Breathing and heartbeat detection method and system based on frequency modulated continuous wave radar

Abstract

The invention discloses a breath and heartbeat detection method and system based on an FMCW (Frequency Modulated Continuous Wave) radar. The method comprises the following steps: acquiring a continuous wave signal sent by an FMCW millimeter wave radar to detect a target object so as to obtain a reflected echo signal; the echo signals are preprocessed; separating a respiration signal from a heartbeat signal of the preprocessed echo signal based on an EPH multi-stage cooperative signal processing algorithm; and carrying out anomaly correction on the separated instantaneous frequency anomaly data based on a local weighted regression algorithm. Therefore, the vital sign detection accuracy of the radar can be improved under the interference conditions of background noise, signal harmonic waves and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of frequency-modulated continuous wave (FMCW) radar technology, in particular to a breathing and heartbeat detection method and system based on a frequency-modulated continuous wave radar. BACKGROUND

[0002] During the process of breathing and heartbeat, the chest cavity part will appear slight fluctuation. The millimeter wave radar can penetrate clothes and capture signals by detecting the Doppler effect or phase change caused by body surface displacement. The wavelength of the millimeter wave is relatively short, generally in the range of 1 to 10 millimeters, and the radar is very sensitive to very small displacement changes, and can well capture the slight fluctuation of the chest cavity and the body surface. After detecting the phase change of the reflected signal, the phase change signal is subjected to time-frequency analysis and filtering processing, etc., and finally the frequency information of breathing and heartbeat can be extracted.

[0003] In the field of vital sign detection, different types of radar technology have their unique advantages and limitations. Continuous wave radar (CW) is based on the Doppler effect, and realizes breathing and heartbeat detection by detecting the frequency offset caused by chest cavity micro-motion, but cannot distinguish multiple targets or static clutter, and is easily affected by environmental interference. Frequency-modulated continuous wave radar (FMCW) can simultaneously obtain target distance and speed information, and can suppress static clutter and distinguish multiple targets at different distances. Ultra-wideband radar (UWB) can accurately capture the details of chest cavity micro-motion, and has strong ability to penetrate non-metallic obstacles such as clothes and thin walls, but the frequency band occupation conflicts, and the regulations are strictly limited. Compared with the above three kinds of radars, the FMCW millimeter wave radar combines the signal modulation characteristics of FMCW and the advantages of millimeter wave in vital sign detection. Millimeter wave can penetrate non-metallic materials such as clothes and bedding, directly sense chest cavity activity, avoid privacy leakage problem of camera monitoring, and is suitable for home and medical scenarios. At the same time, the millimeter wave radar has excellent anti-interference ability, which can divide the space range and angular domain interval, and use the related method of harmonic correction to improve the detection and tracking of vital signs, thereby improving the accuracy of heart rate detection. In the field of medical monitoring, the millimeter wave radar has unique advantages that other technologies cannot replace.

[0004] Therefore, the non-contact vital sign detection technology using the frequency modulation continuous wave radar has important application value in the fields of home health monitoring, respiratory and heart disease diagnosis and the like. However, the problems of mode aliasing, nonlinearity and non-stationarity existing in the radar signal processing process seriously restrict the accurate extraction of the respiratory and heartbeat signals. Therefore, how to improve the accuracy of radar detection of vital signs under the interference of background noise, signal harmonics and the like and apply the radar detection of vital signs to the detection of respiratory and heart diseases is a problem to be solved. SUMMARY

[0005] The respiratory heartbeat detection method and system based on the frequency modulation continuous wave radar provided by the application can improve the accuracy of radar detection of vital signs under the interference of background noise, signal harmonics and the like.

[0006] In a first aspect, a respiratory heartbeat detection method based on a frequency modulation continuous wave radar is provided, comprising the following steps: An echo signal reflected back by a target object detected by a FMCW millimeter wave radar sending a continuous wave signal is acquired; The echo signal is preprocessed; The respiratory signal and the heartbeat signal of the preprocessed echo signal are separated based on an EPH multi-level collaborative signal processing algorithm; The abnormal data of the separated instantaneous frequency are corrected based on a local weighted regression algorithm.

[0007] In a second aspect, a respiratory heartbeat detection system based on a frequency modulation continuous wave radar is provided, comprising: An echo signal acquisition module is configured to acquire an echo signal reflected back by a target object detected by a FMCW millimeter wave radar sending a continuous wave signal; A preprocessing module is in communication connection with the echo signal acquisition module and is configured to preprocess the echo signal; A separation module is in communication connection with the preprocessing module and is configured to separate the respiratory signal and the heartbeat signal of the preprocessed echo signal based on an EPH multi-level collaborative signal processing algorithm; and An abnormal correction module is in communication connection with the separation module and is configured to correct the abnormal data of the separated instantaneous frequency based on a local weighted regression algorithm.

[0008] Compared with the prior art, the application has the following advantages: the radar echo signal is preprocessed in view of the problems of static clutter and noise interference in the radar signal; the respiratory signal and the heartbeat signal of the preprocessed echo signal are separated based on an EPH multi-level collaborative signal processing algorithm to solve the problems of mode aliasing and non-linear signal separation; and the abnormal data of the separated instantaneous frequency are corrected based on a local weighted regression algorithm to solve the problems of insufficient instantaneous frequency resolution and stability. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating an embodiment of a respiratory and heartbeat detection method based on frequency-modulated continuous wave radar according to the present invention. Figure 2 This is a flowchart illustrating another embodiment of the respiratory and heartbeat detection method based on frequency-modulated continuous wave radar of the present invention. Figure 3 This is a comparison chart of respiratory and heart rate experimental data of the present invention; Figure 4 This is a comparison diagram of the Hilbert spectrum of this invention; Figure 5 This is the Pearson correlation coefficient plot of the present invention; Figure 6 This is a structural diagram of the CNN classification model of this invention; Figure 7 This is a flowchart of the CNN classification model of the present invention; Figure 8 This is a graph showing the change in detection accuracy of the CNN classification model of this invention. Detailed Implementation

[0010] Referring now to specific embodiments of the invention, examples of which are illustrated in the accompanying drawings. Although the invention will be described in conjunction with specific embodiments, it will be understood that it is not intended to limit the invention to the described embodiments. Rather, it is intended to cover variations, modifications, and equivalents included within the spirit and scope of the invention as defined by the appended claims. It should be noted that the method steps described herein can be implemented by any functional block or functional arrangement, and any functional block or functional arrangement can be implemented as a physical entity or a logical entity, or a combination of both.

