be recognized, and based on the cuff fitting and arm movement conditions, distinguish between "usable" and "unusable" signals, and evaluate the quality of the "usable" oscillatory wave signals.

[0029] Step 1.1: Detect errors in the collected oscillatory wave signals to determine whether the oscillatory wave signals collected by the cuff and the sensor are "usable".

[0030] "Unusable" oscillatory wave signals are usually interference caused by factors such as too tight or insecure cuff fitting, movement or force of the user's arm during measurement, or even complete signal loss. Given that the "unusable" signals have lost valid information and are difficult to restore, they must be collected again. "Usable" signals indicate that their signal quality has not been completely damaged by noise or interference, and further distinction between good and bad quality can be made for them subsequently.

[0031] Step 1.2: For the "usable" oscillatory wave signals, comprehensively consider the indicators in both the time domain and the frequency domain to evaluate the signal quality, as Figure 3 shown.

[0032] Step 1.2.1: Conduct time-domain analysis on the oscillatory wave signal, and use the amplitude change degree index, baseline drift degree index, and pulse rate consistency degree index as the signal quality evaluation indexes in the time domain.

[0033] Perform VMD (Variational Mode Decomposition) decomposition and period segmentation on the oscillatory wave signal pulse respectively to obtain the reconstructed signal bPulse and single-period signal Pulse that characterize the baseline drift of the signal. i ; The amplitude change degree index is obtained through the amplitude difference of the signal waveform in multiple cardiac cycles; the amplitude change degree index a(i) within the i-th cycle of the oscillatory wave signal is as shown in the following formula: a(i)=|max(Pulse i )-min(Pulse i ) | (1); Among them, max(Pulse i ) and min(Pulse i ) are respectively the maximum and minimum values of the waveform amplitude of the i-th cycle of the oscillatory wave signal after filtering processing.

[0034] The baseline drift degree index is obtained by windowing the reconstructed signal.

[0035] Perform windowing on the reconstructed signal bPulse to obtain the baseline drift degree index b(i) within the i-th cycle of the oscillatory wave signal, as shown in the following formula: b(i)= max(|bPulse i |)-b (2); Among them, bPulse i is the reconstructed signal of the i-th cycle of the oscillatory wave signal, and b is the baseline standard value of the oscillatory wave signal in the stationary state.

[0036] The pulse rate consistency degree index is calculated by statistically analyzing the time length of the single-period signal after period segmentation and calculating the consistency of the signal time length within multiple cycles.

[0037] Set the period length of the oscillatory wave signal as length(Pulse i ) to obtain the pulse rate consistency degree index r(i) within the i-th cycle of the oscillatory wave signal, as shown in the following formula: (3); Among them, SD(length(Pulse i )) is the standard deviation of the calculated i-th signal cycle length, and Mean(length(Pulse i)) is the mean value of the calculated length of the i-th signal period.

[0038] Step 1.2.2: Perform frequency-domain analysis on the oscillatory wave signal. Select the effective frequency band of the oscillatory wave signal through fast Fourier transform, calculate the power spectral density as the frequency component index, and evaluate the quality of the oscillatory wave signal.

[0039] The frequency component index f(i) within the i-th period of the oscillatory wave signal is: (4); where P is the power of the oscillatory wave signal and f represents frequency.

[0040] Step 1.2.3: Based on the signal quality evaluation indicators in the time domain and frequency domain, use the information entropy theory to calculate the weight coefficients of each index in the time-frequency domain, and perform multi-index weighted fusion of the oscillatory wave signal in the time-frequency domain to obtain the fusion index SQI(i), as shown in the following formula: (5); where w a (i), w b (i), w r (i), w f (i) are the entropy weight values of the evaluation indicators a(i), b(i), r(i), and f(i) respectively.

[0041] Step 1.2.4: Perform threshold discrimination based on the fusion index, and screen out high-quality signals from the "usable" oscillatory wave signals.

[0042] When the fusion index SQI(i) exceeds the preset threshold, it is determined as a "high-quality signal" and enters the subsequent processing flow. Signals that do not meet the standard will be marked as "sub-optimal signals", and the system will prompt for re-measurement.

[0043] Step 2: Preprocess the oscillatory wave signals with good quality, including noise removal, filtering, and downsampling operations to remove the noise in the oscillatory wave signals.

[0044] During the acquisition process, the oscillatory wave signals are subject to various noise interferences, including environmental electromagnetic interference, artifacts generated by human movement, physiological baseline drift, and electronic noise of the sensor itself. These noise components will significantly reduce the signal fidelity and analysis reliability. Based on this, this embodiment designs a multi-stage preprocessing process: first, eliminate high-frequency noise and pulse interference through a composite filtering algorithm, then use dynamic baseline correction technology to compensate for the low-frequency drift caused by respiratory movement, and finally retain the key feature information through an adaptive downsampling strategy. This preprocessing scheme uses time-frequency domain joint noise reduction technology to suppress noise interference while maximizing the retention of the physiological characteristics of the signal, providing a high-quality data basis for subsequent feature extraction and pattern recognition.

[0045] Step 2.1: Adopt cascaded digital filtering technology. First, filter out high-frequency noise through an 8th-order Butterworth low-pass filter (cutoff frequency 15 Hz), and then apply a 3-point non-linear median filtering algorithm to eliminate burst pulse interference, forming a preliminarily noise-reduced signal sequence to make subsequent processing more accurate and reliable.

[0046] Step 2.2: Use a 5-point moving window mean filter combined with a second-order high-pass filter (cutoff frequency 0.1 Hz) to construct a dynamic baseline correction system. While retaining the characteristic information of the pulse wave, this system effectively eliminates baseline drift and low-frequency noise interference caused by respiratory movement and outputs a smooth signal waveform.

[0047] Step 2.3: Implement selective downsampling processing. First, restore the continuity of the original signal through cubic spline interpolation technology, and then adopt a multi-stage decimation algorithm to reduce the sampling rate from 200 Hz to 50 Hz while retaining more than 95% of the effective characteristic information, significantly reducing the data volume of subsequent processing.

[0048] Step 3: Extract time-domain, frequency-domain, and time-frequency characteristics from the preprocessed oscillatory wave signals respectively to obtain multi-index characteristics of the oscillatory wave signals, such as Figure 4 shown.

[0049] Step 3.1: Use an improved Pan-Tompkins algorithm to construct a period recognition system. Locate the waveform starting point through adaptive threshold detection technology to achieve per-period segmentation processing. For a single-period waveform, extract relevant contour characteristic parameters.

