Method and system for measuring blood pressure, electronic equipment and storage medium

Abstract

The invention discloses a blood pressure measurement method and system, electronic equipment and a storage medium. The method comprises the following steps: extracting a target fingertip pulse signal from video data of a fingertip covering camera of a detected person; performing variational mode decomposition reconstruction operation and data enhancement operation on the target fingertip pulse signal to obtain an effective fingertip pulse signal; inputting the effective fingertip pulse signal into a meta-learning model for blood pressure measurement meta-learning to obtain a trained meta-learning model; and inputting the target fingertip pulse signal into the trained meta-learning model for blood pressure measurement to obtain a blood pressure measurement result. By means of the scheme, personalized and accurate blood pressure estimation can be achieved through a small amount of data, the method has the advantages of being low in cost, capable of rapidly adapting to individual differences, resistant to environmental interference and the like, and a practical scheme is provided for ubiquitous health monitoring.

Description

Technical Field

[0001] This application generally relates to the technical field of blood pressure measurement. More specifically, this application relates to a method, system, electronic device, and computer-readable storage medium for blood pressure measurement. Background Art

[0002] Blood pressure, as a core indicator of cardiovascular health, its convenient monitoring is crucial for disease prevention and management. Although traditional cuff-based blood pressure measurement techniques (such as oscillometry, auscultation) have relatively high accuracy, they rely on dedicated equipment and are cumbersome to operate, making it difficult to meet the needs of daily continuous monitoring. With the popularization of smart devices, blood pressure monitoring technology based on wearable devices (such as smart watches) integrates sensors such as photoplethysmogram (PPG) and electrocardiogram (ECG), and uses principles such as pulse transit time (PTT) and change in pulse wave amplitude (CPA) to achieve non-invasive blood pressure estimation, significantly improving the measurement convenience. However, such technologies are still limited by the device form (such as wristbands, ear-worn devices) and the cost of dedicated sensors, and can only be implemented on specific wearable devices, unable to fully utilize the camera resources of ubiquitous smart devices such as smartphones and tablets. Using the camera of a smart device to collect the signal of the change in fingertip blood volume provides a new direction for low-cost, non-contact blood pressure monitoring, but this technical path faces key challenges such as signal quality, individual adaptability, and model generalization.

[0003] The current camera-based blood pressure monitoring methods have the following core problems: First, signal acquisition is easily interfered by environmental noise. The camera is sensitive to factors such as environmental light changes and the movement of the subject being examined, resulting in a large amount of noise in the collected pulse signal, affecting the accuracy of subsequent blood pressure estimation. Second, individual differences lead to insufficient model generalization ability. The skin type (such as skin color depth), physiological characteristics (such as blood vessel elasticity), and device parameters (such as camera resolution, frame rate) of different subjects being examined will all cause signal feature differences. Traditional machine learning models need to rely on large-scale and diverse datasets for training, making it difficult to quickly adapt to new subjects being examined in resource-constrained actual scenarios and prone to overfitting problems. In addition, existing technologies mostly rely on a single sensor or a fixed device architecture, lacking sufficient exploration of the universality of smart device cameras, resulting in limitations in the hardware compatibility of the measurement system and the experience of the subject being examined.

[0004] In view of this, there is an urgent need to provide a solution for blood pressure measurement, so as to achieve personalized and accurate blood pressure estimation through a small amount of data, with advantages such as low cost, rapid adaptation to individual differences, and anti-environmental interference, providing a practical solution for ubiquitous health monitoring. Summary of the Invention

[0005] In order to solve at least one or more of the above-mentioned technical problems, this application proposes solutions for blood pressure measurement in multiple aspects.

[0006] In a first aspect, the present application provides a method for blood pressure measurement, including: extracting a target fingertip pulse signal from video data of a camera covering the fingertip of a subject; performing a variational mode decomposition reconstruction operation and a data augmentation operation on the target fingertip pulse signal to obtain an effective fingertip pulse signal; inputting the effective fingertip pulse signal into a meta-learning model for blood pressure measurement meta-learning to obtain a trained meta-learning model; and inputting the target fingertip pulse signal into the trained meta-learning model for blood pressure measurement to obtain a blood pressure measurement result.

[0007] In some embodiments, extracting the target fingertip pulse signal from the video data of the camera covering the fingertip of the subject includes: segmenting the video data into single-frame data; decomposing the single-frame data into color channels to obtain multiple color channels in each frame of data; and calculating the average value of the target color channel in each frame of data to extract the target fingertip pulse signal, where the target color channel includes the R color channel.

[0008] In some other embodiments, the variational mode decomposition reconstruction operation is performed by the following operations: decomposing the target fingertip pulse signal into a set of intrinsic mode functions; selecting a first intrinsic mode function representing low-frequency heartbeat and a second intrinsic mode function representing high-frequency heartbeat from the set of intrinsic mode functions; and performing superposition reconstruction on the first intrinsic mode function and the second intrinsic mode function.

[0009] In still some other embodiments, decomposing the target fingertip pulse signal into a set of intrinsic mode functions includes: constructing a variational constraint model based on the target fingertip pulse signal; introducing a second-order penalty factor and a Lagrange multiplier into the variational constraint model to form an augmented Lagrangian function; and using the alternating direction multiplier method to iterate the augmented Lagrangian function to decompose it into the set of intrinsic mode functions.

[0010] In still some other embodiments, the data augmentation operation is performed by the following operations: performing a time-frequency transform on the target fingertip pulse signal or the fingertip pulse signal after superposition reconstruction to separate the heartbeat frequency; performing a scaling operation on the heartbeat frequency; and performing an inverse time-frequency transform on the scaled heartbeat frequency.

[0011] In still some other embodiments, it further includes: performing a normalization operation on the fingertip pulse signal after superposition reconstruction or the fingertip pulse signal after inverse time-frequency transform to obtain the effective fingertip pulse signal.

[0012] In still some other embodiments, the meta-learning model includes a plurality of convolutional layers, a plurality of bidirectional long short-term memory layers, a plurality of fully connected layers, and an output layer, and each convolutional layer includes a sub-convolutional layer, a batch normalization layer, and a pooling layer, and the output layer includes a single fully connected layer.

