Non-contact electrocardiosignal monitoring reconstruction method and system based on time-frequency domain combination

Abstract

The invention discloses a non-contact electrocardiosignal monitoring reconstruction method and system based on time-frequency domain combination, and relates to the technical field of radar signal processing, and the method comprises the following steps: obtaining a chest wall vibration signal S of a to-be-detected person by adopting a frequency-modulated continuous wave radar; separating a heart mechanical motion signal SH from the chest wall vibration signal S; and converting the heart mechanical motion signal SH into an electrocardiosignal ECG through a pre-established ResNet architecture neural network model. Time domain loss and frequency domain loss are used for training a ResNet architecture neural network model, and the error between the calculated heart rate and the real heart rate is quantified through the mean absolute error and the root-mean-square error. According to the method, the neural network capable of being used for ECG signal conversion is designed based on the ResNet architecture, after the neural network is subjected to data training, more detailed signal features can be learned, the fitting accuracy is improved, then the ECG waveform can be accurately predicted, and non-contact electrocardiosignal monitoring and reconstruction are achieved.

Description

Technical Field

[0001] The present invention relates to the field of radar signal processing technology, and in particular to a non-contact electrocardiogram signal monitoring reconstruction method and system based on time-frequency domain combination. Background Art

[0002] Electrocardiogram (ECG) is a technology and instrument used to measure and record the electrical activity of the heart. It collects cardiac electrical signals by placing electrodes at specific locations on the patient's body and converts them into visual graphs or digital data.

[0003] With the development of computer vision and radar technology, non-contact measurement of ECG signals has received increasing attention. rPPG (Remote Photoplethysmography, non-contact heart rate monitoring) technology collects heart rate information through face videos, but is sensitive to lighting conditions and skin type. In contrast, radar signals are not affected by light, have the advantages of high sampling frequency and small data volume, and can more accurately perceive cardiac mechanical motion signals. However, traditional signal processing methods are sensitive to noise and interference signals, and the signal characteristics of different individuals may be different. Although some deep learning-based methods have been used for heart rate and respiratory rate measurement, the lack of high-quality annotated ECG datasets makes it difficult to train these algorithms. Summary of the invention

[0004] The purpose of the present invention is to address the problems in the above-mentioned prior art and provide a non-contact ECG signal monitoring reconstruction method and system based on the combination of time and frequency domains, which can realize the conversion of cardiac mechanical motion signals into ECG signals, and then complete high-precision prediction of ECG activity, thereby improving the comfort and convenience of ECG monitoring.

[0005] In order to achieve the above object, the present invention has the following technical solutions:

[0006] In a first aspect, a non-contact electrocardiogram signal monitoring reconstruction method based on time-frequency domain combination is provided, comprising:

[0007] A frequency modulated continuous wave radar is used to obtain a chest wall vibration signal S of the subject;

[0008] Separate the cardiac mechanical motion signal S from the chest wall vibration signal S H ;

[0009] The cardiac mechanical motion signal S is transformed into H Converted into ECG signal.

[0010] Preferably, the cardiac mechanical motion signal S is converted intoH In the step of converting to ECG, a one-dimensional convolutional neural network (1D CNN) with a kernel of 3 is used to extract the cardiac mechanical motion signal S H The extracted data features enter the backbone network after normalization and activation function; the backbone network consists of three reused models, each of which contains four residual blocks ResNetBlock. In each residual block ResNetBlock, the input signal passes through a convolution layer with a kernel of 3 for feature extraction, and then undergoes normalization and activation function processing; thereafter, it passes through another convolution layer with a kernel of 3, undergoes normalization and activation function processing again, and is finally output to the next residual block ResNetBlock; when the number of input channels is the same as the number of output channels, or the stride is 1, the data is jump-connected through a shortcut block; the shortcut block includes a convolution layer with a kernel of 1 and a normalization layer, which is used to restore the information lost in the convolution process and add the obtained features to the learned features.