[0050] The waveform of the oscillatory wave signal mainly contains two types of waveforms, namely the pre-peak and the late-peak. Its waveform morphological parameters (including amplitude height, rising slope, and time interval) contain rich physiological information. These parameters reflect the rhythm characteristics, vascular elasticity, and hemodynamic state of the cardiovascular system, providing important bases for clinical diagnosis. Specifically, the pre-peak amplitude is directly related to the heart's ejection ability, the morphological changes of the dicrotic wave can indicate changes in vascular compliance, and the combination of various time parameters can comprehensively evaluate the heart's diastolic function and vascular resistance. The effective extraction of this physiological characteristic information has important clinical value for the early screening and dynamic monitoring of cardiovascular diseases such as hypertension.

[0051] For single-cycle waveforms, the following contour feature parameters are extracted: the vertical distance between the main peak vertex and the baseline, the main peak amplitude H, the vertical distance between the late peak vertex and the baseline, the main peak amplitude H1, the maximum slope value of the waveform rising edge, the maximum gradient K, the vertical distance between the dicrotic notch point and the baseline, the notch depth D, time parameters: the duration from the main peak to the starting point T1, the duration from the main peak to the dicrotic notch point T2, the duration from the dicrotic notch point to the late peak T3, the duration from the late peak to the end point T4, the full cycle duration T, area parameters: the waveform area A1 in the T1 period, the waveform area A2 in the T2 period, the waveform area A3 in the T3 period, the waveform area A4 in the T4 period, the full cycle area A, as Figure 5 shown.

[0052] Step 3.2: Construct a multi-resolution spectrum analysis system based on the spectrum analysis method, and use the spectral energy ratio of different sub-band frequencies as the frequency-domain feature of the oscillatory wave signal.

[0053] The frequency-domain feature analysis of the oscillatory wave signal can effectively reflect the complexity and individual differences of the dynamic regulation of the cardiovascular system by revealing the distribution law of signal energy in different frequency bands. These characteristic parameters not only reflect the frequency response characteristics of the cardiac pumping function, but also can sensitively capture pathological states such as vascular elasticity changes and abnormal blood flow resistance. The frequency-domain feature extraction method is based on the power spectrum analysis of the oscillatory wave signal, and realizes identity recognition by exploring the specificity of the spectral energy distribution among different individuals. First, the effective information concentrated frequency band is equally divided into multiple sub-band intervals, and the method combining downsampling and Fourier transform is used to calculate the power spectral density of each sub-band. Through normalization processing, spectral energy ratio characteristic parameters are generated, and these parameters reflect the distribution pattern of signal energy in different frequency intervals. Although in some common analyses, the frequency band of low-frequency signals may be focused on, in fact, meaningful features may also exist in the oscillatory wave at higher frequencies. There may be information related to certain physiological processes of the cardiovascular system at higher frequencies, and these high-frequency features may be related to aspects such as the elasticity of blood vessels, the dynamic characteristics of blood flow, and the function of the heart. The study of the high-frequency features of the oscillatory wave helps to more comprehensively understand the physiological state of the cardiovascular system and provide more information for the diagnosis and monitoring of related diseases. Therefore, the frequency band containing certain high-frequency signal bands in the effective information concentration is used for frequency-domain feature extraction.

[0054] For the spectrum analysis of the oscillatory wave signal, in order to prevent the omission of effective information, the method of sub-band division is used to divide the frequency band; the calculation formula for the power spectral energy E in the c-d frequency band range is: (6); where P (Pulse) is the power spectral density of the oscillatory wave signal in the c-d frequency band, as shown in the following formula: (7); Among them, FFT(Pulse) is the spectrum obtained by applying FFT to the oscillatory wave signal, and fs is the sampling frequency of the oscillatory wave signal.

[0055] Combined with the main information distribution characteristics of the oscillatory wave signal, the multi-spectrum energy of the oscillatory wave signal is analyzed multiple times. At the same time, in order to prevent the omission of effective information, the frequency band where the effective information is concentrated is equally divided into six sub-bands, covering the continuous frequency intervals of each frequency band in turn. Through equal ratio division, the low-frequency band information related to blood pressure feature recognition can be more effectively extracted, and at the same time, the high-frequency band signals with more noise but also containing hemodynamic characteristics can be covered. The system calculates the power spectrum energy value of each frequency band by the method described in Equation (6), and generates the power spectrum energy ratio (PowerSpectrum Energy Ratio, PSER) characteristic parameter through normalization processing. This scheme ensures the effective suppression of high-frequency noise interference while retaining key physiological information by dynamically adjusting the spectrum analysis window and multi-stage energy calculation.

[0056] PSER m =E m / E all (8); Among them, m represents the number of sub-band frequency bands, E m is the power spectrum energy value of the m-th sub-band frequency band, and E all is the total energy of all frequency bands.

[0057] Step 3.3: Extract the time-frequency diagram features of the oscillatory wave signal based on the short-time Fourier transform.

[0058] The time-frequency joint analysis technology realizes an in-depth analysis of the complex patterns of the dynamic evolution of the oscillatory wave by constructing a two-dimensional time-frequency distribution map of the signal. By converting the one-dimensional time series into a time-frequency matrix through the short-time Fourier transform, the frequency component changes at different time points can be captured. This multi-dimensional feature analysis not only reveals the dynamic characteristics of cardiovascular system regulation but also can capture subtle changes that are difficult to detect by traditional time-domain or frequency-domain analysis. By extracting the energy distribution profile and local extreme point features in the time-frequency diagram, the system can construct a spatio-temporal feature map reflecting the cardiovascular function state, providing more comprehensive information support for early disease warning and personalized diagnosis and treatment.

[0059] Perform short-time Fourier transform (Short-timeFourier Transform, STFT) on the oscillatory wave signal Pulse(t) using a sliding window w(t) to obtain the time-frequency distribution Figure X (t,f) of the oscillatory wave signal, where t represents time.

[0060] Detect the time-frequency distribution Figure XThe main energy distribution region on (t, f), extract the edge contour of this region, and represent it with a closed curve L(t, f).

[0061] Calculate the maximum value f of the closed curve L(t, f) on the frequency f axis max and the minimum value f min , and obtain the contour height range feature h of the oscillatory wave signal: h = f max - f min (9); According to the contour height of the oscillatory wave signal, locate the local maximum points in the time-frequency distribution Figure X (t, f) based on the energy gradient detection algorithm, select the N extreme values with the largest amplitudes, and form a feature sequence with the time and frequency coordinates of these points to form the time-frequency local maximum point feature M of the oscillatory wave signal: (10); where, S(t n , f n ) is the local maximum point, n = 1, 2,..., N.

