Millimeter-wave radar vital sign detection method for suppressing body motion effects

Abstract

The present invention discloses a millimeter-wave radar vital sign detection method for suppressing the effect of body motion, comprising the following steps: extracting phase data containing vital sign information from the radar signal, performing multi-channel averaging processing, and then extracting the heartbeat signal using a high-pass filter; detecting body motion using a dual-energy threshold detection method on the heartbeat signal, determining the body motion pattern, and selecting the initial distribution parameters of the maximum expectation algorithm; filtering out motion artifacts using the maximum expectation algorithm; and obtaining the heart rate of the target being measured using a fast Fourier transform spectrum analysis method. The non-contact heart rate detection framework of MCA processing + APSEM method + FFT spectrum analysis method proposed in the present invention suppresses strong interference caused by body motion during the detection process, and is conducive to obtaining accurate heart rate measurement results in a detection environment where the subject is experiencing body motion.

Description

Technical Field

[0001] The present invention relates to a millimeter wave radar vital sign detection method for suppressing body motion effects, and belongs to the field of non-contact vital sign detection in radar signal processing. Background Art

[0002] Non-contact vital sign detection is a key research area in radar signal processing, with promising applications in numerous fields, including healthcare monitoring, disaster relief, and military security. Radar-based vital sign detection technology works as follows: A radar emits electromagnetic waves of a specific waveform, which strike the moving chest wall, generating an echo. This echo, modulated by the chest wall, contains information about chest displacement. By processing the intermediate frequency signal after radar mixing, the human respiratory and heart rates are determined.

[0003] At the same time, the integration of millimeter-wave technology has further advanced development in this area. Due to millimeter-wave radar's 24 / 7 operation characteristics, as well as its advantages such as narrow beamwidth, high resolution, and compact size for portability, its application in contactless vital sign detection has become a hot research area in recent years. Respiration and heartbeat cause localized vibrations on the body surface with amplitudes of approximately 5 mm and 200-500 microns, respectively. Millimeter-wave radar's ability to detect small-amplitude vibrations allows for accurate vital sign detection. Furthermore, the radar signals emitted by millimeter-wave radar can penetrate non-metallic materials such as clothing and debris, detecting micro-movements generated by breathing and heartbeat.

[0004] However, during the actual detection process, the amplitude of the subject's body movement is generally much larger than the amplitude of chest displacement caused by breathing and heartbeat activities, thereby forming an extremely strong interference signal, greatly reducing the accuracy of vital signs detection. Summary of the Invention

[0005] The present invention proposes a millimeter-wave radar vital sign detection method for suppressing the body motion effect, the purpose of which is to suppress the interference of body motion on vital sign detection, thereby ensuring extremely high-accuracy vital sign estimation.

[0006] The technical solutions of the present invention are as follows:

[0007] A millimeter wave radar vital sign detection method for suppressing body motion effects, comprising the following steps:

[0008] Step S1, radar signal preprocessing: extracting phase data containing vital sign information from the radar signal, then performing multi-channel averaging on the phase data, and then extracting the heartbeat signal using a high-pass filter;

[0009] Step S2, selecting parameters: using a dual energy threshold detection method to detect body motion on the heartbeat signal and determine the body motion pattern; adaptively selecting initial distribution parameters of the maximum expectation algorithm based on the body motion pattern to which the heartbeat signal belongs;

[0010] Step S3: Filter out motion artifacts using the maximum expectation algorithm: Model the heartbeat signal disturbed by body motion as an observation sequence, and define the latent variable state sequence corresponding to the observation sequence; introduce a scalar weight for the variance of each sample in the observation sequence; preset the prior distribution of the weight to a Gamma distribution, and the initial distribution parameters selected in step S2 are the initial Gamma distribution parameters; based on the established model, use the maximum expectation algorithm to solve the latent variables and output the heartbeat signal after filtering out body motion artifacts;

[0011] Step S4, using the fast Fourier transform spectrum analysis method to obtain the heart rate of the target being measured: perform fast Fourier transform on the heartbeat signal after filtering out body motion artifacts to obtain the spectrum of the heartbeat signal; find the maximum peak in the spectrum within the human heartbeat frequency range, and the position corresponding to the maximum peak is the heartbeat frequency of the target being measured.

