Vital sign detection method based on WOA-SVMD

Abstract

The invention discloses a WOA-SVMD-based vital sign detection method, which comprises the following steps of: firstly, carrying out self-adaptive optimization on balance parameters of successive variational mode decomposition (SVMD) by using a whale optimization algorithm (WOA), and then decomposing signals into a series of intrinsic mode functions by using the successive variational mode decomposition based on the optimal balance parameters; the intrinsic mode functions are screened twice through the energy ratio and the correlation coefficient, and finally the intrinsic mode function with the maximum correlation coefficient is determined to reconstruct the heartbeat signal. According to the WOA-SVMD-based vital sign detection method, accurate estimation of the heart rate can be achieved, the separated heartbeat signals are closer to actual signals, and the problem that noise interference and breathing harmonic waves easily interfere with the heartbeat signals in radar vital sign detection can be effectively solved.

Description

Technical Field

[0001] The present invention relates to a vital sign detection method, in particular to a vital sign detection method based on WOA-SVMD, and belongs to the technical field of radar. Background Art

[0002] The detection of vital signs such as breathing and heartbeat is of great significance in disaster relief, home monitoring, medical treatment and other scenarios. Currently, there are two main detection methods: contact and non-contact. Contact detection requires direct contact between the device and the human body, which cannot adapt to scenarios such as disaster rescue life detection and large-area burns on patients during medical treatment. Non-contact detection can perform wireless detection of target objects. Common methods include millimeter wave radar, infrared, video, ultrasound, electromagnetic waves, etc. Among them, the frequency modulated continuous wave (FMCW) millimeter wave radar has the advantage of high resolution in vital sign detection and has broad application prospects in non-contact vital sign detection.

[0003] In the detection of life signals based on FMCW radar, on the one hand, the displacement caused by breathing and heartbeat is mostly in the millimeter level, which is easily submerged by background clutter and noise. On the other hand, the displacement caused by the heartbeat signal is much smaller than that caused by the breathing signal, and the harmonics of the heartbeat signal and the breathing signal may have similar frequencies and amplitudes, which makes it difficult for conventional frequency domain filtering methods to distinguish between the heartbeat signal and the breathing harmonics. Therefore, how to separate the heartbeat signal is a key issue in current industry research. Methods based on modal decomposition can be used to separate heartbeat signals, but the commonly used methods based on modal decomposition still have defects, such as: the decomposition based on empirical mode decomposition is prone to mode overlap and endpoint effects; the decomposition based on variational mode decomposition must set the appropriate number of decomposition layers and balance parameters to ensure the accuracy of decomposition; the decomposition algorithm based on adaptive noise complete empirical mode decomposition has problems such as low computational efficiency and large amount of calculation; the decomposition based on wavelet transform depends on the wavelet basis function, and it is difficult to achieve accurate division of frequency bands; successive variational mode decomposition is an efficient signal decomposition algorithm based on variational mode decomposition, and does not require the number of decomposition layers to be known in advance, which can better capture the different frequency components in the signal, but the balance parameters of successive variational mode decomposition still need to be optimized. In addition, there are also problems such as noise interference and respiratory harmonics that easily interfere with heartbeat signals when radar detects vital signs. Summary of the invention

[0004] In view of the problems existing in the above-mentioned prior art, the present invention provides a vital sign detection method based on WOA-SVMD, which can solve the problem of successive variational modal decomposition balance parameter optimization in life sign detection technology, the problem of noise interference when radar detects vital signs, and the problem of respiratory harmonics easily interfering with heartbeat signals, thereby realizing high-precision respiratory rate and heart rate detection.

