Non-contact electrocardiogram detection method and device, electronic equipment and storage medium

Abstract

The invention discloses a non-contact electrocardio detection method which comprises the following steps: transmitting a frequency modulation continuous wave signal to the heart part of a human body by using an MIMO millimeter wave radar so as to obtain an echo signal reflected by the human body and pre-process the echo signal to obtain a discrete three-dimensional radar signal; enhancing the discrete three-dimensional radar signal by using a 2D beam forming technology to obtain a 2D signal; extracting a target area signal from the 2D signal, and performing phase extraction on the extracted target area signal to obtain phase data; the method comprises the following steps: constructing a WaveGRU-Net network comprising an MODWT module, a CNN module and a Bi-GRU module; and training the WaveGRU-Net network by using the phase data so as to output a reconstructed electrocardiogram by using the trained WaveGRU-Net network, thereby realizing non-contact electrocardiogram detection. According to the method, accurate extraction of the electrocardiosignal is realized, and meanwhile, the signal reconstruction precision is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of electrocardiogram detection, and particularly relates to a non-contact electrocardiogram detection method, device, electronic device, and storage medium. Background Art

[0002] Vital signs such as heart rate, respiratory rate, and blood pressure are important indicators reflecting the health status of the human body, and their abnormalities often indicate potential health risks or sudden medical events. In the field of cardiovascular and cerebrovascular diseases, cardiovascular and cerebrovascular diseases have become a major public health problem globally. Changes in modern social lifestyles, including high-pressure work, unhealthy eating habits, and lack of exercise, have further exacerbated the high risk of cardiovascular and cerebrovascular diseases. At the same time, the long-term management needs of chronic diseases and the early warning and intervention of sudden acute cardiovascular events have become the core concerns in the medical field. As an important parameter for evaluating cardiac physiological activities, effective detection of parameters such as heart rate (HR) and heart rate variability (HRV) can reflect the health status of the cardiovascular system.

[0003] In traditional methods, the detection of HR and HRV mainly relies on contact technologies such as electrocardiogram (ECG) and photoplethysmography (PPG). ECG is mainly a technology that uses an electrocardiograph to record the electroactivity change graph generated by the heart in each cardiac cycle from the body surface. PPG mainly monitors the change of the pulse wave through optical means, can measure heart rate and blood oxygen saturation, and is suitable for portable and wearable devices. However, these contact detection technologies have certain limitations in practical applications. For example, wearing devices by burn patients or people with sensitive skin may cause discomfort or irritation, some users lack compliance due to privacy concerns or poor wearing experience, and the deviation of the device wearing position may lead to inaccurate data, making it difficult to achieve continuous and reliable monitoring.

[0004] With the continuous integration of radar technology and biomedical engineering, non-contact detection technology based on bio-radar has emerged. This technology can non-contact and remotely penetrate non-metallic obstacles such as wooden boards and clothes to detect physiological signals such as the heartbeat and respiration of the human body, and has the characteristics of significant penetration ability and protection of personal privacy.

[0005] However, current non-contact bio-radar technology still faces many challenges in the detection process of heartbeat signals. For example, the amplitude of the chest wall fluctuation signal caused by respiration is much larger than that of the heartbeat signal, which makes the detection of the heartbeat signal weak and even difficult to distinguish in some cases. In addition, respiratory harmonics and other physiological noises may seriously interfere with the accurate extraction of the heartbeat signal, increasing the complexity of signal processing.

[0006] In addition, there are also high requirements for the subsequent processing of the collected Doppler signals. Existing technologies generally use fast Fourier transform to preliminarily determine the target signal, and then use filters and signal decomposition to remove body shaking and various types of noise interference to obtain relatively pure signals for subsequent analysis. In terms of reconstructing electrocardiograms, technology development has gone through multiple stages. In the early days, the extracted Doppler signals were used to use convolutional neural networks (CNN) to try to restore heart sounds, and then there were explorations of using CNN to restore heart shocks. As for the reconstruction of ECG signals, there are currently several main methods: First, a convolutional long short-term memory network is used to take the time-frequency representation of the radar signal as input, the ECG signal after bandpass filtering is reconstructed, and the heart rate is calculated with the help of peak detection technology to obtain the relevant characteristics of the ECG signal; second, a model is constructed by combining CNN with a recurrent neural network, and the model is used to accurately estimate the R peak and S peak signals based on the radar signal, providing an important basis for the analysis of ECG signals; third, there are also studies that use CNN to realize the reconstruction from frequency-modulated continuous wave radar signals to ECG signals, which opens up a new way to obtain high-quality ECG signals.

[0007] However, the above processing method still has the following shortcomings: 1. The signal is easily affected by physiological noise, body movement and other noises, and the existing technology cannot effectively remove such interference for different individuals, resulting in low accuracy and consistency of heartbeat signal reconstruction; 2. The existing technology uses deep learning and often only considers the heartbeat intervals, but cannot consider all the interval details contained in the ECG, resulting in the reconstructed ECG being difficult to present the semantic connotation of different interval information; 3. When reconstructing ECG signals, existing methods often focus on the estimation of basic features such as heartbeat intervals, while ignoring important detail features in the ECG signals such as PQRST waveforms, resulting in the reconstructed ECG signals being unable to fully and accurately reflect subtle changes in heart activity, and are not suitable for scenarios that require precise analysis of subtle heart lesions.

