Blood pressure measurement method, apparatus, and related devices

Abstract

The application discloses a blood pressure measurement method and device, computer equipment and a storage medium. The method comprises the following steps: obtaining a face video and individual information of a user to be measured; inputting the face video and individual information of the user to be measured into a blood pressure prediction model to obtain systolic pressure and diastolic pressure of the user to be measured, wherein the blood pressure prediction model comprises a space-time feature extraction module and a blood pressure signal prediction module, the space-time feature extraction module is used for performing feature extraction on the face video of the user to be measured to obtain a feature vector containing global space information and time information; the blood pressure signal prediction module is used for fusing the feature vector and the individual information, and predicting the systolic pressure and diastolic pressure of the user to be measured based on the fused feature vector; and outputting the systolic pressure and diastolic pressure of the user to be measured. The application can reduce the blood pressure measurement complexity and cost, improve the blood pressure measurement efficiency and measurement accuracy without direct contact with the measured person.

Description

Technical Field

[0001] This invention relates to the fields of facial video feature processing and blood pressure measurement technology, and in particular to a blood pressure measurement method, device, computer equipment, and storage medium. Background Technology

[0002] Blood pressure refers to the pressure exerted by the heart on the blood vessels as it pumps blood to the body. It is a crucial health indicator, playing a vital role in medical diagnosis and health monitoring. Mercury sphygmomanometers and cuff-type electronic blood pressure monitors are currently the most common and mainstream blood pressure measurement devices. Mercury sphygmomanometers provide relatively accurate results, but require contact with the person being measured, making operation more complex and demanding a higher level of expertise. Electronic blood pressure monitors also require contact with the person being measured, but they have several limitations, such as accuracy being affected by sensor position and the inability to perform continuous measurements.

[0003] Photoplethysmography (PPG) is a method that uses photoelectric technology to detect human pulse waves. It is based on the principle that the subcutaneous blood volume changes periodically with the periodic ejection of blood from the heart. A photodetector is used to detect this pulsating change to obtain the volumetric pulse wave signal. With the development of this technology, many blood pressure measurement devices based on photoplethysmography (such as blood pressure monitoring wristbands) have appeared on the market. However, this has also increased the cost of blood pressure measurement. These devices generally have problems such as low measurement accuracy and insufficient ability to capture abnormal blood pressure.

[0004] Therefore, improving the flexibility of blood pressure measurement, reducing its complexity and cost, and enhancing its efficiency and accuracy are urgent problems to be solved. Summary of the Invention

[0005] This invention provides a blood pressure measurement method, device, computer equipment, and storage medium that do not require direct contact with the subject, thereby improving the flexibility of blood pressure measurement, reducing the complexity and cost of blood pressure measurement, and improving the efficiency and accuracy of blood pressure measurement.

[0006] To address the aforementioned technical problems, this application provides a blood pressure measurement method, comprising the following steps: acquiring a facial video and individual information of a user to be tested, wherein the facial video includes multiple facial image sequences; inputting the facial video and individual information of the user to be tested into a blood pressure prediction model to obtain the systolic and diastolic blood pressure of the user to be tested, wherein the blood pressure prediction model includes a spatiotemporal feature extraction module and a blood pressure signal prediction module, wherein the spatiotemporal feature extraction module is used to extract features from the facial video of the user to be tested to obtain a feature vector containing global spatial information and temporal information; the blood pressure signal prediction module is used to fuse the feature vector and the individual information, and predict the systolic and diastolic blood pressure of the user to be tested based on the fused feature vector; and outputting the systolic and diastolic blood pressure of the user to be tested.

[0007] In one possible implementation, the facial video of the user to be tested is input into a shallow feature extraction network for convolutional processing to output a shallow feature map. The shallow feature extraction network includes multiple 2D convolutional blocks. Based on the shallow feature map and a multi-scale kernel spatial attention mechanism, an attention feature map is obtained. The attention feature map is input into an encoder based on step-by-step spatiotemporal self-attention for encoding to obtain the feature vector containing global spatial information and temporal information.

[0008] In another possible implementation, multiple convolutional kernels of different sizes are used to convolve the shallow feature map to extract a multi-scale feature map; the multi-scale feature maps are concatenated and global max pooling and global average pooling are performed on the channel dimension respectively to obtain a global max pooling feature map and a global average pooling feature map; an attention weight map is obtained based on the global max pooling feature map and the global average pooling feature map; and an attention feature map is obtained based on the attention weight map and the shallow feature map.

