CN119326393A

CN119326393A - Blood pressure measurement method, wearable device and cloud device

Info

Publication number: CN119326393A
Application number: CN202411780887.0A
Authority: CN
Inventors: 刘泽新
Original assignee: Honor Device Co Ltd
Current assignee: Honor Device Co Ltd
Priority date: 2024-12-05
Filing date: 2024-12-05
Publication date: 2025-01-21
Anticipated expiration: 2044-12-05
Also published as: CN119326393B

Abstract

The present application provides a blood pressure measurement method, a wearable device and a cloud device. In this method, in response to the user's blood pressure measurement operation, the user's photoplethysmography (PPG) signal is collected to obtain PPG signal data; based on the PPG signal data and the trained blood pressure prediction model, the user's blood pressure measurement result is obtained; the target training sample is obtained by sampling the original training sample set, and the difference between any two sample PPG signal data in the target training sample set is greater than a preset threshold, and the target training sample set contains the target sample PPG signal data, and the original training sample set does not contain the target sample PPG signal data. In this way, by sampling the training set, the data redundancy of the training set is reduced and the gaps between the data in the training set are filled, so as to achieve the effect of improving the generalization of the blood pressure prediction model, thereby improving the blood pressure prediction accuracy of the blood pressure prediction model for unseen individuals.

Description

Blood pressure measurement method, wearable device and cloud device

Technical Field

The application relates to the technical field of intelligent terminals, in particular to a blood pressure measurement method, wearable equipment and cloud equipment.

Background

Blood pressure measurement is a common health assessment means, and along with the development of technology, blood pressure measurement equipment is also smaller and lighter. For example, the blood pressure measurement device may be a smart wearable device. The user who needs to carry out blood pressure measurement can wear intelligent wearing equipment for a long time to use intelligent wearing equipment at any time to carry out blood pressure measurement.

The existing intelligent wearable equipment generally acquires relevant physiological signals of a user and inputs the relevant physiological signals into a trained blood pressure prediction model to obtain a blood pressure value of the user. Blood pressure prediction models are usually trained on a large-scale training set, and tend to have strong memory for individuals who have seen, so that the prediction accuracy is high. However, this memory is associated with the individual and not the blood pressure, which results in a blood pressure predictive model that has very limited blood pressure predictive capabilities for the unseen user.

Disclosure of Invention

The application provides a blood pressure measurement method, wearable equipment and cloud equipment. According to the method, the training set is sampled, so that the data redundancy of the training set is reduced, gaps among all data in the training set are filled, the effect of improving generalization of the blood pressure prediction model is achieved, and further the blood pressure prediction accuracy of the blood pressure prediction model on individuals who do not see the blood pressure prediction model is improved.

The first aspect of the application provides a blood pressure measurement method which is applied to wearable equipment and comprises the steps of responding to blood pressure measurement operation of a user, collecting photoplethysmography (PPG) signals of the user to obtain PPG signal data, obtaining blood pressure measurement results of the user based on the PPG signal data and a trained blood pressure prediction model, and displaying the blood pressure measurement results, wherein the blood pressure prediction model is obtained by training based on a target training sample set, the target training sample is obtained by sampling an original training sample set, the difference degree between any two sample PPG signal data in the target training sample set is larger than a preset threshold value, the target training sample set contains target sample PPG signal data, and the original training sample set does not contain target sample PPG signal data.

Through the scheme, after the photoplethysmography (PPG) signal data of the user to be measured is obtained, the PPG signal data is input into a trained blood pressure prediction model, and a blood pressure measurement result of the user to be measured is obtained. Because the blood pressure prediction model is obtained by training based on the target training sample set, the blood pressure prediction model is not obtained by training the original training sample set. The target training sample set is obtained by sampling the original training sample, and the method effectively screens and amplifies the original training sample set, so that the target training sample set has more abundant and representative data compared with the original training sample set. Therefore, compared with the prior art that the original training sample is directly used for training the blood pressure prediction model, the method can inhibit the overfitting phenomenon of the blood pressure prediction model, enhance the generalization of the blood pressure prediction model and further improve the accuracy of the blood pressure prediction model through optimizing the training sample.

In a possible implementation mode, the wearable device is provided with a blood pressure prediction model, and a blood pressure measurement result of a user is obtained based on PPG signal data and the trained blood pressure prediction model, wherein the method comprises the steps of inputting the PPG signal data into the blood pressure prediction model to obtain the blood pressure measurement result of the user to be measured.

Through the scheme, after the wearable device acquires the PPG signal data of the user, the PPG signal data can be directly input into a locally deployed blood pressure prediction model, and then the effect of rapidly measuring the blood pressure is achieved.

In a possible implementation manner, a blood pressure measurement result of a user is obtained based on PPG signal data and a trained blood pressure prediction model, and the method comprises the steps of sending a blood pressure measurement request containing the PPG signal data to a cloud, deploying the blood pressure prediction model to the cloud, receiving the blood pressure measurement result sent by the cloud, and inputting the PPG signal data into the blood pressure prediction model by the cloud.

Through the scheme, the wearable device utilizes the cloud deployment model to realize blood pressure measurement of the user, so that the local deployment pressure of the wearable device is reduced, and the computing power of the wearable device is saved.

The method comprises the steps of receiving a blood pressure measurement request containing PPG signal data sent by a wearable device, inputting the PPG signal data into a trained blood pressure prediction model to obtain a blood pressure measurement result, and sending the blood pressure measurement result to the wearable device, wherein the blood pressure prediction model is obtained by training based on a target training sample set, the target training sample is obtained by sampling an original training sample set, the difference degree between PPG signal data of any two samples in the target training sample set is larger than a preset threshold, the target training sample set contains target sample PPG signal data, and the original training sample set does not contain target sample PPG signal data.

Through the scheme, after the photoplethysmography PPG signal data of the user to be measured is obtained, the PPG signal data is input into a trained blood pressure prediction model, and a blood pressure measurement result of the user to be measured is obtained. Because the blood pressure prediction model is obtained by training based on the target training sample set, the blood pressure prediction model is not obtained by training the original training sample set. The target training sample set is obtained by sampling the original training sample, and the method effectively screens and amplifies the original training sample set, so that the target training sample set has more abundant and representative data compared with the original training sample set. Therefore, compared with the prior art that the original training sample is directly used for training the blood pressure prediction model, the method can inhibit the overfitting phenomenon of the blood pressure prediction model, enhance the generalization of the blood pressure prediction model and further improve the accuracy of the blood pressure prediction model through optimizing the training sample.

In a possible implementation manner, the method further comprises the steps of obtaining an original training sample set, sampling the original training sample set to obtain a target training sample set, taking sample PPG signal data in the target training sample set as input, taking blood pressure corresponding to the sample PPG signal data in the target training sample set as a label, and training to obtain a blood pressure prediction model.

In a possible implementation manner, a primary training sample set is sampled to obtain a target training sample set, the method comprises the steps of carrying out first sampling on the primary training sample set based on a first sampling strategy to obtain a first training sample set, wherein the number of sample PPG signal data in the first training sample set is smaller than that of sample PPG signal data in the primary training sample set, the difference degree between any two sample PPG signal data in the first training sample set is larger than a preset threshold value, carrying out second sampling on the first training sample set based on a second sampling strategy to obtain a second training sample set, the second training sample set comprises target sample PPG signal data, and the primary training sample set and the first training sample set do not comprise target sample PPG signal data, and taking the second training sample set as the target training sample set.

Through the scheme, the cloud server firstly performs first sampling on the original training sample set through the first sampling strategy, and selects samples with larger difference, so that redundant information among the samples can be reduced, overfitting caused by excessive redundant information is avoided, original feature space is covered as much as possible, personal information is not lost, and training efficiency can be fully improved. Meanwhile, when the unbalanced data set is processed, the method can help to select few types of samples, and improves the recognition capability of the model on the few types. Furthermore, the cloud server performs second sampling on the first training sample set through a second sampling strategy, and fills samples lacking in the original training sample set, so that generalization of the model can be enhanced during subsequent training.

In a possible implementation manner, a primary training sample set is sampled to obtain a target training sample set, the method comprises the steps of performing first sampling on the primary training sample set based on a first sampling strategy to obtain a first training sample set, enabling the number of sample PPG signal data in the first training sample set to be smaller than that of sample PPG signal data in the primary training sample set, enabling the difference degree between any two sample PPG signal data in the first training sample set to be larger than a preset threshold value, performing second sampling on the primary training sample set based on a second sampling strategy to obtain a second training sample set, enabling the second training sample set to contain target sample PPG signal data, enabling the primary training sample set to not contain target sample PPG signal data, and enabling the second training sample set and the second training sample set to serve as the target training sample set.

Through the scheme, the cloud server can respectively perform first sampling and second sampling on the original training samples. Samples with larger differences can be selected through the first sampling, redundant information among the samples can be reduced, and overfitting caused by excessive redundant information is avoided. Some samples lacking in the original training sample set may be filled in by the first sampling. Further, the cloud server takes the training sample set with two advantages as the target training sample set, so that the generalization capability of the model can be better enhanced.