[0008] In summary, how to accurately extract ECG signals under non-contact conditions, eliminate respiratory and other physiological noise interference, and improve signal reconstruction accuracy, especially to achieve accurate semantic expression of ECG information at different intervals, is a core technical problem that needs to be solved urgently. Summary of the invention

[0009] In order to solve the above problems existing in the prior art, the present invention provides a non-contact electrocardiogram detection method, device, electronic device, and storage medium. The technical problem to be solved by the present invention is achieved through the following technical solutions:

[0010] In a first aspect, the present invention proposes a non-contact electrocardiogram detection method, including:

[0011] Transmitting a frequency-modulated continuous wave signal to the human heart area by using a MIMO (Multiple Input Multiple Output) millimeter-wave radar to obtain an echo signal reflected by the human body and performing preprocessing to obtain a discrete three-dimensional radar signal;

[0012] Performing enhancement processing on the discrete three-dimensional radar signal by using 2D (two-dimensional) beamforming technology to obtain a 2D signal;

[0013] Extracting a target area signal from the 2D signal and performing phase extraction on the extracted target area signal to obtain phase data;

[0014] Constructing a WaveGRU-Net network including a MODWT (Multiresolution Discrete Wavelet Transform) module, a CNN module, and a Bi-GRU (Bidirectional Gated Recurrent Unit) module; wherein, the MODWT module is used to perform multi-scale analysis on the input phase data to decompose the signal into multiple frequency bands; the CNN module is used to perform feature extraction on signals in different frequency bands in intervals; the Bi-GRU module is used to perform time-series based modeling on the features extracted by the CNN module to obtain a reconstructed electrocardiogram;

[0015] Training the WaveGRU-Net network by using the phase data so as to output a reconstructed electrocardiogram by using the trained WaveGRU-Net network, thereby realizing non-contact electrocardiogram detection.

[0016] In a second aspect, the present invention proposes a non-contact electrocardiogram detection device for implementing a non-contact electrocardiogram detection method proposed in the first aspect of the present invention. The device includes:

[0017] A signal acquisition module, configured to transmit a frequency-modulated continuous wave signal to the human heart area by using a MIMO millimeter-wave radar to obtain an echo signal reflected by the human body and perform preprocessing to obtain a discrete three-dimensional radar signal;

[0018] A first signal processing module, configured to perform enhancement processing on the discrete three-dimensional radar signal by using 2D beamforming technology to obtain a 2D signal;

[0019] A second signal processing module, configured to extract a target area signal from the 2D signal and perform phase extraction on the extracted target area signal to obtain phase data;

[0020] A model construction module for constructing a WaveGRU-Net network including a MODWT module, a CNN module, and a Bi-GRU module; wherein, the MODWT module is used to perform multi-scale analysis on the input phase data to decompose the signal into multiple frequency bands; the CNN module is used to extract features for different frequency band signals in intervals; the Bi-GRU module is used to perform time-series-based modeling on the features extracted by the CNN module to obtain a reconstructed electrocardiogram;

[0021] A training module for training the WaveGRU-Net network using phase data, so as to output a reconstructed electrocardiogram using the trained WaveGRU-Net network, thereby realizing non-contact electrocardiogram detection.

[0022] In a third aspect, the present invention proposes an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory complete communication with each other through the communication bus;

[0023] The memory is used to store computer programs;

[0024] The processor is used to execute the programs stored on the memory to implement the method provided in the first aspect of the present invention.

[0025] In a fourth aspect, the present invention proposes a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the method provided in the first aspect of the present invention is implemented.

[0026] The beneficial effects of the present invention:

[0027] 1. A non-contact electrocardiogram detection method provided by the present invention first uses a MIMO millimeter-wave radar to transmit a frequency-modulated continuous-wave signal to the human heart area, obtains the echo signal reflected by the human body and performs preprocessing to obtain a discrete three-dimensional radar signal; then uses 2D beamforming technology to enhance the discrete three-dimensional radar signal, extracts the target area signal from the obtained 2D signal, performs phase extraction on the extracted target area signal to obtain phase data; then constructs a WaveGRU-Net network including a MODWT module, a CNN module, and a Bi-GRU module; finally uses the phase data to train the WaveGRU-Net network; the trained WaveGRU-Net network can realize electrocardiogram reconstruction, thereby realizing non-contact electrocardiogram detection. On the one hand, during the early signal processing, the 2D beamforming technology is used to effectively suppress physiological noise and body movement interference, and accurately extract the heartbeat signal; on the other hand, MODWT is combined with deep learning, and the characteristics of MODWT are used to accurately decompose the signal and provide multi-scale features for the network, realizing multi-scale analysis of the signal and extraction of time-series features, and being able to capture various features in the electrocardiogram signal more comprehensively and meticulously, especially detailed features such as PQRST waveforms, so that the reconstructed electrocardiogram signal has a high consistency with the real ECG signal in terms of morphology, time series, and characteristic intervals, improving the data processing efficiency and model accuracy;

[0028] 2. When constructing the CNN module in the non-contact electrocardiogram detection method provided by the present invention, template matching is used to identify electrocardiogram features, and the convolution kernel size of each convolutional module is set according to the key features of different electrocardiogram signals, so as to accurately extract and distinguish the features of each interval, providing more accurate electrocardiogram information, far exceeding the existing technology that focuses on estimating the heartbeat interval and ignores details. The network model has a unique structure, combines multiple technologies to achieve multi-scale analysis and time-series feature extraction, reduces data requirements and computational complexity, improves generalization and real-time processing capabilities, and the overall performance is efficient and practical.