[0012] As a further improvement to the millimeter-wave radar vital sign detection method for suppressing body motion effects, the specific method of step S1 is as follows: performing a distance-dimensional fast Fourier transform on the radar intermediate frequency signal, then selecting the distance unit where the target is located, extracting the phase data and performing phase unwrapping; performing multi-channel averaging processing on the phase data of the four channels corresponding to the same target to enhance the vital sign characteristics; and then applying a high-pass filter with a cutoff frequency of 0.8 Hz to the processed phase data to extract the heartbeat signal.

[0013] As a further improvement of the millimeter wave radar vital sign detection method for suppressing body motion effects, step S2 includes:

[0014] Step 21: Detect body movement using a dual energy threshold detection method:

[0015] Assumptions represents the amplitude of the i-th sample point in the heartbeat signal. The sum of the energies of the M sample points near the i-th sample point is defined as the energy statistic corresponding to the i-th sample point, which is expressed as:

[0016] ;

[0017] Combine the energy statistics of the i-th sample point with the two energy thresholds 、 For comparison, ; Determine whether body movement has occurred by comparing the results:

[0018] ;

[0019] In the above formula, When it is equal to -1, it means that there is no body movement in this signal. When it is equal to 0, no judgment is made. When it is equal to 1, it means that body movement occurs;

[0020] The above threshold and It is determined by the statistical mean and standard deviation of the signal energy when no body movement occurred in the previous moment:

[0021] ;

[0022] In the above formula, and Respectively represent Two threshold values corresponding to the heartbeat signal sampling points, Before The average energy value corresponding to the heartbeat signal sampling points is Before The standard deviation of the energy values corresponding to the heartbeat signal sampling points; and are two weight coefficients.

[0023] As a further improvement of the millimeter wave radar vital sign detection method for suppressing body motion effects, step S2 further includes:

[0024] Step 22: Body motion pattern identification and selection of adaptive parameters:

[0025] Calculate the percentage of body movement time to the total heartbeat signal duration ; Calculate the maximum energy value in the heartbeat signal and its corresponding second energy threshold Ratio ;

[0026] according to and , which divides body movements into five modes:

[0027] (1) When When , the body movement mode is Model 1: high-intensity body movement;

[0028] (2) When When the body movement mode is Mode2: moderate intensity body movement;

[0029] (3) When When the body movement mode is Mode3: low-intensity body movement;

[0030] (4) When When the body movement mode is Mode 4: slight body movement;

[0031] (5) When When , the body movement mode is Mode 5: almost no body movement occurs;

[0032] Then determine the corresponding initial Gamma distribution parameters according to the body movement pattern :

[0033] .

[0034] As a further improvement of the millimeter wave radar vital sign detection method for suppressing body motion effects, step S3 includes:

[0035] Step 31: Model building:

[0036] Modeling heartbeat signals as observation sequences , Represents the length of the heartbeat signal, element is the observed value; the latent variable state sequence corresponding to the observation sequence is expressed as ,element is the state variable; the relationship between the observation sequence and the latent variable state sequence is expressed as:

[0037] ;

[0038] ;

[0039] In the above formula, is the prediction matrix, is the state transition matrix, represents the observation noise, Represents the noise in the state transition process;

[0040] and is an additive Gaussian white noise with a mean of 0, 、 Respectively and The covariance matrix of is expressed as:

[0041] ;

[0042] Then for each sample in the observation sequence The variance introduces a scalar weight weighted; The prior distribution of is preset to Gamma distribution;

[0043] Based on the above model, the prior relationship between observations, state variables and weights can be expressed as:

[0044] ;

[0045] ;

[0046] ;

[0047] In the above formula, 、 are the shape parameter and scale parameter of the Gamma distribution respectively.

[0048] As a further improvement to the millimeter wave radar vital sign detection method for suppressing body motion effects, step S3 further includes:

[0049] Step 32: Use the maximum expectation algorithm to solve the latent variables:

[0050] E-Step: Calculate the maximum likelihood value and solve the posterior estimate of the hidden variable; when When E-Step starts, iterates to :

[0051] ;

[0052] ;

[0053] ;

[0054] Weight The shape parameter of the Gamma distribution, Weight The scale parameter of the Gamma distribution;

[0055] M-Step: Model parameter update:

[0056] ;

[0057] ;

[0058] ;

[0059] ;

[0060] In the maximum expectation algorithm, the updated model parameters obtained in the M-Step are used to calculate the E-Step in the next iteration;

[0061] The iterative stopping condition of the maximum expectation algorithm is:

[0062] ;

[0063] In the above formula, Represent the covariance matrix of the observation noise and the covariance matrix of the state noise updated by the current M-Step, is the threshold corresponding to the stopping condition; They represent the covariance matrix of the observation noise and the covariance matrix of the state noise updated by the M-Step in the previous iteration respectively; when the stopping condition is met, the maximum expectation algorithm iteration stops, and the heartbeat signal sequence after filtering out the body motion artifacts can be obtained. , expressed as:

[0064] .