[0011] To enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0012] Note: The examples described below are merely specific examples and are not intended to limit the embodiments of the present invention to the specific steps, values, conditions, data, order, etc. Those skilled in the art can utilize the concept of the present invention to construct more embodiments not mentioned herein by reading this specification.

[0013] Non-contact vital sign detection technology using frequency-modulated continuous wave radar has significant application value in areas such as home health monitoring and diagnosis of respiratory and cardiac diseases. However, problems such as mode aliasing, nonlinearity, and non-stationarity in radar signal processing severely limit the accurate extraction of respiratory and heartbeat signals. Furthermore, insufficient frequency resolution and the separation of multiple physiological signals under subtle human movements remain critical technical bottlenecks that urgently need to be overcome. To address these issues, this invention describes the application of Empirical Wavelet Transform (EWT) and Hilbert-Huang Transform (HHT) for respiratory and heartbeat detection. Therefore, this invention provides a respiratory and heartbeat detection method based on frequency-modulated continuous wave radar, specifically including the following steps, see [link to details]. Figure 1 and Figure 2 As shown: S100: Obtain the echo signal reflected back from the target object by the continuous wave signal sent by the FMCW millimeter-wave radar; S200, preprocess the echo signal; the purpose of radar echo signal preprocessing is to extract weak physiological signals related to respiration and heart rate from the echo signal. The preprocessing process mainly includes signal reception and sampling and mixing to obtain intermediate frequency signals, range-dimensional FFT, elimination of static interference, and phase processing (phase extraction, phase expansion, phase difference). The significance of preprocessing is to remove noise, improve signal quality, extract subtle physiological signals, and enhance system robustness. Through the preprocessing process, the original radar echo signal can be processed into a phase difference signal, laying the foundation for subsequent vital sign signal separation and health detection. The specific steps are as follows: S210, The echo signal and the original transmitted signal corresponding to the echo signal are mixed to obtain an intermediate frequency signal; FMCW millimeter-wave radar continuously transmits a continuous wave signal with a linearly varying frequency. When this signal is reflected by a target object, the receiver captures the echo and mixes it with the original transmitted signal. There is a time difference between the echo signal and the current transmitted signal, which is proportional to the target distance. The mixer generates a difference frequency signal, called the intermediate frequency (IF) signal, which is directly related to the target distance and velocity.

[0014] S220, Perform a Fast Fourier Transform (FFT) on the intermediate frequency signal to determine the distance dimension; After the transmit signal and echo signal are mixed to generate an intermediate frequency signal, the intermediate frequency signal is converted into a digital signal by an analog-to-digital converter (ADC). There is a significant time scale difference between the sampling interval (fast time) of the echo signal processed by the ADC and the interval (slow time) of the transmit signal. The fast time dimension refers to the sampling time sequence of the echo signal within a single chirp, corresponding to the time delay from transmission to reception of the radar, usually on the order of microseconds or nanoseconds, and is related to the bandwidth and sampling rate of the radar signal. The slow time dimension refers to the transmission interval sequence between multiple chirps, corresponding to the time period of periodic transmission of the radar chirp, usually on the order of milliseconds, and is related to the pulse repetition interval (PRI) of the radar.

[0015] The fast time FFT is used to determine the distance of the target (such as a human body), and the slow time FFT is used to extract the micro-motion information (such as the chest movement caused by breathing and heart rate) of the target. The fast time FFT and the slow time FFT are essential steps. By combining the two FFTs, the breathing and heart rate signals can be effectively separated.

[0016] For radar data from one of the antennas, the baseband data of multiple chirps form a two-dimensional data matrix. Since the time within a single chirp is very short, it can be considered that the target distance detected by the radar does not change, so performing FFT on the horizontal data of the matrix (fast time, also called distance dimension FFT) can obtain a distance-amplitude spectrum representing the signal strength at different distances. Performing FFT on the vertical data of the matrix (slow time, also called velocity dimension FFT) can be used to calculate the velocity information of the target signal.

[0017] Since the fast time FFT can only provide distance information of the target, it cannot directly extract the micro-motion information (such as the chest movement caused by breathing and heart rate) of the target. The chest movement caused by breathing and heart rate will produce a small Doppler shift, and even a small change in distance or time will cause a large change in phase, so the slow time FFT can extract this frequency shift information.

[0018] S230, based on the slow time dimension phase mean cancellation algorithm to eliminate the static clutter interference of the distance dimension FFT, the specific steps are as follows: S231, calculating the local phase mean of the echo signal using a sliding window method; S232, at each frame of the sliding window, using the local phase mean to perform phase cancellation to obtain a local phase difference signal; S232, using the local phase difference signal to reconstruct the echo signal with static clutter interference eliminated.

[0019] In the application of millimeter wave radar detecting human vital sign signals, static interference such as walls, furniture, beds and other fixed objects can cause serious interference to weak vital sign signals (such as breathing and heartbeat). These interference signals appear as stable strong echoes in the distance dimension, which may mask the weak changes of vital sign signals. Therefore, eliminating static interference is a key step to improve the detection accuracy of vital signs.