In one possible implementation, the first sampling strategy comprises diversity sampling and the second sampling strategy comprises interpolation sampling.

According to the technical defects of the original training sample set, the original training sample set is effectively screened and amplified by designing the targeted sampling strategy and targeted sampling treatment on the original training sample set, so that the target training sample set has more abundant and representative data compared with the original training sample set. Therefore, compared with the prior art that the original training sample is directly used for training the blood pressure prediction model, the method can inhibit the overfitting phenomenon of the blood pressure prediction model, enhance the generalization of the blood pressure prediction model and further improve the accuracy of the blood pressure prediction model through optimizing the training sample.

In a possible implementation manner, the first sampling is performed on an original training sample set based on a first sampling strategy to obtain a first training sample set, the first training sample set comprises sample PPG signal data in the original training sample set, a trained classification model is input to obtain a classification result of the sample PPG signal data, the classification model is used for extracting identity characteristics of each sample PPG signal data and classifying each sample PPG signal data according to the identity characteristics, the sample PPG signal data of the same class in the classification result correspond to the same sample user, and the first sampling is performed on each sample PPG signal data based on the first sampling strategy to obtain the first training sample set.

In a possible implementation manner, first sampling is respectively carried out in sample PPG signal data of each class based on a first sampling strategy to obtain a first training sample set, wherein the first training sample set comprises the steps of collecting sample PPG signal data of a target number from sample PPG signal data of any class to obtain a first sampling data set, wherein cosine similarity between identity features of any two sample PPG signal data in the first sampling data set is larger than first preset similarity, and each first sampling data set is used as the first training sample set.

In one possible implementation, collecting a target number of sample PPG signal data from the sample PPG signal data of the current class to obtain a first sampling data set includes calculating cosine similarity between the sample PPG signal data which is not collected and the sample PPG signal data which is collected respectively for any one of the sample PPG signal data which is collected, and adding the sample PPG signal data with the maximum cosine similarity corresponding to the sample PPG signal data which is not collected to the first sampling data set.

In a possible implementation manner, the method further comprises the steps of dividing sample PPG signal data in an original training sample set according to the belonging sample users to obtain sample PPG signal data of a plurality of sample users, taking the sample PPG signal data of the plurality of sample users as input, setting a cosine margin normalized loss function Angular-margin Softmax, training to obtain a classification model, and enabling the Angular-margin Softmax to be used for indicating that the classification model maximizes the probability of the belonging category and minimizes the probability gap with the adjacent category.

In one possible implementation, the second sampling is performed on the first training sample set based on a second sampling strategy to obtain a second training sample set, and the method comprises the steps of inputting the first training sample set into a trained sampling model to obtain the second training sample set.

In one possible implementation, the sampling model comprises an encoder, a sampler and a decoder, the first training sample set is input into the trained sampling model to obtain a second training sample set, the second training sample set comprises a plurality of first codes obtained by encoding sample PPG signal data in the first training sample set according to a sample user to which the first training sample set belongs by the encoder, one first code corresponds to the sample PPG signal data in the first training sample set under one sample user, interpolation sampling is carried out among the plurality of first codes by the sampler to obtain a plurality of second codes, and the plurality of second codes are decoded by the decoder to obtain the second training sample set.

In a possible implementation manner, interpolation sampling is performed among a plurality of first codes through a sampler to obtain a plurality of second codes, the method comprises the steps of constructing a data structure diagram corresponding to the plurality of first codes, wherein the data structure diagram comprises a plurality of nodes, one node corresponds to one first code, all nodes are connected through edges, the length of the edge between any two nodes is used for reflecting the coding similarity between the two nodes, candidate node pairs in the data structure diagram are determined, the candidate node pairs are nodes with the length of any two edges being larger than a preset length in the data structure diagram, a target node pair is determined according to the candidate node pairs, new nodes are inserted between the target node pairs, and the target node pair is a candidate node pair without other candidate nodes between any two candidate nodes.

In a possible implementation manner, a new node is inserted between target node pairs, wherein the new node comprises a code corresponding to a node to be inserted and a code corresponding to a second target node, the code corresponding to the node to be inserted is determined to obtain the node to be disassembled, the first target node and the second target node are any two target node pairs, and the node to be disassembled is inserted between the first target node and the second target node.

In a possible implementation manner, determining the code corresponding to the node to be inserted according to the code corresponding to the first target node and the code corresponding to the second target node to obtain the node to be disassembled includes weighting the code corresponding to the first target node and the code corresponding to the second target node to obtain the node to be disassembled.

In a third aspect, a wearable device is provided, comprising a processor coupled with a memory, the processor for executing a computer program or instructions stored in the memory to cause the wearable device to implement a blood pressure measurement method as in any of the first aspects.

In a fourth aspect, a cloud device is provided, which includes a processor coupled to a memory, the processor being configured to execute a computer program or instructions stored in the memory, so that the cloud device implements the blood pressure measurement method according to any one of the first aspects.

In a fifth aspect, a chip is provided, the chip being coupled to a memory, the chip being for reading and executing a computer program stored in the memory for implementing the blood pressure measurement method according to any of the first aspects.

In a sixth aspect, a computer readable storage medium is provided, the computer readable storage medium storing a computer program which, when run on a terminal device, causes the terminal device to perform the blood pressure measurement method according to any one of the first aspects.

In a seventh aspect, there is provided a computer program product for, when run on a computer, causing the computer to perform the blood pressure measurement method as in any of the first aspects.

It will be appreciated that the advantages of the second to seventh aspects may be found in the relevant description of the first aspect, and are not described here again.

Drawings

Fig. 1 is a schematic diagram of PPG acquisition provided in an embodiment of the present application;

fig. 2 is a waveform diagram of PPG signals according to an embodiment of the present application;

FIG. 3 is a schematic diagram of training a blood pressure prediction model according to an embodiment of the present application;

FIG. 4 is a schematic diagram of sample data partitioning according to an embodiment of the present application;

FIG. 5 is a second schematic diagram of sample data partitioning according to an embodiment of the present application;

fig. 6 is a schematic diagram one of an application scenario of a blood pressure measurement method according to an embodiment of the present application;

fig. 7 is a schematic diagram two of an application scenario of a blood pressure measurement method according to an embodiment of the present application;

Fig. 8 is a schematic diagram III of an application scenario of a blood pressure measurement method according to an embodiment of the present application;

Fig. 9 is a schematic diagram of a hardware structure of a wearable device according to an embodiment of the present application;

fig. 10 is a schematic diagram of a software structure of a wearable device according to an embodiment of the present application;

fig. 11 is a schematic diagram of a hardware structure of a cloud device according to an embodiment of the present application;

Fig. 12 is a flowchart of a blood pressure measurement method according to an embodiment of the present application;

fig. 13 is a second schematic flow chart of a blood pressure measurement method according to an embodiment of the present application;

fig. 14 is a schematic diagram of a distribution flow of a blood pressure prediction model according to an embodiment of the present application;

FIG. 15 is a flowchart illustrating a training process of a blood pressure prediction model according to an embodiment of the present application;

FIG. 16 is a schematic diagram of a classification model according to an embodiment of the present application;

FIG. 17 is a schematic diagram of classification effect of a classification model according to an embodiment of the present application;

FIG. 18 is a schematic diagram of a sampling model according to an embodiment of the present application;

FIG. 19 is a schematic diagram of interpolation conditions according to an embodiment of the present application;

FIG. 20 is a graph of effects obtained after interpolation and sampling by a sampling model according to an embodiment of the present application;

Fig. 21 is a schematic block diagram of a wearable device according to an embodiment of the present application.

Detailed Description

The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.

The term "and/or" is merely an association relationship describing the associated object, and means that three relationships may exist, for example, a and/or B may mean that a exists alone, while a and B exist together, and B exists alone.

The terms first and second and the like in the description and in the claims of embodiments of the application, are used for distinguishing between different objects and not necessarily for describing a particular sequential order of objects. For example, the first target object and the second target object, etc., are used to distinguish between different target objects, and are not used to describe a particular order of target objects.

In embodiments of the application, words such as "exemplary" or "such as" are used to mean serving as an example, instance, or illustration. Any embodiment or design described herein as "exemplary" or "e.g." in an embodiment should not be taken as preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "such as" is intended to present related concepts in a concrete fashion.

In the description of the embodiments of the present application, unless otherwise indicated, the meaning of "a plurality" means two or more. For example, a plurality of processing units refers to two or more processing units, and a plurality of systems refers to two or more systems.

It is to be understood that related art terms and related arts related to the embodiments of the present application are first described below.

1. Blood Pressure (BP).

Blood pressure is the lateral pressure of blood acting on the wall of a blood vessel per unit area when flowing in the blood vessel, and is the motive force for pushing the blood to flow in the blood vessel. Wherein the blood vessels are divided into arteries, capillaries and veins. Accordingly, blood pressure can be divided into arterial blood pressure, capillary blood pressure, and venous blood pressure. The blood pressure in daily life refers to the blood pressure of the brachial artery in the artery. It is understood that measuring blood pressure herein is also the blood pressure of the brachial artery.

