Multi-person blood pressure measurement system, medium, and equipment based on ultrasonic signals from smart speakers

Abstract

The present invention relates to the technical field of blood pressure detection, and discloses a multi-person blood pressure measurement system, medium, and equipment based on intelligent speaker ultrasonic signals, including: intelligent speakers, emitting ultrasonic signals of different frequencies and receiving corresponding signals; a phase compensation module, combining spatial domain beamforming and frequency domain beamforming to construct the phase difference of the subcarrier of the microphone compared to the subcarrier emitted by the speaker and perform phase compensation; a multi-user signal separation module, separating user signals at different angles; an ultrasonic heartbeat signal extraction module, extracting heartbeat signals from user signals at each angle; a blood pressure measurement network model construction module, constructing a blood pressure measurement network model; a blood pressure prediction module, training the blood pressure measurement network model, and inputting the extracted heartbeat signal into the trained model to obtain a blood pressure prediction result. The present invention can realize non-contact, real-time and effective measurement of blood pressure of multiple people, and has high ease of use and accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of blood pressure detection, and in particular to a multi-person blood pressure measurement system, medium, and equipment based on intelligent speaker ultrasonic signals. Background Art

[0002] Hypertension is a serious global health problem. Approximately 1.28 billion adults aged 30 to 79 suffer from hypertension worldwide, causing more than 7.5 million deaths each year. Hypertension is usually asymptomatic, making early detection and monitoring of hypertension crucial.

[0003] Common blood pressure measurement methods include contact and non-contact. Contact blood pressure measurements include the mercury sphygmomanometer method and the automatic oscillation method. The mercury sphygmomanometer method is currently the gold standard for contact blood pressure measurement, which uses a blood pressure cuff and a stethoscope to gradually measure systolic and diastolic blood pressure. Another contact-based blood pressure measurement technology available in the consumer market is the automatic oscillation method, which involves tying a cuff on the arm, injecting gas, and then systematically releasing the gas to simultaneously monitor arterial pressure to measure blood pressure. The non-contact method mainly relies on changes in wireless signals such as millimeter waves and continuous waves, as well as ultra-wideband signals to detect subtle changes in the skin caused by pulse activity, and can also effectively obtain blood pressure measurements.

[0004] However, contact measurement methods are highly dependent on wearable devices and are easily affected by the use and wearing errors of the devices. Although non-contact measurement methods can reduce the dependence on wearable devices, non-contact measurement devices are affected by distance and position. As the distance increases, the wireless signal rapidly attenuates, which greatly reduces the measurement accuracy. In addition, due to the influence of signal interference, non-contact measurement methods cannot effectively detect the blood pressure of multiple people at the same time. Summary of the invention

[0005] To this end, the technical problem to be solved by the present invention is to overcome the deficiencies in the prior art and provide a multi-person blood pressure measurement system based on smart speaker ultrasonic signals, which can effectively measure the blood pressure of multiple people while reducing dependence on wearable devices, and has high ease of use and accuracy.

[0006] In order to solve the above technical problems, the present invention provides a multi-person blood pressure measurement system based on intelligent speaker ultrasonic signals, comprising:

[0007] A smart speaker, comprising a speaker and a linear microphone array, wherein the speaker transmits ultrasonic signals of different frequencies, and the linear microphone array receives the ultrasonic signals after passing through the user;

[0008] A phase compensation module, used to combine spatial domain beamforming and frequency domain beamforming to construct a phase difference between each subcarrier of each microphone in the linear microphone array and the subcarrier emitted by the speaker, and to perform phase compensation on each subcarrier of each microphone in the linear microphone array according to the phase difference;

[0009] The multi-user signal separation module is used to track the main dynamic components of the phase-compensated signal at multiple consecutive instants, and separate the user signals at different angles by combining the static components;

[0010] Ultrasonic heartbeat signal extraction module, used to extract heartbeat signals from user signals at each angle position;

[0011] A blood pressure measurement network model construction module, used to construct a blood pressure measurement network model, wherein the blood pressure measurement network model includes an input module, a depthwise separable convolution module, and a channel attention module;

[0012] The blood pressure prediction module trains the blood pressure measurement network model to obtain a trained blood pressure measurement network model, and inputs the extracted heartbeat signal into the trained blood pressure measurement network model to obtain a blood pressure prediction result.