[0020] In order to further improve the effect of static filtering algorithm, the slow time dimension phase cancellation algorithm is introduced into the local mean calculation and dynamic adjustment mechanism, so as to better adapt to the change of signal. The specific steps are as follows: instead of using global mean, local phase mean is calculated in the way of sliding window. Assuming that the size of sliding window is frame, the local phase mean at the kth frame can be expressed as:

[0021] The local mean calculation can better adapt to the non-stationary characteristics of the signal. According to the dynamic characteristics of the signal, the size of the sliding window is adaptively adjusted . For example, when a sudden motion is detected, the window size can be reduced to improve the response speed of the algorithm. When the signal is stable, the window size can be increased to improve the accuracy of mean estimation. At each frame, the local phase mean is used for phase cancellation to obtain the local phase difference signal :

[0022] Then, the radar echo signal removed from static clutter is reconstructed using the local phase difference signal, as follows:

[0023] , which represents the amplitude of the echo signal at the th distance unit. Through local mean calculation and dynamic adjustment of window size, the improved algorithm can better adapt to non-stationary signals, effectively suppress local interference, and improve the robustness of the algorithm.

[0024] S240, the distance dimension FFT for eliminating static clutter interference is sequentially subjected to phase extraction, phase unfolding and phase difference.

[0025] The phase modulation characteristics of FMCW millimeter wave radar enable it to detect small motion changes, such as human breathing and heartbeat. When the target moves slightly, the phase of the echo signal will change accordingly, and this phase change is proportional to the displacement of the target. By extracting the phase information, high-precision detection of small motion can be realized.

[0026] When the target undergoes slight motion (such as the fluctuation of the chest), the phase of the reflected signal will change. The phase change , the distance change of the micro-motion between the target and the radar is related as:

[0027] where, is the wavelength of the radar signal. For millimeter-level chest motion, the phase change is very obvious, and the radar can infer the slight motion of the target by detecting the phase change.

[0028] The phase folding phenomenon will cause the phase value to be discontinuous, affecting the accurate detection of target motion. Therefore, phase unwrapping is needed, and the phase unwrapping formula is as follows:

[0029] is an integer, used to compensate for phase jumps. Through phase unwrapping, the folded phase value is restored to a continuous phase curve, accurately reflecting the actual motion of the target.

[0030] The signal received by the radar not only contains the motion information of the target, but also may contain static clutter (such as the reflection signal of stationary objects such as walls and furniture) and noise. Direct analysis of the phase value may be affected by these disturbances, leading to inaccurate detection results. Phase difference is to extract the motion information of the target by calculating the change of the phase value at adjacent time instants. The idea of phase difference is to calculate the phase change at adjacent time instants to extract the motion information of the target. The formula is as follows:

[0031] where, is the phase value at the current time instant, is the phase value at the previous time instant.

[0032] Since the phase value of the static target is almost unchanged at adjacent time instants, the result of phase difference is close to zero, and it can be filtered out by phase difference. The phase value of the moving target will change at adjacent time instants, and phase difference can highlight these changes.

[0033] In radar echo signal processing, the extracted phase information reflects the motion state of the human chest cavity, mainly including components such as respiration, heartbeat, and random body tremors. Therefore, the phase signal can be regarded as a mixture of respiratory signals, heartbeat signals, and environmental noise. Since the heartbeat signal is usually weak, it is easily masked by high-order harmonics of the respiratory signal or environmental noise, making its extraction difficult. To solve this problem, existing technologies provide various signal separation algorithms, including methods based on time-frequency joint representation (such as Short-Time Fourier Transform (STFT) and Wavelet Transform (WT), which locate and extract target components by characterizing the time-frequency distribution of signal energy. STFT provides an intuitive time-frequency spectrum of the signal using a sliding Fourier transform with a fixed window function, but its resolution is limited by the choice of window length, making it difficult to balance the expression requirements of respiratory slowing components and transient heartbeat features. Wavelet transform overcomes the fixed resolution limitation through multi-scale convolution. For example, the Gaussian modulation characteristics of Morlet wavelets can adapt to the quasi-periodic fluctuations of physiological signals, but it is easily affected by mode aliasing in scenarios with overlapping frequency bands (such as respiratory harmonics interfering with heartbeat).

[0034] Mode decomposition (MD) is a data-driven modal decomposition method (such as VMD, EMD, EEMD, EWT, and HHT) that autonomously separates intrinsic signal components by generating eigenmode functions (IMFs). VMD, based on variational optimization constraining modal bandwidth, can accurately separate the respiratory fundamental frequency from the high-frequency components of the heartbeat and suppress harmonic interference. EMD extracts locally stationary modes through iterative screening without the need for preset basis functions, but it is sensitive to noise and susceptible to endpoint effects. HHT, combining EMD and Hilbert transform, further endows time-varying frequency analysis with analytical capabilities, providing support for dynamic physiological parameter analysis.

[0035] Short-time Fourier transform is easily limited by time-frequency resolution, and the choice of basis functions in wavelet transform directly affects the decomposition effect. While empirical mode decomposition (EMD) methods can adaptively process nonlinear signals, they suffer from mode aliasing and endpoint effects. Variational mode decomposition (VMD) improves stability by pre-setting the number of modes, but its parameter sensitivity is difficult to balance with the non-stationary characteristics of physiological signals.

[0036] To address the aforementioned issues, the aim is to find an efficient and reliable method to accurately extract weak heartbeat signals and reduce interference from respiratory signal harmonics and environmental noise.

[0037] In S300, the preprocessed echo signal is separated into respiratory and heartbeat signals based on the EPH multi-level collaborative signal processing algorithm; the specific steps are as follows: S310, based on the empirical wavelet transform algorithm, removes harmonics and noise from the preprocessed echo signal; The Empirical Wavelet Transform (EWT) is an adaptive wavelet transform method used to process non-stationary and nonlinear signals. The EWT algorithm adaptively constructs wavelet filters based on the characteristics of the signal, decomposes the signal, and extracts the heart rate and respiratory frequency band signals using a bandpass filter.