[0013] In a second aspect, the present application provides a system for blood pressure measurement, including: an acquisition module configured to acquire video data of a fingertip covering a camera of a subject and extract a target fingertip pulse signal from the video data of the fingertip covering the camera of the subject; a preprocessing module configured to perform a variational mode decomposition reconstruction operation and a data augmentation operation on the target fingertip pulse signal to obtain an effective fingertip pulse signal; a meta-learning training module configured to input the effective fingertip pulse signal into a meta-learning model for meta-learning of blood pressure measurement to obtain a trained meta-learning model; and a blood pressure measurement module configured to input the target fingertip pulse signal into the trained meta-learning model for blood pressure measurement to obtain a blood pressure measurement result.

[0014] In a third aspect, the present application provides an electronic device, including: a processor; and a memory storing computer instructions for blood pressure measurement, which when executed by the processor, cause the implementation of multiple embodiments in the foregoing first aspect; or cause the implementation of the operations performed by the system in the foregoing second aspect.

[0015] In a fourth aspect, the present application provides a computer-readable storage medium storing computer program instructions for blood pressure measurement, which when executed by one or more processors, cause the implementation of multiple embodiments in the foregoing first aspect; or cause the implementation of the operations performed by the system in the foregoing second aspect.

[0016] Through the blood pressure measurement solution provided above, in the embodiments of the present application, a target fingertip pulse signal is extracted from the video data of the fingertip covering the camera of the subject. The variational mode decomposition reconstruction operation can remove environmental noise interference, improve the quality of the pulse signal, and the data augmentation operation can increase the diversity of training data, enabling the model to better adapt to different environmental conditions and improving the measurement accuracy and stability. Further, training with a meta-learning model can quickly adapt to individual differences of different subjects, and only a small amount of training data is required to construct a personalized blood pressure estimation model for each subject, solving the problem that traditional machine learning models require a large amount of data to handle individual differences, improving the generalization ability and adaptability of the model, and thus providing a reliable blood pressure measurement result for the subject. Description of the Drawings

[0017] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features, and advantages of the exemplary embodiments of the present application will become readily understood. In the drawings, several embodiments of the present application are shown in an exemplary rather than restrictive manner, and the same or corresponding reference numerals represent the same or corresponding parts, where:

[0018] Figure 1is an exemplary flowchart showing method 100 for blood pressure measurement according to an embodiment of the present application;

[0019] Figure 2 is an exemplary schematic diagram showing blood pressure measurement according to an embodiment of the present application;

[0020] Figure 3 is an exemplary schematic diagram showing visualization of the R channel when blood flow is maximum and minimum in corresponding frame data according to an embodiment of the present application;

[0021] Figure 4 is an exemplary schematic diagram showing the variational mode decomposition reconstruction result according to an embodiment of the present application;

[0022] Figure 5 is an exemplary schematic diagram showing the meta - learning model according to an embodiment of the present application;

[0023] Figure 6 is an exemplary structural block diagram showing system 600 for blood pressure measurement according to an embodiment of the present application;

[0024] Figure 7 is an exemplary Bland - Altman plot showing the systolic and diastolic blood pressures obtained by evaluation according to an embodiment of the present application;

[0025] Figure 8 is an exemplary schematic diagram showing the comparison of the mean absolute error (MAE) of blood pressure measurement for different groups of subjects to be examined according to an embodiment of the present application;

[0026] Figure 9 is an exemplary schematic diagram showing the comparison of the influence of camera parameters on the mean absolute value of blood pressure measurement according to an embodiment of the present application;

[0027] Figure 10 is an exemplary schematic diagram showing the comparison of the mean absolute value of blood pressure measurement under different lighting conditions according to an embodiment of the present application;

[0028] Figure 11 is an exemplary structural block diagram showing electronic device 1100 according to an embodiment of the present application. Detailed implementation manners

[0029] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative efforts fall within the scope of protection of the present application.

[0030] It should be understood that the terms "comprising" and "including" as used in the specification and claims of this application indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0031] It should also be understood that the terms used in the specification of this application are merely for the purpose of describing particular embodiments and are not intended to limit this application. As used in the specification and claims of this application, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms. It should be further understood that the term "and / or" as used in the specification and claims of this application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0032] As used in this specification and the claims, the term "if" can be interpreted, depending on the context, as "when", "once", "in response to determining", or "in response to detecting". Similarly, the phrase "if determined" or "if [the described condition or event] is detected" can be interpreted, depending on the context, as meaning "once determined", "in response to determining", "once [the described condition or event] is detected", or "in response to detecting [the described condition or event]".

[0033] According to the description of the background art, blood pressure, as a core indicator of cardiovascular health, its convenient monitoring is crucial for disease prevention and management. For example, hypertension is a major risk factor for cardiovascular diseases (including heart attacks and strokes), and can also lead to chronic kidney disease and vision loss. Hypotension can cause dizziness, fainting, and in severe cases, shock and organ damage due to insufficient blood flow. Traditional blood pressure measurement methods mainly rely on cuff-based devices, which obtain accurate readings through oscillometry or auscultation. Although these methods are very accurate, they require specialized equipment and trained medical professionals to operate. Therefore, these methods often cause discomfort to patients and are not suitable for frequent use in daily life.

[0034] Electrocardiogram (ECG) and photoplethysmogram (PPG) sensors make it possible to measure blood pressure using smart devices, especially wrist-worn devices. These devices can estimate blood pressure based on the principles of pulse arrival time (PAT) and pulse transit time (PTT). PAT measures the time interval between the peak of the R wave in the ECG signal and the arrival of the pulse wave detected by the PPG sensor at the limb end, while PTT measures the time required for the pulse wave to travel between two arterial sites, usually from the heart to the wrist. By analyzing these time intervals, smart devices can infer blood pressure changes relatively accurately.

[0035] However, electrocardiogram (ECG) and photoplethysmogram (PPG) sensors are typically only found in wearable devices such as smartwatches. Therefore, developing accurate and reliable blood pressure monitoring solutions for smartphones and other devices remains a challenging task. A promising direction is to use cameras to obtain pulse signals, which opens up new possibilities for blood pressure measurement. However, camera-based sensing systems are particularly sensitive to environmental changes and exhibit significant individual differences in terms of appearance (e.g., gender, skin type) and physiology (e.g., blood volume dynamics).

[0036] Specifically, the high sensitivity of cameras to environmental noise and motion makes it vulnerable to environmental noise interference when obtaining pulse signals through cameras, resulting in a large amount of noise in the collected pulse signals and affecting the accuracy of subsequent blood pressure estimation. Existing blood pressure measurement methods are not applicable to the pulse signals obtained through cameras, and it is quite challenging to collect such a large-scale and high-quality physiological dataset.