[0029] The present invention will be further described in detail below with reference to the drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flowchart of a non-contact electrocardiogram detection method provided by an embodiment of the present invention;

[0031] Figure 2 It is an overall framework diagram of a non-contact electrocardiogram detection method provided by an embodiment of the present invention;

[0032] Figure 3 It is a schematic diagram of the antenna arrangement and transmission mode of a MIMO millimeter-wave radar provided by an embodiment of the present invention;

[0033] Figure 4Schematic diagram of the process for processing discrete 3D radar signals provided by an embodiment of the present invention;

[0034] Figure 5 Schematic diagram of the structure of the WaveGRU-Net network provided by an embodiment of the present invention;

[0035] Figure 6 Verification result of the consistency of the reconstructed electrocardiogram in the simulation experiment;

[0036] Figure 7 Absolute error results of different intervals of the electrocardiogram in the simulation experiment;

[0037] Figure 8 Average absolute error results of different intervals of the electrocardiogram in the simulation experiment;

[0038] Figure 9 Block diagram of the structure of a non-contact electrocardiogram detection device provided by an embodiment of the present invention. Detailed implementation manners

[0039] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0040] The first aspect of the present invention provides a non-contact electrocardiogram detection method. By collecting reflected signals and performing motion compensation and noise reduction processing, the signal quality is enhanced by combining 2D beamforming technology. The WaveGRU-Net model is used to extract the temporal features of radar signals and reconstruct electrocardiogram signals. This model combines the maximally overlapping discrete wavelet transform (MODWT), convolutional neural network (CNN), and gated recurrent unit (GRU) to effectively separate respiratory and heartbeat signals and accurately reconstruct each interval of the electrocardiogram signal. This method shows high reconstruction accuracy and morphological consistency in electrocardiogram intervals such as R-R, Q-T, P-R, and QRS.

[0041] Specifically, please refer jointly to Figure 1 and Figure 2 , Figure 1 which is a schematic diagram of the process of a non-contact electrocardiogram detection method provided by an embodiment of the present invention, Figure 2 which is the overall framework diagram of a non-contact electrocardiogram detection method provided by an embodiment of the present invention. This method mainly includes the following steps:

[0042] Step 1: Use the MIMO millimeter-wave radar to transmit a frequency-modulated continuous wave signal to the human heart area, obtain the echo signal reflected by the human body and perform preprocessing to obtain a discrete three-dimensional radar signal.

[0043] In this embodiment, the MIMO millimeter-wave radar can perform signal transmission according to the Figure 3 antenna arrangement and transmission mode shown. Among them, Tx1, Tx2, and Tx3 represent the transmitting arrays, Rx1, Rx2, Rx3, and Rx4 represent the receiving arrays, T represents the frame time, T c represents the chirp period, B represents the bandwidth, θ represents the direction angle of the beam, represents the elevation angle of the beam, and λ represents the wavelength.

[0044] First, use the MIMO millimeter-wave radar to obtain the echo signal reflected by the micro-motion of the human chest cavity, and perform rearrangement, motion compensation, and noise reduction preprocessing to obtain the radar echo signal. Among them, the specific implementation processes of motion compensation and noise reduction processing can refer to existing related technologies. The obtained radar target echo signal is a three-dimensional radar signal matrix composed of the fast time dimension, slow time dimension, and channel dimension, denoted as s(t f , t m , l), and the specific expression is as follows:

[0045]

[0046] In the formula, t f is the fast time, t m is the slow time, l is the receiving channel, A IF is the amplitude of the received signal, f c is the carrier frequency of the transmitted signal, β is the frequency modulation rate, (x l , y l ) is the coordinate of the l-th receiving channel, c is the speed of light, φ l is the phase offset of the antenna channel, and its expression is:

[0047]

[0048] d(t m ) represents the distance between the target and the radar, that is, the distance from the human chest cavity to the radar. This distance includes the distance between the human body and the radar and the micro-motion of the chest cavity. Among them, the micro-motion of the chest cavity includes vibrations caused by heartbeat, composite displacements caused by chest expansion and contraction (breathing), and random jitters of the human body. Then d(t m ) can be expressed as:

[0049] d(t m ) = d0 + r(t m );

[0050]

[0051] In the formula, d0 is the distance between the radar and the human chest cavity, and r(t m ) is the chest cavity micro-vibration, that is, the vibration amplitude caused by the heartbeat and respiration in the chest cavity. A r is the heartbeat amplitude, f r is the heartbeat frequency, is the initial heartbeat phase, A h is the respiration amplitude, f h is the respiration frequency, is the initial respiration phase, σ is the fluctuation intensity of the control speed, and W t is the standard Brownian motion, and t m is the slow time.