[0009] In another possible implementation, the attention feature map is input into the embedding network to obtain the embedding vector token corresponding to the attention feature map; the embedding vector token is input into the longformer-based encoding network to output the feature vector containing global spatial information and temporal information, wherein the longformer-based encoding network is used to apply a self-attention mechanism to the embedding vector token in the temporal and spatial dimensions.

[0010] In another possible implementation, the individual information is input into a linear mapping network for mapping processing to obtain the individual features corresponding to the individual information; the individual features and the feature vector are fused using a self-attention mechanism to generate a fused feature vector; the fused feature vector is input into a fully connected layer network to obtain the predicted values ​​of the systolic and diastolic blood pressure of the user to be tested.

[0011] In another possible implementation, sample data is acquired, including sample facial videos, individual information corresponding to the sample facial videos, and the actual blood pressure values ​​corresponding to the sample facial videos, wherein the actual blood pressure values ​​include systolic and diastolic blood pressure; the sample data is input into an initial blood pressure prediction model; the initial blood pressure prediction model predicts the blood pressure value corresponding to the sample facial videos, and compares the obtained blood pressure value with the actual blood pressure value to calculate a blood pressure loss function; the parameters and weights in the initial blood pressure prediction model are adjusted according to the blood pressure loss function to obtain the blood pressure prediction model.

[0012] To address the aforementioned technical problems, this application also provides a blood pressure measurement device, comprising: a data acquisition module for acquiring a facial video and individual information of a user to be tested, wherein the facial video includes multiple facial image sequences; a prediction module for inputting the facial video and individual information of the user to be tested into a blood pressure prediction model to obtain the systolic and diastolic blood pressure of the user to be tested, wherein the blood pressure prediction model includes a spatiotemporal feature extraction module and a blood pressure signal prediction module, wherein the spatiotemporal feature extraction module is used to extract features from the facial video of the user to be tested to obtain a feature vector containing global spatial information and temporal information; the blood pressure signal prediction module is used to fuse the feature vector and the individual information, and predict the systolic and diastolic blood pressure of the user to be tested based on the fused feature vector; and an output module for outputting the systolic and diastolic blood pressure of the user to be tested.

[0013] To address the aforementioned technical problems, this application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0014] To address the aforementioned technical problems, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method.

[0015] The blood pressure measurement method, apparatus, computer equipment, and storage medium provided in this invention, after acquiring the facial video of the user to be measured, directly predict the user's blood pressure through a pre-built end-to-end blood pressure prediction model. This avoids the problem of decreased measurement accuracy caused by signal processing and manual feature extraction, reducing the complexity and cost of blood pressure measurement. Furthermore, by leveraging the correlation between individual information and blood pressure in different population groups, the individual information of the user to be measured, such as gender, age, height, and weight, is fused with the feature information contained in the facial video to assist in blood pressure prediction, further improving measurement accuracy. Moreover, the entire measurement process does not require direct contact with the user, improving detection flexibility and broadening application scenarios. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is an exemplary system architecture diagram in which this application can be applied.

[0018] Figure 2 This is a flowchart of one embodiment of the blood pressure measurement method of this application.

[0019] Figure 3 This is a schematic diagram of the blood pressure prediction model structure of this application.

[0020] Figure 4 This is a schematic diagram of the multi-scale kernel spatial attention mechanism structure of this application.

[0021] Figure 5 This is a flowchart of an embodiment of the blood pressure prediction model training process of this application.

[0022] Figure 6 This is a schematic diagram of one embodiment of the blood pressure measuring device according to this application.

[0023] Figure 7 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0025] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Please see Figure 1 ,like Figure 1 As shown, system architecture 100 may include terminal device 110, network 120, and server 130. Network 120 is used as a medium to provide a communication link between terminal device 110 and server 130. Network 120 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0028] Users can use terminal device 110 to interact with server 130 via network 120 to receive or send messages, etc.

[0029] Terminal device 110 can be various electronic devices with a display screen and support for web browsing and camera functions, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0030] Server 130 can be a server that provides various services, such as a backend server that supports the pages displayed on terminal device 110.