2. Systolic (sys tolic blood pressure, SBP), diastolic (dias tolic blood pressure, DBP).

Blood pressure further includes systolic (i.e., high pressure as people speak during daily life) and diastolic (i.e., low pressure as people speak during daily life). Systolic pressure refers to the pressure at which the heart contracts by pumping blood into the artery, which expands to cushion the pressure, at which time the blood pressure is systolic. Diastolic pressure refers to the time when the heart is diastolic, the artery contracts so that blood returns to the heart, and the blood pressure at this time is the diastolic pressure.

3. Photoplethysmography (PPG, photoplethysmographic).

Photoplethysmography is a non-invasive, continuous technique for monitoring blood volume changes. It uses photoelectric means to detect the change of blood volume in living tissue, so as to indirectly reflect the pulsation condition of heart. PPG technology is typically implemented by emitting light of a certain wavelength (typically green or red) to the skin and measuring the change in intensity of light absorbed and reflected back by the tissue. When light irradiates the skin, a part of the light is absorbed by the skin, muscle, bone and other tissues, and another part of the light is reflected back. Due to the pulsation of blood in the artery, a periodic change in the blood volume under the skin is caused, thereby affecting the intensity of the reflected light. The PPG sensor obtains a pulse waveform by receiving such a change in intensity of reflected light and converting it into an electrical signal.

4. A PPG sensor.

A PPG sensor is a sensor that irradiates light into the skin and measures light scattering due to blood flow. Referring to fig. 1, ppg sensors typically include a light emitter (typically an LED) and a light detector. The light emitter emits a light beam with a certain wavelength to irradiate the skin, and the light beam is reflected to the light detector after penetrating the skin (such as epidermis, dermis and/or subcutaneous tissue in fig. 1), and the light beam is attenuated to a certain extent. It will be appreciated that the absorption of light by muscles, bones, veins, etc. is substantially unchanged without a substantial change in the measurement site. However, during systole and diastole, the absorption of light by the arteries will be different due to blood flow. Just as the absorption of light by arteries varies while the absorption of light by other tissues (such as the aforementioned muscles, bones, veins, etc.) is substantially unchanged, a waveform reflecting the characteristics of blood flow (referred to herein as a PPG waveform) can be obtained by extracting an alternating component (referred to herein as a PPG signal) in an electrical signal corresponding to an optical signal after measuring the optical signal by a photodetector for a period of time. Thereby being used for determining physiological parameters such as blood oxygen saturation, blood pressure, heart rate and the like.

The PPG sensor irradiates human skin by using LEDs, and measures reflected light intensity change caused by blood flow by using photodiodes, and the obtained periodic waveform contains pulsation change information of blood volume in the heart pulsation period. The blood volume is greatest when the heart contracts and the diastolic volume is smallest when the heart relaxes. The systolic and diastolic pressures can thus be obtained from the relationship of volume pulse flow information and blood pressure. The waveform of a photoplethysmography (PPG) signal is shown in fig. 2, in which the horizontal axis is the time stamp, the vertical axis is the waveform of the PPG signal obtained over time, the peaks correspond to the systole of the human heart and the troughs correspond to the diastole of the human heart. Thus, the peaks of the PPG waveform may correspond to systolic pressure and the troughs of the PPG waveform may correspond to diastolic pressure.

5. And (5) a machine learning model.

The machine learning model refers to a mathematical model that predicts or infers based on data. These models automatically accomplish tasks such as classification, regression, clustering, generation, etc., by learning patterns in the data. In practice, the labeled data may be used for training, where the data includes an input and a corresponding correct output (tag). Common models include linear regression, logistic regression, support Vector Machines (SVMs), decision trees, random forests, and the like.

Therefore, the relationship between the PPG signal and the blood pressure can be utilized to train a blood pressure prediction model so as to realize the blood pressure prediction effect.

As shown in fig. 3, the related art obtains a training set and a verification set by collecting data of PPG signals and blood pressure values of different sample users, and trains by using the training set and the verification set to obtain a blood pressure prediction model. In the training process, PPG signals of a large number of sample users are input, and the blood pressure of each sample user is used as a label to adjust model parameters. When the parameters in the model are basically unchanged, the training reaches a stable state, and the training process is ended.

6. Training Set (TRAINING SET), validation Set (Validation Set), and Test Set (Test Set).

In machine learning and data science, it is a common practice to divide the dataset into a training Set (TRAINING SET), a Validation Set (Validation Set) and a Test Set (Test Set) to help evaluate the performance of the model, prevent overfitting, and optimize the model parameters. The specific roles and methods of use of these datasets are as follows:

training set (TRAINING SET)

The training set is used to train a machine learning model, i.e., the model fits the data distribution and learns the parameters by learning the data in the training set.

The method of use is that data is input into a model for training, and the model is used for minimizing a loss function by adjusting parameters so as to better fit training data.

Verification Set (verification Set)

The validation set is used to evaluate the performance of the model during training, helping to select the best model parameters and preventing overfitting. The validation set does not participate in the training process of the model and is only used to evaluate and adjust the model.

Methods of use during training, the verification set is used periodically to evaluate the performance (e.g., accuracy, loss, etc.) of the model. By observing performance changes on the validation set, early shutdown (Early Stopping), adjustment of learning rate, selection of the best model architecture, etc. operations can be performed. During the model selection and tuning phase, the verification set is used to select the best performing model.

Test Set (Test Set)

The test set is used to evaluate the generalization ability of the model after model training is completed, namely the performance of the model on unseen data. The test set provides an unbiased performance assessment for measuring the performance of the model in practical applications.

The using method comprises the steps of carrying out one-time evaluation by using a test set after model training is completed, and obtaining the final performance index of the model. The results of the test set should not be used to adjust model parameters or make any form of model selection.

In the blood pressure prediction scene, the influence of the data set division mode on the blood pressure prediction result is large. The related art generally includes two division modes, and the two division modes are described below.

Division one (also known as intra-subject), divides the data of each individual into a training set, a validation set, and a test set. As shown in fig. 4, the existing data set includes a first sample group (including sample user 1, sample user 2, sample user 3), a second sample group (including sample user 4, sample user 5, sample user 6), and a second sample group (including sample user 7, sample user 8, sample user 9), and the data of each sample user is divided into a training set, a validation set, and a test set according to a dotted line.

Since the training set and the test set share the same individual data in this way, this partitioning approach is somewhat fraudulent to the model. For example, if the first part of data of the sample user 1 is a training set, the second part of data is a verification set, and the third part of data is a test set, then after the training of the first part of data and the second part of data of the sample user 1, the model can easily make a blood pressure prediction for the third part of data of the sample user 1 because the data is the same user, but the model cannot be considered to reach the standard. If a sample user 10 (the data of the user is not present in the sample data), the test result indicates that the model cannot accurately predict the blood pressure of the sample user 10.

The second division mode (also called inter-subject) is to divide different individual data into a training set, a verification set and a test set, so that the individual data in each data set are ensured not to be crossed. As shown in fig. 5, the existing data set includes a first sample group (including sample user 1, sample user 2, sample user 3), a second sample group (including sample user 4, sample user 5, sample user 6), and a second sample group (including sample user 7, sample user 8, sample user 9), and the data of different sample users is divided into a training set, a validation set, and a test set according to dotted lines.

In the dividing mode, the algorithm and the training mode of the model are kept unchanged, and tests show that the blood pressure prediction performance of the model is poorer than that of the dividing mode. Further, the current blood pressure prediction model has very limited prediction capability for unseen data, namely, the model obtained by training the samples in the related technology on all training sets has more serious overfitting phenomenon and poorer generalization.

In view of the above problems, the embodiments of the present application provide a blood pressure measurement method, which considers performing individual identification training on a large-scale data set by using PPG signals, so as to obtain identity codes of different individuals in a hidden space. On the basis of acquiring the individual hidden space codes of the training set, the data set is subjected to diversity sampling and significant gap interpolation sampling based on the generated model, and the aim is to inhibit the over-fitting phenomenon through the data augmentation strategy, improve the generalization of the blood pressure model and further improve the blood pressure prediction precision of individuals which are not seen.

Before the technical scheme of the embodiment of the application is described, an application scene of the embodiment of the application is described with reference to the attached drawings.

Referring to fig. 6, a schematic diagram of an application scenario is provided in an embodiment of the present application. The application scenario includes a wearable device 11 and a cloud 12. As shown in fig. 6, the wearable device 11 is in communication connection with the cloud 12. In the implementation of the embodiment of the present application, the cloud end 12 may include one or more cloud servers for providing services for the wearable device 11. The wearable device 11 may be a smart watch, a smart bracelet, a smart wristband, a smart jewelry, etc., which is not limited in the embodiment of the present application. The specific technology and the specific equipment form adopted by the wearable equipment are not limited in the embodiment of the application. It should be noted that, in practical application, the number of cloud ends may be one or more, and the number of cloud ends and wearable devices in the application scenario shown in fig. 6 is merely an adaptive example, which is not limited in the present application.