[0052] Then, the above three-dimensional radar echo signal is converted into a discrete form.

[0053] Specifically, assuming that the fast time t f has N f sampling points, and the slow time t m has N m sampling points, and the sampling interval is Δt m ; t f =n·Δt f , where n = 0, 1, …, N f -1; t m =m·Δt m , where m = 0, 1, …, N m -1; then the discrete three-dimensional radar signal matrix is s[n, m, l], and its expression is:

[0054]

[0055] Step 2: Use 2D beamforming technology to enhance the discrete three-dimensional radar signal to obtain a 2D signal.

[0056] Please refer to Figure 4 , Figure 4 which is the schematic flow chart of processing the discrete three-dimensional radar signal provided by the embodiment of the present invention.

[0057] It should be noted that before using 2D beamforming technology to enhance the discrete three-dimensional radar signal, it also includes:

[0058] Performing FFT (Fast Fourier Transform) processing and MTI (Moving Target Indicator) processing on the discrete three-dimensional radar signal in sequence to eliminate the interference of static targets.

[0059] Specifically, first, perform a Fourier transform on the fast-time dimension of the discrete three-dimensional signal matrix obtained in step 1 to obtain S[k,m,l], and its expression is:

[0060]

[0061] In the formula, k is the range dimension index.

[0062] Then, use the existing MTI technology to eliminate the interference of static targets to obtain the signal S MTI [k,m,l].

[0063] Furthermore, since the signal in the direction of the heart is relatively weak in space. To enhance this signal, after performing FFT processing and MTI processing on the discrete three-dimensional radar signal in sequence, 2D beamforming technology is used to process along the range dimension.

[0064] Specifically, the 2D beamforming technology selects a specific angle according to actual needs to strengthen the heart signal while suppressing interference from other directions. After two-dimensional beamforming, the signal from the target direction is enhanced.

[0065] After two-dimensional beamforming, the signal from the target direction is enhanced. The range-Doppler information of the target is obtained through this enhanced signal, that is, the 2D signal y(k,m), which is expressed as:

[0066]

[0067] In the formula, y[k,m] is the output signal after 2D beamforming, L is the number of receiving antennas, w l is the weighting coefficient of the l-th channel, which is used to adjust the phase of the signal received by each channel. λ is the wavelength of the signal, d n is the distance between the l-th antenna and the reference point, θ b is the direction angle focused by the 2D beamformer, is the elevation angle focused by the 2D beamformer.

[0068] The present invention uses 2D beamforming technology to process the input Doppler signal. Subsequently, the moving sliding window technology is also used to dynamically analyze the change trend of the signal, capture the key information in the signal, and combined with 2D beamforming, it can further focus on the signal in the direction of the heart, effectively suppress interference from other directions, and improve the signal quality. Under different scenarios and individual differences, this method can adaptively adjust to ensure the acquisition of high-quality and stable target signals, providing a reliable basis for subsequent analysis.

[0069] Step 3: Extract the target area signal from the 2D signal, and perform phase extraction on the extracted target area signal to obtain phase data.

[0070] Due to the hardware errors existing in different devices, first, the real and imaginary offsets of the 2D signal are adjusted by using non-linear least squares optimization to eliminate the fixed offset in the signal, achieve DC offset correction, and obtain the corrected 2D signal.

[0071] Then, pulse compression and envelope alignment processing are performed on the corrected 2D signal to extract the target region signal.

[0072] It should be noted that the specific implementation methods of pulse compression and envelope alignment processing can refer to the existing related technologies, and this embodiment will not introduce them in detail.

[0073] After pulse compression and envelope alignment processing, the extracted target region signal is a 1D signal, expressed as:

[0074]

[0075] Finally, the phase of the target region signal is obtained and unwrapped to obtain the phase data of the coupled signal including respiration and heartbeat, expressed as:

[0076]

[0077] In the formula, Θ(m) represents the phase data.

[0078] Step 4: Construct a WaveGRU-Net network including a MODWT module, a CNN module, and a Bi-GRU module.

[0079] Please refer to Figure 5 , Figure 5 which is the structural schematic diagram of the WaveGRU-Net network provided by the embodiment of the present invention. This network includes a MODWT module, a CNN module, and a Bi-GRU module; among them, the MODWT module is used to perform multi-scale analysis on the input phase data to decompose the signal into multiple frequency bands; the CNN module is used to extract features for different frequency band signals in intervals; the Bi-GRU module is used to perform time-series based modeling on the features extracted by the CNN module to obtain the reconstructed electrocardiogram.

[0080] Specifically, the main advantage of MODWT lies in its translation invariance, which can effectively retain the detailed features in the signal, especially in the low-frequency band of the electrocardiogram signal, and has strong suppression ability for noise and external interference. As Figure 5 shown, in this embodiment, the MODWT module is set in the first layer of the network. This module performs multi-scale analysis on the input phase data based on the Sym4 wavelet basis and extracts the key features of the electrocardiogram signal to obtain feature data of different frequency bands; among them, the key features include heart rate changes, P waves, and QRS complexes.