[0037] Based on this, the present application provides a method for blood pressure measurement. By extracting target fingertip pulse signals from the video data of a camera covering the fingertip of the subject and performing variational mode decomposition and reconstruction operations on the target fingertip pulse signals, it is possible to effectively reduce environmental noise and motion artifacts and retain the key features of the pulse signals. Further, through data augmentation operations combined with meta-learning methods, the model can quickly adapt and learn from a small amount of data, significantly improving the accuracy and efficiency of blood pressure measurement.

[0038] Multiple embodiments of the embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0039] Figure 1 is an exemplary flowchart showing a method 100 for blood pressure measurement according to an embodiment of the present application. As Figure 1 shown, the method 100 includes: Step S101: Extracting target fingertip pulse signals from the video data of a camera covering the fingertip of the subject; Step S102: Performing variational mode decomposition and reconstruction operations and data augmentation operations on the target fingertip pulse signals to obtain effective fingertip pulse signals; Step S103: Inputting the effective fingertip pulse signals into a meta-learning model for blood pressure measurement meta-learning to obtain a trained meta-learning model; Step S104: Inputting the target fingertip pulse signals into the trained meta-learning model for blood pressure measurement to obtain a blood pressure measurement result.

[0040] First, at step S101, a target fingertip pulse signal is extracted from the video data in which the fingertip of the subject covers the camera. In some implementation scenarios, the video data in which the fingertip of the subject covers the camera can be collected by an intelligent device with a camera, such as collecting the video data in which the fingertip of the subject covers the camera through a smartphone with a camera, a tablet computer, a PC, etc. During the collection process, it is required that the subject gently place the fingertip on the camera and keep still for a certain period of time (such as 8 seconds) to ensure that the subtle pulsations of the fingertip can be fully captured.

[0041] Based on this video data, in some embodiments, the video data can be segmented into single-frame data, and then the single-frame data is decomposed into color channels to obtain multiple color channels in each frame of data, and then the average value of the target color channel in each frame of data is calculated to extract the target fingertip pulse signal. Among them, the target color channel includes the R color channel.

[0042] It can be understood that each frame represents a discrete time point in the sequence, and the frame extraction process is consistent with the frame rate of the video to ensure that each frame can be analyzed consistently. Each frame is a snapshot of the fingertip, containing important data on color changes caused by blood flow changes. Based on the extracted single-frame data, the single-frame data is then decomposed into color channels, that is, the image is decomposed into three basic color channels: red (R), green (G), and blue (B), and the R channel is used as the target color channel. This is because the R channel is more sensitive to blood volume changes during each heartbeat. By focusing on the R channel, the complexity of signal processing can be reduced and the calculation efficiency can be improved. Further, the average value of the target color channel in each frame of data is calculated, and the sequence of average red channel values in each frame forms a time series signal, that is, the target fingertip pulse signal.

[0043] Specifically, in one implementation scenario, the average value of the target color channel (i.e., the R channel) in each frame can be calculated by the following formula:

[0044]

[0045] where, R avg，i represents the average value of the red channel in the i-th frame, R i，j represents the red channel value of the j-th pixel in this frame, and N is the total number of pixels in this frame. The time series signal (i.e., the target fingertip pulse signal) formed by the sequence of average red channel values in each frame can be expressed as S(t) = {R avg，1 , R avg，2 , R avg，3 ,..., R avg，n}, and this target fingertip pulse signal contains the change of blood volume over time, providing a continuous signal that can be analyzed.

[0046] Next, at step S102, a variational mode decomposition reconstruction operation and a data augmentation operation are performed on the target fingertip pulse signal to obtain an effective fingertip pulse signal. It should be understood that the normal heart rate range is between 1 Hz and 2 Hz, and simple filtering cannot effectively eliminate the body movement interference that usually appears in the frequency range of 0.1 to 2 Hz. Therefore, in the embodiments of the present application, a pure fingertip pulse signal is separated through a variational mode decomposition reconstruction operation.

[0047] In some embodiments, the target fingertip pulse signal can be decomposed into a set of intrinsic mode functions (IMFs), the first intrinsic mode function representing the low-frequency heartbeat and the second intrinsic mode function representing the high-frequency heartbeat are selected from the set of intrinsic mode functions, and then the first intrinsic mode function and the second intrinsic mode function are superimposed and reconstructed to obtain a pure superimposed and reconstructed fingertip pulse signal. Among them, in some embodiments, a variational constraint model can be constructed based on the target fingertip pulse signal, a second-order penalty factor and a Lagrange multiplier are introduced into the variational constraint model to form an augmented Lagrangian function, and then the augmented Lagrangian function is iterated using the alternating direction multiplier method to be decomposed into a set of intrinsic mode functions.

[0048] For example, in an exemplary scenario, assuming that the number of a set of intrinsic mode functions to be decomposed is K, its variational constraint model can be expressed as:

[0049]

[0050] Among them, represents the partial derivative with respect to time t, υ(t) represents the unit impulse function, j represents the imaginary unit, * represents the convolution operation, and the function ζ k (t) corresponds to the kth intrinsic mode function, and w k (t) represents the central frequency of ζ k (t).

[0051] In some implementation scenarios, a second-order penalty factor β and a Lagrange multiplier λ(t) can be introduced to form an augmented Lagrangian function, and the augmented Lagrangian function can be expressed as follows:

[0052]

[0053] Then, the augmented Lagrangian function is iterated, for example, using the alternating direction multiplier method to be decomposed into a set of intrinsic mode functions. As an example, assuming that the number of intrinsic mode functions K = 10 is set, the target fingertip pulse signal can be decomposed into 10 intrinsic mode functions. These intrinsic mode functions can be selected through fast Fourier transform (FFT) analysis according to their frequency content. For example, the first intrinsic mode function IMF2 representing the low-frequency heartbeat and the second intrinsic mode function IMF10 representing the high-frequency heartbeat are selected.

[0054] Furthermore, the first intrinsic mode function IMF2 and the second intrinsic mode function IMF10 are superimposed and reconstructed. The fingertip pulse signal after superimposed reconstruction retains the pulse signal with the basic characteristics of the heartbeat, combines relevant high and low frequency components, and improves the fidelity of the heartbeat signal while reducing noise.