[0065] Compared with the prior art, the present invention has the following beneficial effects:

[0066] 1. The present invention comprehensively utilizes multi-channel received data and performs multi-channel averaging (MCA) processing to enhance the characteristics of vital sign signals.

[0067] 2. The present invention proposes a body motion pattern discrimination method based on a dual-energy threshold detection method, which realizes a simple classification of the subject's body motion model in non-contact vital sign detection.

[0068] 3. This paper proposes an adaptive parameter selection for expectation-maximization (APSEM) algorithm to suppress the interference of body motion effects in heartbeat signals. The APSEM method models all data samples contained in the heartbeat signal as observations and uses the EM algorithm to iteratively assign weights to the variances of these observations. Furthermore, to make this method more stable and accurate in filtering out motion artifacts, this paper proposes a body motion pattern discrimination method based on a dual-energy threshold detection method to adaptively select the initial distribution parameters of the APSEM method. Furthermore, the APSEM method utilizes iterative updates to suppress the interference caused by singular values in the heartbeat signal. Specifically, the weights assigned to singular value components decrease with each iteration, thereby suppressing these singular values. By suppressing singular values, this method eliminates the influence of body motion effects, resulting in more accurate and higher-quality heartbeat signals and enhancing the effectiveness and robustness of the EM algorithm in suppressing body motion interference.

[0069] The non-contact heart rate detection framework of MCA processing + APSEM method + FFT spectrum analysis method proposed in the present invention suppresses the strong interference caused by body movement during the detection process, which is conducive to obtaining accurate heart rate measurement results in a detection environment where the subject is in body movement. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0071] Figure 2 Flowchart for radar signal preprocessing;

[0072] Figure 3 A flow chart of adaptive parameter selection according to the present invention;

[0073] Figure 4 Flowchart of the Expectation Maximum Algorithm (EM) of the present invention;

[0074] Figure 5 Flowchart of the Fast Fourier Transform (FFT) spectrum analysis method

[0075] Figure 6 This is a time domain signal diagram verified by the measured data of the present invention;

[0076] Figure 7 Spectrum diagram verified by measured data of the present invention;

[0077] Figure 8 Heart rate curve diagram verified by measured data of the present invention. DETAILED DESCRIPTION

[0078] The technical solution of the present invention is described in detail below with reference to the accompanying drawings:

[0079] like Figure 1 , a millimeter-wave radar vital sign detection method for suppressing body motion effects, the overall processing flow is:

[0080] First, the radar signal preprocessing process extracts phase data containing vital sign information. After multi-channel averaging (MCA) of the resulting phase data from the four channels, a fourth-order Butterworth high-pass filter with a cutoff frequency of 0.8Hz is used to roughly extract the heartbeat signal. Second, a dual-energy threshold detection method is used to detect body motion and determine the motion pattern of the heartbeat signal. Subsequently, the initial distribution parameters of the expectation-maximization (EM) method are adaptively selected based on the motion pattern of the heartbeat signal. Third, after the initial distribution parameters are selected, the EM method is used to filter out body motion artifacts from the heartbeat signal to suppress the effect of body motion. Finally, the heartbeat signal, free of motion artifacts, is subjected to fast Fourier transform (FFT) spectral analysis to obtain the target's heart rate.

[0081] The specific process is as follows:

[0082] S1. Radar signal preprocessing process:

[0083] like Figure 2 As shown, after performing a Range-Fast Fourier Transform (FFT) on the radar's intermediate frequency (IF) signal, the target's range bin is selected, phase data is extracted, and phase unwrapping is performed. By leveraging multiple receiving antennas, multi-channel averaging is performed on the phase data from the four channels corresponding to the same target, enhancing vital sign features. Subsequently, a high-pass filter with a cutoff frequency of 0.8Hz is used to roughly extract the heartbeat signal from the phase data after MCA processing.