In a specific application, after the user wears the wearable device 11, the wearable device 11 may be instructed to perform blood pressure measurement. After receiving the blood pressure measurement instruction of the user, the wearable device 11 may collect the PPG signal of the user, and upload the collected PPG signal to the cloud 12, so as to request the cloud 12 to predict the blood pressure of the user. Correspondingly, the cloud 12 deploys a blood pressure prediction model, and after receiving the PPG signal uploaded by the wearable device 11, the cloud 12 inputs the PPG signal into the blood pressure prediction model to obtain a blood pressure value, and sends the blood pressure value to the wearable device 11. After receiving the blood pressure value sent by the cloud 12, the wearable device 11 can store and display the blood pressure value.

Taking wearable equipment as an intelligent watch as an example, as shown in a of fig. 7, a blood pressure detection APP is installed on the intelligent watch, and a user can click the blood pressure detection APP to enter a blood pressure detection interface. As shown in b of fig. 7, the blood pressure detection interface includes a blood pressure display frame and a measurement button. Wherein the blood pressure display frame includes a high pressure (systolic pressure) display frame and a low pressure (diastolic pressure) display frame. The user can click a measurement button in the blood pressure detection interface to enter the blood pressure measurement interface, and the intelligent watch starts to acquire the PPG signal of the user. As shown in fig. 7 c, the blood pressure measurement interface includes a prompt box for prompting the user to "keep still while measuring, note that the wristwatch is worn correctly", and a time box for displaying the countdown of the blood pressure measurement time. After the measurement is completed, as shown in d of fig. 7, the blood pressure detection interface is automatically returned, and the high pressure (systolic pressure) and the low pressure (diastolic pressure) currently measured by the user are displayed in the blood pressure display frame.

Referring to fig. 8, another application scenario is schematically provided in an embodiment of the present application. The application scene comprises a wearable device 21, a mobile phone 22 and a cloud 23. As shown in fig. 8, the wearable device 21 is in communication connection with the mobile phone 22, and the mobile phone 22 is in communication connection with the cloud 23. In the implementation process of the embodiment of the application, the wearable device 21 and the mobile phone 22 can be connected through bluetooth, and the mobile phone 22 and the cloud 23 can be connected through a network.

In a specific application, the wearable device 21 may establish a bluetooth connection with the mobile phone 22 in response to a bluetooth pairing operation of the user. Or the handset 22 establishes a bluetooth connection with the wearable device 21 in response to a bluetooth pairing operation by the user. After the wearable device 21 and the mobile phone 22 establish bluetooth connection for the first time, the wearable device 21 and the mobile phone 22 automatically establish bluetooth connection as long as both the wearable device 21 and the mobile phone 22 have the bluetooth function turned on and the distance is smaller than a certain threshold. After the user wears the wearable device 21, the user can instruct the wearable device 21 to take blood pressure measurement through the cell phone 22. After receiving the blood pressure measurement indication, the wearable device 21 may collect PPG signals of the user and send the collected PPG signals to the handset 22. After receiving the PPG signal collected by the wearable device 21, the mobile phone 22 uploads the PPG signal to the cloud 23 to request the cloud 23 to predict the blood pressure of the user. Correspondingly, the cloud 23 deploys a blood pressure prediction model, and after receiving the PPG signal uploaded by the mobile phone 22, the cloud 23 inputs the PPG signal into the blood pressure prediction model to obtain a blood pressure value, and sends the blood pressure value to the mobile phone 22. After the mobile phone 22 receives the blood pressure value sent by the cloud 23, the blood pressure value can be stored and displayed.

Of course, the present application may also perform the blood pressure measurement method in other scenarios, which are not limited in the embodiment of the present application.

The hardware structure of the wearable device is described below in conjunction with fig. 9.

Referring to fig. 9, a hardware configuration diagram of a wearable device (such as a smart watch) that is wearable on a wrist is provided in an embodiment of the present application. As shown in fig. 9, the wearable device includes a processor 310, a display screen 320, a ppg sensor 330, a pressure sensor 340, an Acceleration (ACC) sensor 350, a memory 360, and a wireless communication module 370. The processor 310 may include one or more interfaces for interfacing with other components of the electronic device. The one or more interfaces may include, among other things, an Input/Output (I/O) interface (also referred to as an I/O pin), an interrupt pin, a data bus interface, and the like. The data bus interface may include one or more of a serial peripheral interface (SERIAL PERIPHERAL INTERFACE, SPI), an integrated circuit (inter-INTEGRATED CIRCUIT, I2C) interface, and the like.

Processor 310 may include one or more processing units, for example, processor 310 may include an application processor (application processor, AP), a modem processor, a graphics processor (graphics processing unit, GPU), an image signal processor (IMAGE SIGNAL processor, ISP), a controller, a video codec, a digital signal processor (DIGITAL SIGNAL processor, DSP), a baseband processor, and/or a neural network processor (neural-network processing unit, NPU), etc. Wherein the different processing units may be separate devices or may be integrated in one or more processors.

The processor 310 may generate operation control signals according to the instruction operation code and the timing signals to complete instruction fetching and instruction execution control.

A memory may also be provided in the processor 310 for storing instructions and data. In some embodiments, the memory in the processor 310 may be a cache memory. The memory may hold instructions or data that are used or used more frequently by the processor 310. If the processor 310 needs to use the instruction or data, it can be called directly from the memory. Repeated accesses are avoided and the latency of the processor 310 is reduced, thereby improving the efficiency of the system.

In some embodiments, the processor 310 may be used to perform the calculation of the blood pressure prediction. For example, the processor 310 may predict the blood pressure based on a phase difference of the pressure waveform and the PPG waveform, a period of the PPG waveform, and a signal amplitude of the pressure waveform.

The display screen 320 is used to display images, videos, and the like. The display screen 320 includes a display panel. The display panel may employ a Liquid Crystal Display (LCD) CRYSTAL DISPLAY, an organic light-emitting diode (OLED), an active-matrix organic LIGHT EMITTING diode (AMOLED), a flexible light-emitting diode (FLED), miniled, microLed, micro-oLed, a quantum dot LIGHT EMITTING diode (QLED), or the like. In some embodiments, the wearable device may include 1 or more display screens 320.

In some embodiments, the display screen 320 may display predicted blood pressure values, relevant controls for blood pressure measurements, such as a start control, an end control, and the like.

Memory 360 may be used to store computer-executable program code that includes instructions. The memory 360 may include a stored program area and a stored data area. The storage program area may store an application program (such as a sound playing function, an image playing function) required for at least one function of the operating system, etc. The storage data area may store data created during use of the wearable device (e.g., audio data, phonebook, etc.), and so on. In addition, memory 360 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, universal flash memory (universal flash storage, UFS), and the like. The processor 310 performs various functional methods or data processing of the wearable device by executing instructions stored in the memory 360 and/or instructions stored in a memory provided in the processor.

The wireless communication module 370 may support data exchange between the electronic device and other electronic devices including bluetooth, global navigation satellite system (global navigation SATELLITE SYSTEM, GNSS), wireless local area network, frequency modulation (frequency modulation, FM), near Field Communication (NFC), infrared (IR), and the like.

The PPG sensor 330 is used to emit light and detect the reflected light signal and convert the light signal into an electrical signal. The alternating component of the electrical signal may reflect the characteristics of the blood flow and may thus be used to calculate blood pressure.

The pressure sensor 340 is used to sense a pressure signal, and may convert the pressure signal into an electrical signal. The pressure sensor 340 is of a wide variety such as a resistive pressure sensor, an inductive pressure sensor, a capacitive pressure sensor, and the like. The capacitive pressure sensor may be a capacitive pressure sensor comprising at least two parallel plates with conductive material. When a force is applied to the pressure sensor 340, the capacitance between the electrodes changes. The wearable device determines the strength of the pressure from the change in capacitance.

In some embodiments, the wearable device further comprises an acceleration sensor 350. The acceleration sensor 350 may be used to detect the magnitude of acceleration of the wearable device in various directions (typically three axes). Thereby being used to determine the movement of the wearable device. For example, the wearable device can measure blood pressure without movement of the wearable device, so that the accuracy of a measurement result is prevented from being influenced by movement of the wearable device.

It should be noted that, in actual implementation, the hardware components of the wearable device are not limited to those shown in fig. 9. The wearable device may also include, for example, a power management module, a battery, a camera, a motor, keys, and the like.

The software structure of the wearable device is described below in connection with fig. 10.

Fig. 10 is a software structural block diagram of a wearable device provided by an embodiment of the present application. The layered architecture of the wearable device divides the software into several layers, each layer having a distinct role and division of labor. The layers communicate with each other through a software interface. In some embodiments, taking the wearable device as a smart watch as an example, as shown in fig. 10, an operating system of the smart watch may include five layers, namely a UI (User Interface) application layer, a system service layer, an algorithm layer, a hardware abstraction layer and a kernel layer from top to bottom.