[0055] In addition, when the number of training samples is small, the model may not be able to effectively capture potential patterns, resulting in poor generalization ability and overfitting. Therefore, the embodiments of the present application enrich and supplement the data set through data augmentation operations to reduce the need for a large number of blood samples. In some embodiments, time-frequency transformation can be performed on the target fingertip pulse signal or the fingertip pulse signal after superimposed reconstruction to separate the heartbeat frequency, and the heartbeat frequency is scaled; the inverse time-frequency transformation is performed on the scaled heartbeat frequency to implement the data augmentation operation. Among them, the aforementioned time-frequency transformation and inverse time-frequency transformation can be realized by Fourier transform and inverse Fourier transform respectively. By scaling the heartbeat frequency, the changes in heart rate and blood flow can be simulated, and then the signal can be converted back to the time domain through inverse Fourier transform to generate diverse and physiologically relevant fingertip pulse signals, that is, effective fingertip pulse signals. The training set is enriched through data augmentation operations, overfitting is avoided, and the generalization ability of the model is improved.

[0056] In some implementation scenarios, normalization operations can also be performed on the fingertip pulse signal after superimposed reconstruction or the fingertip pulse signal after inverse time-frequency transformation to obtain effective fingertip pulse signals. Based on the normalization operation, the influence of different lighting conditions can be reduced. Subsequently, when the effective fingertip pulse signal is used as the input of the meta-learning model, not only the key information is ensured to be retained, but also the influence of noise is minimized.

[0057] Furthermore, at step S103, the effective fingertip pulse signal is input into the meta-learning model for meta-learning of blood pressure measurement to obtain a trained meta-learning model. Specifically, the embodiments of the present application update the parameters of the meta-learning model through model-agnostic meta-learning (MAML). MAML provides a robust initialization, thus enabling faster and more efficient training on new and unseen tasks, that is, fewer data samples are required.

[0058] In some embodiments, the meta-learning model may include multiple convolutional layers, multiple bidirectional long short-term memory layers, multiple fully connected layers, and an output layer, and each convolutional layer includes a sub-convolutional layer, a batch normalization layer, and a pooling layer, and the output layer includes a fully connected layer. This structured method can effectively perform feature extraction and sequence modeling, which is crucial for accurately predicting blood pressure from pulse signals.

[0059] It can be understood that meta - learning is not a mechanism for training a separate model for each individual. Instead, it learns the representations shared by all subjects and fine - tunes the model using a small amount of personalized data to adapt to new subjects. This enables the model to take into account individual differences while avoiding overfitting to the training data. In the embodiments of the present application, the potential problem of overfitting is also addressed by using support sets and query sets, ensuring that the model learns to adapt rather than memorize individual patterns. Training on different individuals and adapting to new individuals helps maintain the robustness and accuracy of the model. Due to the significant individual differences in blood pressure signals, personalized models are crucial. One - size - fits - all models may not perform well in different populations. The meta - learning framework creates a model that is both adaptable and generalizable, capable of capturing the subtle individual - specific changes that are crucial for accurate estimation.

[0060] Based on the obtained trained meta - learning model, at step S104, the target fingertip pulse signal is input into the trained meta - learning model for blood pressure measurement, and a blood pressure measurement result is obtained. In some implementation scenarios, the blood pressure measurement result may include, for example, systolic blood pressure and diastolic blood pressure. That is, by inputting the target fingertip pulse signal extracted from the video data of the fingertip covering the camera of the subject into the trained meta - learning model, the estimated values of systolic blood pressure and diastolic blood pressure can be directly obtained.

[0061] Combined with the above description, in the embodiments of the present application, by collecting fingertip video data based on the camera of the smart device and extracting the target fingertip pulse signal, the dependence on cuffs or dedicated sensors (such as PPG / ECG) in traditional blood pressure measurement is avoided, significantly reducing the hardware cost and improving the usability. Secondly, the variational mode decomposition reconstruction technique is used to separate noise and reconstruct the target fingertip pulse signal, effectively suppressing the interference of environmental light fluctuations and motion artifacts. At the same time, the diversity of training samples is expanded through data augmentation operations, enhancing the adaptability of the model to individual physiological differences (such as heart rate fluctuations). Further, by combining the fast adaptation ability of the meta - learning model (MAML), a personalized blood pressure prediction model can be generated using a small amount of subject data, solving the limitation of traditional machine learning methods that rely on large - scale labeled data. On the premise of ensuring measurement accuracy, the generalization ability across devices and subjects is achieved, and it is applicable to various intelligent terminals such as smartphones and tablets, providing an efficient and universal technical solution for non - invasive blood pressure monitoring.

[0062] Figure 2 is an exemplary schematic diagram showing blood pressure measurement according to an embodiment of the present application. As Figure 2As shown in the figure, video data 203 of the subject's fingertip covering the camera of the smart device 201 is collected through the camera of the smart device 201. Based on the video data, it is segmented into single-frame data, and the single-frame data is decomposed into three basic color channels of R, G, and B. Among them, the R channel is selected as the target color channel, and the average value of the R channel in each frame of data is calculated based on the above formula (1) to obtain the target fingertip pulse signal 204. Further, by inputting the target fingertip pulse signal 204 into the trained meta-learning model 205 for blood pressure measurement, the estimated values of systolic blood pressure and diastolic blood pressure can be obtained, and the blood pressure measurement result can be read by the smart device 201.

[0063] As can be seen from the foregoing, the meta-learning model 205 can be trained based on the effective fingertip pulse signal obtained after performing variational mode decomposition reconstruction operation and data augmentation operation on the target fingertip pulse signal. Among them, regarding the variational mode decomposition reconstruction operation, first, the target fingertip pulse signal 204 can be decomposed into a group of intrinsic mode functions based on the above formula (2) and formula (3), and the first intrinsic mode function IMF2 of low-frequency heartbeat and the second intrinsic mode function IMF10 of high-frequency heartbeat are selected from them. Then, IMF2 and IMF10 are superimposed and reconstructed to obtain the superimposed and reconstructed fingertip pulse signal. For the data augmentation operation, the heartbeat frequency is separated by, for example, Fourier transform, the heartbeat frequency is scaled, and then the inverse Fourier transform is performed on the scaled heartbeat frequency to implement the data augmentation operation. Based on this, an effective fingertip pulse signal can be obtained, and the effective fingertip pulse signal is used as the input of the meta-learning model 205 for training to obtain the trained meta-learning model 205. Among them, for more details about the foregoing content, reference can be made to Figure 1 the description, which is not elaborated herein in this application.