[0084] S2. Adaptive parameter selection:

[0085] like Figure 3 The double energy threshold method is used to detect body motion and determine the body motion pattern of the heartbeat signal extracted by high-pass filter. The initial distribution parameters of the EM method are adaptively selected according to the body motion pattern of the heartbeat signal.

[0086] Specifically:

[0087] S21. Dual energy threshold detection method to detect body movement.

[0088] Assumptions Represents the amplitude of the i-th sample point in the heartbeat signal. The sum of the energies of the M sample points near the i-th sample point is defined as the energy statistic corresponding to the i-th sample point, expressed as:

[0089] .

[0090] Combine the energy statistics of the i-th sample point with the two energy thresholds , ( ) to compare and determine whether body movement has occurred. In the actual signal acquisition process, if the energy statistics of the sample point is greater than and less than , no judgment is made. The judgment process is expressed as:

[0091] ;

[0092] In the above formula, When it is equal to -1, it means that there is no body movement in this signal. When it is equal to 0, no judgment is made. When it is equal to 1, it means that body movement has occurred.

[0093] Threshold value of dual energy threshold detection , It is determined by the statistical mean and standard deviation of the signal energy when there is no body movement in the previous moment, and is defined as follows:

[0094] ;

[0095] In the above formula, and Respectively represent Two threshold values corresponding to the heartbeat signal sampling points, Before The average energy value corresponding to the heartbeat signal sampling points is Before The standard deviation of the energy values corresponding to the heartbeat signal sampling points. In addition, and are two weight coefficients, which are set to 1.1 and 2.6 respectively based on experience in this embodiment.

[0096] S22. Body movement pattern discrimination and adaptive parameter selection.

[0097] After performing dual energy threshold detection on the heartbeat signal, first calculate the percentage of body movement time to the total heartbeat signal duration. ; Then, calculate the maximum energy value in the heartbeat signal and its corresponding second energy threshold Ratio Taking the above two indicators into consideration and In this embodiment, body movement is divided into five modes. The body movement modes and related evaluation criteria are as follows:

[0098] (1) When When , the body movement mode is Model 1: high-intensity body movement;

[0099] (2) When When the body movement mode is Mode2: moderate intensity body movement;

[0100] (3) When When the body movement mode is Mode3: low-intensity body movement;

[0101] (4) When When the body movement mode is Mode 4: slight body movement;

[0102] (5) When , the body movement mode is Mode 5: almost no body movement occurs.

[0103] After determining the motion mode of the current heartbeat signal, it is necessary to determine the corresponding initial Gamma distribution parameters. We conducted a large number of experiments to study the appropriate values of the initial Gamma distribution parameters corresponding to different body motion patterns, and typical values can be given as:

[0104] .

[0105] S3. EM method.

[0106] like Figure 4 After selecting the initial distribution parameters of the EM method, the heartbeat signal disturbed by body motion is first modeled as an observation sequence, and the latent variable state sequence corresponding to the observation sequence is defined. Subsequently, a scalar weight is introduced for the variance of each sample in the observation sequence for weighting. In order to ensure that the weight is always positive, its prior distribution is preset to a Gamma distribution. The initial distribution parameters selected in step S2 are the initial Gamma distribution parameters. Finally, based on the established model, the EM algorithm is used to solve the latent variables and output the heartbeat signal after filtering out body motion artifacts.

[0107] Specifically:

[0108] S31. Model establishment.

[0109] The heartbeat signal is modeled as a sequence of observations , Represents the length of the heartbeat signal. The latent variable state sequence corresponding to this observation sequence is expressed as The relationship between the observation sequence and the latent variable state sequence is expressed as:

[0110] ;

[0111] ;

[0112] In the above formula, is the prediction matrix, is the state transition matrix, represents the observation noise, Represents the noise in the state transition process.

[0113] In this embodiment, it is assumed and is an additive Gaussian white noise with a mean of 0, , Respectively and The covariance matrix of is expressed as:

[0114] .

[0115] Then, for each sample in the observation sequence, we The variance introduces a scalar weight To ensure the weight is always positive, this embodiment will The prior distribution of is preset to Gamma distribution. Based on the above model, the prior relationship between observations, state variables and weights can be simply expressed as:

[0116] ;

[0117] ;

[0118] ;

[0119] In the above formula, , are the shape parameter and scale parameter of the Gamma distribution respectively.