[0062] Step 4: Perform standardization processing and unsupervised feature fusion on the time domain, frequency domain, and time-frequency features of the oscillatory wave signal obtained in Step 3, as Figure 6 shown.

[0063] Step 4.1: Carry out standardization work on the feature data contained in the oscillatory wave signal to eliminate the differences in dimensions and magnitudes between different features, and provide a unified standard data basis for subsequent analysis.

[0064] The oscillatory wave signal covers multiple types of features, which have large differences in dimensions and numerical ranges. If not processed, during the model training process, the features with large numerical ranges are very likely to dominate the training direction, resulting in unbalanced model training. The standardization operation can map all features to the same scale space, effectively eliminating the interference caused by dimensional differences. This measure ensures that the influence of each feature on the objective function is similar, avoiding the excessive dominance of some features during the parameter update stage. Moreover, standardization is also very beneficial for improving the stability and generalization ability of the model. It alleviates the risk of overfitting during the model training process by standardizing the data distribution, enabling the model to also show good prediction and adaptation capabilities when facing new data that has not participated in training.

[0065] Perform the following standardization processing on the time domain, frequency domain, and time-frequency features of the oscillatory wave signal extracted in Step 3: (11); Among them, x is the original eigenvalue of the oscillation wave signal, μ is the mean value of the feature, σ1 is the standard deviation of the feature, and x1 is the standardized eigenvalue.

[0066] Step 4.2: Use the principal component analysis method to reduce the dimension of the oscillation wave signal feature data after standardization, reduce data redundancy, and extract key information.

[0067] Data dimensionality reduction plays a key role in processing high-dimensional data. It can significantly reduce the number of features and reduce the dimension of the feature space. Through data dimensionality reduction, features containing rich information can be effectively identified and retained, and noise and redundant features mixed in the data can be eliminated, so that the model can understand the data more accurately and significantly improve its prediction performance. When processing the standardized oscillation wave signal feature data, PCA is often used for dimensionality reduction. The core goal of this method is to retain the key information of the original data to the greatest extent while reducing the feature dimension, and to ensure that the main features of the data are not lost during the dimensionality reduction process.

[0068] Step 4.2.1: Calculate the covariance matrix: Calculate the covariance matrix of the standardized blood pressure oscillation wave signal feature data to analyze the correlation between the features.

[0069] Step 4.2.2: Eigenvalue decomposition: Perform eigenvalue decomposition on the obtained covariance matrix to obtain a series of eigenvalues ​​and their corresponding eigenvectors, which provide a basis for the subsequent principal component selection.

[0070] Step 4.2.3: Select principal components: Based on the size of the eigenvalues, retain the eigenvectors corresponding to the Q eigenvalues ​​that contribute more than 90%, determine them as principal components, and focus on key information.

[0071] Step 4.2.4: Data projection: Project the original oscillation wave signal characteristic data onto the selected principal component, and then obtain the oscillation wave signal characteristic data x2 after dimensionality reduction processing.

[0072] Step 4.3: Use deep autoencoders to perform deeper feature learning and fusion on the oscillation wave feature data after dimensionality reduction, achieve unsupervised feature fusion of oscillation wave signals, and explore the potential structure of the data.

[0073] After completing PCA dimensionality reduction, a deep autoencoder is used to conduct more in-depth feature learning and fusion on the dimensionality-reduced features. The working mechanism of the deep autoencoder is to capture the internal structure of the data by exploring an efficient representation of the input data. During operation, the hidden layer of the autoencoder transforms the input data to extract a more compact data representation. This representation can accurately reflect the internal characteristics of the oscillatory wave signal, helping the model better extract the key features of the oscillatory wave and eliminate the interference of redundant information. In subsequent learning tasks, since the model has mastered more representative features, it can analyze and predict more efficiently, thus significantly improving the performance of task completion.

[0074] Step 4.3.1: Construct a deep autoencoder: The constructed deep autoencoder consists of two parts: an encoder and a decoder. The role of the encoder is to further compress the input oscillatory wave features into a low-dimensional space, and the decoder attempts to reconstruct the original oscillatory wave features from this low-dimensional space.

[0075] Map the input feature x2 to the low-dimensional latent space z: (12); Among them, and are the weight matrices of the encoder, which are the parameters of the hidden layer; and represent the bias terms of the encoder, σ is the activation function.

[0076] Reconstruct the output from the latent space z : (13); Among them, and are the weight matrices of the decoder, and are the bias terms of the decoder.

[0077] Step 4.3.2: Model training: Use the oscillatory wave feature data x2 after PCA dimensionality reduction as the input of the deep autoencoder, and adopt the Mean-Square Error (MSE) as the loss function to train the model, aiming to minimize the difference between the input oscillatory wave features and the reconstructed features; The MSE calculation formula is: (14); Among them, n is the number of samples, represents the i-th original oscillatory wave feature data after PCA dimensionality reduction, is the i-th oscillatory wave feature data reconstructed by the deep autoencoder.

[0078] Step 4.3.3: Feature fusion representation: After completing the model training, the output of the decoder part, which is the feature representation in the low-dimensional space, is used as the fused oscillatory wave feature x3.

[0079] Step 5: Perform feature weighted fusion to achieve non-invasive blood pressure measurement and identity score recognition, as Figure 7 shown.

[0080] Based on the fused feature x3 extracted from the deep autoencoder in Step 4, these features already contain the key information of the original data, but have a lower dimension and more concentrated information.

[0081] Step 5.1: Evaluate the importance of the oscillatory wave features based on the contribution of the reconstruction error, and clarify the role of each feature in the whole; When evaluating the importance of the fused features, the specific approach is to analyze the change in the reconstruction error of the deep autoencoder when each fused feature is removed. In this process, if the removal of a certain fused feature leads to a significant increase in the reconstruction error of the deep autoencoder, it means that this feature plays a crucial role in the accurate representation of the data. In this way, the importance of the fused features can be effectively evaluated based on the change in the reconstruction error.

[0082] Step 5.1.1: Feature removal: In the input oscillatory wave data, each fused feature is removed in turn, and then the remaining features are used to reconstruct through the trained deep autoencoder. In this way, the influence of each feature on the reconstruction result can be observed.

[0083] Step 5.1.2: For the oscillatory wave data obtained after each feature removal, calculate its reconstruction error according to Equation (15) and compare it with the reconstruction error of the original model. The change in the reconstruction error can reflect the importance of this feature in the data.

[0084] (15); where represents the reconstruction output of the decoder for the i-th sample after removing the z-th feature.