[0031] It should be noted that the blood pressure measurement method provided in this application embodiment is executed by a server, and correspondingly, the blood pressure measuring device is set in the server.

[0032] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included. The terminal device 110 in this embodiment can specifically correspond to an application system in actual production.

[0033] Please see Figure 2 , Figure 2 This invention illustrates a blood pressure measurement method provided by an embodiment of the present invention, which is applied to... Figure 1 The following is a detailed explanation using the server-side example.

[0034] S201. Obtain the facial video and individual information of the user to be tested.

[0035] Specifically, the user being tested can use a terminal device (such as a mobile phone) to capture a video containing their entire face, and then send the video along with their personal information to the server via the network. Alternatively, they can send a pre-stored video containing their entire face and their personal information to the server, which will then process the video and personal information further. It should be noted that the user's personal information can include basic details such as gender, age, height, and weight.

[0036] Optionally, after receiving the facial video of the user to be tested, the server can preprocess the facial video, such as scaling the facial images contained in the facial video to the same scale, or normalizing and subtracting the mean from the facial images, so as to ensure that all data are consistent in subsequent processing and improve data processing efficiency.

[0037] S202. Input the facial video and individual information of the user to be tested into the blood pressure prediction model to obtain the systolic and diastolic blood pressure of the user to be tested.

[0038] Specifically, after receiving the user's facial video and personal information, the server can input the facial image sequence contained in the facial video, along with the user's personal information, into a pre-trained blood pressure prediction model that has the ability to measure blood pressure, thereby obtaining the systolic and diastolic blood pressure of the user to be tested.

[0039] For example, such as Figure 3As shown, the blood pressure prediction model includes a spatiotemporal feature extraction module and a blood pressure signal prediction module. The spatiotemporal feature extraction module is used to extract features from the facial video of the user to be tested, and obtain a feature vector containing global spatial information and temporal information. Then, the blood pressure signal prediction module fuses the feature vector with the individual information of the user to be tested, and obtains the fused feature vector. Based on the fused feature vector, the systolic blood pressure and diastolic blood pressure of the user to be tested are predicted.

[0040] It should be noted that the blood pressure signal prediction model is a detection model built on a convolutional neural network. Before using it for blood pressure measurement, it needs to be trained to enable it to make predictions. The training process of the blood pressure signal prediction model will be explained in detail later.

[0041] In one possible implementation, the facial video of the user to be tested is input into a shallow feature extraction network for convolution processing, and a shallow feature map is output. The shallow feature map extraction network includes multiple 2D convolutional blocks. Based on the shallow feature map and a multi-scale kernel attention mechanism, an attention feature map is obtained. The attention feature map is input into an encoder based on step-by-step spatiotemporal self-attention for encoding, resulting in a feature vector containing global spatial information and temporal information.

[0042] Specifically, for each facial image frame contained in the input facial video, each 2D convolutional block undergoes a 2D convolution operation and is non-linearized using an activation function, ultimately yielding a shallow feature map, which contains a significant amount of local spatial information. It should be understood that the shallow feature map is essentially an F×W×H×C matrix, where F represents the number of features, W represents the width, H represents the height, and C represents the number of channels.

[0043] In another possible implementation, multiple convolutional kernels of different sizes are used to perform convolution operations on the shallow feature map to extract multi-scale feature maps. The multi-scale feature maps are concatenated and global max pooling and global average pooling are performed on the channel dimension to obtain global max pooling feature maps and global average pooling feature maps. Based on the global max pooling feature maps and the global average pooling feature maps, attention weight maps are obtained. Based on the attention weight maps and the shallow feature maps, attention feature maps are obtained.

[0044] Specifically, after obtaining the shallow feature map, in order to assign greater weight to the space rich in effective signal distribution in the shallow feature map, convolutional operations can be performed on the shallow feature map using convolutional kernels of different sizes to extract multi-scale features under different receptive fields, resulting in multiple multi-scale feature maps at different scales. Then, the multi-scale feature maps are concatenated, and global max pooling and global average pooling are performed in the channel dimension to obtain global max pooling feature maps and global average pooling feature maps. Next, the two pooling feature maps are concatenated again, and then passed through a 1×1 convolutional layer and activation function to obtain the attention weight map. Finally, the generated attention weight map is multiplied element-wise with the shallow feature map, and the attention feature map with enhanced attention weights is obtained.