[0120] S32. EM algorithm solves latent variables.

[0121] In the model established in step S31, only the observations is known, and the state variable and weights All are hidden states, so solving this model becomes a process of solving hidden variables. For the observations, the posterior distribution of the state variables and weights can be estimated by maximizing the log-likelihood, and the model can be solved using the EM algorithm.

[0122] The hidden variables in the model established in this embodiment are state sequences and weights , the maximum likelihood estimate of all data should be calculated for their true posterior distribution. After the model parameters are initialized, the EM algorithm based on the model established in this embodiment solves the latent variables according to the following steps:

[0123] E-Step: Calculate the maximum likelihood and solve for the posterior estimates of the hidden variables. When E-Step starts, iterates to :

[0124] ;

[0125] ;

[0126] .

[0127] Weight The shape parameter of the Gamma distribution, Weight The scale parameter of the Gamma distribution.

[0128] M-Step: Model parameter update.

[0129] ;

[0130] ;

[0131] ;

[0132] ;

[0133] In the EM algorithm, the updated model parameters obtained in the M-Step are used to calculate the E-Step in the next iteration. The update expression of includes the deviation between the predicted value and the observed value, which is expressed in the denominator If the deviation between the predicted value and the observed value is large, the weight corresponding to the variance will decrease. When the deviation becomes infinite, the weight will be infinitely close to zero, which means that the observed value has almost no effect on the optimal estimate of the model parameters. Therefore, the EM algorithm used in this embodiment can effectively filter out singular values generated by body motion and suppress the interference of body motion effects. In addition, this embodiment provides the iterative stopping condition for the EM algorithm applied to singular value suppression. The stopping condition is expressed as:

[0134] ;

[0135] In the above formula, Represent the covariance matrix of the observation noise and the covariance matrix of the state noise updated by the current M-Step, is the threshold corresponding to the stopping condition. They represent the covariance matrix of the observation noise and the covariance matrix of the state noise updated by the M-Step in the previous iteration. When the stopping condition is met, the EM algorithm stops iterating, and the heartbeat signal sequence after filtering out the body motion artifacts can be obtained. , expressed as:

[0136] .

[0137] By executing steps S31 and S32 described above, singular values caused by body motion in the heartbeat signal are suppressed, thereby filtering out body motion artifacts from the heartbeat signal. We refer to the overall process for filtering out body motion artifacts described above as the EM method. Furthermore, we combine steps S21 and S22 with the EM method to create the adaptive parameter selection for expectation-maximization method (APSEM), which enhances the effectiveness and robustness of the EM method for suppressing body motion effects.

[0138] S4. FFT spectrum analysis method.

[0139] like Figure 5 After obtaining the heartbeat signal after filtering out body motion artifacts, perform FFT on it to obtain the spectrum of the heartbeat signal. Then, find the maximum peak of the spectrum within the range of 0.8 to 2.5 Hz (human heartbeat frequency range). The position corresponding to the maximum peak of the spectrum is the heartbeat frequency of the target being measured.

[0140] Experimental verification

[0141] To further illustrate the performance of the present invention, we used MATLAB software to process measured vital sign data containing body motion, primarily verifying the effectiveness of the proposed algorithm in suppressing the effects of body motion. During the acquisition of measured vital sign data, the subject sat still directly in front of the radar and performed small movements such as leaning forward and backward, raising their hands, and shaking their shoulders at different time intervals to simulate body motion during heart rate monitoring.

[0142] Figure 6 The results of a set of experiments using the proposed algorithm to filter out body motion artifacts in heartbeat signals collected during the experiment are compared. Figure 6As shown, the dashed line represents the original heartbeat signal without motion artifact filtering, which contains a large number of singular values caused by the subject's simulated body motion. These singular values are marked with circles in the figure. The solid line represents the heartbeat signal after motion artifact filtering using the algorithm proposed in this invention. It can be seen that the singular values caused by body motion have been effectively suppressed.