[0081] By using the Sym4 wavelet basis, the signal is decomposed into multiple scales, ensuring that the low-frequency information in the electrocardiogram (such as the heart rate signal) is accurately retained.

[0082] After MODWT processing, the signal is decomposed into multiple frequency components, including the main frequency bands of cardiac electrical activity. This step not only effectively removes noise but also extracts key features (such as heart rate changes, P waves, QRS complexes, etc.) without distorting the signal.

[0083] The present invention introduces MODWT and integrates it into deep learning. MODWT has translational invariance and multi-resolution analysis capabilities. The decomposed signal is not limited by length, suitable for non-stationary signal processing, capable of denoising and feature extraction. Its decomposition results serve as the input layer of deep learning, enabling the model to learn different scale features, adapt to different individual data, and enhance the model's adaptability.

[0084] Furthermore, after being processed by the MODWT layer, the signal enters the CNN layer of WaveGRU-Net to further extract local features of the electrocardiogram. Since different waveforms of the electrocardiogram have significant time and frequency characteristics, the CNN layer uses multiple convolutional modules to gradually extract important local features in the signal. Each convolutional module includes a convolutional layer, batch normalization, Tanh activation function, and max pooling; among them, the kernel size of each convolutional module is optimized according to the key features of different electrocardiogram signals and shows a decreasing trend layer by layer to capture signal features in different frequency bands.

[0085] Specifically, please continue to refer to Figure 5 , in this embodiment, the CNN module specifically includes a first convolutional module, a second convolutional module, a third convolutional module, and a convolutional output layer arranged in sequence; among them,

[0086] The first convolutional module uses 12 convolutional kernels of 64 to extract low-frequency signal features such as P waves (with a duration of 0.08 to 0.10 seconds) and PR intervals (0.12 to 0.20 seconds), retaining low-frequency patterns and reducing noise interference.

[0087] The second convolutional module uses 8 convolutional kernels of 64 to extract high-frequency fast-changing features of the QRS complex. This waveform usually lasts for 0.06 to 0.12 seconds, and smaller convolutional kernels help to accurately extract these features.

[0088] The third convolutional module uses 8 convolutional kernels of 16 to remove remaining noise and retain fine features of the signal;

[0089] The output of the convolutional layer is upsampled through a deconvolution module to restore the spatial resolution of the signal in order to capture more high-frequency details and maintain the complex features of the electrocardiogram signal.

[0090] The CNN network designed by the present invention uses a template matching method to identify electrocardiogram features. Templates are designed based on the characteristics of each interval of the electrocardiogram signal (such as P-R, Q-T, QRS, R-R intervals, etc.). By adjusting model parameters (such as the number of layers), key features are accurately matched and extracted to distinguish signals of different intervals, improve the reconstruction accuracy, and provide a basis for the diagnosis and monitoring of cardiovascular diseases.

[0091] Further, please continue to refer to Figure 5 , where the Bi-GRU module processes the input signal through two paths, forward and backward, captures the dependencies of the front and back cycles in the signal respectively, and outputs the reconstructed electrocardiogram;

[0092] Among them, the input signal at each time step is divided into two paths in the Bi-GRU module. One path is processed in the order of the time series, and the other path processes the signal time series in reverse order to capture the reverse dependencies in the time series.

[0093] This two-way modeling method adopted in this embodiment can better capture the complex relationship between the front and back cycles in the electrocardiogram, especially the timing characteristics between the QRS complex and the T wave, thereby improving the accuracy and consistency of signal reconstruction. The output of the Bi-GRU layer undergoes further extraction and fusion of high-dimensional features through multiple fully connected layers.

[0094] It should be noted that in order to avoid overfitting, the WaveGRU-Net network also uses a dropout layer to randomly discard a part of the connections in the neural network, thereby enhancing the generalization ability of the model.

[0095] Step 5: Use the phase data to train the WaveGRU-Net network so as to output the reconstructed electrocardiogram using the trained WaveGRU-Net network, thereby realizing non-contact electrocardiogram detection.

[0096] It should be noted that before using the phase data to train the WaveGRU-Net network, it also includes:

[0097] Dividing the phase data into a training set, a validation set, and a test set according to time; among them, the training set is used to train and fit the WaveGRU-Net network; the validation set is used to adjust the hyperparameters of the WaveGRU-Net network and prevent overfitting; the test set is used to finally evaluate the performance of the WaveGRU-Net network.

[0098] Then, based on the typical heart rate, set a data window and use a windowing method to enhance the training set.

[0099] Optionally, as an implementation, the data can be divided into multiple groups according to time, with 80% as the training set, 10% as the validation set, and 10% as the test set. To ensure that the model can capture sufficient heartbeat cycle information, thereby better learning the regularity and variability of cardiac activities. For typical heart rates, i.e., 60 to 100 beats per minute, a data window of length 1024 (5.12 seconds) is selected to ensure that the data has sufficient periodicity and representativeness. The sliding window form is adopted, and the step size is set to 512 to enhance the data.

[0100] Then, the training parameters are set.