[0011] Preferably, the pre-established ResNet architecture neural network model is trained using supervised learning:

[0012] Take the ECG signal as the label to train the heart shock signal SCG, learn the mapping relationship between the ECG signal and the heart shock signal SCG, and calculate the value L of the ECG signal loss in the time domain according to the following formula T :

[0013]

[0014] Where, L T represents the value of time domain loss, E represents the standard data of ECG signal Ground Truth ECG, P represents the ECG signal obtained by neural network fitting, μ E , μ P They represent the standard data of ECG signal Ground Truth ECG and the mean of ECG signal obtained by neural network fitting, σ E , σ P Represents the standard deviation σ E 2 , σ P 2 represents variance;

[0015] The mean absolute error (MAE) is used to calculate the distance L between the ECG signal and the standard ECG data GroundTruth ECG in the frequency domain. F :

[0016]

[0017] Among them, E F and P F They represent the standard data Ground Truth ECG of the ECG signal and the frequency domain data of the reconstructed ECG signal respectively; the final loss function is as follows:

[0018] L=L F +L T .

[0019] Preferably, the time domain loss and the frequency domain loss are used to train the ResNet architecture neural network model, and the error between the calculated heart rate HR and the actual heart rate HR is quantified by the mean absolute error MAE and the root mean square error RMSE.

[0020] Preferably, an ablation experiment is performed to verify whether the number of models reused in the backbone network is reasonable, including removing the three reused models in the neural network one by one, and then training under the same conditions, calculating the heart rate HR from the R peak in the electrocardiogram signal ECG, and calculating the mean absolute error MAE between the heart rate HR and the true heart rate HR to verify the accuracy of the reconstructed signal.

[0021] Preferably, the method further comprises filtering and denoising the obtained electrocardiogram signal ECG using a Butterworth filter:

[0022] S=butter(N,Wn,x)

[0023] Where N is the order of the filter, Wn is the normalized cutoff frequency, and x is the filtered signal;

[0024] Combining the frequency with the highest energy in the spectrum graph and the normal human heart rate range, the heart rate is estimated:

[0025] HR=max(s)*t

[0026] Where t is a hyperparameter;

[0027] Then mark the R peaks in the signal, calculate the difference between two adjacent R peaks, and then divide the result by the sampling rate of the ECG signal to get the heartbeat interval IBI:

[0028]

[0029] Where fs represents the sampling rate of the electrocardiogram signal ECG.

[0030] Preferably, the method further includes performing frequency domain conversion on the electrocardiogram signal ECG after filtering and noise reduction, and then performing power spectrum density PSD analysis to find important frequency components, frequency changes and energy distribution information in the signal, and then analyzing and obtaining the heart rate.

[0031] In a second aspect, a non-contact electrocardiogram signal monitoring and reconstruction system based on time-frequency domain combination is provided, comprising:

[0032] A signal acquisition module, used for acquiring a chest wall vibration signal S of a subject using a frequency modulated continuous wave radar;

[0033] The cardiac mechanical motion signal separation module is used to separate the cardiac mechanical motion signal S from the chest wall vibration signal S H ;

[0034] The ECG signal conversion module is used to convert the cardiac mechanical motion signal S into H Converted into ECG signal.

[0035] According to a third aspect, an electronic device is provided, including:

[0036] A memory storing at least one instruction; and a processor executing the instruction stored in the memory to implement the non-contact electrocardiogram signal monitoring reconstruction method based on the combination of time and frequency domains.

[0037] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the non-contact electrocardiogram signal monitoring reconstruction method based on the combination of time and frequency domains is implemented.

[0038] Compared with the prior art, the present invention has at least the following beneficial effects:

[0039] The method of the present invention uses millimeter wave radar to collect cardiac mechanical motion signals, and processes them through a deep learning model of the ResNet architecture, thereby realizing the prediction of electrocardiogram (ECG) signals in a non-contact manner. The present invention combines the deep learning method to learn and map ECG signals, and transforms weak cardiac mechanical motion signals S into H The ECG signal is converted into an electrocardiogram (ECG) signal, which solves the problem that the traditional non-contact ECG signal measurement method is difficult to train due to the lack of high-quality annotated ECG data sets. The present invention designs a neural network that can be used for ECG signal conversion based on the ResNet architecture. After data training, the proposed neural network can learn more detailed signal features, improve the accuracy of fitting, and then accurately predict the ECG waveform, thereby realizing non-contact ECG signal monitoring and reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0041] Figure 1 It is a flow chart of the non-contact ECG signal monitoring reconstruction method based on the combination of time and frequency domains of the present invention;

[0042] Figure 2 This is a schematic diagram of an application scenario in which the millimeter wave radar is aimed at the chest of a subject for data collection in the present invention;

[0043] Figure 3 Flowchart for converting the signal of cardiac mechanical motion into ECG signal domain for neural network;

[0044] Figure 4 This is a schematic diagram of ECG reconstructed by a neural network. In the figure, green is the signal of cardiac mechanical movement, red is the ECG signal, and blue is the reconstructed ECG signal;

[0045] Figure 5 This is a schematic diagram of the experimental results of the number of multiplexed modules; orange represents the change of HR accuracy with the increase of the number of modules in the sitting state, and blue represents the change of heart rate accuracy with the number of modules in the sitting state;

[0046] Figure 6 The power spectral density corresponding to the reconstructed ECG signal: (a) ECG signal; (b) power spectral density. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, ordinary technicians in this field can also obtain other embodiments without making creative work.

