Sleep physiological parameter real-time monitoring method based on flexible piezoelectric sensing

Abstract

The invention provides a sleep physiological parameter real-time monitoring method based on flexible piezoelectric sensing, and relates to the technical field of biomedical signal processing.The method comprises the steps that a human body sleep biological mechanical signal is collected through a flexible piezoelectric sensor array, and after the signal is converted through a charge amplifier, phase-space reconstruction denoising and self-adaptive gain control are carried out; wherein the gain is dynamically adjusted based on the weight of the user, the rigidity coefficient of the mattress and the real-time signal amplitude; extracting heartbeat wave crest intervals, respiratory cycles and body movement frequency domain energy characteristics from the preprocessed signals; inputting the features into a cascaded convolutional neural network and a long-short-term memory network model, and synchronously outputting the heart rate, the respiration rate and the heart rate variability; in combination with the historical health data of the user and the cardiovascular health mapping model, sleep quality indexes and disease risk scores are generated through graph convolutional network analysis, and the monitoring frequency is dynamically adjusted and an apnea alarm is triggered.

Description

Technical Field

[0001] The present invention relates to the technical field of biomedical signal processing, and in particular to a method for real-time monitoring of sleep physiological parameters based on flexible piezoelectric sensing. Background Art

[0002] In the field of sleep physiological parameter monitoring, existing technologies face several key challenges. First, traditional sensors mostly use rigid materials or contact electrodes, which can easily cause discomfort to users when worn for a long time, and cannot stably capture weak biomechanical signals (such as micro-movements caused by heartbeats and breathing), resulting in insufficient signal integrity. Secondly, biomechanical signals are easily affected by differences in mattress materials, user body movements, and environmental vibrations. Noise suppression methods mostly rely on fixed threshold filtering and cannot be adaptively adjusted, resulting in effective signal distortion or loss of key features. In addition, when processing multi-scale physiological parameters (such as heart rate, respiratory rate, and heart rate variability), existing algorithms often adopt a step-by-step independent extraction strategy, ignoring the temporal correlation between parameters, resulting in computational redundancy and poor real-time performance.

[0003] Further deficiencies are reflected in the health risk assessment level. Existing systems usually only generate indicators based on real-time data, lacking deep integration with users' historical health records (such as cardiovascular disease history and medication data), making it difficult to achieve personalized risk warnings. At the same time, for the detection of emergencies such as sleep apnea, most solutions use fixed-frequency monitoring, which cannot dynamically adjust the response speed according to the risk level, and may delay the warning time. Finally, in limited label scenarios, data augmentation methods often have a large deviation between the generated samples and the true distribution, resulting in a decrease in the model's generalization ability and affecting the accuracy of physiological parameter extraction. Summary of the Invention

[0004] In order to solve the technical problems in the existing technology, such as poor sensor comfort, insufficient noise suppression capability, low efficiency of real-time extraction of multiple parameters, lack of personalized and dynamic response mechanism for health risk assessment, and weak model generalization under limited labels, the present invention provides a real-time monitoring method for sleep physiological parameters based on flexible piezoelectric sensing.

[0005] The technical solutions provided by the present invention are as follows:

[0006] The present invention provides a method for real-time monitoring of sleep physiological parameters based on flexible piezoelectric sensing, comprising:

[0007] S1. Collecting biomechanical signals generated by the human body during sleep using a flexible piezoelectric sensor array. The sensor is composed of a polyvinylidene fluoride substrate and a metal electrode layer. The output signal is converted into a voltage signal by a charge amplifier.

[0008] S2. Preprocessing the voltage signal, including signal denoising based on phase space reconstruction and adaptive gain control, wherein the adaptive gain control dynamically adjusts the amplification factor according to the user's weight, mattress material, and signal amplitude;

[0009] S3, extracting multi-scale features from the preprocessed BCG signal, including the heartbeat characteristic peak interval, the chest movement period caused by breathing, and the frequency domain energy distribution of the body motion signal;

[0010] S4. Inputting the multi-scale features into a physiological parameter extraction model, wherein the model comprises a cascaded convolutional neural network and a long short-term memory network, and outputting real-time heart rate, respiratory rate, and heart rate variability;

[0011] S5. Based on the cardiovascular health mapping model and combined with the user's historical health data, a sleep quality index and cardiovascular disease risk score are generated, and real-time feedback is provided through the terminal device.