[0013] Furthermore, the phase difference of each subcarrier of each microphone in the linear microphone array compared with the subcarrier emitted by the speaker is constructed by combining spatial domain beamforming and frequency domain beamforming, specifically:

[0014] Using spatial beamforming, the phase difference of the mth microphone relative to the first microphone is expressed in the form of a complex exponential:

[0015] ,

[0016] in, is the phase difference of the mth microphone relative to the first microphone, exp( ) represents the natural exponential function, j is the imaginary unit, is the frequency of the nth subcarrier of the mth microphone, represents the distance between two adjacent microphones, is the incident angle of the ultrasonic signal of the first microphone relative to the first microphone, m=1,2,…,M-1, M is the number of microphones in the linear microphone array; c is the propagation speed of ultrasonic wave in air;

[0017] Using frequency domain beamforming, the phase difference of the nth subcarrier of the microphone relative to the first subcarrier is expressed in the form of a complex exponential:

[0018] ,

[0019] in, is the phase difference of the nth subcarrier of the microphone relative to the 1st subcarrier, Indicates the frequency difference between two adjacent subcarriers received by the microphone. is the propagation path length of the ultrasonic signal received by the first microphone, n=1,2,…,N-1, N is the number of subcarriers of the microphone;

[0020] Combining the phase difference of the mth microphone relative to the first microphone and the phase difference of the nth subcarrier of the microphone relative to the first subcarrier, the phase difference of the nth subcarrier of the mth microphone relative to the first subcarrier of the first microphone is constructed as:

[0021] ,

[0022] in, is the phase difference between the nth subcarrier of the mth microphone and the 1st subcarrier of the 1st microphone.

[0023] Further, phase compensation is performed on each subcarrier of each microphone in the linear microphone array according to the phase difference, specifically:

[0024] The steering vector composed of the phase difference introduced by M microphones and N subcarriers is constructed as:

[0025] ,

[0026] in, is the steering vector composed of the phase differences introduced by M microphones and N subcarriers, and T represents the transposition operation;

[0027] The signal after phase compensation is:

[0028] ,

[0029] in, is the signal after phase compensation at time t, is the ultrasonic signal of different subcarriers received by each microphone in the linear microphone array at time t, For noise.

[0030] Furthermore, the tracking phase-compensated signal changes in a plurality of consecutive instantaneous main dynamic components, combined with the static components, to separate user signals at different angles, specifically:

[0031] When there are ultrasonic signals passing through K users in the environment, the ultrasonic phase signals of N subcarriers of M microphones in the linear microphone array at multiple continuous times are taken to form an M×N×L matrix as a virtual joint array, where L represents the length of the signal received within 1 second;

[0032] The average value of the ultrasonic phase signals of N subcarriers of M microphones in the linear microphone array at multiple consecutive times is taken as the static component, and the static component is subtracted from the virtual joint array to obtain a residual matrix;

[0033] The residual matrix is ​​eigen-decomposed using the MUSIC algorithm to obtain a set of MN-K noise vector subspaces, denoted as , , ; where K represents the number of signals, represents the i-th noise vector that is independent of the user's breathing;

[0034] Construct a spectrum function related to the incident angle and path length, obtain the incident angle and path length of different users, substitute them into the calculation formula of the spectrum function to get the spectrum function value, and use the phase-compensated user signal corresponding to the spectrum function value as the final user signal at different angle positions.

[0035] Furthermore, the spectrum function is:

[0036] ,

[0037] in, represents the spectrum function, θ is the incident angle, and d is the path length.

[0038] Furthermore, the heartbeat signal is extracted from the user signal at each angle position, specifically:

[0039] The ultrasonic signal is decomposed into different eigenmode functions using an improved ensemble empirical mode decomposition method with adaptive noise;

[0040] Locate the extreme points of different intrinsic mode functions and perform peak detection, and filter the data within the peak interval that meets the conditions according to the frequency range of the heartbeat signal;

[0041] The ultrasonic heartbeat signal is obtained by reconstructing the data in the filtered peak interval.