[0101] Specifically, the input data slice is 1*1024 (5.12 seconds). The MODWT layer divides the wavelet sym4 into 5 levels and 6 layers of signals, and the output length is 1*1024 (5.12 seconds). The Adam optimizer is adopted, the maximum number of epochs is set to 800, the mini-batch size is set to 900, the initial learning rate is 0.001, validation is performed every 40 iterations, detailed information is output every 40 iterations, the data is shuffled after each epoch, and the model with the best validation result is saved. The formats of the input and target data are both in the CTB format.

[0102] Then, the training set is input into the network for training until the training is completed, and a trained network is obtained.

[0103] It can be understood that during the training of the network, the validation set can also be used to adjust the hyperparameters of the WaveGRU-Net network and prevent overfitting. After the training is completed, the test set can also be used to evaluate the performance of the WaveGRU-Net network, and finally a trained network with high performance is obtained.

[0104] Specifically, the performance of the WaveGRU-Net network can be evaluated through the following consistency experimental indicators:

[0105] Pearson Correlation Coefficient (CC): It is used to measure the linear correlation between two variables. The value range of PCC is from -1 to 1, where 1 indicates a perfect positive correlation, -1 indicates a perfect negative correlation, and 0 indicates no correlation. In this evaluation, CC is used to evaluate the linear dependence between the reconstructed waveform and the real waveform. The closer the correlation coefficient is to 1, the stronger the correlation between the reconstructed waveform and the real waveform in the overall shape.

[0106] Cosine Similarity (CosSim): It is used to evaluate the similarity between two vectors. The reconstructed heartbeat waveform and the real waveform are regarded as two vectors, and their cosine similarity is calculated. This metric is used to measure the morphological similarity between the reconstructed waveform and the real waveform. The higher the cosine similarity, the stronger the morphological similarity between the two.

[0107] Root Mean Square Error (RMSE): It is used to evaluate the overall amplitude deviation between the reconstructed waveform and the real waveform. It represents the average standard deviation of the errors (i.e., the differences between the reconstructed waveform and the real waveform) at all time points. The smaller the RMSE, the higher the similarity between the reconstructed waveform and the real waveform, and the smaller the difference.

[0108] For the object to be detected, the corresponding phase data can be obtained through the solution in steps 1 - 3, and then the phase data is input into the trained network to obtain the reconstructed electrocardiogram, thus realizing non-contact electrocardiogram detection. The obtained reconstructed electrocardiogram has extremely high consistency with the original real ECG signal in terms of morphology, time sequence, and characteristic intervals.

[0109] A non-contact electrocardiogram detection method provided by the present invention first uses a MIMO millimeter-wave radar to transmit a frequency-modulated continuous-wave signal to the human heart area, and preprocesses the echo signal reflected by the human body to obtain discrete three-dimensional radar signals; then uses 2D beamforming technology to enhance the discrete three-dimensional radar signals, extracts the target area signals from the obtained 2D signals, and performs phase extraction on the extracted target area signals to obtain phase data; then constructs a WaveGRU-Net network including a MODWT module, a CNN module, and a Bi-GRU module; finally, uses the phase data to train the WaveGRU-Net network; the trained WaveGRU-Net network can realize electrocardiogram reconstruction, thus realizing non-contact electrocardiogram detection. On the one hand, during the early signal processing, this method uses 2D beamforming technology to effectively suppress physiological noise and body movement interference and accurately extracts the heartbeat signal; on the other hand, it combines MODWT with deep learning, uses the characteristics of MODWT to accurately decompose the signal and provides multi-scale features for the network, realizes multi-scale analysis of the signal and extraction of time sequence features, can capture various features in the electrocardiogram signal more comprehensively and meticulously, especially details such as the PQRST waveform, so that the reconstructed electrocardiogram signal has high consistency with the real ECG signal in terms of morphology, time sequence, and characteristic intervals, improving the data processing efficiency and model accuracy.

[0110] The effectiveness of the method proposed by the present invention is verified and illustrated through simulation experiments below.

[0111] I. Experimental Parameter Settings

[0112] In this experiment, the AWR1843BOOST millimeter-wave radar system and DCA1000 data capture board from Texas Instruments (TI) were used. The radar system utilized three transmitters (Tx) and four receivers (Rx) to form a 12-channel virtual two-dimensional antenna array. The transmitters employed a time-division multiplexing strategy to sequentially transmit radio frequency signals at intervals of 5 milliseconds, and the baseband signals were captured at each receiver. The system parameters are shown in Table 1.

[0113] Table 1 Experimental Hardware Parameters

[0114] Parameter Value Radar Center Frequency / GHz 77 Frequency Modulation Rate / MHz / ms 65 Pulse Repetition Period / Hz 200 Idle Time / ms 10 Ramp End Time / ms 60 Number of Sampling Points 256 Sampling Rate / MHz 5 Frame Period / ms 5

[0115] II. Experimental Contents and Result Analysis:

[0116] In this experiment, data from 20 individuals were obtained. Based on the method proposed in the present invention, the WaveGRU-Net network was trained, tested, and verified using the data of 10 individuals, and the consistency of the reconstructed electrocardiogram at different intervals was verified for the other 10 individuals. The results are as Figures 6 - 8 shown. Among them, Figure 6 is the consistency verification of the reconstructed electrocardiogram, Figure 7 is the absolute error result of the electrocardiogram at different intervals in the simulation experiment, Figure 8 is the mean absolute error result of the electrocardiogram at different intervals in the simulation experiment.