[0012] The beneficial effects brought about by the technical solution provided by the present invention include at least:

[0013] (1) In the present invention, the stability and comfort of signal acquisition are significantly improved by adopting a flexible piezoelectric sensor array and adaptive gain control technology. The composite structure of the flexible polyvinylidene fluoride (PVDF) substrate and the metal electrode layer has both high sensitivity and flexibility, and can fit the human body surface seamlessly, fully capturing weak biomechanical signals (such as heartbeat and chest movement caused by breathing); at the same time, the adaptive gain control dynamically adjusts the amplification factor according to the user's weight, mattress material, and real-time signal amplitude, effectively suppressing environmental noise and signal saturation problems, solving the signal distortion defects caused by the poor comfort and insufficient noise suppression ability of traditional rigid sensors, and providing a high-quality data foundation for subsequent precise analysis.

[0014] (2) In the present invention, a physiological parameter extraction model of a cascaded convolutional neural network (CNN) and a long short-term memory network (LSTM) is used to achieve efficient real-time extraction of multi-scale physiological parameters. The CNN module extracts local time domain features (such as the heartbeat peak interval), and the LSTM module captures the temporal correlation of the respiratory cycle and body motion signals, avoiding the computational redundancy of the traditional step-by-step independent extraction strategy. Combined with multi-scale feature fusion technology, the model can synchronously output heart rate, respiratory rate, and heart rate variability, and maintain the physiological correlation between parameters, significantly improving the real-time performance and computational efficiency of the monitoring system, and overcoming the limitations of low multi-parameter extraction efficiency and insufficient temporal modeling in the prior art.

[0015] (3) In the present invention, the personalized and dynamic response capabilities of health risk assessment are enhanced through the collaborative optimization of the cardiovascular health mapping model and semi-supervised deep learning. The model integrates the user's historical health data (such as electronic medical records, medication records) with real-time BCG signals, and uses the graph convolutional network (GCN) to mine the nonlinear relationship between features to generate a personalized sleep quality index and disease risk score; at the same time, the semi-supervised learning method uses the generative adversarial network (GAN) to augment the limited label samples and combines the consistency regularization constraint to improve the generalization of the model. This technical solution not only realizes the dynamic frequency adjustment of risk warning (such as apnea events triggering immediate alarms), but also solves the problem of model overfitting in limited label scenarios, significantly improving the clinical application reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 A schematic flow chart of a method for real-time monitoring of sleep physiological parameters based on flexible piezoelectric sensing provided by an embodiment of the present invention;

[0018] Figure 2 A schematic diagram of a process for preprocessing voltage signals in a method for real-time monitoring of sleep physiological parameters based on flexible piezoelectric sensing provided by an embodiment of the present invention;

[0019] Figure 3 A schematic diagram of the cardiovascular health mapping model construction and dynamic risk assessment process in the real-time monitoring method of sleep physiological parameters based on flexible piezoelectric sensing provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0021] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0022] In the embodiments of the present invention, the terms "image" and "picture" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same. The terms "of," "corresponding," and "corresponding" may be used interchangeably. It should be noted that, when the distinction between them is not emphasized, their intended meanings are the same.

[0023] In the embodiments of the present invention, sometimes a subscript such as W1 may be mistakenly written as a non-subscript form such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0024] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0025] Reference Manual Figure 1 , which shows a flow chart of a method for real-time monitoring of sleep physiological parameters based on flexible piezoelectric sensing provided by an embodiment of the present invention.

[0026] The embodiment of the present invention provides a method for real-time monitoring of sleep physiological parameters based on flexible piezoelectric sensing. The processing flow may include the following steps:

[0027] S1. The biomechanical signals generated by the human body in the sleep state are collected through a flexible piezoelectric sensor array. The sensor is composed of a polyvinylidene fluoride substrate and a metal electrode layer. The output signal is converted into a voltage signal through a charge amplifier.