The UI application layer may include a series of application packages, which may be dials, motion recordings, conversations, exercises, etc.

The system services layer may include a series of system services. As shown in fig. 10, the system service layer may include a blood pressure service, where the blood pressure service may provide physiological parameter information of the wearer of the smart watch, such as heart rate, blood pressure, blood oxygen, pulse rate (abbreviated as pulse), respiratory rate, body temperature, and the like, and may also detect physiological parameter change information of the wearer of the smart watch. The system services layer may also include a step counting service, a calorie service, a heart health service.

The algorithm layer may include a series of algorithm models. As shown in fig. 10, the algorithm layer may include a blood pressure algorithm model and an alignment algorithm model of the multi-source data. The blood pressure algorithm model is used for calculating physiological parameters of the wearer of the intelligent watch, and the multi-source data alignment algorithm model is used for collecting data of Ji Duoyuan sensors. The algorithm layer may also include a sleep algorithm model, a wear algorithm model, a dimming algorithm model, and the like.

A hardware abstraction layer (hardware abstraction layer, HAL) is an interface layer between the operating system kernel and the hardware circuitry. As shown in fig. 10, the HAL layer includes a storage HAL, a display HAL, a touch HAL, a bluetooth HAL, a global positioning system (global positioning system, GPS) HAL. Wherein the storage HAL can be used for storing and processing the data acquired by the sensor. The display HAL may be used to display information. The touch HAL may be the same as detecting a user-triggered operation. The bluetooth HAL may be used to implement bluetooth communication functions. GPSHAL may be used to implement positioning functions. The HAL layer may be implemented by a c++ library.

As shown in fig. 10, the kernel layer includes at least an operating system kernel.

It will be appreciated that the layers and the components contained in the layers in the software structure shown in fig. 10 do not constitute a specific limitation on the smart watch. In other embodiments of the present application, the smart watch may include more or fewer layers than shown, and more or fewer components may be included in each layer, as the application is not limited.

Fig. 11 is a hardware structure block diagram of an example of a cloud device according to an embodiment of the present application. The cloud in the embodiment of the application can be one or more server clusters. One or more cloud devices may be included in the cluster. The cloud device may be a cloud server or other device, which is not limited by the present application. As shown in fig. 11, cloud device 400 includes a processor 401 further including one or more processors and memory resources represented by memory 402 for storing instructions, such as applications, executable by processor 401. The application program stored in memory 402 may include one or more modules each corresponding to a set of instructions. Further, the processor 401 is configured to execute instructions to perform the blood pressure measurement method provided by the embodiment of the present application.

Cloud device 400 may also include a power component 403 configured to perform power management of cloud device 400, a wired or wireless network interface 404 configured to connect cloud device 400 to a network, and an input output interface 405. Cloud device 400 may operate an operating system based on memory 402.

The methods in the following embodiments may be implemented in a wearable device and a cloud terminal having the above hardware structures. In the following embodiments, a specific implementation manner of the embodiments of the present application is described in detail by taking an example that the blood pressure measurement system includes a wearable device and a cloud as shown in fig. 6. Specifically, the blood pressure measurement process in the embodiment of the present application may be divided into two parts. The first part is a model generation scene, namely a scene in which the cloud performs blood pressure prediction model training based on a large amount of sample data. The second part is a blood pressure prediction scene, namely a scene in which the wearable device performs blood pressure measurement based on a blood pressure prediction model.

Firstly, for a blood pressure prediction scene, referring to fig. 12, taking a wearable device as a smart watch and a cloud device as a cloud server as an example, the blood pressure measurement method provided by the embodiment of the application may include S501-S505:

S501, responding to blood pressure measurement operation of a user, and acquiring a PPG signal of the user by the intelligent watch through a PPG sensor to obtain PPG signal data.

The blood pressure measurement operation may be an operation of a user triggering a blood pressure measurement button in the smart watch. The blood pressure measurement key can be a key in blood pressure measurement application, and can also be a key capable of realizing blood pressure measurement function in other applications in the intelligent watch.

When the user needs to measure the blood pressure value, the user can measure the blood pressure by using the smart watch and the mobile phone. For example, a user may wear a smart watch, and first collect physical characteristics of the user using the smart watch, thereby obtaining a blood pressure value according to the physical characteristics of the user. Wherein the physical characteristic of the user comprises a PPG signal.

In some examples, the process of the smart watch collecting the PPG signal by the PPG sensor may include first the user activating a blood pressure measurement function of the smart watch. Then, after the blood pressure measurement function is turned on, the smart watch may prompt the user to perform a first preset behavioral action, and detect whether the user has performed the first preset behavioral action. For example, the first preset behavioral action may be wearing the smart watch at a designated location on the user's body, e.g., a wrist, etc. And under the condition that the user is detected to execute the first preset behavior action, the intelligent watch controls the PPG sensor to collect PPG signals. It may be appreciated that, when the user is detected to trigger the first preset behavior action, the smart watch may also prompt the user to perform the first preset behavior action more regularly through the voice module, for example, prompt the user to remain stationary.

S502, the intelligent watch sends a blood pressure measurement request to the cloud server through the wireless communication module.

The blood pressure measurement request is used for requesting the cloud server to determine the blood pressure value of the user.

As a possible implementation manner, when the smart watch obtains PPG signal data of the user, a blood pressure measurement request containing the PPG signal data may be sent to the cloud server through the wireless communication module.

As another possible implementation manner, after obtaining PPG signal data of the user, the smart watch may upload the PPG signal data to the cloud server through the wireless communication module, and then send a blood pressure measurement request to the cloud server.

In some implementations, in order to ensure accuracy of a blood pressure value determined based on the PPG signal, the embodiment of the present application may further perform signal quality detection on the PPG signal after the PPG signal is acquired, and if the PPG signal passes the signal quality detection, further processing may be performed on the continued PPG signal. If the PPG signal does not pass the signal quality detection, S501 to S502 need to be executed again to re-collect the PPG signal until the collected PPG signal meets the requirement of the signal quality detection. Thus, the signal quality and the signal stability of the collected PPG signal can be ensured to meet the requirements.

S503, the cloud server inputs PPG signal data into a trained blood pressure prediction model to obtain a blood pressure measurement result.

The cloud server is deployed with a trained blood pressure prediction model, the blood pressure prediction model is obtained by training based on a target training sample set, the target training sample is obtained by sampling an original training sample set, the degree of difference between PPG signal data of any two samples in the target training sample set is larger than a preset threshold, the target training sample set contains target sample PPG signal data, and the original training sample set does not contain target sample PPG signal data.

As a possible implementation manner, after receiving a blood pressure measurement request including PPG signal data sent by the smart watch, the cloud server inputs the PPG signal data into a trained blood pressure prediction model to obtain a blood pressure measurement result.

As another possible implementation manner, after receiving a blood pressure measurement request sent by the smart watch, the cloud server inputs PPG signal data previously uploaded by the smart watch into a trained blood pressure prediction model, so as to obtain a blood pressure measurement result.

S504, the cloud server sends the blood pressure measurement result to a wireless communication module of the intelligent watch.

As one possible implementation manner, the cloud server sends the blood pressure measurement result to the smart watch, and correspondingly, the smart watch receives the blood pressure measurement result sent by the cloud server through the wireless communication module and transmits the blood pressure measurement result to the display screen.

S505, the intelligent watch displays the blood pressure measurement result through a display screen.

Wherein the blood pressure measurement includes diastolic and systolic pressures. The intelligent watch can display the blood pressure measurement result to the user through the display screen, and can also save the test time of the blood pressure measurement result.

As one possible implementation, the smart watch may display the blood pressure measurement results on the interface through a display screen. After the wearable device displays the blood pressure measurement result, the user can wait for operation, the option of re-measurement can be displayed so as to re-measure the blood pressure, and the option of exiting can be displayed so as to end the blood pressure measurement.

Optionally, if the smart watch detects a triggering operation of the user on the re-measurement option, S501 to S505 may be executed.

For a blood pressure prediction scenario, referring to fig. 13, taking a wearable device as a smart watch and a cloud device as a cloud server as an example, the blood pressure measurement method provided by the embodiment of the application may further include S601-S604:

S601, the cloud server transmits the trained blood pressure prediction model to a blood pressure algorithm module of the intelligent watch.

As a possible implementation manner, the cloud server may send the trained blood pressure prediction model to the smart watch after training the blood pressure prediction model. Correspondingly, the intelligent watch can store and deploy the blood pressure prediction model so as to call the blood pressure prediction model when the user measures the blood pressure.

In some embodiments, the cloud server may send model parameters of the blood pressure prediction model to each smart watch, which performs model updates by updating the model parameters of the local blood pressure prediction model.