[0045] It is worth noting that after multi-scale feature maps are concatenated, there are multiple channels. Global max pooling in the channel dimension means taking the largest value among all the values ​​in different channels at the same spatial location on the feature map to represent that location. Similarly, global average pooling means taking the average value among all the values ​​to represent that location. In this way, the feature map is compressed in the channel dimension, that is, the resulting feature map has only one channel.

[0046] For example, such as Figure 4 As shown, 1×1, 3×3, and 5×5 convolutional kernels are used to perform convolution operations on shallow feature maps, and then the convolution results are concatenated. Global max pooling and global average pooling are then performed on the concatenated feature maps, and the pooling results are concatenated. Finally, an attention weight map is generated through a 1×1 convolutional layer and an activation function.

[0047] In another possible implementation, the attention feature map is input into the embedding network to obtain the embedding vector token corresponding to the attention feature map. Then, the embedding vector token is input into the longformer-based encoding network to output a feature vector containing global spatial and temporal information. The longformer-based encoding network is used to apply a self-attention mechanism to the embedding vector token in the temporal and spatial dimensions.

[0048] Specifically, the encoder based on step-by-step spatiotemporal self-attention consists of an embedding layer and multiple encoding blocks containing longformer-based encoding. The embedding layer represents the input attention feature map as a vector and encodes it in combination with spatial and temporal positions to obtain a set of input tokens. Then, the longformer-based encoding blocks sequentially apply a self-attention mechanism to the input tokens in the temporal and spatial dimensions to obtain feature vectors containing global spatial and temporal information.

[0049] It should be understood that the essence of the embedding layer is to represent the input as a sequence. An input image is divided into a group of image patches, and a linear transformation is applied to each image patch to map it to a low-dimensional feature space. This results in a serialized feature representation. Each image patch (including spatial and temporal positions) corresponds to a token. Finally, the self-attention mechanism contained in the longformer encoding block can obtain the spatiotemporal relationship between image patches in the temporal and spatial dimensions, respectively. These relationships are reflected by weights, and the final generated feature vector is the result after being assigned different weights.

[0050] In another possible implementation, the individual information corresponding to the facial video of the user to be tested is input into a linear mapping network for mapping processing to obtain the individual features corresponding to the individual information. Then, the individual features and feature vectors are fused using a self-attention mechanism to generate a fused feature vector. Finally, the fused feature vector is input into a fully connected layer network to obtain the predicted values ​​of the systolic and diastolic blood pressure of the user to be tested.

[0051] Specifically, after obtaining the individual information of the user to be tested, a linear mapping network is used to map this individual information to a high-dimensional space as individual features. The specific mapping process can be represented by the following formula:

[0052] h d =w×x d +b

[0053] Where, x d Let w represent individual information, w represent the mapping weight matrix, b represent the bias term, and h represent the bias term. d This represents the individual characteristics after spatial mapping.

[0054] After obtaining individual characteristics, the individual characteristics and feature vectors are fused using a self-attention mechanism to generate a fused feature vector. Finally, the fused feature vector is passed through a fully connected layer to obtain the predicted values ​​of the systolic blood pressure (SBP) and diastolic blood pressure (DBP) of the user under test.

[0055] It can be seen that the blood pressure signal prediction model directly predicts blood pressure through the constructed end-to-end blood pressure prediction network, which can effectively avoid the problem of decreased blood pressure measurement accuracy caused by signal processing and manual feature extraction, and reduce the complexity and cost of blood pressure measurement. In addition, by fusing the individual information of the user to be tested with the feature information contained in the facial video through the correlation between individual information of different groups and blood pressure, the measurement accuracy can be further improved. Moreover, the entire measurement process does not require direct contact with the user being tested, which improves the detection flexibility and broadens the application scenarios.

[0056] S203, outputs the systolic and diastolic blood pressure of the user to be tested.

[0057] Specifically, after obtaining the systolic and diastolic blood pressure of the user through the blood pressure prediction model, it can be sent to the terminal device associated with the user, such as a mobile phone, for display, thereby helping the user to clearly know their own blood pressure information.

[0058] The training process of the blood pressure prediction model will be explained below. Please refer to [link / reference]. Figure 5 .