[0028] It should be noted that the biomechanical signals of human sleep are collected through a flexible piezoelectric sensor array. The flexible piezoelectric sensor array is composed of a polyvinylidene fluoride (PVDF) film as a substrate, and its upper and lower surfaces are respectively compounded with a metal electrode layer composed of silver nanowires. The sensor array is arranged in a matrix form on the surface of the mattress, covering the chest to waist area to ensure that the contact area with the human body is maximized. The output signal of each sensor unit is converted into a voltage signal by a charge amplifier. The feedback capacitance of the charge amplifier is set to 10pF to 100pF, and the bandwidth is adjusted to 0.1Hz to 100Hz to suppress high-frequency noise and retain the effective components of the BCG (Ballistocardiogram) signal.

[0029] In one possible implementation, the flexible piezoelectric sensor array is arranged in a honeycomb pattern with a unit spacing of 5 cm to balance spatial resolution and signal crosstalk; the embedding dimension d of phase space reconstruction is determined by the false nearest neighbor method, and the dimension increase is stopped when the proportion of false neighbors is less than 5%; the generator of the Generative Adversarial Network (GAN) adopts a U-Net structure, the discriminator is a 5-layer convolutional network, and the learning rate is set to 0.0002 during adversarial training; in the topological alignment process, the dynamic time warping (DTW) algorithm is used to align the R peak positions of BCG and ECG, and the maximum bending window is set to 10 sampling points.

[0030] S2. Preprocessing the voltage signal, including signal denoising based on phase space reconstruction and adaptive gain control, wherein the adaptive gain control dynamically adjusts the amplification factor according to the user's weight, mattress material, and signal amplitude.

[0031] It should be noted that the voltage signal is preprocessed. First, the phase space reconstruction method is used to denoise the original signal: the one-dimensional BCG signal sequence {x(t)} is mapped to the d-dimensional phase space to generate the trajectory point set {X i =(x(t i ),x(t i +τ),...,x(t i +(d-1)τ))}, where d is the embedding dimension and τ is the delay time. By calculating the local density distribution of the trajectory points, the noise points that deviate from the main cluster are eliminated. Subsequently, adaptive gain control is performed: according to the user weight m (obtained by user input or the built-in pressure sensor of the mattress) and the mattress stiffness coefficient k (measured by the material mechanics tester), the basic gain coefficient G0 is calculated, where the reference mass m0 is set to 70kg and the sensor cutoff frequency f c The actual gain G is dynamically adjusted based on the standard deviation σ of the signal amplitude within the sliding window (the window length is set to 5 seconds). When ω exceeds the threshold σ th When the gain is increased nonlinearly according to the hyperbolic tangent function (preset to 0.2V), the gain is increased nonlinearly to suppress signal saturation.

[0032] like Figure 2As shown, voltage signal preprocessing begins with the input voltage signal, first performing phase space reconstruction and denoising. This method converts the original one-dimensional biomechanical signal (BCG) into trajectory points in a multidimensional space. By analyzing the distribution density of these points, it automatically identifies and removes noise points (such as environmental vibration interference), retaining signal components that reflect true physiological activity. Next, the adaptive gain control stage begins: the system calculates the base signal amplification factor based on parameters such as user weight and mattress material to ensure consistent data amplitude for users of different body types. Simultaneously, the signal fluctuation intensity is monitored in real time. If the signal amplitude changes significantly (such as due to sudden body movement), the gain is dynamically adjusted to avoid signal overload or weak signal loss. The final output is a stable, denoised signal for use in the subsequent feature extraction module.

[0033] In a possible implementation manner, the adaptive gain control in S2 specifically includes:

[0034] S201. Calculate the basic gain coefficient G0 based on the user's weight m and the mattress stiffness coefficient k, satisfying the formula:

[0035]

[0036] Where m0 is the reference mass in kg, f c is the sensor cutoff frequency, f s is the current signal sampling frequency;

[0037] S202, dynamically adjust the actual gain G based on the standard deviation σ of the signal amplitude in the sliding window to meet

[0038] G=G0·(1+tanh(σ / σ th ))

[0039] Among them, σ th is the preset amplitude threshold.