[0143] Figure 7 The spectrograms corresponding to the original heartbeat signal and the heartbeat signal after motion artifacts are filtered out are shown. Figure 7 The dotted curve in the figure represents the spectrum of the original heartbeat signal, while the solid curve represents the spectrum of the heartbeat signal after motion artifacts are filtered out using the algorithm proposed in the present invention. The average heart rate of the subjects during the experimental period was measured using a Polar H10 heart rate sensor to be 78.4 beats per minute (bpm), corresponding to a heart rate of approximately 1.31Hz. Based on the reference heart rate value provided by the heart rate sensor, we marked the true heartbeat spectrum peak with a circle on the right in the spectrum diagram, which corresponds to a heart rate of 1.32Hz. In the spectrum of the original heartbeat signal, the true heartbeat spectrum peak is submerged by the spectrum peak corresponding to 1.22Hz, resulting in an error of approximately 5.2bpm when the heart rate is measured using the spectrum analysis method. It can be seen from the spectrum of the heartbeat signal after motion artifacts are filtered out that the motion artifact filtering process effectively suppresses the false heartbeat peak corresponding to 1.22Hz in the original heartbeat signal, thereby allowing the true heartbeat spectrum peak to occupy the center position. Furthermore, compared to the reference heart rate provided by the heart rate sensor, the heart rate obtained using the proposed algorithm for motion artifact removal, using spectral analysis, had an error of only 0.8 bpm. These experimental results demonstrate the effectiveness of the proposed algorithm for motion artifact removal.

[0144] In the comparative analysis of the experimental results shown in Table 1, we compared the motion artifact filtering performance of the EM method (abbreviated as the EM method in Table 1) and the APSEM method with the default initial Gamma distribution parameters. The performance evaluation indicator is the Mean Absolute Error (MAE). In this experiment, the default initial Gamma distribution parameters are uniformly set to To ensure a fair performance comparison, the same set of heartbeat signals was subjected to both methods to remove motion artifacts, and then heart rate was estimated using a unified FFT spectrum analysis method. Table 1 shows the heart rate estimation performance (MAE) of the EM method and the APSEM method using the default initial Gamma distribution parameters for five body motion patterns.

[0145] Table 1 Heart rate detection results under five body movement modes

[0146] .

[0147] As shown in the results in Table 1, when the body motion pattern is mode 1 (high-intensity physical activity), the interference with the heartbeat signal is significant. Without motion artifact filtering, the MAE for heart rate estimation using only spectral analysis is as high as 25.97 bpm. After using the EM method with the default initial Gamma distribution parameters to filter out motion artifacts, the heart rate estimation results for this group of heartbeat signals improved by 1.26 bpm compared to the unfiltered method, but the MAE remained as high as 24.71 bpm. However, after using the APSEM method (i.e., the EM method with adaptive parameter selection) to filter out motion artifacts, the MAE for heart rate estimation for this group of signals was only 2.11 bpm. Compared to the EM method with the default initial Gamma distribution parameters, APSEM improved heart rate estimation accuracy by approximately 22.6 bpm. When the body motion model is mode 2 (moderate-intensity body motion) or mode 3 (low-intensity body motion), both the EM and APSEM methods using the default initial distribution parameters are able to effectively filter out motion artifacts and improve the accuracy of heart rate estimation. However, the APSEM method outperforms the EM method using the default initial distribution parameters. When the body motion model is mode 2, the heart rate estimation result obtained by the EM method using the default initial distribution parameters has a MAE of 4.51 bpm, while the MAE obtained by the APSEM method is 2.31 bpm, an improvement of 2.2 bpm over the EM method using the default initial distribution parameters. When the body motion model is mode 3, the heart rate estimation result obtained by the EM method using the default initial distribution parameters has a MAE of 4.11 bpm, while the MAE obtained by the APSEM method is 2.83 bpm, an improvement of 1.28 bpm over the EM method using the default initial distribution parameters.

[0148] When the body motion model is mode 4 (weak body motion), the two methods have the same performance. However, when the body motion model is mode 5 (almost no body motion), the EM method with the default initial Gamma distribution parameters suffers a loss of approximately 1 bpm after motion artifact filtering compared to the method without motion artifact filtering, while the APSEM method does not suffer any loss.