[0059] S501. Obtain sample data.

[0060] Specifically, the acquired sample data includes sample facial videos, the individual information corresponding to the sample facial videos, and the real blood pressure values ​​corresponding to the sample facial videos, including systolic and diastolic blood pressure. Furthermore, based on existing image acquisition terminals, sample facial videos are acquired in a real, natural, and open environment, or facial videos are obtained from a facial video database. The video duration can be selected as needed; for example, a continuous 15-second facial video can be selected as the sample facial video, and this application does not limit this selection. For each facial video, the individual information corresponding to the facial video is recorded, including gender, age, height, weight, etc., and the real-time blood pressure values ​​(including systolic and diastolic blood pressure) of the subject corresponding to the facial video can be obtained through an electronic blood pressure monitor.

[0061] It should be noted that in the process of acquiring sample face videos, it is necessary to collect face videos of different qualities in order to improve the model's ability to generalize data and expand the model's applicability.

[0062] S502. Input the sample data into the initial blood pressure prediction model for training.

[0063] Specifically, after inputting the sample face video, the initial blood pressure prediction model extracts features through the time-matter feature extraction module to obtain feature vectors. Then, the blood pressure signal prediction module fuses the feature vectors and individual information, and makes predictions based on the fused feature vectors to obtain the predicted systolic and diastolic blood pressure. The predicted systolic and diastolic blood pressures are then compared with the actual systolic and diastolic blood pressures to calculate the blood pressure loss function. Here, the mean square error between the predicted and actual blood pressures can be used as the loss function.

[0064] S503. Adjust the parameters and weights in the initial blood pressure prediction model according to the blood pressure loss function to obtain the blood pressure prediction model.

[0065] Specifically, after calculating the blood pressure loss function, the parameters and weights in the initial blood pressure prediction model are adjusted according to the blood pressure loss function. Through continuous iteration, that is, by continuously adjusting the parameters and weights in the network, until the error between the predicted blood pressure output by the prediction model and the actual blood pressure is less than the threshold (or the blood pressure loss function converges), the blood pressure prediction model is trained and has the ability to measure blood pressure.

[0066] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0067] Figure 6 A block diagram illustrating the principle of a blood pressure measuring device corresponding to the blood pressure measurement methods described in the above embodiments is shown. Figure 6 As shown, the multi-blood pressure measurement device 600 includes a data acquisition module 610, a prediction module 620, and an output module 630. Detailed descriptions of each functional module are as follows.

[0068] The data acquisition module 610 is used to acquire facial videos and individual information of the user to be tested, wherein the facial videos include multiple facial image sequences.

[0069] The prediction module 620 is used to input the facial video of the user to be tested and the individual information into the blood pressure prediction model to obtain the systolic and diastolic blood pressure of the user to be tested. The blood pressure prediction model includes a spatiotemporal feature extraction module and a blood pressure signal prediction module. The spatiotemporal feature extraction module is used to extract features from the facial video of the user to be tested to obtain a feature vector containing global spatial information and temporal information. The blood pressure signal prediction module is used to fuse the feature vector and the individual information, and predict the systolic and diastolic blood pressure of the user to be tested based on the fused feature vector.

[0070] The output module 630 is used to output the systolic and diastolic blood pressure of the user under test.

[0071] For specific limitations regarding the blood pressure measuring device, please refer to the limitations on blood pressure measurement methods above, which will not be repeated here. Each module in the aforementioned blood pressure measuring device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independently of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0072] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 7 , Figure 7 This is a basic structural block diagram of the computer device in this embodiment.

[0073] The computer device 700 includes a memory 710, a processor 720, and a network interface 730 that are interconnected via a system bus. It should be noted that the figure only shows a computer device 700 with components connected to the memory 710, processor 720, and network interface 730; however, it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0074] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0075] The memory 710 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or D-interface display memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 710 may be an internal storage unit of the computer device 700, such as the hard disk or memory of the computer device 700. In other embodiments, the memory 710 may also be an external storage device of the computer device 700, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. Of course, the memory 710 may also include both internal storage units and external storage devices of the computer device 700. In this embodiment, the memory 710 is typically used to store the operating system and various application software installed on the computer device 700, such as program code for controlling electronic documents. Furthermore, the memory 710 can also be used to temporarily store various types of data that have been output or will be output.