[0117] From Figure 6 it can be seen that the RMSE value of the WaveGRU-Net network proposed in the present invention is relatively low, ranging from 0.05 to 0.15, and the median is close to 0.1, which reflects relatively small and consistent reconstruction errors. The PCC and CosSim values of WaveGRU-Net are both close to 1, and the interquartile range is between 0.9 and 0.95, indicating a high degree of consistency in waveform reconstruction and very few outliers. These results highlight the excellent performance of WaveGRU-Net in terms of accuracy, stability, and consistency, and all evaluation indicators show more accurate and reliable waveform reconstruction effects.

[0118] In addition, WaveGRU-Net also has excellent performance in generating electrocardiogram waveforms. Figure 7 and Figure 8 present the mean absolute error of the WaveGRU-Net network proposed in the present invention at four ECG duration intervals of R-R, QRS, P-R, and Q-T. Among them, the average errors of the R-R interval and Q-T interval are both 5 milliseconds; the average error of the P-R interval is 12.25 milliseconds; the average error of the QRS interval is 10.25 milliseconds.

[0119] Thus, the effectiveness of the present invention was verified.

[0120] Based on the same inventive concept, the second aspect of the present invention further provides a non-contact electrocardiogram detection device. Please refer to Figure 9 , Figure 9 which is a structural block diagram of a non-contact electrocardiogram detection device provided by an embodiment of the present invention. The device includes:

[0121] A signal acquisition module, configured to transmit a frequency-modulated continuous wave signal to the human heart part by using a MIMO millimeter-wave radar, so as to acquire an echo signal reflected by the human body and perform preprocessing to obtain a discrete three-dimensional radar signal;

[0122] A first signal processing module, configured to perform enhancement processing on the discrete three-dimensional radar signal by using a 2D beamforming technology to obtain a 2D signal;

[0123] A second signal processing module, configured to extract a target area signal from the 2D signal and perform phase extraction on the extracted target area signal to obtain phase data;

[0124] A model construction module, configured to construct a WaveGRU-Net network including a MODWT module, a CNN module, and a Bi-GRU module; wherein the MODWT module is configured to perform multi-scale analysis on the input phase data to decompose the signal into multiple frequency bands; the CNN module is configured to perform feature extraction on signals in different frequency bands in intervals; the Bi-GRU module is configured to perform time-series-based modeling on the features extracted by the CNN module to obtain a reconstructed electrocardiogram;

[0125] A training module, configured to train the WaveGRU-Net network by using the phase data, so as to output a reconstructed electrocardiogram by using the trained WaveGRU-Net network, thereby realizing non-contact electrocardiogram detection.

[0126] Based on the same inventive concept, the third aspect of the present invention further provides an electronic device, which includes a processor, a communication interface, a memory, and a communication bus; wherein, the processor, the communication interface, and the memory complete communication with each other through the communication bus;

[0127] The memory is used to store a computer program;

[0128] When the processor is configured to execute the program stored in the memory, it realizes the method steps provided by the first aspect of the present invention.

[0129] Based on the same inventive concept, the fourth aspect of the present invention further proposes a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, it realizes the method steps provided by the first aspect of the present invention.

[0130] It should be noted that for the embodiments of the device, electronic device, and storage medium, since they are basically similar to the method embodiments, the description is relatively simple. For the relevant parts, please refer to the partial description of the method embodiments.

[0131] The device, electronic device, and storage medium of the embodiments of the present invention are respectively the device, electronic device, and storage medium applying the above non-contact electrocardiogram detection method. Then all the embodiments of the above non-contact electrocardiogram detection method are applicable to the device, electronic device, and storage medium, and can achieve the same or similar beneficial effects.

[0132] Although the present application is described in conjunction with various embodiments herein, however, in the process of implementing the claimed present application, those skilled in the art can understand and implement other variations of the disclosed embodiments by viewing the drawings, the disclosure content, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "one" does not exclude a plurality of cases. A single processor or other unit can implement several functions recited in the claims. Certain measures are recited in mutually different dependent claims, but this does not mean that these measures cannot be combined to produce good results.

[0133] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a device (equipment), or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects, and here they are all collectively referred to as "modules" or "systems". Moreover, the present application can take 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 code. The computer program is stored / distributed in a suitable medium, provided together with other hardware or as part of the hardware, and can also be in other distribution forms, such as through the Internet or other wired or wireless telecommunication systems.

[0134] The above content is a further detailed description of the present invention in combination with specific preferred implementation manners, and it cannot be determined that the specific implementation of the present invention is only limited to these descriptions. For those of ordinary skill in the technical field to which the present invention pertains, without departing from the concept of the present invention, several simple deductions or substitutions can still be made, and all should be regarded as belonging to the protection scope of the present invention.