[0048] See also Figure 1 The embodiment of the present invention provides a non-contact ECG signal monitoring reconstruction method based on the time-frequency domain combination, including:

[0049] Step 1: Use millimeter wave radar to collect data, including the following steps:

[0050] Step 1, the present invention collects data by finding 30 healthy volunteers. All subjects in this embodiment are from Asia, weighing 50kg to 80kg, and the male to female ratio is 4:6. The relevant information of the subjects is shown in Table 1.

[0051] Table 1

[0052]

[0053] Step 2, using frequency modulated continuous wave radar to obtain cardiac mechanical motion data. In order to obtain data in real time, the embodiment of the present invention adopts a serial communication method for real-time data transmission. In order to determine the accuracy of the collected data, the embodiment of the present invention uses BIOPAC equipment certified by the FDA (U.S. Food and Drug Administration) and AcqKnowledge5.0 data analysis software to synchronize the collection of ECG and heart rate signals, and then complete the comparison and verification of the experimental results.

[0054] Data collection scenarios such as Figure 1 As shown. During the collection process, the subject was in a supine position and was required to stay relaxed and still as much as possible. The radar was located just above the subject's chest, and the distance between the chest and the radar was between 0.15m and 0.45m. The BIOPAC electrodes used in the control experiment were attached to the inner side of the subject's left and right ankles and right wrist. Each group of data was collected for 50 seconds and divided into five groups. During the collection process, two staff members pressed the collection buttons of the radar and BIOPAC at the same time to achieve the purpose of synchronous collection.

[0055] Step 2: Design a neural network to convert the cardiac mechanical motion signal detected by radar into ECG signal:

[0056] Step 1. The embodiment of the present invention designs a neural network that can be used for ECG signal conversion based on the ResNet architecture. First, the collected heart vibration signal is input into the network, and then a 1D CNN with a kernel of 3 is used to extract data features. After the extracted features are normalized and activated, they enter the backbone network. The backbone network consists of three multiplexed models, each of which contains four residual blocks. In each residual block, the input signal is subjected to feature extraction through a convolutional layer with a kernel of 3, and then normalized and activated. After that, it passes through a convolutional layer with a kernel of 3, is normalized and activated again, and is finally output to the next residual block. The specific network structure is as follows: Figure 3 shown.

[0057] Step 2, design the residual block ResNetBlock to solve the problem of gradient disappearance and gradient explosion, and promote the training and convergence of the model. When the number of input channels is the same as the number of output channels, or when the stride is 1, the data will be skipped through the shortcut block. The shortcut block includes a convolution layer with a kernel of 1 and a normalization layer, which is used to restore the information lost in the convolution process and add the obtained features to the learned features to ensure that the learned results will not deteriorate.

[0058] Step 3: Train the neural network model using supervised learning.

[0059] Using the ECG signal as the label, we train the SCG signal and learn the mapping relationship between the ECG and SCG signals. We use the negative Pearson coefficient to calculate the fitted ECG signal and the ECG gt Distance in time domain:

[0060]

[0061] Where, L T represents the value of time domain loss, E represents Ground Truth ECG, P represents ECG obtained by neural network fitting, μ E , μ P Represent the mean of the two signals, σ E , σ P represents the standard deviation, σ E 2 , σ P 2 Represents variance.

[0062] At the same time, the mean absolute error is used to calculate the distance between the fitted ECG signal and the groundtruth in the frequency domain:

[0063]

[0064] Among them, E F and P F Represent the frequency domain data of ground truth ECG and reconstructed ECG respectively.

[0065] So the final loss function is described by the following formula:

[0066] L=L F +L T

[0067] Step 4: Perform an ablation experiment to verify whether the number of multiplexing modules in the backbone network is reasonable.

[0068] The three identical feature extraction models in the neural network were removed one by one, and then trained under the same conditions. The HR was calculated from the ECGre signal, and the MAE between the real HR was calculated to illustrate the accuracy of the reconstructed signal. To ensure fairness, the input radar signal size was 10s, and the model training was performed under the same conditions.