[0005] To achieve the above purpose, the WOA-SVMD-based vital sign detection method specifically includes the following steps:

[0006] Step 1, separation of respiratory signal and clutter heartbeat signal: pre-process the raw data collected by the radar to obtain the respiratory rate and clutter heartbeat signal;

[0007] Step 2, decomposing the clutter heartbeat signal: Based on the optimal parameter α obtained by WOA, the clutter heartbeat signal is decomposed by SVMD to obtain a series of intrinsic mode functions IMF;

[0008] Step 3, obtain the initial reconstructed signal: calculate the energy E(i) of each IMF and the energy E(i) of each IMF in the heartbeat signal frequency range of 0.8 to 2 Hz h (i), according to E(i) and E h (i) Calculate the energy ratio and select the IMF whose energy ratio reaches the set energy ratio threshold ε for reconstruction to ensure the accuracy of decomposition, and obtain the initial reconstructed signal S h1 ;

[0009] Step 4, obtain the secondary reconstruction signal: the initial reconstruction signal S h1 Perform secondary decomposition, calculate the correlation coefficient between each IMF after decomposition and the initial reconstructed signal, reconstruct the IMF whose correlation coefficient is greater than the set correlation coefficient threshold σ, and obtain the secondary reconstructed signal S h2 ;

[0010] Step 5, heart rate solution: The secondary reconstructed signal is envelope demodulated and then fast Fourier transformed to obtain the heart rate f H , then the heart rate HR is: HR = f H ×60.

[0011] Furthermore, Step 1 is as follows:

[0012] Step 1-1, introduce MTI for processing, and perform subtraction operation on the signals with the same amplitude and phase;

[0013] Step 1-2, after determining the distance unit where the person is located according to the radar, the extracted phase is unwrapped;

[0014] Step 1-3, perform phase differentiation on the unwrapped signal, and use IIR to separate the differentiated signal to obtain the respiratory frequency and clutter heartbeat signal.

[0015] Furthermore, the specific steps of Step 2 are as follows:

[0016] Step 2-1, decompose the clutter heartbeat signal using SVMD;

[0017] Step 2-2, use the WOA algorithm to determine the optimal balance parameter α of SVMD.

[0018] Furthermore, Step 2-1 is as follows:

[0019] SVMD decomposes the clutter heartbeat signal into multiple modal functions, updates the modal function by minimizing the difference between the signal and the modal function in each iteration, and extracts a frequency component from the signal until all frequency components in the signal can be captured;

[0020] Assume that the time domain signal f(t) is decomposed into L-order mode functions u L (t) and the residual part f r (t):

[0021] f(t)=u L (t)+f r (t)

[0022] The residual signal f r (t) includes the first L-1 order mode functions and the remaining unprocessed signal f u (t), add the first constraint J1 to ensure the accuracy of the decomposition:

[0023]

[0024] Where: w L is the center frequency of the Lth-order mode function, is the quadratic penalty factor, represents the convolution calculation; δ(t) is the impulse function;

[0025] Define the second constraint J2 to make the residual signal f r (t) has the least spectral overlap with the energy of the L-order mode function:

[0026]

[0027] Among them, β L (t) is the impulse response of the filter selected by the second constraint J2, and its frequency response The expression is:

[0028]

[0029] Define the third constraint J3 to distinguish the L-order mode function from the L-1-order mode function, and use the frequency response of the filter The expression is:

[0030]

[0031] Then the expression of the third constraint J3 is:

[0032]

[0033] Where: β i (t) is the impulse response of the filter selected by the third constraint J3;

[0034] The fourth constraint f(t) is defined so that the L-order mode function, the L-1 mode function and the residual signal can be added to reconstruct the signal. The expression of the fourth constraint f(t) is:

[0035]

[0036] When the L-1 order modal function is known, solving the L order modal function is transformed into a constrained minimization problem:

[0037]

[0038] Where: α is the balance parameter used to balance constraints J1, J2, and J3.