[0085] Step 5.1.3: Importance evaluation: According to the increase in the reconstruction error after feature removal, reflect the importance of this feature in the oscillatory wave data. The greater the increase, the more important the feature.

[0086] Step 5.2: Perform weighted fusion on the oscillatory wave features according to the results of the feature importance evaluation.

[0087] Step 5.2.1: Feature weight assignment. The weight of a feature is positively correlated with the importance of the feature. Divide the importance score of each oscillatory wave feature by the sum of the importance scores of all features to obtain the weight of the feature, which can ensure that important features receive more attention during the fusion process. The fused feature vector is x3 = [x3(1), x3(2), …, x3(k)], and its corresponding weight is w = [w(1), w(2), …, w(k)].

[0088] Step 5.2.2: Feature weighted fusion based on weight assignment. Apply the calculated weight to each feature vector to obtain the weighted feature vector X weighted = [x3(1)w(1), x3(2)w(2), …, x3(k)w(k)].

[0089] Step 5.3: Non-invasive blood pressure measurement based on multi-index feature weighted fusion. Based on the weighted feature vector X obtained in the above Step 5.2 weighted , input it into the pre-trained model for non-invasive blood pressure prediction to achieve non-invasive blood pressure measurement of the test subject.

[0090] Step 5.3.1: Based on the pre-stored oscillatory wave database, use the Extreme Gradient Boosting (XGBoost) algorithm to optimize for individual differences. Mix the oscillatory wave data of various population categories to train the model, and control the class bias through cross-validation to complete the pre-training of the non-invasive blood pressure prediction model. At the same time, continuously use the newly collected oscillatory wave data to enable the non-invasive blood pressure prediction model to adapt to the individual characteristics of the test object and iteratively update the model. The objective function of XGBoost consists of a loss function (mean squared error is used for regression tasks) and a regularization term, and the formula is as follows: (16); where, y i is the true blood pressure value, is the blood pressure value predicted by the model, T k is the number of leaf nodes of the k-th tree, w k is the leaf weight of the k-th tree, γ and λ are the regularization coefficients for controlling the model complexity.

[0091] Use stratified K-fold cross-validation to ensure that the class distribution of each fold of data is consistent. The model evaluation formula is: (17); where, K is the number of cross-validation folds, is the k-th fold validation data set.

[0092] When new data arrives, the incremental training formula is: (18); where K is the number of trees in the existing model, M is the number of newly added trees, f K+m ( x i ) is the predicted value of the m-th newly added tree, is the predicted value of the original model, is the predicted value of the updated model.

[0093] Step 5.3.2: Non-invasive blood pressure measurement of the subject. The weighted oscillatory wave feature vector X weighted is used as the input data and input into the pre-trained model for non-invasive blood pressure prediction. The model prediction operation is performed, and two blood pressure values, systolic blood pressure and diastolic blood pressure, are output, thus completing the non-invasive blood pressure measurement of the subject.

[0094] Step 5.4: Identity score recognition based on multi-index feature weighted fusion. The weighted feature vector X weighted obtained in the above Step 5.2 is used as the feature template of the currently detected object; the Euclidean distance metric algorithm is adopted to perform identity recognition in the pre-stored oscillatory wave feature database and finally output the result.

[0095] Step 5.4.1: Calculate the weighted Euclidean distance between the sample to be recognized and each known identity sample. Using the weighted feature vector X weighted , calculate the weighted Euclidean distance between the sample to be recognized and each known identity sample. For the sample to be recognized Y and the known identity sample R, its weighted Euclidean distance is expressed as: (19); where D(Y,R) is the weighted Euclidean distance between sample Y and sample R, x3(z,Y) and x3(z,R) are the values of sample Y and sample R on the z-th feature respectively, and w(z) is the weight of the z-th feature.

[0096] Step 5.4.2: Calculate multiple identity recognition scores based on the weighted Euclidean distance between the sample to be recognized and each known identity sample. The identity recognition score is defined as the reciprocal of the weighted Euclidean distance between the sample to be recognized Y and the known identity sample R: (20); where Score(Y) represents the identity recognition score of the sample to be recognized Y. The higher the score, the higher the similarity between the samples.

[0097] By comparing the identity recognition score Score(Y) between the sample Y to be recognized and all known identity samples, the higher the score, the closer the sample Y is to a certain known identity sample, and thus the higher the credibility of the identity recognition. Effectively utilize the information in the weighted feature vector and select the known identity sample with the highest identity recognition score as the final identity recognition result of the sample to be recognized.

[0098] Meanwhile, continuously update the known identity samples of this member with the newly collected oscillatory wave data, so that the identity recognition process can adapt to the characteristics of the test object under different physiological states and iteratively update the existing identity samples. The characteristics of the oscillatory wave signals under different physiological states will also be different. For example, in a tense state represented by holding urine, by continuously using the newly collected oscillatory wave data to iteratively update the model, the accuracy and adaptability of identity recognition can be greatly improved. Considering that the overall heterogeneity of oscillatory waves in a large population is low, but they have high specificity within a small sample size and small group, especially for individuals of different ages and genders, as Figure 8 , Figure 9 shown, even different physiological states of the same individual will also cause changes in the characteristics of oscillatory waves, as Figure 10 , Figure 11 shown. Therefore, oscillatory waves are suitable for identity recognition and health monitoring within family members.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.