[0042] Furthermore, the blood pressure measurement network model processes the input signal as follows:

[0043] In the input module, the input signal is instance normalized and sliced ​​into blocks. The sliced ​​embedding operation divides the 1×L input signal into different blocks and then assembles them into a 1×P×D slice matrix, where L is the length of the input signal and D is the number of channels. , P is the length of the slice, It means rounding down, and the slice matrix is ​​used as the input of the depth-wise separable convolution module;

[0044] The depth-wise separable convolution module includes depth-wise separable convolution, activation function and batch normalization operations. The depth-wise separable convolution includes depth-wise convolution and point-wise convolution. The slice matrix is ​​subjected to activation function and batch normalization operations after depth-wise separable convolution. In the depth-wise convolution, each input channel is subjected to a separate convolution kernel for convolution operation to extract feature information of each channel respectively. In the point-wise convolution, the output of the depth-wise convolution is subjected to convolution operation by using the convolution kernel to realize integration and fusion between channels. A shortcut connection is added between the input of the depth-wise separable convolution, the activation function and the output of the batch normalization operation.

[0045] The features obtained after the depth-wise separable convolution module are input into the channel attention module. The channel attention module performs the following operations:

[0046] ,

[0047] Among them, F is the feature obtained after the depth-separable convolution module, The operation performed by the channel attention module, is the activation function, It means that after two fully connected layers, is the maximum pooling operation, is the average pooling operation; After scaling the data of each channel of F, the characteristics after passing through the channel attention module are:

[0048] ,

[0049] in, represents the features after the channel attention module, represents the Hadamard product operator;

[0050] By adding shortcut connections before the maximum pooling operation and the average pooling operation and after the two fully connected layers, the final output feature is ;

[0051] After the final output features are input into two fully connected layers, the blood pressure prediction value is obtained.

[0052] Furthermore, when training the blood pressure measurement network model, the loss function constructed is:

[0053] ,

[0054] in, is the loss function, represents the true label, is the predicted blood pressure value, is a hyperparameter.

[0055] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the function of the multi-person blood pressure measurement system based on intelligent speaker ultrasonic signals is realized.

[0056] The present invention also provides a multi-person blood pressure measurement device based on smart speaker ultrasonic signals, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the function of the multi-person blood pressure measurement system based on smart speaker ultrasonic signals is realized.

[0057] The above technical solution of the present invention has the following beneficial effects compared with the prior art:

[0058] The present invention enhances ultrasonic signals by compensating the phases of subcarriers of different frequencies received by a microphone, and on this basis separates user signals at different angular positions; extracts heartbeat signals from the ultrasonic signals of user signals at each angular position, and uses a blood pressure measurement network model to predict blood pressure, thereby solving the problems of inconvenient operation, inability to measure over long distances, and low accuracy in existing blood pressure measurement methods, and achieving non-contact, real-time and effective measurement of blood pressure of multiple people. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to make the content of the present invention more clearly understood, the present invention is further described in detail below according to specific embodiments of the present invention in conjunction with the accompanying drawings, wherein:

[0060] Figure 1 This is a flow chart of the system predicting blood pressure of multiple people in a preferred embodiment of the present invention.

[0061] Figure 2 Schematic diagram of a multi-user model in a two-dimensional scene.

[0062] Figure 3 It is a schematic diagram of the result of multi-user signal separation in a preferred embodiment of the present invention.

[0063] Figure 4 Graph showing the relationship between a heartbeat signal and an actual electrocardiogram signal in a preferred embodiment of the present invention.

[0064] Figure 5 Schematic diagram of the structure of the blood pressure measurement network model in the preferred embodiment of the present invention. DETAILED DESCRIPTION

[0065] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention. Embodiment 1

[0066] The present invention discloses a multi-person blood pressure measurement system based on intelligent speaker ultrasonic signals, including an intelligent speaker, a phase compensation module, a multi-user signal separation module, an ultrasonic heartbeat signal extraction module, a blood pressure measurement network model construction module and a blood pressure prediction module. Figure 1 As shown in the figure, the specific process of predicting blood pressure of multiple people through each module is as follows:

[0067] S1: The smart speaker includes a speaker and a linear microphone array. The speaker emits ultrasonic signals of different frequencies, and the linear microphone array receives ultrasonic signals after passing through the user. In this embodiment, the microphone receives ultrasonic signals in the form of orthogonal frequency division multiplexing (OFDM). Due to path differences and frequency differences, different linear microphone arrays have different phases when acquiring different subcarriers.