Illustratively, a schematic diagram of the distribution flow of the exemplary illustrated blood pressure prediction model is shown in fig. 14. After training to obtain the blood pressure prediction model, the cloud 62 may send model parameters of the blood pressure prediction model to the smart watch 61, so that the smart watch 61 may perform blood pressure measurement based on the blood pressure prediction model.

S602, responding to blood pressure measurement operation of a user, and acquiring a PPG signal of the user through a PPG sensor by the intelligent watch to obtain PPG signal data.

For a specific implementation of this step, reference may be made to S501, which is not described herein.

S603, the intelligent watch inputs PPG signal data into a blood pressure prediction model to obtain a blood pressure measurement result of the user to be measured.

As a possible implementation manner, the smart watch inputs PPG signal data into a locally deployed blood pressure prediction model, so as to obtain a blood pressure measurement result of the user to be measured.

S604, the intelligent watch displays the blood pressure measurement result through a display screen.

The specific implementation of this step may refer to S505, which is not described herein.

The training process of the blood pressure prediction model is described in detail below with reference to the flowchart of the method shown in fig. 15, and specifically includes:

S701, the cloud server acquires an original training sample set.

The original training sample set can be obtained by collecting a plurality of real sample users, and comprises blood pressure values of the plurality of real sample users and PPG signal data corresponding to the blood pressure values.

As one possible implementation, the cloud server may obtain the data of these real sample users from the database, and use these data as the original training sample set.

S702, the cloud server samples the original training sample set to obtain a target training sample set.

As a possible implementation manner, the cloud server may perform first sampling on the original training sample set based on the first sampling policy, to obtain a first training sample set. The number of the sample PPG signal data in the first training sample set is smaller than that of the sample PPG signal data in the original training sample set, and the degree of difference between any two sample PPG signal data in the first training sample set is larger than a preset threshold. Further, the cloud server performs second sampling on the first training sample set based on a second sampling strategy to obtain a second training sample set, wherein the second training sample set contains target sample PPG signal data, and the original training sample set and the first training sample set do not contain target sample PPG signal data. The cloud server may take the second training sample set as the target training sample set.

It can be understood that the cloud server firstly performs first sampling on the original training sample set through the first sampling strategy, and selects samples with larger difference, so that redundant information among the samples can be reduced, overfitting caused by excessive redundant information is avoided, original feature space is covered as much as possible, personal information is not lost, and training efficiency can be fully improved. Meanwhile, when the unbalanced data set is processed, the method can help to select few types of samples, and improves the recognition capability of the model on the few types. Further, the cloud server performs second sampling on the first training sample set through a second sampling strategy, and fills samples lacking in the original training sample set, so that generalization of the model is enhanced in subsequent training.

As another possible implementation manner, the cloud server may perform first sampling on the original training sample set based on the first sampling policy to obtain a first training sample set, where the number of sample PPG signal data in the first training sample set is smaller than the number of sample PPG signal data in the original training sample, and the degree of difference between any two sample PPG signal data in the first training sample set is greater than a preset threshold. Further, the cloud server performs second sampling on the original training sample set based on a second sampling strategy to obtain a second training sample set, wherein the second training sample set contains target sample PPG signal data, and the original training sample set does not contain target sample PPG signal data. The cloud server takes a training sample set and a second training sample set as target training sample sets.

It can be appreciated that the cloud server can perform the first sampling and the second sampling on the original training samples, respectively. Samples with larger differences can be selected through the first sampling, redundant information among the samples can be reduced, and overfitting caused by excessive redundant information is avoided. Some samples lacking in the original training sample set may be filled in by the first sampling. Further, the cloud server takes the training sample set with two advantages as the target training sample set, so that the generalization capability of the model can be better enhanced.

Alternatively, the first sampling strategy may be any downsampling strategy, e.g. the first sampling strategy comprises diversity sampling. The second sampling strategy may be any up-sampling strategy, for example the second sampling strategy comprises interpolation sampling. Wherein the diversity sampling aims at picking samples with different characteristics or attributes from the data set so as to ensure that the machine learning model can comprehensively and uniformly learn the data distribution. These selected samples may have features that are rare in the training data, or they may represent true statistics that are currently not representative enough in the model. Through diversity sampling, the generalization capability and the robustness of the model can be improved. Interpolation sampling refers to the process of estimating an unknown data point by some algorithm or model based on the known data point.

It can be appreciated that the embodiment of the application aims at the technical defects of the original training sample set, and the original training sample set is effectively screened and amplified by designing a targeted sampling strategy and targeted sampling treatment on the original training sample set, so that the target training sample set has more abundant and representative data compared with the original training sample set. Therefore, compared with the prior art that the original training sample is directly used for training the blood pressure prediction model, the method can inhibit the overfitting phenomenon of the blood pressure prediction model, enhance the generalization of the blood pressure prediction model and further improve the accuracy of the blood pressure prediction model through optimizing the training sample.

The following describes the first sampling strategy and the second sampling strategy in detail, taking the first sampling strategy as diversity sampling and the second sampling strategy as interpolation sampling as an example.

In some embodiments, for the diversity sampling policy, before the cloud server performs first sampling on the original training sample set based on the diversity sampling policy, it is required to classify PPG signal data in the original training sample set, and then perform diversity sampling on PPG signal data in each classification result.

For example, the cloud server may classify PPG signal data in the original training sample set according to the user to which the PPG signal data corresponds, so that PPG signal data of each sample user may be separated. Further, the PPG signal data of each sample user is sampled in a diversity manner.

As a possible implementation manner, the cloud server may input the sample PPG signal data in the original training sample set into a trained classification model to obtain a classification result of the sample PPG signal data, where the classification model is used to extract identity features of each sample PPG signal data and classify each sample PPG signal data according to the identity features, and the sample PPG signal data of the same class in the classification result corresponds to the same sample user. Further, the cloud server may perform first sampling in each class of sample PPG signal data based on a first sampling policy, to obtain a first training sample set.

A specific classification model is described below. In order to implement classifying PPG signal data in the original training sample set, the cloud server may train a classification model. Specifically, the cloud server may first divide PPG signal data in a large-scale dataset (e.g., a public dataset) by individuals and perform class encoding on the individuals. Further, the cloud server may extract individual identity-related features using the feature extractor as an individual identity classifier based on PPG signal data.

The feature extractor may be a small volume simple feature extractor, such as a simple convolutional neural network (Convolutional Neural Networks, CNN) multi-layer stack, among others.

In the training process, since the test set individuals never see by default and classification (namely zero-shot) problem) can be required, the embodiment of the application can use the cosine margin normalized loss function (Angular-margin Softmax) based on elastic penalty to replace the normalized index function (Softmax) commonly used in classification tasks when designing the loss function. The aim is to make the identity code in the hidden space have partitionability, more compact in class, more distinguishing between classes and flexibility through the additive angle margin penalty. For example, the loss function formula can be designed as follows:

Wherein θ is the included angle between the weight vector W and the feature vector x of the full connection layer in the classifier network. By means of the included angle constraint, a margin between decision boundaries is generated, and the larger the margin is, the stricter the classification boundary is. S is a fixed scale factor for enlarging the feature space, E is an added random Gaussian elasticity penalty, m is the mean value, and sigma is the standard deviation.

It will be appreciated that the Angular-margin Softmax enhances the classification effect by enhancing the weight and feature vector angle constraint, adding an elastic penalty to assign different margin to each sample, in order to derive class centers with greater penalty for samples farther away, and giving less penalty attention to samples nearer.

For example, referring to fig. 16, the training set of the classification model may include training set PPG data and training set individual class labels, and the feature extractor of the classification model is composed of a plurality of convolutional layers (Conv) stacked. After the cloud server inputs the training set PPG data into the feature extractor, embeadding is carried out on the training set PPG data through the feature extractor so as to extract individual identity features corresponding to the training set PPG data, and parameters of all convolution layers are corrected by taking a loss function Angular-margin Softmax as an evaluation basis.

As shown in fig. 17, after the training of the classification model is completed, the cloud server may also test the classification effect of the classification model on the test set. The cloud server can utilize a dimension reduction visualization algorithm (for example, tSNE algorithm) to reduce dimension of the identity coding features extracted by the feature extractor and visualize the dimension reduction, so that the clustering degree of the identity coding features from the same individual can be obtained. In addition, the cloud server can perform classification test on the identity coding features by using a random forest machine learning algorithm, and the classification effect confirmation of the classifier on the test set is completed. The random forest machine learning algorithm is an integrated learning method, and classification and regression tasks are realized by constructing a plurality of decision trees.

After the classification result of the sample PPG signal data is obtained through the classification model, the cloud server can respectively sample the sample PPG signal data of each category in a diversity mode, namely the diversity sampling is carried out in one classification result. The diversity sampling can define the difference based on cosine similarity distance, and samples which cover the characteristic space as much as possible and have larger difference are selected through a greedy algorithm.