Claims

1. A home blood pressure measurement and identity recognition system based on the characteristics of oscillatory waves, characterized in that: It includes an oscillatory wave signal acquisition and "usable" signal discrimination module, a preprocessing module, a multi-index feature extraction module, a multi-index feature normalization processing and unsupervised fusion module, and a non-invasive blood pressure measurement and identity recognition module; The oscillatory wave signal acquisition and "usable" signal discrimination module is used to continuously and real-time collect the original oscillatory wave signals through a cuff and a pressure sensor during blood pressure measurement. According to the cuff fitting and arm movement conditions, it evaluates the influence of various interferences on the signal quality, and further fuses the time-domain and frequency-domain features to evaluate the quality of the "usable" oscillatory wave signals, and screens out high-quality signals from the "usable" oscillatory wave signals; the "unusable" oscillatory wave signals are signals with distorted waveforms or completely disappeared waveforms; the signal quality of the "usable" oscillatory wave signals is not completely obscured by noise or interference, and the distinction between good and poor quality can be further made; The preprocessing module is used to preprocess the oscillatory wave signals with good quality, remove the noise in the oscillatory wave signals, and facilitate subsequent feature extraction; The multi-index feature extraction module extracts the time-domain features of the oscillatory wave signals based on the Pan-Tompkins feature point detection algorithm, extracts the frequency-domain features of the oscillatory wave signals based on the spectrum analysis method, and extracts the time-frequency features of the oscillatory wave signals based on the short-time Fourier transform method; In the extraction of the time-domain features of the oscillatory wave signals, an improved Pan-Tompkins algorithm is used for period recognition, and the starting point of the waveform is located through the adaptive threshold detection technology to achieve per-period segmentation processing; for a single-period waveform, the following contour feature parameters are extracted: the vertical distance between the main peak vertex and the baseline main peak amplitude H, the vertical distance between the late peak vertex and the baseline main peak amplitude H1, the maximum slope value of the waveform rising edge maximum slope K, the vertical distance between the dicrotic notch point and the baseline notch depth D, time parameters: the duration from the main peak to the starting point T1, the duration from the main peak to the dicrotic notch point T2, the duration from the dicrotic notch point to the late peak T3, the duration from the late peak to the end point T4, the complete cycle duration T, area parameters: the waveform area A1 in the T1 period, the waveform area A2 in the T2 period, the waveform area A3 in the T3 period, the waveform area A4 in the T4 period, the complete cycle area A number; In the extraction of the frequency-domain features of the oscillatory wave signals, first, the method of sub-band division is used to equally divide the effective information concentrated frequency band into multiple sub-band intervals, and the method of combining downsampling and Fourier transform is used to calculate the power spectral density of each sub-band; the calculation formula for the power spectral energy E in the c-d frequency band range is: （6）； Among them, P(Pulse) is the power spectral density of the oscillatory wave signal in the c-d frequency band, as shown in the following formula: （7）； Among them, FFT(Pulse) is the spectrum obtained by the oscillatory wave signal using FFT, and fs is the sampling frequency of the oscillatory wave signal; combined with the main information distribution characteristics of the oscillatory wave signal, the multi-spectrum energy of the oscillatory wave signal is analyzed multiple times, and the power spectral energy ratio feature parameters are generated through normalization processing: PSER m =E m / E all (8); where m represents the number of sub-band frequencies, and E m is the power spectral energy value of the m-th sub-band frequency, and E all is the total energy of all frequencies; the spectral energy ratio of different sub-band frequencies is used as the frequency domain feature of the oscillatory wave signal; The multi-index feature standardization and unsupervised fusion module is responsible for standardizing the oscillatory wave signal feature data. First, perform a standardization operation on the oscillatory wave signal feature data to eliminate the differences in dimension and magnitude between different features. Subsequently, use PCA technology to perform dimensionality reduction on the standardized oscillatory wave signal feature data. Then, with the help of a deep autoencoder, further feature learning and fusion are carried out on the oscillatory wave feature data after PCA dimensionality reduction to achieve unsupervised feature fusion of the oscillatory wave signal. The non-invasive blood pressure measurement and identity recognition module is based on the fusion features extracted by the deep autoencoder. On the one hand, it evaluates the importance of features based on the contribution of reconstruction error, that is, analyzes the change in the reconstruction error of the autoencoder after each fusion feature is removed. The greater the contribution of the feature to the reconstruction error, the more important the feature is for the representation of the data. Calculate the reconstruction error for the data after each feature removal and compare it with the reconstruction error of the original model when no feature is removed. The increase in the reconstruction error after feature removal reflects the importance of the feature. Features with a significant increase in reconstruction error are considered more important for the model and data representation. On the other hand, according to the results of feature importance evaluation, weighted fusion of oscillatory wave features is carried out. First, feature weight allocation is performed. The weight size of a feature is positively correlated with the importance of the feature. Divide the importance score of each oscillatory wave feature by the sum of the importance scores of all features to obtain the weight of this feature. Apply the calculated weight to each oscillatory wave feature vector to obtain the weighted oscillatory wave feature vector. On this basis, use the pre-trained model for non-invasive blood pressure prediction to achieve non-invasive blood pressure measurement, and use the obtained weighted oscillatory wave feature vector as the feature template of the current detected object. Adopt the Euclidean distance metric algorithm to perform identity recognition in the pre-stored oscillatory wave feature database, that is, match it with the known identity template library, and finally output the result to complete identity recognition. The implementation process of the non-invasive blood pressure measurement by the non-invasive blood pressure measurement and identity recognition module is as follows: Based on the pre-stored oscillatory wave database, use the extreme gradient boosting algorithm to optimize for individual differences. The objective function of XGBoost consists of a loss function and a regularization term, and the formula is as follows: （16）； Among them, y i is the true blood pressure value, is the blood pressure value predicted by the model, T k is the number of leaf nodes of the k-th tree, w k is the leaf weight of the k-th tree, γ and λ are regularization coefficients for controlling the model complexity; Mix the oscillatory wave data of various categories of people to train the model, and control the category bias through cross-validation to complete the pre-training of the non-invasive blood pressure prediction model. Use stratified K-fold cross-validation to ensure that the category distribution of each fold of data is consistent. The model evaluation formula is: （17）； where K is the number of cross-validation folds, is the k-th fold validation dataset; Meanwhile, continuously use the newly collected oscillometric wave data to enable the non-invasive blood pressure prediction model to adapt to the individual characteristics of the test subject and iteratively update the model; when the new data arrives, the incremental training formula is: （18）； Among them, K is the number of trees in the existing model, and M is the number of newly added trees. f K+m ( x i ) is the predicted value of the m-th newly added tree. is the predicted value of the original model. is the predicted value of the updated model. Input the weighted oscillatory wave feature vector as input data into the pre-trained model for non-invasive blood pressure prediction, perform the model prediction operation, and output two blood pressure values, systolic blood pressure and diastolic blood pressure, thus completing the non-invasive blood pressure measurement of the test subject. The process of identity recognition by the non-invasive blood pressure measurement and identity recognition module is as follows: Calculate the weighted Euclidean distance between the sample to be recognized and each known identity sample. Use the weighted oscillatory wave feature vector to calculate the weighted Euclidean distance between the sample to be recognized and each known identity sample. For the sample Y to be identified and the known identity sample R, their weighted Euclidean distance is expressed as: （19）； Among them, D(Y, R) is the weighted Euclidean distance between sample Y and sample R, x3(z, Y) and x3(z, R) are the values of sample Y and sample R on the z-th feature respectively, and w(z) is the weight of the z-th feature; Based on the weighted Euclidean distance between the sample to be identified and each known identity sample, multiple identity recognition scores are calculated; The identity recognition score is defined as the reciprocal of the weighted Euclidean distance between the sample Y to be identified and the known identity sample R, as shown in the following formula: （20）； Among them, Score(Y) represents the identity recognition score of the sample Y to be identified. The higher the score, the higher the similarity between the samples; By comparing the identity recognition scores Score(Y) between the sample Y to be identified and all known identity samples, the higher the score, the closer the sample Y is to a certain known identity sample, and thus the higher the credibility of the identity recognition. Select the known identity sample with the highest identity recognition score as the final identity recognition result of the sample to be identified; At the same time, continuously update the newly collected oscillatory wave data to the known identity samples of this member, so that the identity recognition process can adapt to the characteristics of the test object under different physiological states, and iteratively update the existing identity samples.