[0038] To separate the dominant respiratory frequency, dominant heartbeat frequency, harmonics, and noise components, the EWT algorithm can remove harmonics and noise. The EWT algorithm separates these components in the frequency domain; if the respiratory harmonics do not overlap with the dominant heartbeat frequency, they can be directly removed using EWT's frequency band segmentation. If the harmonics partially overlap with the dominant heartbeat frequency, EWT's dynamic adjustment strategy can allocate the mixed components to independent sub-bands, avoiding mode aliasing.

[0039] S320, based on particle swarm optimization algorithm and Shannon entropy algorithm, finds the final empirical mode decomposition termination condition of the signal after removing harmonics and noise; S330: Select the IMF signal within the respiratory and heartbeat frequency bands from the final empirical mode decomposition signal and perform Hilbert-Huang transform to obtain the respiratory signal and heartbeat signal.

[0040] The optimal EMD termination condition is found by using optimization algorithm (PSO) and Shannon entropy (SE) to remove noise and harmonics and select signals within a specific frequency band, thereby reducing the complexity and noise interference during EMD decomposition.

[0041] Particle Swarm Optimization (PSO) algorithm simulates the foraging behavior of bird flocks, achieving rapid convergence through a particle position-velocity update mechanism. It boasts advantages such as few parameters and ease of implementation. It searches for optimal solutions through information sharing between individuals and the group. Each solution continuously compares itself with its own experience and the group's optimal solution, adjusting its search strategy to move closer to the optimal solution. Each particle represents a potential solution, constantly adjusting its position and velocity to find the optimal solution.

[0042] Shannon entropy (SE) measures the degree of uncertainty or disorder in information and is used to evaluate the complexity of EMD decomposition results. A higher entropy value means that the decomposed IMF signal is more complex and contains more information. Conversely, a smaller SE value indicates that the current IMF frequency is cleaner, which is more conducive to extracting purer respiratory and heart rate signals. Compared with other entropies, such as fuzzy entropy and sample entropy, Shannon entropy does not require preset parameters, avoiding the sensitivity of the results to parameter selection, and is suitable for real-time signal processing.

[0043] Empirical Mode Decomposition (EMD) is an adaptive decomposition method for non-stationary and nonlinear signals. Its core idea is to decompose complex signals into several intrinsic mode functions (IMFs), each IMF representing an oscillation mode at a different time scale in the signal. EMD does not require preset basis functions and relies entirely on the inherent characteristics of the signal for decomposition, making it particularly suitable for the separation and analysis of physiological signals (such as respiration and heartbeat).

[0044] A standard deviation (SD) termination condition is added to the EMD termination criteria. The standard deviation measures the degree of change in the current IMF between two iterations. During decomposition, the decomposition stops when the standard deviation of the IMF falls below a preset threshold. During the iteration of EMD, as local means are continuously removed, the IMF gradually converges to a stable state. The shape and amplitude changes of the IMF decrease, and the standard deviation of these changes also decreases. Optimizing the preset threshold can prevent excessive decomposition of high-frequency noise.

[0045] The core idea of ​​this algorithm is to find the most suitable preset threshold for the standard deviation through PSO optimization, using Shannon entropy as the optimization criterion. In other words, the optimization objective is to reduce signal complexity and make the signal purer. The formula for standard deviation is expressed as follows:

[0046] In the formula, SD is the standard deviation threshold parameter; It is the first The IMF signal of the next iteration; The number of sampling points for the signal; For the first One particle.

[0047] in, It is the first The next iteration of the IMF, It is the number of sampling points for the signal.

[0048] The specific process of the PSO-HHT algorithm is as follows: Step 1: Initialize all parameters of the particle swarm optimization algorithm: Set the number of iterations to [value missing]. The maximum number of iterations is Randomly initialize the initial position of each particle. and speed The position represents the standard deviation parameter added to the termination condition of the EMD algorithm in this paper. The particle velocity is the speed at which the particle finds the optimal position. If it is too large, it will skip the optimal solution, and if it is too small, it will easily get trapped in a local optimum.

[0049] Step 2: Perform EMD decomposition on the signal and calculate the Shannon entropy value for each IMF signal. The minimum Shannon entropy value is taken as the local minimum. Determine whether the modal component is within the range of resting heart rate and respiratory rate. If not, remove the extreme point. The Shannon entropy is calculated as follows:

[0050] Step 3: Update the particle's velocity and position.

[0051] The velocity of each particle is updated as shown in the following formula: The position of each particle is updated as shown in the following formula:

[0052] In the formula, For the first The particle in the first The speed of each iteration; For the first The particle in the first The position of the next iteration; Inertial weights; and For learning factors; and It is a random number; For the first The best historical position of each particle; To obtain the global optimal position S330, the IMF signal within the respiratory and heartbeat frequency bands is selected from the final empirical mode decomposition signal and subjected to Hilbert-Huang transform to obtain the respiratory and heartbeat signals.

[0053] Step 4: Update Individual and Global Best Positions: Compare the current local minimum of each particle with its historical minimum. If the current value is better, update the particle's historical best position. Similarly, update the global best position.

[0054] Step 5: Check if the maximum number of iterations has been reached or the change in the global minimum value is less than the preset threshold. If the conditions are met, stop the iteration, output the optimal standard deviation parameter, and use the parameter for the final EMD decomposition.