As one possible implementation, for any class of sample PPG signal data, the cloud server may collect a target number of sample PPG signal data from the current class of sample PPG signal data, resulting in a first sample dataset. The cosine similarity between the identity features of any two sample PPG signal data in the first sampling data set is larger than the first preset similarity. Further, the cloud server takes each first sampling data set as a first training sample set.

Specifically, for any one of the collected sample PPG signal data, the cloud server may calculate the cosine similarity between the non-collected sample PPG signal data and the collected sample PPG signal data, respectively. Further, the cloud server may add, to the first sampling dataset, sample PPG signal data having a maximum cosine similarity corresponding to the non-collected sample PPG signal data.

It can be understood that the diversity sampling strategy can reduce redundant information among samples by selecting samples with larger difference, avoid overfitting caused by excessive redundant information, and cover original feature space as much as possible, so that personal information is not lost, and training efficiency can be fully improved. In addition, the diversity sampling strategy can also help to select few types of samples when the unbalanced data set is processed, so that the recognition capability of the model to the few types is improved.

As can be seen from the classification results of the classification model, a significant GAP (GAP) exists between the clustering results, and the GAP corresponds to an individual sample which is not seen by the blood pressure prediction model, so that the generalization of the model can be affected by the presence of the GAP in the training stage. Therefore, the embodiment of the application needs to interpolate among the clustering results through an interpolation sampling strategy to fill up the GAP so as to enhance the generalization of the blood pressure prediction model, and a specific interpolation sampling strategy is introduced below.

In some embodiments, in order to make up GAPs between clustering results, the cloud server may input a first training sample set obtained through a diversity sampling strategy into a trained sampling model to obtain a second training sample set.

As shown in fig. 18, the sampling model includes an encoder, a sampler, and a decoder. The encoder is used for encoding the sample PPG signal data in the first training sample set according to the sample user to obtain a plurality of first codes. One first code corresponds to sample PPG signal data under one sample user in the first training sample set. The sampler is used for carrying out interpolation sampling among the plurality of first codes to obtain a plurality of second codes, and the decoder is used for decoding the plurality of second codes to obtain a second training sample set.

The cloud server can conduct interpolation sampling among the first codes through the sampler to obtain second codes. Specifically, as shown in fig. 19, the cloud server may construct a data structure diagram corresponding to a plurality of first encodings by using a model. The data structure diagram comprises a plurality of nodes, one node corresponds to one first code, all the nodes are connected through edges, and the length of the edges between any two nodes is used for reflecting the similarity of codes between the two nodes. Further, the cloud server determines candidate node pairs in the data structure diagram by adopting a model, wherein the candidate node pairs are nodes with any two sides longer than a preset side length in the data structure diagram. And further, the target node pair can be determined according to the candidate node pairs by adopting the model, and new nodes are inserted between the target node pairs, wherein the target node pair is a candidate node pair without other candidate nodes between any two candidate nodes.

The interpolation process of the sampling model is that firstly, an individual representative pair is selected according to distance judgment (the individual representative pair can be generated through all characteristic sample clusters of the individual), and for the individual representative pair meeting the sufficient distance, whether other individuals exist among individual clusters or not is judged through a strategy formulation during each interpolation. If the GAP exists, the GAP is filled, interpolation processing is not needed, and if the GAP does not exist, the GAP is provided with a significance GAP, and a significance interpolation formula is used for carrying out significance interpolation sampling. The interpolation formula may be z= (1- α) ×z1+α×z2, where z is a newly inserted value, z1 and z2 are a pair of individual representative pairs satisfying a sufficient distance, and α is a constant between 0 and 1.

As a possible implementation manner, the cloud server may determine, according to the code corresponding to the first target node and the code corresponding to the second target node, the code corresponding to the node to be inserted, to obtain the node to be disassembled, where the first target node and the second target node are any two target node pairs. Further, the cloud server inserts the node to be torn down between the first target node and the second target node.

For example, the cloud server may perform weighting processing on the code corresponding to the first target node and the code corresponding to the second target node, to obtain the node to be torn down.

The line between any two points shown in fig. 19 represents candidate individual representative pairs screened out according to the distance determination, wherein the solid line represents that no other individual samples exist between individual clusters, and represents that the individual clusters have significance gaps, and significance interpolation and sampling are performed by using an interpolation formula. The dashed lines indicate that there are other individual samples (in this case, white pellet individual representations) between individuals, without interpolation.

Fig. 20 shows an effect diagram after interpolation sampling by the sampling model. Where a of fig. 20 represents a training sample set that is not interpolated, white nodes in b of fig. 20 represent target nodes that satisfy interpolation conditions, and c of fig. 20 represents a training sample set after inserting nodes between the target nodes. As can be seen from fig. 20, for the input first training sample set, the sampling model performs interpolation sampling for the target nodes therein, and inserts new nodes between the target nodes, so that the number of nodes in the output second training sample set is greater than that of the input first training sample set, and the goal of supplementing the first training sample set is reached.

It can be understood that after the sampling model collects all the individual representative pairs meeting the conditions, the new feature vector and the individual label code are obtained through saliency interpolation, and a large number of interpolation sampling samples are generated through a decoder of the sampling model. Furthermore, the cloud server can train the blood pressure prediction model by taking the samples sampled by the two sampling strategies as a training sample set.

S703, the cloud server builds a blood pressure prediction model, and trains the model by using a target training sample set.

The constructed blood pressure prediction model can be any neural network model, such as a transducer model. The constructed blood pressure prediction model consists of a plurality of hidden layers, each hidden layer comprises a linear processing unit and a nonlinear processing unit, and the units are connected in an end-to-end mode so as to realize the task of blood pressure prediction.

As a possible implementation manner, after the cloud server builds the network structure of the blood pressure prediction model, the cloud server takes sample PPG signal data in the target training sample set as input, takes blood pressure values corresponding to the sample PPG signal data as labels, trains the model until the training times reach preset times, and determines that the training is completed.

It can be understood that the blood pressure prediction model trained by the embodiment is deployed on the cloud for predicting the blood pressure of the user of the wearable device, and the traditional blood pressure prediction model is distinguished.

In some embodiments, the training process of the blood pressure prediction model of the present application comprises two parts, meta model training and calibration. The meta model is a general model trained according to a large amount of data (such as a public data set). And the calibration is to perform personalized calibration on the meta-model according to small-batch data of the individual to be tested. After the cloud server trains the meta-model by adopting the method, the trained meta-model can be calibrated according to the specific user group, so that the blood pressure prediction capability of the blood pressure prediction model for the specific user group is improved.

For example, for some older user populations, blood pressure is more likely to be higher than for younger populations. Therefore, after the cloud server is trained to obtain the meta-model, the output result of the meta-model can be finely adjusted upwards so as to achieve the effect of better adapting to the user group with the older age.

It can be appreciated that the training sample is improved by the embodiment of the application, so that the representativeness and coverage of the improved training sample are stronger, and the generalization of the meta-model can be optimized, thereby improving the recognition effect of the meta-model. And in the subsequent fine tuning process of the meta-model, the difficulty of fine tuning is reduced, and the pressure of fine tuning work is reduced.

In addition, the wearable device includes a hardware structure or a software module corresponding to each function, or a combination of both, in order to implement the above functions. Those of skill in the art will readily appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as hardware or combinations of hardware and computer software. Whether a function is implemented as hardware or computer software driven hardware depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.

The embodiment of the application can divide the functional modules of the wearable device according to the method example, for example, each functional module can be divided corresponding to each function, and two or more functions can be integrated in one processing module. The integrated modules may be implemented in hardware or in software functional modules. It should be noted that, in the embodiment of the present application, the division of the modules is schematic, which is merely a logic function division, and other division manners may be implemented in actual implementation. The following description will take an example of dividing each function module into corresponding functions.

Fig. 21 is a schematic structural diagram of a wearable device according to an embodiment of the present application. As shown in fig. 21, the apparatus may include an acquisition unit 801 and a processing unit 802.

The acquisition unit 801 is configured to acquire a photoplethysmographic PPG signal of a user in response to a blood pressure measurement operation of the user, and obtain PPG signal data.

The processing unit 802 is configured to obtain a blood pressure measurement result of a user based on the PPG signal data and a trained blood pressure prediction model, and display the blood pressure measurement result, where the blood pressure prediction model is obtained by training based on a target training sample set, the target training sample is obtained by sampling an original training sample set, a degree of difference between PPG signal data of any two samples in the target training sample set is greater than a preset threshold, the target training sample set includes target sample PPG signal data, and the original training sample set does not include the target sample PPG signal data.

The present embodiment also provides a computer storage medium having stored therein computer instructions which, when executed on an electronic device, cause the electronic device to perform the above-described related method steps to implement the blood pressure measurement method in the above-described embodiments.

The present embodiment also provides a computer program product which, when run on a computer, causes the computer to perform the above-mentioned related steps to implement the blood pressure measurement method in the above-mentioned embodiments.