[0149] Based on the above experimental results and analysis, it can be concluded that the EM method with the default initial distribution parameters is suitable for processing heartbeat signals affected by low and medium-intensity body motion, but it cannot eliminate the interference caused by high-intensity body motion, and there will be some loss in heart rate detection in a static state. In contrast, the proposed APSEM method achieved better motion artifact filtering performance in all body motion modes processed in the experiments conducted in this embodiment, and it was particularly effective in suppressing interference caused by high-intensity body motion. The above experimental results fully demonstrate that the APSEM method is superior to the EM method with the default initial distribution parameters in its ability to suppress body motion effects. The proposed adaptive parameter selection method enhances the effectiveness and robustness of the EM method in suppressing body motion effects.

[0150] Figure 8 The heart rate change curves obtained by using the EM method with the default initial Gamma distribution parameters and the APSEM method for the same set of heartbeat signals in the presence of body motion are shown. Figure 8 middle:

[0151] Curve 1: The thin solid line curve represents the reference heart rate change curve of the subject recorded using the Polar H10 chest heart rate sensor.

[0152] Curve 2: The curve marked with an "x" indicates that no motion artifact filtering was performed and only the spectrum analysis method was used to obtain the heart rate variation curve.

[0153] Curve 3: The curve marked with a “+” represents the heart rate variation curve obtained by using the EM method with the default initial Gamma distribution parameters to filter out motion artifacts and then using the spectral analysis method.

[0154] Curve 4: The curve marked with a circle represents the heart rate variation curve obtained by using the spectrum analysis method after filtering out motion artifacts using the APSEM method.

[0155] Through analysis Figure 8 From the changing trends of the heart rate curves, it can be seen that the heart rate change curve 4 obtained using the APSEM method is most similar to the reference heart rate change curve 1, while the heart rate change curve 2 without motion artifact filtering and the heart rate change curve 3 obtained using the default initial Gamma distribution parameters have large deviations from the reference heart rate change curve 1.

[0156] A further comparison of heart rate variation curve 2, which does not undergo motion artifact filtering, and heart rate variation curve 3, which uses the default initial Gamma distribution parameters, shows that the EM method with the default initial Gamma distribution parameters has some ability to suppress motion artifacts, but the effect is not significant. Compared to the reference heart rate, the MAE of heart rate variation curve 2 without motion artifact filtering is 12.34 bpm, while the MAE of curve 3 with the default initial Gamma distribution parameters is 10.57 bpm. Compared to the case without motion artifact filtering, the EM method with the default initial Gamma distribution parameters achieves only a MAE improvement of approximately 1.8 bpm in heart rate detection.

[0157] However, heart rate curve 4 obtained using the APSEM method is significantly closer to the reference heart rate curve 1, with a corresponding MAE of 1.49 bpm. This represents an improvement of approximately 9 bpm in heart rate detection performance compared to curve 1, and demonstrates the superiority of the APSEM method in detecting heart rate in the presence of body motion, compared to the EM method with the default initial Gamma distribution parameters. These experimental results and analysis further demonstrate that the proposed APSEM method can effectively suppress interference caused by body motion, ensuring accurate heart rate estimation.

Claims

1. A millimeter wave radar vital sign detection method for suppressing body motion effects, characterized in that The steps are: Step S1, radar signal preprocessing: extracting phase data containing vital sign information from the radar signal, then performing multi-channel averaging on the phase data, and then extracting the heartbeat signal using a high-pass filter; Step S2, selecting parameters: using a dual energy threshold detection method to detect body motion on the heartbeat signal and determine the body motion pattern; Adaptively select the initial distribution parameters of the maximum expectation algorithm based on the body motion pattern to which the heartbeat signal belongs; Step S2 includes: Step 21: Detect body movement using a dual energy threshold detection method: Assumptions represents the amplitude of the i-th sample point in the heartbeat signal. The sum of the energies of the M sample points near the i-th sample point is defined as the energy statistic corresponding to the i-th sample point, which is expressed as: ； Combine the energy statistics of the i-th sample point with the two energy thresholds 、 For comparison, ; Determine whether body movement has occurred by comparing the results: ； In the above formula, When it is equal to -1, it means that there is no body movement in this signal. When it is equal to 0, no judgment is made. When it is equal to 1, it means that body movement occurs; The above threshold and It is determined by the statistical mean and standard deviation of the signal energy when no body movement occurred in the previous moment: ； In the above formula, and Respectively represent Two threshold values corresponding to the heartbeat signal sampling points, Before The average energy value corresponding to the heartbeat signal sampling points is Before The standard deviation of the energy values corresponding to the heartbeat signal sampling points; and are two weight coefficients; Step S3: Filter out motion artifacts using the maximum expectation algorithm: Model the heartbeat signal disturbed by body motion as an observation sequence, and define the latent variable state sequence corresponding to the observation sequence; introduce a scalar weight for the variance of each sample in the observation sequence; preset the prior distribution of the weight to a Gamma distribution, and the initial distribution parameters selected in step S2 are the initial Gamma distribution parameters; based on the established model, use the maximum expectation algorithm to solve the latent variables and output the heartbeat signal after filtering out body motion artifacts; Step S4, using the fast Fourier transform spectrum analysis method to obtain the heart rate of the target being measured: perform fast Fourier transform on the heartbeat signal after filtering out body motion artifacts to obtain the spectrum of the heartbeat signal; find the maximum peak in the spectrum within the human heartbeat frequency range, and the position corresponding to the maximum peak is the heartbeat frequency of the target being measured.