2. The home blood pressure measurement and identity recognition system based on the characteristics of the oscillatory wave according to claim 1, wherein: The deep autoencoder includes two parts: an encoder and a decoder. The function of the encoder is to further compress the input oscillatory wave feature data into a low-dimensional space, and the decoder attempts to reconstruct the original oscillatory wave features from this low-dimensional space; The specific process of unsupervised feature fusion is as follows: Use the oscillatory wave feature data after PCA dimensionality reduction as the input of the deep autoencoder, and use the mean square error as the loss function to train the model. The purpose is to minimize the difference between the input oscillatory wave features and the reconstructed features; after the model training is completed, the output of the decoder part, that is, the feature representation in the low-dimensional space, is used as the fused oscillatory wave features; Map the input feature x2 to the low-dimensional latent space z: （12）； Among them, and are the encoder weight matrices and are hidden layer parameters; and represent the encoder bias terms, σ is the activation function; Reconstruct the output from the latent space z : （13）； Among them, and are the decoder weight matrices, and are the decoder bias tops.

3. A method for home blood pressure measurement and identity recognition based on the characteristics of oscillatory waves, characterized in that: It is realized through the above-mentioned home blood pressure measurement and identity recognition system based on oscillatory wave features, including the following steps: Step 1: During the blood pressure measurement process, continuously collect the original oscillatory wave signal of the measured person through the cuff and the pressure sensor; the system first detects the cuff fitting state and the arm movement condition, synchronously evaluates the influence degree of each interference factor on the signal quality, and discriminates the "usable" and "unusable" signals; for the "usable" signals that meet the basic acquisition requirements, the system will further carry out multi-dimensional quality grading evaluation; Step 2: Preprocess the oscillatory wave signal with good quality, including noise removal, filtering and downsampling operations to remove the noise in the oscillatory wave signal; Step 3: Use the Pan-Tompkins feature point detection algorithm to extract the time-domain features of the oscillatory wave signal, the frequency-domain analysis method to extract the frequency-domain features of the oscillatory wave signal, and the short-time Fourier transform to extract the time-frequency features of the oscillatory wave signal from the preprocessed oscillatory wave signal respectively, and obtain the multi-index features of the oscillatory wave signal; The specific method is: Step 3.1: Use the improved Pan-Tompkins algorithm for cycle identification. Locate the starting point of the waveform through adaptive threshold detection technology to achieve per-cycle segmentation processing; For a single-cycle waveform, extract the following contour feature parameters: the vertical distance between the main peak vertex and the baseline, the main peak amplitude H; the vertical distance between the late peak vertex and the baseline, the main peak amplitude H1; the maximum slope value of the waveform rising edge, the maximum slope K; the vertical distance between the dicrotic notch point and the baseline, the notch depth D; time parameters: the duration from the main peak to the starting point T1, the duration from the main peak to the dicrotic notch point T2, the duration from the dicrotic notch point to the late peak T3, the duration from the late peak to the end point T4, the complete cycle duration T; area parameters: the waveform area A1 in the T1 period, the waveform area A2 in the T2 period, the waveform area A3 in the T3 period, the waveform area A4 in the T4 period, the number of the complete cycle area A; Step 3.2: Conduct multi-resolution spectrum analysis based on the spectrum analysis method, and use the spectral energy ratio of different sub-band frequencies as the frequency-domain feature of the oscillatory wave signal; First, use the sub-band division method to equally divide the concentrated frequency band of the effective information into multiple sub-band intervals, and calculate the power spectral density of each sub-band by combining downsampling and Fourier transform; the calculation formula for the power spectral energy E in the c-d frequency band range is: （6）； where P (Pulse) is the power spectral density of the oscillatory wave signal in the c-d frequency band, as shown in the following formula: （7）； where FFT(Pulse) is the spectrum obtained by using FFT for the oscillatory wave signal, and fs is the sampling frequency of the oscillatory wave signal; Combine the main information distribution characteristics of the oscillatory wave signal to conduct multiple analyses on the multi-spectrum energy of the oscillatory wave signal, and generate the power spectral energy ratio feature parameter through normalization processing: PSER m =E m / E all (8); Among them, m represents the number of sub - band frequency bands, and E m is the power spectrum energy value of the m - th sub - band frequency band, and E all is the total energy of all frequency bands; Step 3.3: Extract the time-frequency diagram features of the oscillatory wave signal based on the short-time Fourier transform; Convert the one-dimensional time series into a time-frequency matrix through the short-time Fourier transform, and use the dynamic threshold segmentation technology to extract the core area with an energy ratio of more than 90%; obtain the frequency axis range through the contour tracking algorithm, calculate the contour height, locate the local maximum points in the time-frequency diagram based on the energy gradient detection algorithm, and select the N extreme points with the largest amplitude to form the time-frequency feature vector, where each coordinate point corresponds to the significant feature position of the energy distribution in the time-frequency diagram; Perform short-time Fourier transform on the oscillatory wave signal Pulse(t) using the sliding window w(t) to obtain the time-frequency distribution diagram X(t,f) of the oscillatory wave signal, where t represents time; Detect the main energy distribution area on the time-frequency distribution diagram X(t,f), extract the edge contour of this area, and represent it with the closed curve L(t,f); Calculate the maximum value \(f\) of the closed curve \(L(t, f)\) on the frequency \(f\) axis max and the minimum value \(f\) min , and obtain the contour height range characteristic \(h\) of the oscillatory wave signal: h = f max -f min (9); Based on the contour height of the oscillatory wave signal, locate the local maximum points in the time-frequency distribution diagram X(t,f) based on the energy gradient detection algorithm, select the N extreme values with the largest amplitude, and form the feature sequence with the time and frequency coordinates of these points to form the time-frequency local maximum point feature M of the oscillatory wave signal; （10）； where S(t n , f n ) is a local maximum point, n = 1, 2, …, N; Step 4: Perform standardization processing and dimensionality reduction operations on the time domain, frequency domain, and time-frequency characteristics of the oscillatory wave signal obtained in Step 3. Use a deep autoencoder to perform more in-depth feature learning and fusion on the oscillatory wave feature data x2 after dimensionality reduction processing, achieving unsupervised feature fusion of the oscillatory wave signal and obtaining the fused oscillatory wave feature x3; Step 5: Based on the fused oscillatory wave feature x3 extracted from the deep autoencoder in Step 4, perform feature weighted fusion, and then realize non-invasive blood pressure measurement and identity score recognition; Step 5.1: Evaluate the importance of oscillatory wave features based on the contribution of reconstruction error, and clarify the role of each feature in the whole; Step 5.2: Perform weighted fusion on the oscillatory wave features according to the results of feature importance evaluation; Step 5.2.1: Feature weight assignment; The weight size of a feature is positively correlated with the importance of the feature. Divide the importance score of each oscillatory wave feature by the sum of the importance scores of all features to obtain the weight of the feature; the fused feature vector is x3 = [x3(1), x3(2), …, x3(k)], and its corresponding weight is w = [w(1), w(2), …, w(k)]; Step 5.2.2: Perform feature weighted fusion based on weight assignment; Apply the calculated weights to each oscillatory wave eigenvector to obtain the weighted oscillatory wave eigenvector X weighted =[x3(1)w(1), x3(2)w(2),…, x3(k)w(k)]; Step 5.3: Realize non-invasive blood pressure measurement based on multi-index feature weighted fusion; The weighted oscillometric feature vector X obtained based on Step 5.2 weighted is input into the pre-trained model for non-invasive blood pressure prediction to achieve non-invasive blood pressure measurement of the subject. The specific method is as follows: Step 5.3.1: Based on the pre-stored oscillatory wave database, use the Extreme Gradient Boosting algorithm to optimize for individual differences; mix the oscillatory wave data of various population categories to train the model, and control the category bias through cross-validation to complete the pre-training of the non-invasive blood pressure prediction model; at the same time, continuously use the newly collected oscillatory wave data to enable the non-invasive blood pressure prediction model to adapt to the individual characteristics of the test object and perform iterative updates on the model; The objective function of XGBoost consists of a loss function and a regularization term. In the regression task, the mean squared error is used as the loss function. The formula for the objective function of XGBoost is as follows: （16）； Among them, y i is the true blood pressure value, is the blood pressure value predicted by the model, T k is the number of leaf nodes of the k-th tree, w k is the leaf weight of the k-th tree, γ and λ are the regularization coefficients for controlling the model complexity; Use stratified K-fold cross-validation to ensure that the class distribution of each fold of data is consistent. The model evaluation formula is: （17）； where K is the number of cross-validation folds, is the k-th fold validation dataset; When new data arrives, the incremental training formula is as follows: （18）； Among them, K is the number of trees in the existing model, and M is the number of newly added trees. f K+m ( x i ) is the predicted value of the m-th newly added tree. is the predicted value of the original model. is the predicted value of the updated model. Step 5.3.2: Non-invasive blood pressure measurement of the test subject; Input the weighted oscillometric wave feature vector X weighted as input data into the pre-trained model for non-invasive blood pressure prediction, perform the model prediction operation, and output two blood pressure values, systolic blood pressure and diastolic blood pressure, thereby completing the non-invasive blood pressure measurement of the subject; Step 5.4: Perform identity score recognition based on multi-index feature weighted fusion; Take the weighted oscillatory wave feature vector obtained in Step 5.2 as the feature template of the current detected object; use the Euclidean distance metric algorithm to perform identity recognition in the pre-stored oscillatory wave feature database and finally output the result; the specific method is: Step 5.4.1: Calculate the weighted Euclidean distance between the sample to be recognized and each known identity sample; Using the weighted feature vector X weighted , calculate the weighted Euclidean distance between the sample to be recognized and each known identity sample; For the sample to be recognized Y and the known identity sample R, its weighted Euclidean distance is expressed as: （19）； where D(Y, R) is the weighted Euclidean distance between sample Y and sample R, x3(z, Y) and x3(z, R) are the values of sample Y and sample R on the z-th feature respectively, and w(z) is the weight of the z-th feature; Step 5.4.2: Calculate multiple identity recognition scores based on the weighted Euclidean distance between the sample to be recognized and each known identity sample; The identity recognition score is defined as the reciprocal of the weighted Euclidean distance between the sample Y to be recognized and the known identity sample R: （20）； Among them, Score(Y) represents the identity recognition score of the sample Y to be recognized. The higher the score, the higher the similarity between the samples; By comparing the identity recognition scores Score(Y) between the sample Y to be recognized and all known identity samples, the higher the score, the closer the sample Y is to a certain known identity sample, and thus the higher the credibility of the identity recognition. Select the known identity sample with the highest identity recognition score as the final identity recognition result of the sample Y to be recognized; At the same time, continuously update the newly collected oscillatory wave data to the known identity samples of this member, so that the identity recognition process can adapt to the characteristics of the test object under different physiological states and iteratively update the existing identity samples.