In addition, the embodiment of the application also provides a device which can be a chip, a component or a module, and the device can comprise a processor and a memory which are connected, wherein the memory is used for storing computer execution instructions, and when the device is operated, the processor can execute the computer execution instructions stored in the memory so as to enable the chip to execute the blood pressure measuring method in each method embodiment.

The electronic device (such as a wearable device, a cloud device, etc.), a computer storage medium, a computer program product, or a chip provided in this embodiment are used to execute the corresponding method provided above, so that the beneficial effects that can be achieved by the electronic device can refer to the beneficial effects in the corresponding method provided above, and are not repeated herein.

It will be appreciated by those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional modules is illustrated, and in practical application, the above-described functional allocation may be performed by different functional modules according to needs, i.e. the internal structure of the apparatus is divided into different functional modules to perform all or part of the functions described above.

In the several embodiments provided by the present application, it should be understood that the disclosed apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of modules or units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another apparatus, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.

While the application has been described in detail with reference to the foregoing embodiments, it will be understood by those skilled in the art that the foregoing embodiments may be modified or equivalents may be substituted for some of the features thereof, and that the modifications or substitutions do not depart from the spirit of the embodiments.

Claims

1. A method of blood pressure measurement, applied to a wearable device, the method comprising:

responding to blood pressure measurement operation of a user, collecting photoplethysmography (PPG) signals of the user, and obtaining PPG signal data;

The blood pressure prediction model is obtained by training based on a target training sample set, the target training sample is obtained by sampling an original training sample set, the difference degree between PPG signal data of any two samples in the target training sample set is larger than a preset threshold, the target training sample set contains target sample PPG signal data, and the original training sample set does not contain the target sample PPG signal data.

2. The method of claim 1, wherein the wearable device is deployed with the blood pressure prediction model, wherein the obtaining a blood pressure measurement of the user based on the PPG signal data and the trained blood pressure prediction model comprises:

and inputting the PPG signal data into the blood pressure prediction model to obtain a blood pressure measurement result of the user.

3. The method according to claim 1, wherein the obtaining a blood pressure measurement of the user based on the PPG signal data and a trained blood pressure prediction model comprises:

sending a blood pressure measurement request containing the PPG signal data to a cloud, wherein the cloud is provided with the blood pressure prediction model;

And receiving a blood pressure measurement result sent by the cloud, wherein the blood pressure measurement result is obtained by inputting the PPG signal data into the blood pressure prediction model by the cloud.

4. A blood pressure measurement method, applied to a cloud, the method comprising:

receiving a blood pressure measurement request containing PPG signal data sent by a wearable device;

The method comprises the steps of inputting PPG signal data into a trained blood pressure prediction model to obtain a blood pressure measurement result, and sending the blood pressure measurement result to the wearable device, wherein the blood pressure prediction model is obtained by training based on a target training sample set, the target training sample is obtained by sampling an original training sample set, the difference degree between PPG signal data of any two samples in the target training sample set is larger than a preset threshold, the target training sample set contains target sample PPG signal data, and the original training sample set does not contain the target sample PPG signal data.

5. The method according to claim 4, wherein the method further comprises:

acquiring the original training sample set;

Sampling the original training sample set to obtain the target training sample set;

And taking sample PPG signal data in the target training sample set as input, taking blood pressure corresponding to the sample PPG signal data in the target training sample set as a label, and training to obtain the blood pressure prediction model.

6. The method of claim 5, wherein sampling the original training sample set to obtain the target training sample set comprises:

The method comprises the steps of carrying out first sampling on an original training sample set based on a first sampling strategy to obtain a first training sample set, wherein the number of sample PPG signal data in the first training sample set is smaller than that in the original training sample set, and the degree of difference between any two sample PPG signal data in the first training sample set is larger than the preset threshold value;

Performing second sampling on the first training sample set based on a second sampling strategy to obtain a second training sample set, wherein the second training sample set contains the target sample PPG signal data, and the original training sample set and the first training sample set do not contain the target sample PPG signal data;

and taking the second training sample set as the target training sample set.

7. The method of claim 5, wherein sampling the original training sample set to obtain the target training sample set comprises:

Performing second sampling on the original training sample set based on a second sampling strategy to obtain a second training sample set, wherein the second training sample set contains the target sample PPG signal data, and the original training sample set does not contain the target sample PPG signal data;

And taking the first training sample set and the second training sample set as the target training sample set.

8. The method of claim 6 or 7, wherein the first sampling strategy comprises diversity sampling and the second sampling strategy comprises interpolation sampling.

9. The method of claim 6, wherein the performing a first sampling on the original training sample set based on a first sampling strategy to obtain a first training sample set comprises:

The method comprises the steps of inputting sample PPG signal data in an original training sample set into a trained classification model to obtain a classification result of the sample PPG signal data, wherein the classification model is used for extracting identity characteristics of each sample PPG signal data and classifying each sample PPG signal data according to the identity characteristics, and the sample PPG signal data of the same class in the classification result corresponds to the same sample user;

and respectively carrying out first sampling in sample PPG signal data of each class based on the first sampling strategy to obtain the first training sample set.

10. The method according to claim 9, wherein the performing, based on the first sampling strategy, first sampling in each class of sample PPG signal data, respectively, to obtain the first training sample set includes:

for sample PPG signal data of any category, collecting sample PPG signal data of a target number from sample PPG signal data of a current category to obtain a first sampling data set, wherein the cosine similarity between the identity features of any two sample PPG signal data in the first sampling data set is larger than a first preset similarity;

each first sampling data set is used as the first training sample set.

11. The method according to claim 10, wherein the collecting a target number of sample PPG signal data from the current class of sample PPG signal data, resulting in a first sample data set, comprises:

for any one collected sample PPG signal data, respectively calculating cosine similarity between the non-collected sample PPG signal data and the collected sample PPG signal data;

And adding the sample PPG signal data with the maximum cosine similarity corresponding to the sample PPG signal data which is not acquired to the first sampling data set.

12. The method according to claim 9, wherein the method further comprises:

Dividing the sample PPG signal data in the original training sample set according to the belonging sample users to obtain sample PPG signal data of a plurality of sample users;

And taking sample PPG signal data of the plurality of sample users as input, setting a cosine margin normalized loss function Angular-margin Softmax, and training to obtain the classification model, wherein the Angular-margin Softmax is used for indicating the classification model to maximize the probability of the category and minimize the probability gap with the adjacent category.

13. The method of claim 6, wherein performing a second sampling on the first training sample set based on a second sampling strategy to obtain a second training sample set comprises:

And inputting the first training sample set into a trained sampling model to obtain the second training sample set.

14. The method of claim 13, wherein the sampling model comprises an encoder, a sampler, and a decoder, and wherein inputting the first training sample set into the trained sampling model to obtain the second training sample set comprises:

the encoder is used for encoding the sample PPG signal data in the first training sample set according to the sample user to obtain a plurality of first codes, wherein one first code corresponds to the sample PPG signal data in one sample user in the first training sample set;

performing interpolation sampling among the plurality of first codes through the sampler to obtain a plurality of second codes;

And decoding the plurality of second codes through the decoder to obtain the second training sample set.

15. The method of claim 14, wherein interpolating samples between the plurality of first codes by the sampler to obtain a plurality of second codes comprises:

Constructing a data structure diagram corresponding to the plurality of first codes, wherein the data structure diagram comprises a plurality of nodes, one node corresponds to one first code, all the nodes are connected through edges, and the length of the edge between any two nodes is used for reflecting the similarity of codes between the two nodes;

Determining candidate node pairs in the data structure diagram, wherein the candidate node pairs are nodes with any two sides longer than a preset side length in the data structure diagram;

And determining a target node pair according to the candidate node pairs, and inserting new nodes between the target node pairs, wherein the target node pair is a candidate node pair without other candidate nodes between any two candidate nodes.

16. The method of claim 15, wherein the inserting new nodes between the target node pair comprises:

Determining codes corresponding to nodes to be inserted according to codes corresponding to a first target node and codes corresponding to a second target node to obtain nodes to be disassembled;

And inserting the node to be disassembled between the first target node and the second target node.

17. The method of claim 16, wherein determining the code corresponding to the node to be inserted according to the code corresponding to the first target node and the code corresponding to the second target node, to obtain the node to be disassembled, comprises:

And weighting the codes corresponding to the first target node and the codes corresponding to the second target node to obtain the node to be disassembled.

18. A wearable device, comprising:

one or more processors;

A memory;

and one or more computer programs, wherein the one or more computer programs are stored on the memory, which when executed by the one or more processors, cause the wearable device to perform the blood pressure measurement method of any of claims 1-3.

19. A cloud device, comprising:

one or more processors;

A memory;

And one or more computer programs, wherein the one or more computer programs are stored on the memory, which when executed by the one or more processors, cause the cloud device to perform the blood pressure measurement method of any of claims 4-17.

20. A computer readable storage medium comprising a computer program, characterized in that the computer program, when run on an electronic device, causes the electronic device to perform the blood pressure measurement method of any one of claims 1-3 or the blood pressure measurement method of any one of claims 4-17.