[0069] Table 2 Experiment on the number of reused models

[0070]

[0071] The embodiment of the present invention records the MAE error of HR under different conditions. When only one feature extraction module is used, the MAE of the subject is 22.75 in the lying state and 27.29 in the sitting state. This shows that in this case, the model has not converged. When the number of feature extraction modules is increased to two, the MAE of the subject in the sitting and lying states is 7.97 and 8.81 respectively. Compared with the case of one feature extraction module, the MAE of this state is significantly reduced, and the model further converges. Therefore, the feature extraction module is increased to 3. At this time, the MAE in the two states is 0.15 and 0.20 respectively, and when 4 feature extraction modules are used, the MAE of the two states is 1.83 and 2.46 respectively. Compared with the previous state, the error increases, and the model is overfitting. Therefore, the experimental results show that it is more reasonable to use three multiplexed feature extraction modules.

[0072] Step 5: Use time domain loss and frequency domain loss to train the neural network. MAE and RMSE are used to quantify the error between the HR and the true HR, and to illustrate the effectiveness of the two loss functions.

[0073] Table 3

[0074]

[0075] From the experimental results, when only the time domain loss is used, the MSE of the HR of the subjects in the sitting and supine states is 1.43 and 2.16, respectively, while when only the frequency domain loss function is used, the errors are 2.10 and 3.17, respectively, that is, the data accuracy obtained by using only the time domain loss function is greater than the accuracy of using only the frequency domain loss function. This is because the time domain loss is used to calculate the time domain error between the reconstructed ECG signal and the real ECG signal, which can help the model to learn a continuous waveform similar to the ECG signal, and can also help the model learn accurate amplitude, phase and timing information. The frequency domain loss is used to measure the frequency domain error of the two. It pays more attention to the distribution of the ECG signal in the frequency domain, which may cause the generated signal to have large jitter or discontinuous parts in the time domain, and then reduce the accuracy of the final result. When two loss functions are used for training at the same time, the MAE error and RMSE error of the heart rate in the supine state are 0.15 and 0.46, respectively, and the corresponding data in the sitting state are 0.20 and 0.53, respectively. This is much higher than the data accuracy obtained by using only one loss function.

[0076] This is because, when both frequency domain and time domain loss functions are used, the embodiments of the present invention can use information from two different levels for training and evaluation. The time domain loss function can help the model capture the instantaneous change characteristics of the signal, while the frequency domain loss function can help the model learn the frequency components and harmonic structure of the signal, which will increase the difficulty of model training and also promote the model to learn more detailed signal characteristics and improve the accuracy of fitting.

[0077] Step 6: Measure heart rate and heart rate variability.

[0078] After obtaining the ECG signal, the embodiment of the present invention first performs frequency domain conversion on the ECG signal, and then performs power spectral density (PSD) analysis to find important frequency components, frequency changes and energy distribution information in the signal, and finally obtains the heart rate.

[0079] Considering that the ECG signal reconstructed by the neural network may contain noise in some results, resulting in high-frequency noise in the signal, the embodiment of the present invention performs low-pass filtering on the signal to reduce noise.

[0080] The Butterworth filter has a flat passband and a steep stopband. It is widely used in fields where you want to retain signal energy while filtering out unwanted noise or interference. Therefore, the Butterworth filter is used to filter the ECG signal:

[0081] S=butter(N,Wn,x)

[0082] Where N is the order of the filter, Wn is the normalized cutoff frequency, and x is the signal to be filtered.

[0083] Because the experiment did not involve exercise, the normal heart rate range should be between 40bmp and 120bmp. Therefore, the heart rate is estimated by combining the frequency with the highest energy in the spectrum and the normal human heart rate range:

[0084] HR=max(s)*t

[0085] Among them, t is a hyperparameter.

[0086] Then mark the R peaks in the signal and calculate the difference between two adjacent R peaks. Then divide the result by the sampling rate of the ECG signal to get the IBI:

[0087]

[0088] Wherein, Fs represents the sampling rate of the ECG signal.

[0089] Next, the embodiment of the present invention calculates the changes in heart rate on different time scales, quantifies HRV (heart rate variability) through indicators such as MAE and RMSE, and compares it with the real ECG signal to verify the validity of the results obtained by the present invention.