[0068] Principle Figure 2 As shown in the figure, for two-dimensional spatial positioning, the influence of the path length and the incident angle in the horizontal direction on the received ultrasonic signal is fully considered. When two adjacent microphones receive signals of the same frequency, the difference in propagation distance between the signals received by the two microphones is ,in, Indicates the distance between two adjacent microphones. is the incident angle of the microphone in the horizontal direction, and the incident angle of two adjacent microphones in the horizontal direction Approximately equal. Figure 2 In the figure, the incident angle of the microphone corresponding to user 1 in the horizontal direction is , the incident angle of the microphone corresponding to user 2 in the horizontal direction is , the incident angle of the microphone corresponding to user 3 in the horizontal direction is In OFDM, ultrasonic signals are transmitted in different subcarriers, and these subcarriers are orthogonal, that is, they do not interfere with each other. However, since the frequencies of different subcarriers are not equal, their corresponding wavelengths are are also not equal. According to the phase With wavelength The relationship , is the path length when receiving ultrasonic signals; when the same microphone obtains different subcarriers, there are differences between the initial phases of the corresponding subcarriers.

[0069] S2: The phase compensation module combines spatial domain beamforming and frequency domain beamforming (i.e., dual beamforming technology) to construct the phase difference of each subcarrier of each microphone in the linear microphone array compared to the subcarrier emitted by the speaker, and performs phase compensation on each subcarrier of each microphone in the linear microphone array according to the phase difference to achieve superposition of ultrasonic signals and thus signal enhancement.

[0070] Assuming that the linear microphone array in the smart speaker has M microphones and the number of subcarriers emitted by the speaker is N, M×N ultrasonic signals can be obtained. Considering that the initial phases of these ultrasonic signals are different, the present invention proposes a new dual-beamforming technology that combines spatial domain beamforming and frequency domain beamforming, which can significantly enhance the sensing signal-to-noise ratio (SSNR) of multiple object positions in two-dimensional space. The dual-beamforming technology derives the phase corresponding to each subcarrier of each microphone in the array by understanding the positions of multiple users, including the path length and incident angle relative to the transceiver, and then achieves constructive superposition by compensating for the phase difference between multiple subcarriers and microphones, thereby achieving fine-grained perception at a farther distance.

[0071] S2-1: Using spatial beamforming, the phase difference of the mth microphone relative to the first microphone is expressed in the form of a complex exponential:

[0072] ,

[0073] in, is the phase difference of the mth microphone relative to the first microphone, exp( ) represents the natural exponential function, j is the imaginary unit, is the frequency of the nth subcarrier of the mth microphone, represents the distance between two adjacent microphones, is the incident angle of the ultrasonic signal of the first microphone relative to the first microphone, m=1,2,…,M-1, M is the number of microphones in the linear microphone array; c is the propagation speed of ultrasonic wave in air.

[0074] S2-2: Using frequency domain beamforming, the phase difference of the nth subcarrier of the microphone relative to the first subcarrier is expressed in the form of a complex exponential:

[0075] ,

[0076] in, is the phase difference of the nth subcarrier of the microphone relative to the 1st subcarrier, Indicates the frequency difference between two adjacent subcarriers received by the microphone. is the propagation path length of the ultrasonic signal received by the first microphone, n=1,2,…,N-1, and N is the number of subcarriers of the microphone.

[0077] S2-3: Combining the phase difference of the mth microphone relative to the first microphone and the phase difference of the nth subcarrier of the microphone relative to the first subcarrier, the phase difference of the nth subcarrier of the mth microphone relative to the first subcarrier of the first microphone is constructed as:

[0078] ,

[0079] in, is the phase difference between the nth subcarrier of the mth microphone and the 1st subcarrier of the 1st microphone.

[0080] S2-4: Perform phase compensation on each subcarrier of each microphone in the linear microphone array according to the phase difference.

[0081] S2-4-1: Construct the steering vector composed of the phase difference introduced by M microphones and N subcarriers as:

[0082] ,

[0083] in, is the steering vector composed of the phase differences introduced by M microphones and N subcarriers, and T represents the transposition operation.

[0084] S2-4-2: The signal after phase compensation can be enhanced. The signal after phase compensation is:

[0085] ,

[0086] in, is the signal after phase compensation at time t, is the ultrasonic signal of different subcarriers received by each microphone in the linear microphone array at time t, The size of is MN*1, For noise.

[0087] By compensating the phase difference between the nth subcarrier of the mth microphone and the first subcarrier of the first microphone to the phase of each subcarrier of each microphone, the perceived signal-to-noise ratio can be increased by M*N times. Taking heartbeat as an example in practical application, the present invention can achieve an average heartbeat measurement error of less than 1 beat per minute within a range of 10 m, which proves that the present invention can effectively improve the perception distance of ultrasound and well guarantee the measurement accuracy.