[0064] Figure 3 is an exemplary schematic diagram showing the visualization of the R channel when the blood flow is maximum and minimum in the corresponding frame data according to an embodiment of the present application. As Figure 3 shown in the (a) figure of, the visualization result of the R channel when the corresponding frame data has the maximum blood flow, the visualization result of the R channel when the corresponding frame data has the minimum blood flow shown in the (b) figure, and the curves a and b in the (c) figure are respectively the average value of the R channel when the blood flow is maximum and the average value of the R channel when the blood flow is minimum. The abscissa represents the frame index, and the ordinate represents the average value. Since the R channel has higher sensitivity to the change in blood volume during each heartbeat, the embodiment of the present application focuses on the R channel analysis, which can not only improve the accuracy of fingertip pulse signal analysis, but also reduce the complexity of signal processing and improve the calculation efficiency.

[0065] Figure 4 is an exemplary schematic diagram showing the variational mode decomposition reconstruction result according to an embodiment of the present application. As Figure 4Figures (a), (b), (c), and (d) in

[0066] Figure 5 successively show the signal after variational mode decomposition, the IMF2 signal, the IMF10 signal, and the fingertip pulse signal after superimposed reconstruction. Among them, the abscissa represents time, and the ordinate represents the signal amplitude. As mentioned above, through variational mode decomposition and reconstruction, the pulse signal with the basic characteristics of the heartbeat can be retained, which combines relevant high and low frequency components, and can improve the fidelity of the heartbeat signal while reducing noise. Figure 5 is an exemplary schematic diagram showing the meta-learning model according to an embodiment of the present application. As

[0067] shown in it, the meta-learning model may include a plurality of convolutional layers 501, a plurality of bidirectional long short-term memory layers 502, a plurality of fully connected layers 503, and an output layer 504. For example, 5 convolutional layers 501, 3 bidirectional long short-term memory layers 502, 3 fully connected layers 503, and an output layer 504 are exemplarily shown in the figure. Among them, each convolutional layer 501 may include a sub-convolutional layer, a batch normalization layer, and a pooling layer, and the output layer 504 may include a fully connected layer. In an implementation scenario, by inputting the target fingertip pulse signal 204 into the meta-learning model, and successively passing through the convolutional layer 501, the bidirectional long short-term memory layer 502, the fully connected layer 503, and the output layer 504, a blood pressure detection result 505 is output, including, for example, systolic blood pressure (SBP) and diastolic blood pressure (DBP). Among them, is the personalized parameter, θ is the global initial parameter, and α is the learning rate (for example, 0.001). Then, the loss is calculated through the query set to reversely optimize the global parameter: where β is the learning rate (for example, 0.001), and the training batch contains 30 subject tasks. In this way, the model can quickly adapt to the individual differences of different subjects, and only a small amount of data is required to construct a personalized blood pressure estimation model.

[0068] Figure 6 is an exemplary structural block diagram showing a system 600 for blood pressure measurement according to an embodiment of the present application. As Figure 6As shown in the figure, the system 600 may include an acquisition module 601, which is used to acquire video data of the fingertips of the subject covering the camera and extract the target fingertip pulse signal from the video data of the fingertips of the subject covering the camera. A preprocessing module 602, which is used to perform variational mode decomposition reconstruction operation and data enhancement operation on the target fingertip pulse signal to obtain an effective fingertip pulse signal. A meta-learning training module 603, which is used to input the effective fingertip pulse signal into the meta-learning model for meta-learning of blood pressure measurement to obtain a trained meta-learning model. A blood pressure measurement module 604, which is used to input the target fingertip pulse signal into the trained meta-learning model for blood pressure measurement to obtain a blood pressure measurement result.

[0069] In some implementation scenarios, the acquisition module 601 may include devices such as a smartphone with a camera, a tablet computer, a PC, etc. The preprocessing module 602, the meta-learning training module 603, and the blood pressure measurement module 604 may be implemented through an overall data processing module. Specifically, they may be implemented through existing data processing software or through software code. Among them, the specific operations of each module correspond to the Figure 1 method, so the relevant details can be referred to the above Figure 1 description, and will not be elaborated herein.

[0070] To evaluate the reliability of the blood pressure measurement results of the embodiments of the present application, a camera module and intelligent devices (mobile phones, tablets, computers) with a frame rate of 30 Hz and a resolution of 1920×1080 are used, and a fingertip video and blood pressure ground truth of 30 subjects (24 males / 6 females, covering different BMIs, skin tones, and blood pressure levels) are obtained in combination with an arm-type electronic blood pressure monitor to ensure data diversity.

[0071] Figure 7 is an exemplary Bland-Altman plot showing the systolic and diastolic blood pressures obtained from the evaluation according to the embodiments of the present application. As shown on the left and right in Figure 7 , the average systolic blood pressure and the average diastolic blood pressure are respectively shown exemplarily. The abscissa represents the reference blood pressure value, and the ordinate represents the prediction error. Among them, ME represents the average value of the difference between the predicted blood pressure value and the true blood pressure value, and is used to quantify the overall deviation direction and degree of the prediction result. If ME is positive, it means that the predicted value is generally on the high side; if it is negative, it is generally on the low side. For example, for systolic blood pressure, ME = 1.37 mmHg, indicating that the predicted systolic blood pressure is on average 1.37 mmHg higher than the true value. For diastolic blood pressure, ME = 0.82 mmHg, indicating that the predicted diastolic blood pressure is on average 0.82 mmHg higher than the true value. Based on this, it can be shown that the average deviation between the prediction results of the embodiments of the present application and the true blood pressure values is small (the ME of SBP and DBP are both close to 0), and 95% of the errors fall within the clinically acceptable range, verifying the accuracy and reliability of the model.

[0072] Furthermore, in the embodiments of the present application, the cumulative distribution function percentage of the experimental results is also compared with the BHS standard, as shown in Table 1 below.

[0073] Table 1 Comparison of the cumulative distribution function percentage of the experimental results and the BHS standard

[0074]

[0075] It can be understood that Table 1 shows the cumulative percentages at different error thresholds (5 mmHg, 10 mmHg, 15 mmHg), that is, the proportion of samples with a prediction error less than or equal to the corresponding threshold. BHS defines the requirements for blood pressure prediction errors for different grades (Grade A / B / C), which is an important clinical basis for evaluating the accuracy of blood pressure measurement devices, and divides grades through the cumulative percentages of the three error thresholds. For example, Grade C: 110% of the error < 5 mmHg, 65% < 10 mmHg, 85% < 15 mmHg; Grade B: 50% of the error < 5 mmHg, 75% < 10 mmHg, 90% < 15 mmHg; Grade A (the highest grade): 60% of the error < 5 mmHg, 85% < 10 mmHg, 95% < 15 mmHg, indicating extremely high clinical practicability.