[0040] The entire process achieves noise suppression and signal enhancement through a closed-loop feedback mechanism, which not only solves the problem that traditional rigid sensors are susceptible to interference, but also overcomes the defect that fixed gain design cannot adapt to individual differences, laying the foundation for accurate monitoring of sleep physiological parameters.

[0041] S3. Extract multi-scale features from the preprocessed BCG signal, including the heartbeat characteristic peak interval, the chest movement period caused by breathing, and the frequency domain energy distribution of the body motion signal.

[0042] It should be noted that multi-scale features were extracted from the preprocessed BCG signal. For heartbeat features, wavelet transform was used to detect the J-wave peak position, and the interval between adjacent peaks was calculated as the heart rate. For respiratory signals, a bandpass filter (0.1Hz to 0.5Hz) was used to separate the chest motion component, and the zero-crossing point method was used to count the respiratory cycles. For body motion signals, short-time Fourier transform was used to analyze the frequency domain energy, and the energy integral in the 1Hz to 10Hz frequency band was extracted as an indicator of body motion intensity.

[0043] In one possible implementation, the extraction of the respiratory rate in S3 includes:

[0044] S301, reconstructing the phase space of the BCG signal to obtain a multi-dimensional phase trajectory;

[0045] S302, calculating the Euclidean distance D of adjacent phase trajectory points ij , filter to meet D ij <R th The feature point pairs, where R th is the reconstruction radius;

[0046] S303. Calculate the respiratory rate RR using maximum likelihood estimation based on the time interval distribution of the feature point pairs, satisfying the formula:

[0047]

[0048] Among them, N peak is the number of breathing-related feature points, t k is the timestamp of the kth feature point.

[0049] In a possible implementation, the phase space reconstruction in S301 adopts a delayed coordinate method, which specifically includes:

[0050] Its embedding dimension d and delay time τ are determined by the mutual information method, satisfying:

[0051]

[0052] Where C(d,τ) is the correlation integral of the reconstructed phase space and ∈ is the convergence threshold.

[0053] S4. Input the multi-scale features into a physiological parameter extraction model, wherein the model includes a cascaded convolutional neural network and a long short-term memory network, and outputs real-time heart rate, respiratory rate, and heart rate variability.

[0054] It should be noted that the multi-scale features are input into the physiological parameter extraction model. The model consists of a cascaded convolutional neural network (CNN) and a long short-term memory network (LSTM): the CNN part contains 3 convolutional layers (the convolution kernel sizes are 5×1, 3×1, and 3×1 respectively) to extract local time domain features; the LSTM part contains 2 hidden layers (128 units per layer) to capture temporal dependencies. The model outputs real-time heart rate (HR), respiratory rate (RR) and heart rate variability (HRV). The calculation of HRV further includes: extracting a continuous RR interval sequence {RR1, RR2, ..., RR n ,}, and calculate the standard deviation SDNN through the improved time domain analysis model. Among them, the value of the physiological state coefficient α(T) is related to the sleep stage T. The sleep stage is divided by the energy distribution of the body motion signal. Specifically: when the body motion energy is lower than 0.1mV 2 It is judged as deep sleep stage, 0.1mV 2 to 0.5mV 2 It is a light sleep stage, higher than 0.5mV 2 When accompanied by irregular peaks, it is judged as the wakefulness or rapid eye movement (REM) stage.

[0055] In a possible implementation, the HRV calculation in S4 includes:

[0056] S401, extract the continuous RR interval sequence {RR1, RR2, ..., RR n ,};

[0057] S402. Calculate the standard deviation SDNN based on the improved HRV time domain analysis model, satisfying:

[0058]

[0059] in, is the average RR interval, and α(T) is the physiological state coefficient related to sleep stage T.

[0060] In one possible implementation, the training of the physiological parameter extraction model adopts a semi-supervised deep learning method, including:

[0061] S411. Use limited labeled samples to construct an initial training set and generate augmented samples through a generative adversarial network;

[0062] S412. Use consistency regularization constraints to ensure that the model maintains similarity in feature mapping between augmented samples and original samples.