[0055] The Hilbert Huang Transform (HHT) is a time-frequency analysis method for analyzing nonlinear and non-stationary signals. HHT consists of two parts: Empirical Mode Decomposition (EMD) and Hilbert Spectral Analysis (HSA). While EMD can separate the signal, it suffers from mode aliasing, and the number of Intrinsic Mode Functions (IMFs) and the stopping criterion significantly impact the accuracy of the results. The number of IMFs in EMD depends on the termination condition. Applying the Hilbert Transform (HT) to each IMF yields its analytical solution. The analytical solution includes a real part and an imaginary part, corresponding to the instantaneous frequency and amplitude of the IMF, ultimately resulting in the time-frequency energy distribution (Hilbert spectrum) of the signal.

[0056] The Hilbert-Huang transform includes empirical mode decomposition and Hilbert spectral analysis. Empirical mode decomposition aims to find all local maxima and local minima of the original signal. Spline interpolation is then used to connect all local maxima and local minima to obtain the upper envelope. and lower envelope Calculate the mean of the upper and lower envelopes, then subtract the mean from the original signal to obtain a new signal. .

[0057] determination Whether a convergence criterion is met can be considered as follows: It is an IMF: the difference between the number of zeros and the number of extreme points is at most 1; mean sequence =0. Otherwise, reconstruct the envelope to extract the components. Then... As an IMF, the residual signal is obtained by subtracting the IMF from the original signal.

[0058] The EMD process is repeated on the residual signal until the current IMF standard deviation is less than a threshold or the set number of iterations is reached. The decision to stop EMD decomposition is based on the threshold parameter obtained from PSO optimization. The output consists of several IMFs and a residual signal. Hilbert transforms are performed on the IMFs within the respiratory and heart rate frequency bands respectively. For each IMF... Performing a Hilbert transform yields the analytic signal: where, yes The Hilbert transform is used to calculate the instantaneous amplitude and phase of the analytic signal, thereby determining the instantaneous frequency and ultimately obtaining the Hilbert spectrum.

[0059] This invention proposes an innovative framework for respiratory and heartbeat extraction that combines Empirical Wavelet Decomposition (EWT) with Optimized Shannon Entropy and Hilbert-Huang Transform (HHT). EWT, through adaptive spectrum segmentation, precisely divides radar signals into respiratory, heartbeat, and noise frequency bands, effectively filtering out high-frequency interference and respiratory harmonics, providing a high signal-to-noise ratio signal basis for subsequent processing. Based on this, Shannon entropy is introduced as a quantification metric, and combined with Particle Swarm Optimization (PSO) algorithm, the modal screening threshold of HHT is dynamically adjusted to ensure the complete preservation of effective physiological components and the automatic removal of noise components.

[0060] This method, based on particle swarm optimization (PSO) algorithm and Shannon entropy SE-optimized Hilbert-Huang transform (PSO-HHT), employs a multi-stage processing scheme known as the EPH method. Its purpose is to further separate the intrinsic mode functions (IMFs) contained in the signal and, through frequency band selection, extract the IMFs corresponding to respiratory and heart rate signals, thereby obtaining the instantaneous frequency and instantaneous energy. This collaborative mechanism not only overcomes the inherent shortcomings of traditional methods in parameter adaptation and noise suppression but also significantly improves frequency resolution.

[0061] In S400, anomaly correction is performed on the separated instantaneous frequency anomaly data based on a local weighted regression algorithm; S410, Set a sliding window, calculate the mean and average of the data within the window for each sliding window time point, and determine the instantaneous frequency abnormal data based on the mean and average. S420, Based on the local weighted regression algorithm, the instantaneous frequency anomaly data is processed by local weighted regression to obtain a weighted least squares regression model. The instantaneous frequency anomaly data is replaced using the weighted least squares regression model to obtain smooth respiratory and heartbeat signals.

[0062] This embodiment employs the Locally Weighted Regression (LWR) method to process the separated instantaneous frequency data. The LWR algorithm is superior when dealing with relatively concentrated outlier datasets. Outliers often occur during signal separation, and these outliers can affect the accuracy of the instantaneous spectrum.

[0063] Traditional global regression methods often struggle to handle local fluctuations and outliers, while locally weighted regression, by applying weighted regression to each data point and its neighbors, can effectively smooth and correct the data. The core idea of ​​the algorithm is to assign higher weights to points near the point to be predicted, with the weights decreasing as distance increases.

[0064] Outlier detection is achieved by setting a sliding window with 5 points. For each time point, the mean and standard deviation of the data within the window are calculated. Based on statistical methods, outliers are then identified. All points within the specified range are considered outliers. Specifically, this paper sets... This is because in a normal distribution, approximately 95% of the data points will fall into the range of 0. If the range is cleared, then other points can be considered outliers. Local weighted regression is then applied to these outliers.

[0065] In locally weighted regression, the goal is to fit a linear model. Here, a and b are model parameters that change based on the target point t. To estimate parameters a and b, the objective function of weighted least squares is to minimize the weighted sum of squared errors. Weighting function The goal is to ensure that points closer to the target point t have a higher influence on the fitting process; therefore, a Gaussian kernel function is chosen as the weighting function. To solve for parameters a and b, the problem can be transformed into matrix form. For outliers, weights are adjusted to 0. The resulting regression model... This is used to estimate and replace outliers. Finally, heart rate or respiratory rate is calculated by windowing the smoothed instantaneous frequency.

[0066] Locally weighted regression (LWR) can adaptively adjust based on the local characteristics of data points to more accurately capture the true trend of the signal. LWR calculates the weights of each data point and its neighbors by selecting appropriate kernel functions and bandwidth parameters, and uses these weights to construct a weighted least squares regression model, thereby reducing the impact of outliers. Finally, the smoothed instantaneous frequency data is used to calculate heart rate more accurately. This method not only improves the robustness of data processing but also significantly improves the accuracy of heart rate calculation, providing a solid foundation for subsequent physiological parameter analysis.