[0076] As can be seen from Table 1, the actual cumulative percentages of the systolic blood pressure (SBP) and diastolic blood pressure (DBP) prediction errors obtained based on the blood pressure measurement method of the present application at each threshold all reach or far exceed the requirements of Grade A, proving that its measurement accuracy meets the clinical use standard. Combining the above Figure 7 which shows that 95% of the errors fall within the range of ME ± 1.96×STD (SBP: -13.11 to 15.85 mmHg, DBP: -10.61 to 12.25 mmHg), and the cumulative percentage at the 15 mmHg threshold in Table 1 (SBP 95.88%, DBP 97.98%) is close to the 95% confidence interval, indicating that the experimental data distribution conforms to the statistical expectation, further verifying the stability of the blood pressure measurement of the present application.

[0077] Figure 8 is an exemplary schematic diagram showing the comparison of the mean absolute value (MAE) of blood pressure measurement for different groups of subjects to be examined according to the embodiments of the present application. As Figure 8Figure (a) shows the result graph demonstrating gender differences (male vs. female). The horizontal axis represents gender (male, female), and the vertical axis represents MAE (mmHg), divided into systolic blood pressure (SBP) and diastolic blood pressure (DBP). Among them, the male SBP MAE is 8.18 mmHg and the DBP is 6.15 mmHg; the female SBP MAE is 6.41 mmHg and the DBP is 4.10 mmHg, with the error significantly lower than that of males (about 21.6% lower). This is because the female skin is usually thinner, with better light penetration, and the change in blood volume captured by the camera is clearer, verifying the influence of signal acquisition quality on measurement accuracy and indirectly reflecting the enhancement effect of variational mode decomposition reconstruction denoising on weak signals.

[0078] As Figure 8 Figure (b) shows age differences (<25 years old vs. ≥25 years old). The horizontal axis represents age groups, and the vertical axis represents MAE (mmHg), with colors distinguishing SBP / DBP. Among them, for the <25-year-old group: SBP MAE is 7.82 mmHg and DBP is 5.23 mmHg; for the ≥25-year-old group: SBP MAE is 7.95 mmHg and DBP is 5.31 mmHg, and the error difference between groups is <1 mmHg. The influence of age on vascular elasticity is not significantly reflected in the measurement error, indicating that the meta-learning model can effectively capture the common physiological characteristics of different age groups and reduce the interference of age-related individual differences.

[0079] As Figure 8 Figure (c) shows skin tone differences (light / normal / dark, Fitzpatrick classification). The horizontal axis represents skin tone (light type III, normal type IV, dark type V), and the vertical axis represents MAE (mmHg). Among them, for light skin: SBP is 5.27 mmHg and DBP is 5.50 mmHg; for normal skin: SBP is 7.32 mmHg and DBP is 5.89 mmHg; for dark skin: SBP is 8.63 mmHg and DBP is 7.59 mmHg (with the highest error, 63.8% higher than the light skin group). Dark skin absorbs more light, resulting in a lower signal-to-noise ratio of the signals collected by the camera. The variational mode decomposition reconstruction denoising and the personalized adaptation of meta-learning compensate for this difference to a certain extent, but the signal preprocessing algorithm for the dark skin scenario still needs to be further optimized.

[0080] As Figure 8Figure (d) shows the BMI differences (<20 / 20 - 25 / >25). The horizontal axis represents the BMI groups (underweight, normal, overweight), and the vertical axis represents the MAE (mmHg). Among them, for BMI < 20: SBP is 6.15 mmHg and DBP is 4.82 mmHg; for BMI 20 - 25 (normal): SBP is 7.21 mmHg and DBP is 5.63 mmHg; for BMI > 25 (overweight): SBP is 8.47 mmHg and DBP is 6.91 mmHg. In overweight people, due to the relatively thick subcutaneous fat, it may affect the light penetration depth, resulting in a decrease in signal quality. However, the error is still controlled within the clinically acceptable range (<9 mmHg), demonstrating the robustness of the model to different body types.

[0081] Figure 9 is an exemplary schematic diagram showing the comparison of the influence of camera parameters on the mean absolute value of blood pressure measurement according to an embodiment of the present application. As Figure 9 shown in Figure (a), it is for Auto - White - Balance (AWB) on vs. off. The horizontal axis represents the AWB state (on, off), and the vertical axis represents the MAE (mmHg). The colors distinguish SBP / DBP. Among them, for AWB on: SBP is 7.12 mmHg and DBP is 5.45 mmHg; for AWB off: SBP is 8.36 mmHg and DBP is 6.72 mmHg (the error increases by approximately 17.4%). AWB ensures the video color restoration by automatically correcting the ambient light color temperature, avoiding the deviation of the R - channel pixel values caused by color cast of light, and is a key pre - step for signal pre - processing.

[0082] As Figure 9 shown in Figure (b), it is for Auto - Exposure (AE) on vs. off. The horizontal axis represents the AE state, and the vertical axis represents the MAE (mmHg). Among them, for AE on: SBP is 6.98 mmHg and DBP is 5.31 mmHg; for AE off (fixed low exposure): SBP is 9.25 mmHg and DBP is 7.89 mmHg (the error increases significantly by approximately 32.5%). AE controls the light input by dynamically adjusting the exposure time. Turning off AE will cause the video to be too dark, reducing the pixel value difference corresponding to the blood volume change and increasing the difficulty of signal extraction, verifying the direct influence of the exposure parameters on the quality of the original signal.

[0083] As Figure 9As shown in Figure (c), the frame rate is presented (30FPS vs. 60FPS). The horizontal axis represents the frame rate, and the vertical axis represents the MAE (mmHg). Among them, for 30FPS: SBP is 7.23 mmHg, and DBP is 5.56 mmHg; for 60FPS: SBP is 7.19 mmHg, and DBP is 5.48 mmHg. There is no significant difference in the error between the two groups (p > 0.05). 30FPS already meets the Nyquist sampling requirements for the heartbeat signal (0.5 - 2Hz). A higher frame rate does not improve the accuracy. It is recommended to use 30FPS to reduce the consumption of computing resources and be compatible with low-end devices.