[0039] Furthermore, Step 2-2 is as follows:

[0040] ① Input the preprocessed signal, initialize the population size, maximum number of iterations, upper and lower bounds of the search space of WOA, and determine the value range of the balance parameter α;

[0041] ② Initialize the parameters set in the whale optimization algorithm, calculate the fitness value of each IMF decomposed by SVMD, and select the individual whale position corresponding to the minimum fitness value as the optimal solution;

[0042] ③ Iterate within the set iteration number range. When the random number p in WOA is less than 0.5, go directly to the next step. Otherwise, perform spiral update on the individual position of the whale. The spiral update formula is as follows:

[0043]

[0044] Where: p is the probability of predation mechanism, which is a random number between [0,1]; A is the coefficient variable; X * (t) is the position vector of the optimal solution; b is a constant used to define the shape of the logarithmic spiral; D *is the distance between the current search individual and the current optimal solution; l is a random number with a uniform distribution in the range [-1,1];

[0045] ④ Calculate the value of the coefficient variable A. If |A| < 1, perform encirclement and contraction. The mathematical formula for encirclement and contraction is the same as the spiral update formula above. If |A| > 1, perform random search. The random search formula is as follows:

[0046] D r =|C·X r (t)-X(t)|

[0047] X(t+1)=X r (t)-AD r

[0048] Where: X r (t) is the position vector of a randomly selected individual whale; D r is the distance between the currently selected individual whale and its prey; C is the coefficient variable;

[0049] ⑤ Determine whether the iteration is completed. When the number of iterations T<T max When , the fitness of each whale is calculated. When the fitness function value is better than the previous optimal solution, the optimal solution is updated. Otherwise, the iteration is terminated and the optimal parameter α is output.

[0050] Furthermore, the fitness calculation based on MIC in step ② is as follows:

[0051] The definition of mutual information I(X;Y) is as follows:

[0052]

[0053] Where: p(·) is the joint density probability;

[0054] When solving the joint density probability p(·), the two variables X and Y are divided into a grid diagram with i rows and j columns, and the mutual information value between X and Y is calculated. The maximum mutual information value obtained is normalized. The calculation formula of MIC is:

[0055]

[0056] Where: B is the upper limit of the grid graph;

[0057] Fitness f based on the maximum mutual information coefficient fitness The calculation formula is as follows:

[0058]

[0059] Where: x(t) is the input signal; α is the balance parameter in SVMD.

[0060] Furthermore, in Step 4, the correlation coefficient between each IMF after decomposition and the initial reconstructed signal is calculated, and the calculation formula is as follows:

[0061]

[0062] Where r represents the correlation coefficient between the IMF and the original signal; ui and f represent the decomposed modal components and the original signal, respectively; E(·) and D(·) represent the expectation and variance of the random variable, respectively.

[0063] Furthermore, the energy ratio threshold ε is set to 0.9 in Step 3; and the correlation coefficient threshold σ is set to 0.3 in Step 4.

[0064] Compared with the existing technology, the WOA-SVMD-based vital sign detection method first uses the whale optimization algorithm (WOA) to adaptively optimize the balance parameters of the successive variational mode decomposition (SVMD), and then uses the successive variational mode decomposition based on the optimal balance parameters to decompose the signal into a series of intrinsic mode functions, and then screens the intrinsic mode functions twice by energy ratio and correlation coefficient, and finally determines the intrinsic mode function with the largest correlation coefficient to reconstruct the heartbeat signal. Experiments show that compared with the existing empirical mode decomposition (EMD), complete empirical mode decomposition with adaptive noise (CEEMDAN), and bandpass filtering algorithms, the root mean square error of the WOA-SVMD-based vital sign detection method is only 1.25, which can achieve accurate estimation of the heart rate, and the separated heartbeat signal is closer to the actual signal, which can effectively solve the noise interference and respiratory harmonics in radar vital sign detection. The problem of interference with the heartbeat signal is easy. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 is a flow chart of the present invention;

[0066] Figure 2 is a diagram showing the bandpass filtering results in the data preprocessing process of the present invention;

[0067] Figure 3 is a flow chart of WOA-SVMD parameter optimization of the present invention;

[0068] Figure 4 It is a time domain diagram and a frequency domain diagram of the initial decomposition of the heartbeat signal of the present invention;

[0069] Figure 5 It is a frequency domain comparison diagram of the heartbeat signal before and after the initial decomposition of the signal of the present invention;

[0070] Figure 6 It is a time domain diagram and a frequency domain diagram of the secondary decomposition of the heartbeat signal of the present invention;

[0071] Figure 7It is the spectrum diagram corresponding to the heartbeat signal before and after decomposition of the present invention;

[0072] Figure 8 This is a performance comparison chart of heartbeat signal reconstruction using four different methods. DETAILED DESCRIPTION

[0073] The present invention is further described below by taking the actual collection of raw radar signal data as an example in conjunction with the accompanying drawings.