[0067] To verify the effectiveness of this invention, the experimental data processed using the method of this invention were compared with the data detected and exported by the monitor. The results of respiration and heart rate were as follows: Figure 3 As shown, the estimated value is the result obtained through the process of this invention, while the actual value is the data exported from the Meilingsi 12-lead medical monitor. Both the estimated and actual values ​​represent the changes in respiratory and heart rate per second over time. It can be seen that the estimation accuracy of respiratory rate is very high with no significant error. The estimation error of heart rate is relatively larger than that of respiratory rate, but it still reflects the actual situation fairly well.

[0068] Combining EWT and HHT to process radar signals, such as Figure 4 The left figure shows the Hilbert spectrum of the heartbeat signal obtained by using PSO-HHT alone and filtering. Figure 4The right figure shows the Hilbert spectrum of the heartbeat signal obtained by combining EWT with PSO-HHT in this experiment. A comparison of the two figures reveals that the processing flow proposed in this invention, i.e., after using EWT for denoising, results in more concentrated energy, smoother frequency changes, and increased continuity with fewer frequency jumps. It is evident that the proposed algorithm performs better, displaying instantaneous frequency changes more stably and clearly, reducing the impact of frequency jumps and noise, and exhibiting better frequency stability.

[0069] Depend on Figure 4 Although the separated signal spectrum is relatively stable, it still contains a small number of outliers. Frequency outliers are detected by setting frequency and energy thresholds, and finally, outliers are supplemented through local weighted regression and smoothed. Figure 5 The figures show the Pearson correlation coefficients of heart rate values ​​obtained from three different algorithms. Figures (a) and (b) show the Pearson correlation coefficients obtained from direct decomposition using the VMD and EWT algorithms, respectively. Figure 5 (c) shows the Pearson correlation coefficient of the subject's heart rate under the algorithm of this invention. It can be seen that the correlation coefficient is 0.98, indicating that the true value and the estimated value are highly correlated. Compared with the correlation coefficients of 0.94 and 0.93 of VMD and EWT algorithms, respectively, the processing scheme of the algorithm proposed in this invention has better performance and can estimate the heart rate more accurately.

[0070] This invention also included experiments comparing the performance of different algorithms in separating respiratory and heart rate signals when detecting a target at a radar distance of 0.75 m. These included EMD, EEMD, EWT, VMD, and the proposed EWT combined with the PSO-HHT algorithm (the method of this invention). The HHT algorithm calculates the instantaneous frequency and uses a sliding window combined with outlier processing to calculate the changes in heart rate and respiration. In this experiment, EMD, EEMD, EWT, and VMD constructed the changes in heart rate and respiration values ​​over time using a sliding window. By substituting the estimated and true values ​​obtained from the algorithms into the formulas for MSE (mean squared error) and MAPE (mean absolute percentage error), a comparison of the experimental algorithm performance can be obtained. Specific experimental results are shown in Table 1.

[0071] Table 1 shows a comparison of the effects of different algorithms.

[0072] As shown in Table 1, the proposed EWT combined with PSO-HHT algorithm significantly outperforms other conventional algorithms in both MSE and MAPE. Specifically, the EWT combined with PSO-HHT algorithm achieves an MSE of 4.79 and a MAPE of 3.11%, which are significantly lower than traditional algorithms such as EMD and VMD. This result demonstrates that by introducing particle swarm optimization (PSO) and combining it with EWT to remove noise and harmonics, the separation accuracy of radar signals can be effectively improved, and the error can be significantly reduced, thus enabling more accurate detection of respiratory and heart rate signals. Therefore, the EWT combined with PSO-HHT algorithm has significant advantages in radar applications for detecting physiological signals.

[0073] Specifically, after the anomaly correction step in S400, which involves separating the instantaneous frequency anomaly data based on a local weighted regression algorithm, the present invention includes: A CNN classification model was trained using the frequency-energy spectrum of respiratory signals after anomaly correction, data labeled with obstructive sleep apnea (OSA), and a weighted cross-entropy loss function. The trained CNN classification model is used to detect whether a target object suffers from obstructive sleep apnea.

[0074] The Hilbert spectrum is essentially a time-frequency matrix, where each element represents the energy intensity at the corresponding time and frequency. The time-frequency matrix has two dimensions: a time axis and a frequency axis. The time axis is 20 minutes (1200 seconds). The frequency axis focuses on the respiratory frequency range of 0.01-0.5 Hz with a step size of 0.01 Hz, resulting in 50 frequency points. Before building the model, time windows are segmented, with each window being 30 seconds and a sliding step size of 15 seconds. This effectively preserves the characteristics of obstructive sleep apnea (OSA) within each window. To ensure the training efficiency and stability of the model, the frequency energy within each window is Z-score standardized per frequency point to eliminate differences among different subjects, enabling the model to learn more effectively.

[0075] A CNN classification model based on Hilbert's spectrum determines the probability of each window and finally calculates the total number of OSA windows to determine the OSA severity level. The core structure of the CNN model is as follows: Figure 6As shown. The input layer is a slice of the time-frequency matrix, with a shape of (30, 50, 1), representing a two-dimensional matrix of time, frequency, and energy intensity. Two convolutional layers are used: the first layer has 16 filters, a 3×3 kernel, and ReLU activation; this layer captures local time-frequency energy abrupt changes, such as the sudden drop in low-frequency energy during apnea. The second layer has 32 filters, a 3×3 kernel, and ReLU activation, extracting higher-order features. Each convolutional layer is followed by 2×2 max pooling to reduce the dimension to (15, 25, 32). A fully connected layer, flattened, has 128 neurons, ReLU activation, and Dropout=0.5. The output layer has one neuron; since the model determines whether an event is OSA, the sigmoid activation function is chosen, and the output window represents the probability of whether it is an OSA event. This paper sets a threshold of 0.5; a probability greater than 0.5 is considered a positive window.