4. The method for home blood pressure measurement and identity recognition based on oscillometric wave characteristics according to claim 3, wherein: The specific method of step 1 is as follows: Step 1.1: Adopt a multi-parameter joint error detection mechanism to detect errors in the collected oscillatory wave signal, monitor the signal integrity in real time, and judge whether the oscillatory wave signal collected by the cuff and the sensor is "usable"; When detecting abnormal situations such as the cuff being too tight or not firmly attached, interference caused by the user's arm movement or force during the measurement, or severe baseline drift, determine that this section of the signal is "unusable" and trigger a re-collection instruction; for signals that do not trigger the error detection mechanism, the system marks them as "usable" and transfers them to the next evaluation link; Step 1.2: For the "usable" oscillatory wave signals, a quality assessment model based on the fusion of time-frequency domain features is constructed, and indicators in both the time domain and the frequency domain are comprehensively considered to evaluate the signal quality; the specific method is as follows: Step 1.2.1: Conduct a time-domain analysis of the oscillatory wave signal, and use the amplitude change degree index, baseline drift degree index, and pulse rate consistency degree index as the signal quality assessment indicators in the time domain; The variational mode decomposition and period segmentation are respectively performed on the oscillatory wave signal pulse to obtain the reconstructed signal bPulse and the single-period signal Pulse that characterize the baseline drift of the signal i ; The amplitude change degree index is obtained through the amplitude difference of the signal waveforms in multiple cardiac cycles; the amplitude change degree index a(i) in the i-th cycle of the oscillatory wave signal is as shown in the following formula: a(i) = |max(Pulse i ) - min(Pulse i )| (1); where, max(Pulse i ) and min(Pulse i ) are respectively the maximum value and the minimum value of the waveform amplitude of the i-th cycle of the oscillating wave signal after filtering processing; The baseline drift degree index is obtained by windowing the reconstructed signal; Window the reconstructed signal bPulse to obtain the baseline drift degree index b(i) in the i-th cycle of the oscillatory wave signal, as shown in the following formula: b(i)= max( |bPulse i |)-b(2); Among them, bPulse i is the reconstructed signal of the i-th cycle of the oscillatory wave signal, and b is the baseline standard value of the oscillatory wave signal in the stationary state; The pulse rate consistency degree index is calculated by statistically analyzing the time length of the single-cycle signal after cycle segmentation and calculating the consistency of the signal time lengths in multiple cycles; Set the period length of the oscillating wave signal to length(Pulse i ) to obtain the pulse rate consistency index r(i) within the i-th period of the oscillating wave signal, as shown in the following formula: （3）； where SD(length(Pulse i )) is the standard deviation of the calculated length of the i-th signal period, and Mean(length(Pulse i )) is the mean of the calculated length of the i-th signal period; Step 1.2.2: Conduct a frequency-domain analysis of the oscillatory wave signal, select the effective frequency band of the oscillatory wave signal through fast Fourier transform, calculate the power spectral density as the frequency component index, and evaluate the oscillatory wave signal quality; The frequency component index f(i) in the i-th cycle of the oscillatory wave signal is: （4）； Among them, P is the power of the oscillatory wave signal, and f represents the frequency; Step 1.2.3: Based on the signal quality assessment indicators in the time domain and the frequency domain, use the information entropy theory to calculate the weight coefficients of each index in the time-frequency domain, and perform multi-index weighted fusion in the time-frequency domain for the oscillatory wave signal to obtain the fusion index SQI(i), as shown in the following formula: （5）； Among them, w a (i), w b (i), w r (i), w f (i) are the entropy weights of the evaluation indicators a(i), b(i), r(i), and f(i) respectively; Step 1.2.4: Perform threshold discrimination based on the fusion index to screen out high-quality signals from the "usable" oscillatory wave signals; When the fusion index SQI(i) exceeds the preset threshold, it is determined as a "high-quality signal" and enters the subsequent processing flow. Signals that do not meet the standard will be marked as "sub-optimal signals", and the system will prompt for re-measurement.