[0090] Specifically, the application scenarios of using AWR 1843 millimeter wave radar to collect data from the chest of the person being tested are as follows: Figure 2 As shown, by detecting the chest wall of the subject by millimeter wave radar, the chest wall vibration signal S of the subject can be obtained, and the signal S generated by the heart beat can be separated from it. H . H Data Entry Figure 3 The neural network shown in Figure 1 is trained under the supervision of ground truth ECG. The number of multiplexed modules in the backbone network is verified by designing ablation experiments and loss functions, as shown in Figure 1. Figure 4 shown.

[0091] To facilitate the evaluation of the results, the three signals are placed in the same coordinate system, and the reconstructed ECG signal is misaligned with the ground truth ECG for easy distinction. Figure 5 The results show that the reconstructed ECG signal is highly consistent with the ground truth ECG in time domain morphology. The R peak position of the reconstructed ECG signal is the same as the maximum peak of the mechanical motion signal and the R peak position of the ground truth ECG, and the T peak position is also consistent, which shows that the reconstructed ECG signal has achieved a high time accuracy.

[0092] After obtaining the reconstructed ECG signal, the signal is analyzed in the time domain and frequency domain. Figure 6(a) shows the ECG signal, and (b) shows the power spectrum density diagram corresponding to the signal. It can be seen from the ECG signal that the energy of the R peak is the largest, and the RR interval represents a heartbeat process. The heart rate (HR) can be obtained through the power spectrum density, and the heart beat interval (IBI) can be obtained from the time domain data. The HR of this sample is 75bpm, and the IBI is 0.8 seconds. The final verification shows that after data training, the neural network proposed in the embodiment of the present invention can accurately predict the ECG waveform and realize non-contact electrocardiogram signal monitoring and reconstruction.

[0093] Another embodiment of the present invention provides a non-contact ECG signal monitoring and reconstruction system based on time-frequency domain combination, comprising:

[0094] A signal acquisition module, used for acquiring a chest wall vibration signal S of a subject using a frequency modulated continuous wave radar;

[0095] The cardiac mechanical motion signal separation module is used to separate the cardiac mechanical motion signal S from the chest wall vibration signal S H ;

[0096] The ECG signal conversion module is used to convert the cardiac mechanical motion signal S into H Converted into ECG signal.

[0097] Another embodiment of the present invention further provides an electronic device, including:

[0098] A memory storing at least one instruction; and a processor executing the instruction stored in the memory to implement the non-contact electrocardiogram signal monitoring reconstruction method based on the combination of time and frequency domains.

[0099] Another embodiment of the present invention further proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the non-contact electrocardiogram signal monitoring reconstruction method based on the combination of time and frequency domains is implemented.

[0100] Exemplarily, the instructions stored in the memory may be divided into one or more modules / units, which are stored in a computer-readable storage medium and executed by the processor to complete the non-contact ECG signal monitoring reconstruction method based on the time-frequency domain combination of the present invention. The one or more modules / units may be a series of computer-readable instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the server.

[0101] The electronic device may be a computing device such as a smart phone, a notebook, a PDA, and a cloud server. The electronic device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the electronic device may also include more or fewer components, or a combination of certain components, or different components, for example, the electronic device may also include an input / output device, a network access device, a bus, etc.

[0102] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0103] The memory may be an internal storage unit of the server, such as a hard disk or memory of the server. The memory may also be an external storage device of the server, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the server. Furthermore, the memory may include both an internal storage unit of the server and an external storage device. The memory is used to store the computer-readable instructions and other programs and data required by the server. The memory may also be used to temporarily store data that has been output or is to be output.

[0104] It should be noted that the information interaction, execution process and other contents between the above-mentioned module units are based on the same concept as the method embodiment. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0105] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0106] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device that can carry the computer program code to the camera device / terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a disk or an optical disk.

[0107] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0108] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A non-contact ECG signal monitoring reconstruction method based on time-frequency domain combination, characterized in that: include: A frequency modulated continuous wave radar is used to obtain a chest wall vibration signal S of the subject; Separate the cardiac mechanical motion signal S from the chest wall vibration signal S H ; The cardiac mechanical motion signal S is transformed into H Converted into ECG signal.