[0084] As shown in Figure 9 Figure (d), the device type (mobile phone / tablet / computer) is presented. The horizontal axis represents different types of devices, and the vertical axis represents the MAE (mmHg). Among them, for the mobile phone: SBP is 6.89 / 7.02 mmHg, and DBP is 5.21 / 5.35 mmHg; for the tablet: SBP is 8.11 mmHg, and DBP is 6.42 mmHg; for the PC camera: SBP is 8.97 mmHg, and DBP is 7.18 mmHg (the highest error). When holding the mobile phone, the fingertips are more stable in contact with the camera, reducing motion artifacts; due to the larger volume of the tablet and PC camera, slight jitters are likely to occur during user operation, resulting in an increase in signal noise, reflecting the impact of device portability on measurement stability.

[0085] Figure 10 It is an exemplary schematic diagram showing the comparison of the average absolute values of blood pressure measurements under different lighting conditions according to an embodiment of the present application. As shown in Figure 10 Figure (a), the corresponding MAE (mmHg) under LED, bright light, and dim light without a flash is presented. Among them, for LED light: SBP MAE is 7.02 mmHg, and DBP is 5.38 mmHg; for bright light: SBP MAE is 6.95 mmHg, and DBP is 5.29 mmHg (the lowest error because the signal contrast is the highest under strong light); for dim light: SBP MAE is 9.21 mmHg, and DBP is 7.78 mmHg (the error increases significantly because the pixel values fluctuate greatly due to insufficient light intensity). In a strong light environment (LED / natural light), the dynamic range of the R-channel pixel values captured by the camera is large (about 100 - 200 / 255), and the signal characteristics corresponding to blood volume changes are obvious; in a dim light environment, the pixel values are concentrated in a low range (about 50 - 100 / 255), and the proportion of noise increases, and signal preprocessing is required to extract effective components.

[0086] As shown in Figure 10Figure (b) shows the LED under the flash, and the corresponding MAE (mmHg) under bright and dim light. Among them, for LED light: SBP MAE is 7.15 mmHg, DBP is 5.51 mmHg (the error is close to that without the flash because the LED light is sufficient and the flash does not significantly improve the signal); for bright light: SBP MAE is 7.22 mmHg, DBP is 5.63 mmHg (the error slightly increases because the strong light + flash causes some pixels to be overexposed and the signal is saturated); for dim light: SBP MAE is 6.83 mmHg, DBP is 5.62 mmHg (the error is significantly reduced, which is 25.8% lower than the dim light without the flash in Figure (a) of Figure 10 , and the flash fill light effectively improves the signal quality). In a bright light environment, the flash, as an active light source, raises the light intensity to the range suitable for signal acquisition (600 lux). The light reflection differences caused by blood volume changes are more obvious, and the high-frequency flicker of the flash (about 100 Hz) can be effectively separated by VMD (the heartbeat signal is 0.5 - 2 Hz, without frequency overlap); turning on the flash in a bright light environment will cause the light intensity to be too strong (>1500 lux), and some pixel values will be saturated (>240 / 255), which instead introduces non-linear distortion. Therefore, the MAE is slightly higher than the condition without the flash.

[0087] The above evaluation experiments verify the high precision, strong robustness, and generalization ability of the blood pressure measurement method of the embodiment of the present application. Relying only on the camera of the intelligent device, through variational mode decomposition reconstruction operation, data augmentation operation, and meta-learning personalized adaptation, it maintains reliable performance under complex environments and individual differences, providing a practical solution for low-cost and popular blood pressure monitoring.

[0088] Figure 11 is an exemplary structural block diagram showing the electronic device 1100 according to the embodiment of the present application. It can be understood that the electronic device 1100 may include the device of the embodiment of the present application, and the device implementing the solution of the present application may be a single device (such as a computing device) or a multifunctional device including various peripheral devices.

[0089] Such as Figure 11As shown, the electronic device of the present application may further include a central processing unit or central processing unit (“CPU”) 1111, which may be a general-purpose CPU, a dedicated CPU, or other information processing and program execution units. Further, the electronic device 1100 may also include a mass storage 1112 and a read-only memory (“ROM”) 1113, where the mass storage 1112 may be configured to store various types of data, including various video data, target fingertip pulse signals, valid fingertip pulse signals, blood pressure measurement results, algorithm data, intermediate results, and various programs required to operate the electronic device 1100. The ROM 1113 may be configured to store data and instructions for power-on self-test of the electronic device 1100, initialization of each functional module in the system, basic input / output drivers of the system, and data and instructions required to boot the operating system.

[0090] Optionally, the electronic device 1100 may further include other hardware platforms or components, such as the shown tensor processing unit (“TPU”) 1114, graphics processing unit (“GPU”) 1115, field programmable gate array (“FPGA”) 1116, and machine learning unit (“MLU”) 1117. It can be understood that although various hardware platforms or components are shown in the electronic device 1100, these are merely exemplary and not restrictive, and those skilled in the art can add or remove corresponding hardware according to actual needs. For example, the electronic device 1100 may include only a CPU, related storage devices, and interface devices to implement the method for blood pressure measurement of the present application.

[0091] In some embodiments, for the convenience of data transfer and interaction with an external network, the electronic device 1100 of the present application further includes a communication interface 1118, so that it can be connected to a local area network / wireless local area network (“LAN / WLAN”) 1105 through the communication interface 1118, and then can be connected to a local server 1106 or connected to the Internet (“Internet”) 1107 through the LAN / WLAN. Alternatively or additionally, the electronic device 1100 of the present application may also be directly connected to the Internet or a cellular network based on wireless communication technology through the communication interface 1118, such as based on the 3rd generation (“3G”), 4th generation (“4G”), or 5th generation (“5G”) wireless communication technology. In some application scenarios, the electronic device 1100 of the present application may also access servers 1108 and databases 1109 of an external network as needed to obtain various known algorithms, data, and modules, and may remotely store various data, such as various types of data or instructions for presenting video data, target fingertip pulse signals, valid fingertip pulse signals, blood pressure measurement results, etc.

[0092] The peripheral devices of the electronic device 1100 may include a display device 1102, an input device 1103, and a data transmission interface 1104. In one embodiment, the display device 1102 may include, for example, one or more speakers and / or one or more visual displays, which are configured to provide voice prompts and / or display image videos for blood pressure measurement in this application. The input device 1103 may include, for example, other input buttons or controls such as a keyboard, a mouse, a microphone, a gesture capture camera, etc., which are configured to receive input of audio data and / or instructions from the subject. The data transmission interface 1104 may include, for example, a serial interface, a parallel interface, or a Universal Serial Bus interface ("USB"), a Small Computer System Interface ("SCSI"), Serial ATA, FireWire ("FireWire"), PCI Express, and a High-Definition Multimedia Interface ("HDMI"), etc., which are configured for data transmission and interaction with other devices or systems. According to the solution of this application, the data transmission interface 1104 may receive video data collected by a smart device camera and transmit to the electronic device 1100 data or results including the video data or various other types of data.