2. The millimeter-wave radar vital sign detection method for suppressing body motion effects according to claim 1, wherein: The specific method of step S1 is as follows: performing a range-dimensional fast Fourier transform on the radar intermediate frequency signal, then selecting the distance unit where the target is located, extracting the phase data and performing phase unwrapping; performing multi-channel averaging on the phase data of the four channels corresponding to the same target to enhance the vital sign characteristics; Then, a high-pass filter with a cutoff frequency of 0.8 Hz is applied to the processed phase data to extract the heartbeat signal.

3. The millimeter wave radar vital sign detection method for suppressing body motion effects according to claim 1, characterized in that: Step S2 further includes: Step 22: Body motion pattern identification and selection of adaptive parameters: Calculate the percentage of body movement time to the total heartbeat signal duration ; Calculate the maximum energy value in the heartbeat signal and its corresponding second energy threshold Ratio ; according to and , which divides body movements into five modes: (1) When When , the body movement mode is Model 1: high-intensity body movement; (2) When When the body movement mode is Mode2: moderate intensity body movement; (3) When When the body movement mode is Mode3: low-intensity body movement; (4) When When the body movement mode is Mode 4: slight body movement; (5) When When , the body movement mode is Mode 5: almost no body movement occurs; Then determine the corresponding initial Gamma distribution parameters according to the body movement pattern : 。 4. The millimeter wave radar vital sign detection method for suppressing body motion effects according to claim 1, wherein: Step S3 includes: Step 31: Model building: Modeling heartbeat signals as observation sequences , Represents the length of the heartbeat signal, element is the observed value; the latent variable state sequence corresponding to the observation sequence is expressed as ,element is the state variable; the relationship between the observation sequence and the latent variable state sequence is expressed as: ； ； In the above formula, is the prediction matrix, is the state transition matrix, represents the observation noise, Represents the noise in the state transition process; and is an additive Gaussian white noise with a mean of 0, 、 Respectively and The covariance matrix of is expressed as: ； Then for each sample in the observation sequence The variance introduces a scalar weight weighted; The prior distribution of is preset to Gamma distribution; Based on the above model, the prior relationship between observations, state variables and weights can be expressed as: ； ； ； In the above formula, 、 are the shape parameter and scale parameter of the Gamma distribution respectively.

5. The millimeter wave radar vital sign detection method for suppressing body motion effects according to claim 4, characterized in that: Step S3 further includes: Step 32: Use the maximum expectation algorithm to solve the latent variables: E-Step: Calculate the maximum likelihood value and solve the posterior estimate of the hidden variable; when When E-Step starts, iterates to : ； ； ； Weight The shape parameter of the Gamma distribution, Weight The scale parameter of the Gamma distribution; M-Step: Model parameter update: ； ； ； ； In the maximum expectation algorithm, the updated model parameters obtained in the M-Step are used to calculate the E-Step in the next iteration; The iterative stopping condition of the maximum expectation algorithm is: ； In the above formula, Represent the covariance matrix of the observation noise and the covariance matrix of the state noise updated by the current M-Step, is the threshold corresponding to the stopping condition; They represent the covariance matrix of the observation noise and the covariance matrix of the state noise updated by the M-Step in the previous iteration respectively; when the stopping condition is met, the maximum expectation algorithm iteration stops, and the heartbeat signal sequence after filtering out the body motion artifacts can be obtained. , expressed as: 。

Images