[0088] S3: This invention proposes a new multi-micro signal classification (MUS 2IC) method, the multi-user signal separation module tracks the main dynamic components of the phase-compensated signal that change in multiple consecutive instants, and separates user signals at different angular positions in combination with the static components.

[0089] If there are multiple users in the space, the collected ultrasonic signal is a mixture of signals from multiple users. The present invention uses a new multiple weak signal classification (MUS 2 IC) to separate different users. Specifically:

[0090] S3-1: The main movement of the user in a static state is the chest rise and fall caused by breathing and heartbeat, which can be considered as a pseudo-periodic behavior. This movement can be sensed by the ultrasonic signal and manifested as a phase change of the subcarrier.

[0091] When there are ultrasonic signals passing through K users in the environment, the steering vector superimposed by these K users is constructed as:

[0092] ,i=1,2,..,K,

[0093] in is the steering vector superimposed by K users, represents the steering vector corresponding to the ultrasonic signal of the i-th user, and K is the total number of users.

[0094] S3-2: All user signals after phase compensation can be obtained as:

[0095] ,

[0096] in, is the phase-compensated signal of all users at time t, S(t) represents the ultrasonic signal caused by all users received, ,i=1,2,…,K; represents the ultrasonic signal caused by the i-th user received at time t, and N(t) is the received noise vector.

[0097] S3-3: Take the ultrasonic phase signals of N subcarriers of M microphones in the linear microphone array at multiple consecutive times to form an M×N×L matrix as a virtual joint array, denoted as X, where L represents the length of the signal received within 1 second.

[0098] S3-4: The average value of the ultrasonic phase signals of the N subcarriers of the M microphones in the linear microphone array at multiple consecutive times is taken as the static component, and the virtual joint array is subtracted from the static component to obtain a residual matrix, recorded as X', then the X' matrix is ​​dominated by the principal component representing dynamic changes.

[0099] S3-5: Use the MUSIC algorithm to perform eigendecomposition on the residual matrix to obtain a set consisting of MN-K noise vector subspaces, denoted as , , ; K represents the number of signals, that is, the number of users, represents the i-th noise vector that is independent of the user's breathing.

[0100] S3-6: Since the noise subspace is orthogonal to the user's ultrasonic signal, the spectrum function related to the incident angle θ and the path length d is constructed as follows:

[0101] ,

[0102] in, represents the spectrum function; the spectrum function expansion can be transformed to obtain the expression about θ and d, where θ is the incident angle and d is the path length. Because the column vector related to the user's breathing behavior is also orthogonal to the noise subspace, when the steering vector in the spectrum function Taken from When , the spectrum function can obtain the maximum value.

[0103] S3-7: In practical applications, obtaining the incident angle of different users and the path length d are substituted into the calculation formula of the spectrum function to obtain the spectrum function value, and the compensated user signal corresponding to the spectrum function value is used as the final user signal at different angular positions.

[0104] The result of converting the spectrum function value into a spectrum graph is as follows Figure 3 As shown, the effect of using the MUS2IC algorithm in the present invention to separate user signals at different angles. The larger the spectrum value, the corresponding bright spot appears in the spectrum diagram. Specifically observing the effect of ultrasonic multi-user signal separation, Figure 3 (a) shows the result of separation using the traditional MUSIC algorithm, which cannot identify the two users at 10m. Figure 3 (b) is the result of separation using the MUS2IC algorithm of the present invention. At 10 meters, two bright spots can be clearly distinguished, that is, two users can be identified, indicating that the chest fluctuation signal caused by the user's breathing is well identified. This shows that the MUS2IC algorithm used in the present invention 2 The IC algorithm can effectively separate the multi-user signals at 10m and detect the movements caused by the user's breathing, thereby determining the user's position in two-dimensional space.

[0105] S4: The ultrasonic heartbeat signal extraction module extracts the heartbeat signal from the user signal at each angle position. The compensated user signal at each angle position, that is, after obtaining the corresponding angle distance, is extracted according to the formula The calculated signal is used to extract the ultrasonic heartbeat signal using the improved Iterative Cumulative Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN).