5. The method for home blood pressure measurement and identity recognition based on oscillation wave characteristics according to claim 3, characterized in that: The specific method of Step 2 is as follows: Step 2.1: Adopt a cascaded digital filtering technique. First, filter out high-frequency noise through a low-pass filter, and then apply a non-linear median filtering algorithm to eliminate burst pulse interference, forming a preliminarily noise-reduced signal sequence; Step 2.2: Use a sliding window mean filter combined with a high-pass filter to filter, smooth the signal, and remove baseline drift and low-frequency noise; Step 2.3: Perform selective downsampling processing. First, restore the continuity of the original signal through spline interpolation technology, and then adopt a multi-stage decimation algorithm to retain key feature information.

6. The method for home blood pressure measurement and identity recognition based on oscillometric wave characteristics according to claim 3, characterized in that: The specific method of Step 4 is as follows: Step 4.1: Standardize the characteristic data contained in the oscillatory wave signal to eliminate the differences in dimension and magnitude between different characteristics; the standardization formula is as follows: （11）； where x is the original characteristic value of the oscillatory wave signal, μ is the average value of the characteristic, σ1 is the standard deviation of the characteristic, and x1 is the standardized characteristic value; Step 4.2: Use the principal component analysis method to perform dimensionality reduction on the standardized oscillatory wave signal characteristic data; the specific method is as follows: Step 4.2.1: Calculate the covariance matrix: For the standardized blood pressure oscillatory wave signal characteristic data, calculate its covariance matrix and analyze the correlation between each characteristic; Step 4.2.2: Eigenvalue decomposition: Perform eigenvalue decomposition on the obtained covariance matrix to obtain a series of eigenvalues and their corresponding eigenvectors, providing a basis for subsequent principal component selection; Step 4.2.3: Select the principal components: According to the magnitude of the eigenvalues, retain the eigenvectors corresponding to Q eigenvalues whose contribution exceeds 90%, and determine them as the principal components to focus on key information; Step 4.2.4: Data projection: Project the original oscillatory wave signal characteristic data onto the selected principal components to obtain the oscillatory wave signal characteristic data x2 after dimensionality reduction processing; Step 4.3: Use a deep autoencoder to perform more in-depth feature learning and fusion on the oscillatory wave characteristic data after dimensionality reduction processing to achieve unsupervised feature fusion of the oscillatory wave signals; the specific method is as follows: Step 4.3.1: Construct a deep autoencoder, including an encoder and a decoder; the role of the encoder is to further compress the input oscillatory wave features into a low-dimensional space, and the decoder attempts to reconstruct the original oscillatory wave features from this low-dimensional space; Map the input feature x2 to the low-dimensional latent space z: （12）； Among them, and are the encoder weight matrices and are hidden layer parameters; and represent the encoder bias terms, σ is the activation function; Reconstruct the output from the latent space z : （13）； Among them, and are the decoder weight matrices, and are the decoder bias tops; Step 4.3.2: Model training: Use the oscillatory wave characteristic data x2 after PCA dimensionality reduction processing as the input of the deep autoencoder, and adopt the mean square error MSE as the loss function to train the model, aiming to minimize the difference between the input oscillatory wave features and the reconstructed features; the MSE calculation formula is: （14）； where n is the number of samples, represents the i-th original oscillatory wave feature data after PCA dimensionality reduction, is the i-th oscillatory wave feature data after reconstruction by the deep autoencoder; Step 4.3.3: Feature fusion representation: After the model training is completed, the output of the decoder part is the feature representation in the low-dimensional space, which serves as the fused oscillatory wave feature x3.

7. The method for home blood pressure measurement and identity recognition based on the characteristics of oscillatory waves according to claim 6, characterized in that: The specific method of step 5.1 is as follows: Step 5.1.1: Feature removal: In the input oscillatory wave data, each fused feature is removed in sequence, and then the remaining features are used for reconstruction through the trained deep autoencoder. Step 5.1.2: For the oscillatory wave data obtained after each feature removal, calculate its reconstruction error and compare it with the reconstruction error of the original model. The reconstruction error formula is as follows: （15）； Among them, represents the reconstructed output of the decoder for the i-th sample after removing the z-th feature; Step 5.1.3: Importance evaluation: According to the increase in the reconstruction error after feature removal, reflect the importance of this feature in the oscillatory wave data. The greater the increase, the more important the feature.

Images