[0063] In a possible implementation, the calculation of the physiological state coefficient specifically includes:

[0064] S421. Divide the sleep stage T∈{awake, light sleep, deep sleep, REM} according to the energy distribution of the body motion signal;

[0065] S422. Fitting based on historical HRV data of stage T

[0066] α(T)=1+λ·exp(-|ΔHRV T | / σ T )

[0067] Where λ is the adjustment factor, ΔHRV T is the deviation of the current HRV from the stage baseline value.

[0068] S5. Based on the cardiovascular health mapping model and combined with the user's historical health data, a sleep quality index and cardiovascular disease risk score are generated, and real-time feedback is provided through the terminal device.

[0069] It should be noted that the sleep quality index and disease risk score are generated based on the cardiovascular health mapping model. The model is implemented through a graph convolutional network (GCN): BCG features are topologically aligned with electrocardiogram (ECG) and blood oxygen saturation (SpO2) data, a node feature matrix is ​​constructed, and the nonlinear relationship between features is learned through a three-layer GCN. Disease risk probability P risk The output of the real-time feedback module uses the Sigmoid activation function, and the threshold is set to 0.5. When the real-time feedback module detects an apnea event (respiratory rate RR < 8 times / minute and lasts for more than 10 seconds), it triggers an audible and visual alarm; at the same time, according to P risk Dynamically adjust the monitoring frequency, the basic update frequency f0 is set to 1Hz, when P risk When >0.7, the frequency increases to To increase monitoring density.

[0070] Furthermore, the integration of historical health data is achieved by building a personal digital health space: integrating diagnostic records, medication history and real-time BCG data in electronic medical records, and using the attention mechanism to weight multi-source features. Specifically, the attention score of each feature is calculated. where h i is the eigenvector, W q is the trainable weight matrix, dk The weighted feature vector is input into the fully connected layer to generate a personalized health report, which includes sleep quality score, abnormal event statistics and long-term trend analysis.

[0071] like Figure 3 As shown, the construction of the cardiovascular health mapping model and dynamic risk assessment begins with multi-source data input, including real-time BCG features (heartbeat intervals, respiratory cycles), electrocardiogram (ECG) and blood oxygen (SpO2) data, and the user's historical health records (such as cardiovascular disease history). First, the dynamic time warping (DTW) algorithm is used to align the R peak positions of the BCG signal and the ECG to address the acquisition delay between the two. After synchronizing the timestamps, physiological parameters (heart rate, respiratory rate, etc.) and user attributes (age, BMI) are constructed as graph nodes. The physiological coupling relationships between the parameters (such as the correlation between heart rate and blood oxygen) are defined as edge weights, forming a heterogeneous medical graph.

[0072] The graph convolutional network (GCN) processes data in three stages: the first layer aggregates features of adjacent nodes (e.g., integrating blood oxygen data with heart rate nodes), the second layer learns cross-modal associations (e.g., the relationship between sleep stages and HRV), and the third layer outputs node embedding vectors. Simultaneously, a semi-supervised module uses a generative adversarial network (GAN) to synthesize augmented samples (simulating BCG signals for users with different physical conditions) and constrains the feature distribution of the generated samples to be consistent with the real data to improve model generalization.

[0073] The dynamic risk assessment module maps the embedding vector output by GCN to the disease risk probability (P_risk). When P_risk exceeds the threshold, two feedbacks are triggered: 1) Real-time adjustment of monitoring frequency, the formula is f_update = basic frequency × (1 + P_risk 2 ), the frequency index increases when the risk is high; 2) If the respiratory rate is detected to be continuously lower than 8 times / minute for more than 10 seconds, the sound and light alarm will be activated immediately.

[0074] In one possible implementation, the steps of constructing the cardiovascular health mapping model include:

[0075] S501, topologically aligning the multi-scale features of the BCG signal with the electrocardiogram and blood oxygen saturation data;

[0076] S502: Based on the graph convolutional network, learn the nonlinear relationship between features and output the disease risk probability P. risk .

[0077] In a possible implementation manner, the real-time feedback in S5 specifically includes:

[0078] S511, triggering an audible and visual alarm when an apnea event is detected, wherein the apnea time is less than 8 breaths per minute and the duration is more than 10 seconds;

[0079] S512, dynamically adjust monitoring frequency according to risk score to meet Where f0 is the basic update frequency.