2. According to claim 1, the non-contact ECG signal monitoring reconstruction method based on time-frequency domain combination is characterized in that: In the above, the cardiac mechanical motion signal S is transformed into H In the step of converting to ECG, a one-dimensional convolutional neural network (1D CNN) with a kernel of 3 is used to extract the cardiac mechanical motion signal S H The extracted data features enter the backbone network after normalization and activation function; the backbone network consists of three reused models, each of which contains four residual blocks ResNetBlock. In each residual block ResNetBlock, the input signal passes through a convolution layer with a kernel of 3 for feature extraction, and then undergoes normalization and activation function processing; thereafter, it passes through another convolution layer with a kernel of 3, undergoes normalization and activation function processing again, and is finally output to the next residual block ResNetBlock; when the number of input channels is the same as the number of output channels, or the stride is 1, the data is jump-connected through a shortcut block; the shortcut block includes a convolution layer with a kernel of 1 and a normalization layer, which is used to restore the information lost in the convolution process and add the obtained features to the learned features.

3. According to claim 2, the non-contact ECG signal monitoring reconstruction method based on time-frequency domain combination is characterized in that: The pre-established ResNet architecture neural network model is trained using supervised learning: Take the ECG signal as the label to train the heart shock signal SCG, learn the mapping relationship between the ECG signal and the heart shock signal SCG, and calculate the value L of the ECG signal loss in the time domain according to the following formula T : Where, L T represents the value of time domain loss, E represents the standard data of ECG signal Ground Truth ECG, P represents the ECG signal obtained by neural network fitting, μ E , μ P They represent the standard data of ECG signal Ground Truth ECG and the mean of ECG signal obtained by neural network fitting, σ E , σ P Represents the standard deviation σ E 2 , σ P 2 represents variance; The mean absolute error (MAE) is used to calculate the distance L between the ECG signal and the standard ECG data GroundTruth ECG in the frequency domain. F : Among them, E F and P F They represent the standard data Ground Truth ECG of the ECG signal and the frequency domain data of the reconstructed ECG signal respectively; the final loss function is as follows: L=L F +L T .

4. According to claim 3, the non-contact ECG signal monitoring reconstruction method based on time-frequency domain combination is characterized in that: The time domain loss and frequency domain loss are used to train the ResNet architecture neural network model respectively, and the error between the calculated heart rate HR and the actual heart rate HR is quantified by the mean absolute error MAE and root mean square error RMSE.

5. According to claim 2, the non-contact ECG signal monitoring reconstruction method based on time-frequency domain combination is characterized in that: Ablation experiments are conducted to verify whether the number of models reused in the backbone network is reasonable, including removing the three reused models in the neural network one by one, and then training under the same conditions. The heart rate HR is calculated from the R peak in the electrocardiogram signal ECG, and the mean absolute error MAE between the heart rate HR and the true heart rate HR is calculated to verify the accuracy of the reconstructed signal.

6. The non-contact ECG signal monitoring reconstruction method based on time-frequency domain combination according to claim 1 is characterized in that: It also includes filtering and denoising the obtained ECG signal using a Butterworth filter: S=butter(N,Wn,x) Where N is the order of the filter, Wn is the normalized cutoff frequency, and x is the filtered signal; Combining the frequency with the highest energy in the spectrum graph and the normal human heart rate range, the heart rate is estimated: HR=max(s)*t Where t is a hyperparameter; Then mark the R peaks in the signal, calculate the difference between two adjacent R peaks, and then divide the result by the sampling rate of the ECG signal to obtain the heartbeat interval IBI: Where fs represents the sampling rate of the electrocardiogram signal ECG.

7. The non-contact ECG signal monitoring reconstruction method based on time-frequency domain combination according to claim 1 is characterized in that: It also includes frequency domain conversion of the ECG signal after filtering and noise reduction, followed by power spectral density PSD analysis to find important frequency components, frequency changes and energy distribution information in the signal, and then analyze and obtain the heart rate.

8. A non-contact ECG signal monitoring and reconstruction system based on time-frequency domain combination, characterized in that: include: A signal acquisition module, used for acquiring a chest wall vibration signal S of a subject using a frequency modulated continuous wave radar; The cardiac mechanical motion signal separation module is used to separate the cardiac mechanical motion signal S from the chest wall vibration signal S H ; The ECG signal conversion module is used to convert the cardiac mechanical motion signal S into H Converted into ECG signal.

9. An electronic device, characterized in that: include: A memory storing at least one instruction; and a processor, executing instructions stored in the memory to implement the non-contact electrocardiogram signal monitoring reconstruction method based on time-frequency domain combination as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the non-contact electrocardiogram signal monitoring reconstruction method based on time-frequency domain combination as claimed in any one of claims 1 to 7 is implemented.

Images