[0106] S4-1: Use ICEEMDAN to decompose the ultrasonic signal into different intrinsic mode functions (IMFs);

[0107] S4-2: locate the extreme points of different intrinsic mode functions and perform peak detection, and filter the data within the peak interval that meets the conditions according to the frequency range of the heartbeat signal. The heartbeat frequency range set in this embodiment is 1-2Hz, and the interval between two adjacent peaks in the eigenfunction component corresponding to the heartbeat signal should be within the range of 0.5-1 second (equivalent to the heartbeat cycle 1-Hz) to filter out the data.

[0108] S4-3: Reconstruct the data in the filtered peak interval to obtain the ultrasonic heartbeat signal:

[0109] ,

[0110] in, Indicates the ultrasonic heartbeat signal finally extracted. Represents the data within the filtered peak range.

[0111] In this embodiment, the comparison result between the finally extracted ultrasonic heartbeat signal and the actual electrocardiogram signal is as follows: Figure 4 As shown, Figure 4 (a) is the extracted ultrasonic heartbeat signal. Figure 4 (b) is the actual ECG signal. By aligning the timestamps of the heartbeat signal with the actual ECG signal, it can be found that the two have a strong correlation, indicating that the user's heartbeat signal extracted from ultrasound can capture heart activity for further analysis.

[0112] S5: Blood pressure measurement network model construction module constructs a blood pressure measurement network model, such as Figure 5 As shown, the blood pressure measurement network model includes an input module, a depthwise separable convolution module (Depthwise Separable Convolution, DSConv) and a channel attention module.

[0113] The blood pressure measurement network model processes the input signal as follows:

[0114] In the input module, the input signal is instance normalized and sliced ​​(Patch Emdeding) is performed. The slice embedding operation converts The input signal is split into different blocks and then pieced together The slice matrix is ​​denoted as H, where L is the length of the input signal and D is the number of channels. , P is the slice length, It means rounding down, and the slice matrix is ​​used as the input of the depth-wise separable convolution module;

[0115] The depth separable convolution module includes depth separable convolution, activation function and batch normalization operation. The depth separable convolution includes depth convolution (DwConv) and pointwise convolution (PwConv). The slice matrix is ​​subjected to activation function and batch normalization operation after depth separable convolution. In this embodiment, the activation function here is Relu. In depth convolution, each input channel applies a separate convolution kernel for convolution operation, extracts feature information of each channel respectively, and realizes feature extraction and processing between channels. Depth separable convolution can significantly reduce the number of parameters and reduce computational costs. In pointwise convolution, the output of depth convolution is convolved by using a 1x1 convolution kernel to realize integration and fusion between channels, thereby obtaining the final output feature map. Pointwise convolution can help increase nonlinearity and improve the expression ability of the network. Add shortcut connections between the input of depth separable convolution, activation function and output of batch normalization operation to avoid gradient disappearance during training.

[0116] Figure 5 The “×2” in denoted that the operation is repeated twice, assuming that the operations of depthwise separable convolution, activation function, batch normalization, and shortcut connection are recorded as functions , then the final output result is .

[0117] The features obtained after the depth-wise separable convolution module are input into the channel attention module. The channel attention module performs the following operations:

[0118] ,

[0119] Among them, F is the feature obtained after the depth-separable convolution module, The operation performed by the channel attention module, is the activation function, It means that after two fully connected layers, is the maximum pooling operation, is the average pooling operation; After scaling the data of each channel of F, the characteristics after passing through the channel attention module are:

[0120] ,in, represents the features after the channel attention module, represents the Hadamard product operator;

[0121] By adding shortcut connections before the maximum pooling operation and the average pooling operation and after the two fully connected layers, the final output feature is ;

[0122] After the final output features are input into two fully connected layers, the blood pressure prediction value is obtained.

[0123] S6: The blood pressure prediction module trains the blood pressure measurement network model to obtain a trained blood pressure measurement network model, and inputs the extracted heartbeat signal into the trained blood pressure measurement network model to obtain a blood pressure prediction result.

[0124] When training the blood pressure measurement network model, the loss function constructed is:

[0125] ,

[0126] in, is the loss function, represents the true label, is the predicted blood pressure value, is a hyperparameter, a hyperparameter The default setting is 1. Therefore, when the absolute difference between the predicted value and the actual value is less than or equal to 1, the loss function takes the form of square error; when the absolute difference is greater than 1, the loss function takes the form of absolute error. The present invention uses Huber loss when constructing the loss function. Compared with the traditional MSE, Huber loss is more robust to outliers; Huber loss is similar to mean square error (MSE) when the error is small, and similar to mean absolute error (MAE) when the error is large. Embodiment 2

[0127] The present invention also discloses a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the function of the multi-person blood pressure measurement system based on intelligent speaker ultrasonic signals in embodiment one is realized. Embodiment 3

[0128] The present invention also discloses a multi-person blood pressure measurement device based on smart speaker ultrasonic signals, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the functions of the multi-person blood pressure measurement system based on smart speaker ultrasonic signals in Example 1 are implemented.