[0080] In one possible implementation, the integration of historical health data specifically includes:

[0081] S521. Build a personal digital health space that integrates electronic medical records, medication records, and real-time BCG data;

[0082] S522. Use the attention mechanism to weight multi-source data features and generate personalized health reports.

[0083] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0084] (1) In the present invention, the stability and comfort of signal acquisition are significantly improved by adopting a flexible piezoelectric sensor array and adaptive gain control technology. The composite structure of the flexible polyvinylidene fluoride (PVDF) substrate and the metal electrode layer has both high sensitivity and flexibility, and can fit the human body surface seamlessly, fully capturing weak biomechanical signals (such as heartbeat and chest movement caused by breathing); at the same time, the adaptive gain control dynamically adjusts the amplification factor according to the user's weight, mattress material, and real-time signal amplitude, effectively suppressing environmental noise and signal saturation problems, solving the signal distortion defects caused by the poor comfort and insufficient noise suppression ability of traditional rigid sensors, and providing a high-quality data foundation for subsequent precise analysis.

[0085] (2) In the present invention, a physiological parameter extraction model of a cascaded convolutional neural network (CNN) and a long short-term memory network (LSTM) is used to achieve efficient real-time extraction of multi-scale physiological parameters. The CNN module extracts local time domain features (such as the heartbeat peak interval), and the LSTM module captures the temporal correlation of the respiratory cycle and body motion signals, avoiding the computational redundancy of the traditional step-by-step independent extraction strategy. Combined with multi-scale feature fusion technology, the model can synchronously output heart rate, respiratory rate, and heart rate variability, and maintain the physiological correlation between parameters, significantly improving the real-time performance and computational efficiency of the monitoring system, and overcoming the limitations of low multi-parameter extraction efficiency and insufficient temporal modeling in the prior art.

[0086] (3) In the present invention, the personalized and dynamic response capabilities of health risk assessment are enhanced through the collaborative optimization of the cardiovascular health mapping model and semi-supervised deep learning. The model integrates the user's historical health data (such as electronic medical records, medication records) with real-time BCG signals, and uses the graph convolutional network (GCN) to mine the nonlinear relationship between features to generate a personalized sleep quality index and disease risk score; at the same time, the semi-supervised learning method uses the generative adversarial network (GAN) to augment the limited label samples and combines the consistency regularization constraint to improve the generalization of the model. This technical solution not only realizes the dynamic frequency adjustment of risk warning (such as apnea events triggering immediate alarms), but also solves the problem of model overfitting in limited label scenarios, significantly improving the clinical application reliability of the system.

[0087] The above content is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

[0088] There are a few points to note:

[0089] (1) The drawings of the embodiments of the present invention only relate to the structures related to the embodiments of the present invention. Other structures may refer to conventional designs.

[0090] (2) For the sake of clarity, the thickness of layers or regions in the drawings used to describe the embodiments of the present invention are exaggerated or reduced, that is, these drawings are not drawn to scale. It is understood that when an element such as a layer, film, region, or substrate is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element or intervening elements may be present.

[0091] (3) In the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other to form new embodiments.

[0092] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A real-time monitoring method for sleep physiological parameters based on flexible piezoelectric sensing, characterized in that: include: S1. Collecting biomechanical signals generated by the human body during sleep using a flexible piezoelectric sensor array. The sensor is composed of a polyvinylidene fluoride substrate and a metal electrode layer. The output signal is converted into a voltage signal by a charge amplifier. S2. Preprocessing the voltage signal, including signal denoising based on phase space reconstruction and adaptive gain control, wherein the adaptive gain control dynamically adjusts the amplification factor according to the user's weight, mattress material, and signal amplitude; S3, extracting multi-scale features from the preprocessed BCG signal, including the heartbeat characteristic peak interval, the chest movement period caused by breathing, and the frequency domain energy distribution of the body motion signal; S4. Inputting the multi-scale features into a physiological parameter extraction model, wherein the model comprises a cascaded convolutional neural network and a long short-term memory network, and outputting real-time heart rate, respiratory rate, and heart rate variability; S5. Based on the cardiovascular health mapping model and combined with the user's historical health data, a sleep quality index and cardiovascular disease risk score are generated, and real-time feedback is provided through the terminal device.