[0076] Since the healthy breathing window is generally much longer than the apnea window during the entire detection process, meaning there are far more negative samples than positive samples, the choice of loss function during training should increase the loss contribution of the apnea window, forcing the model to pay more attention to the minority class. This invention uses weighted cross-entropy as the loss function, as follows:

[0077] in, It is the first The true label of a sample, that is, the true category. If the window is positive, meaning an apnea event, Record it as 1, otherwise, A value of 0 indicates that the window shows normal breathing and the window is negative. and These are the weights of the positive and negative samples, respectively. The model optimizer used is Adam, with a learning rate set to 0.001.

[0078] Because the same apnea event may be covered by multiple windows, leading to duplicate classifications as apnea events, this model records the start time of each positive window. If two positive windows overlap, they are merged into one event. The AHI index is calculated based on the number of windows per hour to ultimately determine whether the subject has OSA.

[0079] Five-fold cross-validation was used to evaluate model performance, ensuring that the ratio of apnea windows to healthy windows in each fold was consistent with the overall data. During training, class weights were independently calculated for each fold to address class imbalance, and the loss was continuously compared to the validation set. Training was terminated if there was no improvement after five consecutive folds. The final model performance was the average result of the five folds. The model was trained for 100 epochs, and early stopping was used to terminate training early upon model convergence. Figure 7 The flowchart illustrates the training and detection process for a CNN classification model. This invention utilizes the frequency-energy spectrum of anomaly-corrected respiratory signals and data labeled with obstructive sleep apnea (OSA) as input to a CNN classification model. After training the model using the TensorFlow 2.10 framework, the accuracy of this model changes as follows: Figure 8 As shown, the model gradually converges after 35 epochs, achieving an accuracy of 94.3% on the validation set. (See details...) Figure 8 As shown.

[0080] Meanwhile, the present invention provides a respiratory and heartbeat detection system based on frequency-modulated continuous wave radar, comprising: The echo signal acquisition module is used to acquire the echo signal reflected back from the target object detected by the continuous wave signal sent by the FMCW millimeter wave radar; The preprocessing module is communicatively connected to the echo signal acquisition module and is used to preprocess the echo signal. A separation module, communicatively connected to the preprocessing module, is used to separate respiratory and heartbeat signals from the preprocessed echo signal based on the EPH multi-level collaborative signal processing algorithm; and... An anomaly correction module, which is communicatively connected to the separation module, is used to correct anomalies in the separated instantaneous frequency anomaly data based on a local weighted regression algorithm.

[0081] The innovative achievements of this invention are as follows: 1. To address the issues of static clutter and noise interference in radar signals, preprocessing of radar echo signals is performed. First, Fourier transform is used in the fast time dimension to determine the target range. Second, to address the insufficient low-speed target detection capability of moving target indication algorithms when removing static clutter, an algorithm is proposed that incorporates local mean calculation and dynamic adjustment mechanisms in the slow time dimension phase cancellation, thus more effectively eliminating static interference. Third, phase expansion is performed to address the phase jump problem of the echo signal. Finally, phase difference denoising is used to extract phase change information.

[0082] 2. To address the problems of modal aliasing and nonlinear signal separation, this paper proposes an EPH multi-level collaborative signal processing method. First, by comparing and analyzing the characteristics and shortcomings of mainstream signal separation algorithms, the EWT algorithm is used to adaptively eliminate noise and respiratory harmonic interference. Based on this, a PSO-HHT method is constructed, using Particle Swarm Optimization (PSO) and Shannon entropy to optimize the parameters of HHT, effectively solving the modal aliasing problem of HHT and significantly improving the detection accuracy of respiration and heartbeat.

[0083] 3. To address the issues of insufficient instantaneous frequency resolution and stability, an outlier correction strategy combining instantaneous frequency and local weighted regression is designed by detecting abnormal fluctuations in instantaneous frequency, thereby achieving continuous high-resolution analysis of respiratory rate and heart rate.

[0084] 4. To address the problem of obstructive sleep apnea detection, a joint OSA detection method based on instantaneous energy spectrum and neural network is proposed. The respiratory signal frequency-energy spectrum and OSA-labeled data calculated by the EPH algorithm framework are input into a lightweight CNN network model. Experiments show that this method can maintain an OSA recognition accuracy of 94.3% even under static and minimally invasive conditions.

[0085] Specifically, this embodiment corresponds one-to-one with the above method embodiments. The functions of each module have been described in detail in the corresponding method embodiments, so they will not be repeated here.

[0086] Based on the same inventive concept, embodiments of this application also provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements all or part of the method steps of the above method.

[0087] The present invention can implement all or part of the processes in the above methods, or it can be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content contained in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0088] Based on the same inventive concept, embodiments of this application also provide an electronic device, including a memory and a processor. The memory stores a computer program that runs on the processor. When the processor executes the computer program, it implements all or part of the method steps described above.

[0089] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the computer device, connecting all parts of the computer device through various interfaces and lines.

[0090] Memory can be used to store computer programs and / or modules. The processor performs various functions of the computer device by running or executing the computer programs and / or modules stored in the memory, and by accessing data stored in the memory. Memory can primarily include a program storage area and a data storage area. The program storage area can store the operating system and at least one application program required for a function (e.g., sound playback, image playback, etc.); the data storage area can store data created based on the use of the mobile phone (e.g., audio data, video data, etc.). Furthermore, memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disks, RAM, plug-in hard disks, SmartMedia Cards (SMC), Secure Digital (SD) cards, Flash Cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0091] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, servers, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0092] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), servers, and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0093] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0094] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0095] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for detecting respiratory and heartbeat based on frequency-modulated continuous wave radar, characterized in that, Includes the following steps: The FMCW millimeter-wave radar transmits continuous wave signals to detect target objects and obtains the reflected echo signals. The echo signal is preprocessed; The preprocessed echo signal is separated into respiratory and heartbeat signals based on the EPH multi-level collaborative signal processing algorithm. Anomaly correction is performed on the isolated instantaneous frequency anomaly data based on the local weighted regression algorithm.