[0129] The present invention enhances ultrasonic signals by compensating the phases of subcarriers of different frequencies received by a microphone, and on this basis separates user signals at different angular positions; extracts heartbeat signals from the ultrasonic signals of user signals at each angular position, and uses a blood pressure measurement network model to predict blood pressure, thereby solving the problems of inconvenient operation, inability to measure over long distances, and low accuracy in existing blood pressure measurement methods, and achieving non-contact, real-time and effective measurement of blood pressure of multiple people.

[0130] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0131] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0132] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0133] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1The steps for the functions specified in one or more boxes.

[0134] Obviously, the above embodiments are merely examples for clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived from these are still within the protection scope of the invention.

Claims

1. A multi-person blood pressure measurement system based on intelligent speaker ultrasonic signals, characterized in that: include: A smart speaker, comprising a speaker and a linear microphone array, wherein the speaker transmits ultrasonic signals of different frequencies, and the linear microphone array receives the ultrasonic signals after passing through the user; A phase compensation module, used to combine spatial domain beamforming and frequency domain beamforming to construct a phase difference between each subcarrier of each microphone in the linear microphone array and the subcarrier emitted by the speaker, and to perform phase compensation on each subcarrier of each microphone in the linear microphone array according to the phase difference; The multi-user signal separation module is used to track the main dynamic components of the phase-compensated signal at multiple consecutive instants, and separate the user signals at different angles by combining the static components; Ultrasonic heartbeat signal extraction module, used to extract heartbeat signals from user signals at each angle position; A blood pressure measurement network model construction module, used to construct a blood pressure measurement network model, wherein the blood pressure measurement network model includes an input module, a depthwise separable convolution module, and a channel attention module; A blood pressure prediction module, which trains the blood pressure measurement network model to obtain a trained blood pressure measurement network model, and inputs the extracted heartbeat signal into the trained blood pressure measurement network model to obtain a blood pressure prediction result; The phase difference of each subcarrier of each microphone in the linear microphone array compared with the subcarrier emitted by the speaker is constructed by combining spatial domain beamforming and frequency domain beamforming, specifically: Using spatial beamforming, the phase difference of the mth microphone relative to the first microphone is expressed in the form of a complex exponential: ， in, is the phase difference of the mth microphone relative to the first microphone, exp( ) represents the natural exponential function, j is the imaginary unit, is the frequency of the nth subcarrier of the mth microphone, represents the distance between two adjacent microphones, is the incident angle of the ultrasonic signal of the first microphone relative to the first microphone, m=1,2,…,M-1, M is the number of microphones in the linear microphone array; c is the propagation speed of ultrasonic wave in air; Using frequency domain beamforming, the phase difference of the nth subcarrier of the microphone relative to the first subcarrier is expressed in the form of a complex exponential: ， in, is the phase difference of the nth subcarrier of the microphone relative to the 1st subcarrier, Indicates the frequency difference between two adjacent subcarriers received by the microphone. is the propagation path length of the ultrasonic signal received by the first microphone, n=1,2,…,N-1, N is the number of subcarriers of the microphone; Combining the phase difference of the mth microphone relative to the first microphone and the phase difference of the nth subcarrier of the microphone relative to the first subcarrier, the phase difference of the nth subcarrier of the mth microphone relative to the first subcarrier of the first microphone is constructed as: ， in, is the phase difference between the nth subcarrier of the mth microphone and the 1st subcarrier of the 1st microphone; Phase compensation is performed on each subcarrier of each microphone in the linear microphone array according to the phase difference, specifically: The steering vector composed of the phase difference introduced by M microphones and N subcarriers is constructed as: ， in, is the steering vector composed of the phase differences introduced by M microphones and N subcarriers, and T represents the transposition operation; The signal after phase compensation is: ， in, is the signal after phase compensation at time t, is the ultrasonic signal of different subcarriers received by each microphone in the linear microphone array at time t, For noise.