2. The method for real-time monitoring of sleep physiological parameters based on flexible piezoelectric sensing according to claim 1, characterized in that: The adaptive gain control in S2 specifically includes: S201. Calculate the basic gain coefficient G0 based on the user's weight m and the mattress stiffness coefficient k, satisfying the formula: Where m0 is the reference mass in kg, f c is the sensor cutoff frequency, f s is the current signal sampling frequency; S202, dynamically adjust the actual gain G based on the standard deviation σ of the signal amplitude in the sliding window to meet G=G0·(1+tanh(σ / σ th )) Among them, σ th is the preset amplitude threshold.

3. The method for real-time monitoring of sleep physiological parameters based on flexible piezoelectric sensing according to claim 1, characterized in that: The extraction of respiratory rate in S3 includes: S301, reconstructing the phase space of the BCG signal to obtain a multi-dimensional phase trajectory; S302, calculating the Euclidean distance D of adjacent phase trajectory points ij , filter to meet D ij <R th The feature point pairs, where R th is the reconstruction radius; S303. Calculate the respiratory rate RR using maximum likelihood estimation based on the time interval distribution of the feature point pairs, satisfying the formula: Among them, N peak is the number of breathing-related feature points, t k is the timestamp of the kth feature point.

4. The method for real-time monitoring of sleep physiological parameters based on flexible piezoelectric sensing according to claim 1, characterized in that: The HRV calculation in S4 includes: S401, extract the continuous RR interval sequence {RR1, RR2, ..., RR n ,}; S402. Calculate the standard deviation SDNN based on the improved HRV time domain analysis model, satisfying: in, is the average RR interval, and α(T) is the physiological state coefficient related to sleep stage T.

5. The method for real-time monitoring of sleep physiological parameters based on flexible piezoelectric sensing according to claim 4, characterized in that: The training of the physiological parameter extraction model adopts a semi-supervised deep learning method, including: S411. Construct an initial training set using limited labeled samples and generate augmented samples through a generative adversarial network. S412. Use consistency regularization constraints to ensure that the model maintains similarity in feature mapping between augmented samples and original samples.

6. The method for real-time monitoring of sleep physiological parameters based on flexible piezoelectric sensing according to claim 1, characterized in that: The steps of constructing the cardiovascular health mapping model include: S501, topologically aligning the multi-scale features of the BCG signal with the electrocardiogram and blood oxygen saturation data; S502: Based on the graph convolutional network, learn the nonlinear relationship between features and output the disease risk probability P. risk .

7. The method for real-time monitoring of sleep physiological parameters based on flexible piezoelectric sensing according to claim 3, characterized in that: The phase space reconstruction in S301 adopts the delayed coordinate method, which specifically includes: Its embedding dimension d and delay time τ are determined by the mutual information method, satisfying: Where C(d,τ) is the correlation integral of the reconstructed phase space and ∈ is the convergence threshold.

8. The method for real-time monitoring of sleep physiological parameters based on flexible piezoelectric sensing according to claim 4, characterized in that: The calculation of the physiological state coefficient specifically includes: S421. Divide the sleep stage T∈{awake, light sleep, deep sleep, REM} according to the energy distribution of the body motion signal; S422. Fitting based on historical HRV data of stage T α(T)=1+λ·exp(-|ΔHRV T | / s T ) Where λ is the adjustment factor, ΔHRV T is the deviation of the current HRV from the stage baseline value.

9. The method for real-time monitoring of sleep physiological parameters based on flexible piezoelectric sensing according to claim 6, characterized in that: The real-time feedback in S5 specifically includes: S511, triggering an audible and visual alarm when an apnea event is detected, wherein the apnea time is less than 8 breaths per minute and the duration is more than 10 seconds; S512, dynamically adjust monitoring frequency according to risk score to meet Where f0 is the basic update frequency.

10. The method for real-time monitoring of sleep physiological parameters based on flexible piezoelectric sensing according to claim 1, characterized in that: The integration of historical health data specifically includes: S521. Build a personal digital health space that integrates electronic medical records, medication records, and real-time BCG data; S522. Use the attention mechanism to weight multi-source data features and generate personalized health reports.

Images