[0076] In some embodiments, the processor 720 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 720 is typically used to control the overall operation of the computer device 700. In this embodiment, the processor 720 is used to run program code stored in the memory 710 or process data, such as program code controlling electronic documents.

[0077] The network interface 730 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 700 and other electronic devices.

[0078] This application also provides another embodiment, namely, a computer-readable storage medium storing an interface display program that can be executed by at least one processor to cause the at least one processor to perform the steps of the multi-physiological index detection method described above.

[0079] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0080] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A method for measuring blood pressure, characterized in that, The method includes: Acquire facial videos and individual information of the user to be tested, wherein the facial videos include multiple facial image sequences; The facial video of the user to be tested is input into a shallow feature extraction network for convolution processing, and a shallow feature map is output. The shallow feature extraction network includes multiple 2D convolutional blocks. Based on the shallow feature map and the multi-scale kernel spatial attention mechanism, an attention feature map is obtained; The attention feature map is input into an encoder based on step-by-step spatiotemporal self-attention for encoding to obtain a feature vector containing global spatial information and temporal information; The feature vector and the individual information are fused, and the systolic and diastolic blood pressure of the user under test are predicted based on the fused feature vector. Output the systolic and diastolic blood pressure of the user being tested.

2. The method as described in claim 1, characterized in that, The attention feature map obtained based on the shallow feature map and the multi-scale kernel spatial attention mechanism includes: The shallow feature map is convolved using multiple convolution kernels of different sizes to extract a multi-scale feature map; The multi-scale feature maps are concatenated and global max pooling and global average pooling are performed on the channel dimension respectively to obtain global max pooling feature maps and global average pooling feature maps. Based on the global max pooling feature map and the global average pooling feature map, an attention weight map is obtained; The attention feature map is obtained based on the attention weight map and the shallow feature map.

3. The method as described in claim 1, characterized in that, The step of inputting the attention feature map into an encoder based on stepwise spatiotemporal self-attention for encoding to obtain the feature vector containing global spatial information and temporal information includes: The attention feature map is input into the embedding network to obtain the embedding vector token corresponding to the attention feature map; The embedding vector token is input into a longformer-based encoding network, which outputs a feature vector containing global spatial and temporal information. The longformer-based encoding network is used to apply a self-attention mechanism to the embedding vector token in the temporal and spatial dimensions.

4. The method as described in claim 1, characterized in that, The process of fusing the feature vector and the individual information, and predicting the systolic and diastolic blood pressure of the user under test based on the fused feature vector, includes: The individual information is input into a linear mapping network for mapping processing to obtain the individual features corresponding to the individual information; The individual features and the feature vector are fused using a self-attention mechanism to generate a fused feature vector; The fused feature vector is input into a fully connected layer network to obtain the predicted values ​​of the systolic and diastolic blood pressure of the user under test.

5. The method as described in claim 1, characterized in that, The method further includes: Acquire sample data, which includes sample face videos, individual information corresponding to the sample face videos, and real blood pressure values ​​corresponding to the sample face videos, wherein the real blood pressure values ​​include systolic blood pressure and diastolic blood pressure; The sample data is input into the initial blood pressure prediction model; The initial blood pressure prediction model predicts the blood pressure value corresponding to the sample face video, and compares the obtained blood pressure value with the real blood pressure value to calculate the blood pressure loss function; The parameters and weights in the initial blood pressure prediction model are adjusted according to the blood pressure loss function to obtain the blood pressure prediction model.

6. A blood pressure measuring device, characterized in that, The device includes: The data acquisition module is used to acquire facial videos and individual information of the user to be tested, wherein the facial videos include multiple facial image sequences; The prediction module is used to input the facial video of the user to be tested into a shallow feature extraction network for convolutional processing, and output a shallow feature map. The shallow feature extraction network includes multiple 2D convolutional blocks. Based on the shallow feature map and a multi-scale kernel spatial attention mechanism, an attention feature map is obtained. The attention feature map is input into an encoder based on stepwise spatiotemporal self-attention for encoding to obtain a feature vector containing global spatial information and temporal information. The feature vector and the individual information are fused, and the systolic and diastolic blood pressure of the user to be tested are predicted based on the fused feature vector. The output module is used to output the systolic and diastolic blood pressure of the user under test.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 5.

Images