2. The multi-person blood pressure measurement system based on intelligent speaker ultrasonic signals according to claim 1 is characterized in that: The tracking phase-compensated signal changes in a plurality of consecutive instantaneous main dynamic components, combined with the static components to separate user signals at different angles, specifically: When there are ultrasonic signals passing through K users in the environment, the ultrasonic phase signals of N subcarriers of M microphones in the linear microphone array at multiple continuous times are taken to form an M×N×L matrix as a virtual joint array, where L represents the length of the signal received within 1 second; The average value of the ultrasonic phase signals of N subcarriers of M microphones in the linear microphone array at multiple consecutive times is taken as the static component, and the static component is subtracted from the virtual joint array to obtain a residual matrix; The residual matrix is ​​eigen-decomposed using the MUSIC algorithm to obtain a set of MN-K noise vector subspaces, denoted as , , ; where K represents the number of signals, represents the i-th noise vector that is independent of the user's breathing; Construct a spectrum function related to the incident angle and path length, obtain the incident angle and path length of different users, substitute them into the calculation formula of the spectrum function to get the spectrum function value, and use the phase-compensated user signal corresponding to the spectrum function value as the final user signal at different angle positions.

3. The multi-person blood pressure measurement system based on intelligent speaker ultrasonic signals according to claim 2 is characterized in that: The spectrum function is: ， in, represents the spectrum function, θ is the incident angle, and d is the path length.

4. The multi-person blood pressure measurement system based on intelligent speaker ultrasonic signals according to claim 1 is characterized in that: The heartbeat signal extraction of the user signal at each angle position is specifically performed as follows: The ultrasonic signal is decomposed into different eigenmode functions using an improved ensemble empirical mode decomposition method with adaptive noise; Locate the extreme points of different intrinsic mode functions and perform peak detection, and filter the data within the peak interval that meets the conditions according to the frequency range of the heartbeat signal; The ultrasonic heartbeat signal is obtained by reconstructing the data in the filtered peak interval.

5. The multi-person blood pressure measurement system based on intelligent speaker ultrasonic signals according to claim 1 is characterized in that: The blood pressure measurement network model processes the input signal as follows: In the input module, the input signal is subjected to instance normalization and slice embedding operations. The slice embedding operation converts The input signal is split into different blocks and then pieced together The slice matrix, where L is the length of the input signal, D is the number of channels, , P is the length of the slice, It means rounding down, and the slice matrix is ​​used as the input of the depth-wise separable convolution module; The depth-wise separable convolution module includes depth-wise separable convolution, activation function and batch normalization operations. The depth-wise separable convolution includes depth-wise convolution and point-wise convolution. The slice matrix is ​​subjected to activation function and batch normalization operations after depth-wise separable convolution. In the depth-wise convolution, each input channel is subjected to a separate convolution kernel for convolution operation to extract feature information of each channel respectively. In the point-wise convolution, the output of the depth-wise convolution is subjected to convolution operation by using the convolution kernel to realize integration and fusion between channels. A shortcut connection is added between the input of the depth-wise separable convolution, the activation function and the output of the batch normalization operation. The features obtained after the depth-wise separable convolution module are input into the channel attention module. The channel attention module performs the following operations: ， Among them, F is the feature obtained after the depth-separable convolution module, The operation performed by the channel attention module, is the activation function, It means that after two fully connected layers, is the maximum pooling operation, is the average pooling operation; After scaling the data of each channel of F, the characteristics after passing through the channel attention module are: ， in, represents the features after the channel attention module, represents the Hadamard product operator; By adding shortcut connections before the maximum pooling operation and the average pooling operation and after the two fully connected layers, the final output feature is ; After the final output features are input into two fully connected layers, the blood pressure prediction value is obtained.

6. The multi-person blood pressure measurement system based on intelligent speaker ultrasonic signals according to claim 1 is characterized in that: When training the blood pressure measurement network model, the loss function constructed is: ， in, is the loss function, represents the true label, is the predicted blood pressure value, is a hyperparameter.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the functions of the multi-person blood pressure measurement system based on smart speaker ultrasonic signals as described in any one of claims 1 to 6 are realized.

8. A multi-person blood pressure measurement device based on intelligent speaker ultrasonic signals, characterized in that: It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the function of the multi-person blood pressure measurement system based on the ultrasonic signal of the smart speaker as described in any one of claims 1 to 6 is realized.

Images