[0093] The above-mentioned CPU 1111, mass storage 1112, ROM 1113, TPU 1114, GPU 1115, FPGA 1116, MLU 1117, and communication interface 1118 of the electronic device 1100 of this application may be interconnected with each other through a bus 1119 and achieve data interaction with the peripheral devices through this bus. In one embodiment, through the bus 1119, the CPU 1111 may control other hardware components and their peripheral devices in the electronic device 1100.

[0094] The above combination Figure 11 has described the electronic device that can be used to execute this application. It should be understood that the device structure or architecture here is only exemplary, and the implementation manner and implementation entity of this application are not limited by it, but can be changed without departing from the spirit of this application.

[0095] According to the above description in combination with the accompanying drawings, those skilled in the art can also understand that the embodiments of this application can also be implemented through software programs. Therefore, this application also provides a computer-readable storage medium, on which computer-readable instructions for blood pressure measurement are stored. When the computer-readable instructions are executed by one or more processors, they can be used to implement the method for blood pressure measurement described in this application in combination with the attached Figure 1 drawings.

[0096] It should be noted that although the operations of the method of the present application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the order of execution of the steps depicted in the flowchart can be changed. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step for execution, and / or one step may be decomposed into multiple steps for execution.

[0097] It should be understood that when terms such as "first", "second", "third", and "fourth" are used in the claims, the specification, and the accompanying drawings of the present application, they are only used to distinguish different objects and not to describe a specific order. The terms "comprising" and "including" used in the specification and claims of the present application indicate the presence of the described features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0098] It should also be understood that the terms used in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the specification and claims of the present application, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms. It should be further understood that the term "and / or" used in the specification and claims of the present application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0099] Although the embodiments of the present application are as described above, the above content is only an example used for easy understanding of the present application and is not intended to limit the scope and application scenarios of the present application. Any person skilled in the technical field related to the present application may make any modifications and changes in the form of implementation and details without departing from the spirit and scope disclosed by the present application. However, the scope of patent protection of the present application shall still be subject to the scope defined by the appended claims.

[0100] In addition, the collection and acquisition of various data in the present application comply with relevant laws and regulations and are authorized by the data provider. Any organization or individual that needs to obtain external data should obtain authorization in accordance with the law and ensure data security, and shall not illegally collect, use, process, or transmit unauthorized or unprotected data, nor illegally buy, provide, or disclose unauthorized or unprotected data.

Claims

1. A method for measuring blood pressure, comprising: Extracting the target fingertip pulse signal from the video data of the camera covering the fingertip of the subject; Performing a variational mode decomposition reconstruction operation and a data enhancement operation on the target fingertip pulse signal to obtain a valid fingertip pulse signal; Inputting the effective fingertip pulse signal into the meta-learning model to perform blood pressure measurement meta-learning to obtain a trained meta-learning model; The target fingertip pulse signal is input into the trained meta-learning model to measure blood pressure and obtain a blood pressure measurement result.

2. The method according to claim 1, wherein extracting the target fingertip pulse signal from the video data of the subject's fingertip covering the camera comprises: Splitting the video data into single frame data; Decomposing the single frame data into color channels to obtain multiple color channels in each frame data; The average value of the target color channel in each frame of data is calculated to extract the target fingertip pulse signal, wherein the target color channel includes the R color channel.

3. The method according to claim 1, wherein the variational mode decomposition reconstruction operation is performed by: Decomposing the target fingertip pulse signal into a set of intrinsic mode functions; Selecting a first intrinsic modal function representing a low-frequency heartbeat and a second intrinsic modal function representing a high-frequency heartbeat from the set of intrinsic modal functions; The first intrinsic mode function and the second intrinsic mode function are superimposed and reconstructed.

4. The method of claim 3, wherein decomposing the target fingertip pulse signal into a set of intrinsic mode functions comprises: Constructing a variational constraint model based on the target fingertip pulse signal; Introducing a second-order penalty factor and a Lagrangian multiplier into the variational constraint model to form an augmented Lagrangian function; The augmented Lagrangian function is iterated using an alternating direction multiplier method to decompose into the set of intrinsic mode functions.

5. The method according to claim 3, wherein the data enhancement operation is performed by: Performing time-frequency transformation on the target fingertip pulse signal or the superimposed and reconstructed fingertip pulse signal to separate the heartbeat frequency; Scaling the heartbeat frequency; Perform inverse time-frequency transform on the scaled heartbeat frequency.

6. The method according to claim 5, further comprising: A normalization operation is performed on the fingertip pulse signal after superposition and reconstruction or the fingertip pulse signal after time-frequency inverse transformation to obtain the effective fingertip pulse signal.

7. The method according to claim 1, wherein the meta-learning model comprises multiple convolutional layers, multiple bidirectional long short-term memory layers, multiple fully connected layers and an output layer, and each of the convolutional layers comprises a sub-convolutional layer, a batch normalization layer and a pooling layer, and the output layer comprises a fully connected layer.

8. A system for measuring blood pressure, comprising: An acquisition module, which is used to acquire video data of the subject's fingertips covering the camera, and extract a target fingertip pulse signal from the video data of the subject's fingertips covering the camera; A preprocessing module, which is used to perform a variational mode decomposition reconstruction operation and a data enhancement operation on the target fingertip pulse signal to obtain a valid fingertip pulse signal; A meta-learning training module, which is used to input the effective fingertip pulse signal into the meta-learning model to perform blood pressure measurement meta-learning to obtain a trained meta-learning model; The blood pressure measurement module is used to input the target fingertip pulse signal into the trained meta-learning model to perform blood pressure measurement and obtain a blood pressure measurement result.

9. An electronic device, comprising: processor; as well as A memory having computer instructions for blood pressure measurement stored thereon, wherein when the computer instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented; or the operations performed by the system according to claim 8 are implemented.

10. A computer-readable storage medium having stored thereon computer program instructions for blood pressure measurement, wherein when the computer program instructions are executed by one or more processors, the method according to any one of claims 1 to 7 is implemented; or the operations performed by the system according to claim 8 are implemented.

Images