2. The respiratory and heartbeat detection method based on frequency-modulated continuous wave radar as described in claim 1, characterized in that, The preprocessing step for the echo signal includes: The echo signal and the corresponding original transmitted signal are mixed to obtain an intermediate frequency signal; Perform a Fast Fourier Transform (FFT) on the intermediate frequency signal to determine the distance dimension. The static clutter interference of the range-dimensional FFT is eliminated based on the slow-time dimension phase mean cancellation algorithm. The range-dimensional FFT used to eliminate static clutter interference is sequentially subjected to phase extraction, phase expansion, and phase difference.

3. The respiratory and heartbeat detection method based on frequency-modulated continuous wave radar as described in claim 2, characterized in that, The step of eliminating static clutter interference based on the slow-time-dimensional phase mean cancellation algorithm to eliminate the range-dimensional FFT includes: The local phase mean of the echo signal is calculated using a sliding window method; At each frame of the sliding window, phase cancellation is performed using the local phase mean to obtain the local phase difference signal; The echo signal, after eliminating static clutter interference, is reconstructed using the local phase difference signal.

4. The respiratory and heartbeat detection method based on frequency-modulated continuous wave radar as described in claim 1, characterized in that, The step of separating respiratory and heartbeat signals from the preprocessed echo signal using the EPH multi-level collaborative signal processing algorithm includes: Harmonics and noise in the preprocessed echo signal are removed based on the empirical wavelet transform algorithm. The final empirical mode decomposition termination condition of the signal after removing harmonics and noise is found based on the particle swarm optimization algorithm and the Shannon entropy algorithm. In the final empirical mode decomposition signal, the IMF signal within the respiratory and heartbeat frequency bands is selected and subjected to Hilbert-Huang transform to obtain the respiratory and heartbeat signals.

5. The respiratory and heartbeat detection method based on frequency-modulated continuous wave radar as described in claim 4, characterized in that, The step of finding the final empirical mode decomposition termination condition of the signal after removing harmonics and noise based on particle swarm optimization algorithm and Shannon entropy algorithm includes: The particle swarm optimization algorithm is used to initialize the signal after removing harmonics and noise, generating an initial particle swarm. Each particle includes position and velocity information, where the position information corresponds to the standard deviation parameter. The initial particle swarm is decomposed into modes based on the empirical mode decomposition algorithm, and the Shannon entropy value of each IMF signal obtained by decomposition is calculated based on the Shannon entropy algorithm. Update the position and velocity information of each particle, compare the current minimum Shannon entropy value of each particle with the historical minimum Shannon entropy value, and if the current minimum Shannon entropy value is less than the historical minimum Shannon entropy value, update the historical minimum Shannon entropy value of each particle to the current minimum Shannon entropy value, update the historical best position of each particle, and update the best particle corresponding to the global best position of the particle swarm. The optimal particle is calculated iteratively until the condition is met to terminate the iteration. The standard deviation parameter corresponding to the optimal particle position is obtained, and the signal is finally decomposed using the standard deviation parameter.

6. The respiratory and heartbeat detection method based on frequency-modulated continuous wave radar as described in claim 5, characterized in that, The standard deviation parameter is shown in the following formula: In the formula, SD is the standard deviation threshold parameter; It is the first The IMF signal of the next iteration; The number of sampling points for the signal; For the first One particle.

7. The respiratory and heartbeat detection method based on frequency-modulated continuous wave radar as described in claim 5, characterized in that, The velocity of each particle is updated as shown in the following formula: The position of each particle is updated as shown in the following formula: In the formula, For the first The particle in the first The speed of each iteration; For the first The particle in the first The position of the next iteration; Inertial weights; and For learning factors; and It is a random number; For the first The best historical position of each particle; This is the optimal position globally.

8. The respiratory and heartbeat detection method based on frequency-modulated continuous wave radar as described in claim 1, characterized in that, The step of correcting anomalies in the separated instantaneous frequency anomaly data based on the local weighted regression algorithm includes: Set up a sliding window, calculate the mean and average of the data within the window at each time point of the sliding window, and determine the instantaneous frequency abnormal data based on the mean and average. The instantaneous frequency anomaly data is processed by a locally weighted regression algorithm to obtain a weighted least squares regression model. The instantaneous frequency anomaly data is then replaced using the weighted least squares regression model to obtain smooth respiratory and heartbeat signals.

9. The respiratory and heartbeat detection method based on frequency-modulated continuous wave radar as described in claim 1, characterized in that, After the step of correcting the anomalies in the separated instantaneous frequency anomaly data based on the local weighted regression algorithm, the following steps are included: A CNN classification model was trained using the frequency-energy spectrum of respiratory signals after anomaly correction, data labeled with obstructive sleep apnea (OSA), and a weighted cross-entropy loss function. The trained CNN classification model is used to detect whether a target object suffers from obstructive sleep apnea.

10. A respiratory and heartbeat detection system based on frequency-modulated continuous wave radar, characterized in that, include: The echo signal acquisition module is used to acquire the echo signal reflected back from the target object detected by the continuous wave signal sent by the FMCW millimeter wave radar; The preprocessing module is communicatively connected to the echo signal acquisition module and is used to preprocess the echo signal. The separation module is communicatively connected to the preprocessing module and is used to separate the respiratory signal and heartbeat signal from the preprocessed echo signal based on the EPH multi-level collaborative signal processing algorithm. as well as, An anomaly correction module, which is communicatively connected to the separation module, is used to correct anomalies in the separated instantaneous frequency anomaly data based on a local weighted regression algorithm.

Images