Claims

1. A non-contact electrocardiogram detection method, characterized in that, Including: Using an MIMO millimeter-wave radar to transmit a frequency-modulated continuous-wave signal to the human heart area, obtaining an echo signal reflected by the human body and performing preprocessing to obtain a discrete three-dimensional radar signal; Using 2D beamforming technology to perform enhancement processing on the discrete three-dimensional radar signal to obtain a 2D signal; Extracting a target area signal from the 2D signal and performing phase extraction on the extracted target area signal to obtain phase data; Constructing a WaveGRU-Net network including a MODWT module, a CNN module, and a Bi-GRU module; wherein, the MODWT module is used to perform multi-scale analysis on the input phase data to decompose the signal into multiple frequency bands; the CNN module is used to perform feature extraction on signals in different frequency bands in intervals; the Bi-GRU module is used to perform time-series-based modeling on the features extracted by the CNN module to obtain a reconstructed electrocardiogram; Using the phase data to train the WaveGRU-Net network, so as to use the trained WaveGRU-Net network to output a reconstructed electrocardiogram, thereby realizing non-contact electrocardiogram detection.

2. The non-contact electrocardiogram detection method according to claim 1, characterized in that Extracting a target area signal from the 2D signal and performing phase extraction on the extracted target area signal to obtain phase data, including: Using nonlinear least squares optimization on the 2D signal to adjust the offset of the real part and the imaginary part to eliminate the fixed offset in the signal, realizing DC offset correction, and obtaining a corrected 2D signal; Performing pulse compression and envelope alignment processing on the corrected 2D signal to extract the target area signal; Obtaining the phase of the target area signal and performing phase unwrapping to obtain phase data including the coupled signals of respiration and heartbeat.

3. The non-contact electrocardiogram detection method according to claim 1, wherein The MODWT module performs multi-scale analysis on the input phase data based on the Sym4 wavelet basis and extracts the key features of the electrocardiogram signal to obtain feature data in different frequency bands; wherein, the key features include heart rate changes, P waves, and QRS complexes.

4. A non-contact electrocardiogram detection method according to claim 1, characterized in that, The CNN module includes multiple convolutional modules, and each convolutional module includes a convolutional layer, batch normalization, a Tanh activation function, and max pooling; wherein, the convolutional kernel size of each convolutional module is set according to the key features of different electrocardiogram signals and shows a decreasing trend layer by layer to capture signal features in different frequency bands.

5. The non-contact electrocardiogram detection method according to claim 4, characterized in that, The CNN module includes a first convolutional module, a second convolutional module, a third convolutional module, and a convolutional output layer arranged in sequence; wherein, The first convolutional module uses 12 convolutional kernels of 64 to extract the low-frequency signal features of P waves and PR intervals; The second convolutional module uses 8 convolutional kernels of 64 to extract the high-frequency fast-changing features of QRS complexes; The third convolutional module uses 8 convolutional kernels of 16 to remove the remaining noise and retain the fine features of the signal; The output of the convolutional layer is upsampled through a transposed convolutional module to restore the spatial resolution of the signal, thereby capturing more high-frequency details and maintaining the complex features of the electrocardiogram signal.

6. The non-contact electrocardiogram detection method according to claim 1, characterized in that The Bi-GRU module processes the input signal through two paths, forward and backward, captures the dependencies of the front and back cycles in the signal respectively, and outputs the reconstructed electrocardiogram. Among them, the input signal at each time step is divided into two paths in the Bi-GRU module. One path is processed in the order of the time series, and the other path processes the signal time series in reverse order to capture the reverse dependencies in the time series.

7. A non-contact electrocardiogram detection method according to claim 1, characterized in that Before training the WaveGRU-Net network using the phase data, it further includes: Dividing the phase data into a training set, a validation set, and a test set according to time; setting a data window based on the typical heart rate, and using the windowing method to perform enhancement processing on the training set. Among them, the training set is used to train and fit the WaveGRU-Net network; the validation set is used to adjust the hyperparameters of the WaveGRU-Net network and prevent overfitting; the test set is used to finally evaluate the performance of the WaveGRU-Net network.

8. A non-contact electrocardiogram detection device for implementing the non-contact electrocardiogram detection method according to any one of claims 1-7, characterized in that, The device includes: A signal acquisition module, which is used to transmit a frequency-modulated continuous wave signal to the human heart part by using an MIMO millimeter-wave radar, obtain the echo signal reflected by the human body and perform preprocessing to obtain a discrete three-dimensional radar signal. A first signal processing module, which is used to perform enhancement processing on the discrete three-dimensional radar signal by using 2D beamforming technology to obtain a 2D signal. A second signal processing module, which is used to extract the target area signal from the 2D signal and perform phase extraction on the extracted target area signal to obtain phase data. A model construction module, which is used to construct a WaveGRU-Net network including a MODWT module, a CNN module, and a Bi-GRU module; among them, the MODWT module is used to perform multi-scale analysis on the input phase data to decompose the signal into multiple frequency bands; the CNN module is used to perform feature extraction on different frequency band signals in intervals; the Bi-GRU module is used to perform time series-based modeling on the features extracted by the CNN module to obtain a reconstructed electrocardiogram. A training module, which is used to train the WaveGRU-Net network by using the phase data, so as to output a reconstructed electrocardiogram by using the trained WaveGRU-Net network, thereby realizing non-contact electrocardiogram detection.

9. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete mutual communication through the communication bus. The memory is used to store computer programs. The processor is used to execute the programs stored on the memory to implement the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed, it can implement the method according to any one of claims 1-7.

Images