[0074] like Figure 1 As shown, the WOA-SVMD-based vital sign detection method specifically includes the following steps:

[0075] Step 1, separate the breathing signal and the clutter heartbeat signal, perform distance solution, phase extraction, phase unwrapping, phase difference and bandpass filtering for preprocessing, and obtain the breathing rate and clutter heartbeat signal. The Texas Instruments millimeter wave radar IWR6843 is used as the signal acquisition device to collect the original radar signal data. Its operating frequency band is 60-77GHz, with 3 transmitting channels and 4 receiving channels. The starting frequency of IWR6843 is 60GHz, the bandwidth is 3.25GHz, the number of fast time sampling points N is 200, the number of slow time sampling points is 1024, the fast time sampling rate is 4MHz, the slow time sampling rate is 20Hz, and the sampling interval is 50ms. The person under test sits about 0.79m in front of the radar, and the chest of the person under test is consistent with the radar height. At the same time, the person under test wears an ECG electrocardiogram module, and the measured breathing signal and heartbeat signal are used as reference values ​​for comparing the algorithm performance.

[0076] In order to eliminate background clutter, the Moving Target Indicator (MTI) algorithm is introduced for processing. The principle of MTI filtering is to use pulse signals to repeatedly measure the target and subtract the signals with the same amplitude and phase. Since the phase of background clutter is chaotic, the pulses in adjacent repetition cycles of the same range unit are subtracted to eliminate background clutter.

[0077] The Fast Fourier Transform (FFT) is used to determine the distance unit where the person is located. After the radar determines the distance unit where the person is located, the extracted phase is unwrapped in order to filter the environmental noise in the phase signal and eliminate the phase drift.

[0078] The unwrapped signal is phase-differentiated to enhance the heartbeat signal and suppress the respiratory signal and its harmonics. The differential signal is separated using an infinite impulse response filter (IIR) to preliminarily obtain the respiratory frequency and the cluttered heartbeat signal. The sampling frequency, starting frequency, cutoff frequency and order of the respiratory signal filter are 20Hz, 0.1Hz, 0.5Hz and 4, and the sampling frequency, starting frequency, cutoff frequency and order of the heartbeat signal filter are 20Hz, 0.8Hz, 2Hz and 8. Figure 2 As shown in the figure, since the heartbeat signal is mixed with a lot of clutter and respiratory harmonics, the successive variational mode decomposition (SVMD) is needed to further process the heartbeat signal after bandpass filtering. Since the balance parameters in SVMD will affect the accuracy of decomposition, it is necessary to optimize it through the Whale Optimization Algorithm (WOA) algorithm, and then decompose and reconstruct the heartbeat signal after determining the optimal balance parameters.

[0079] Step 2, decompose the clutter heartbeat signal. Based on the optimal parameter α obtained by the Whale Optimization Algorithm (WOA), the clutter heartbeat signal is decomposed by successive variational mode decomposition (SVMD) to obtain a series of intrinsic mode functions (IMF). Specifically:

[0080] Step 2-1, use SVMD to decompose the clutter heartbeat signal. SVMD decomposes the clutter heartbeat signal into multiple modal functions. In each iteration, the modal function is updated by minimizing the difference between the signal and the modal function, and a frequency component is extracted from the signal until all frequency components in the signal can be captured. Assume that the time domain signal f(t) is decomposed into L-order modes u L (t) and the residual part f r (t):

[0081] f(t)=u L (t)+f r (t)

[0082] The residual signal f r (t) includes the first L-1 modes and the remaining unprocessed signal f u (t), in order to ensure the accuracy of decomposition, it is necessary to add constraint J1:

[0083]

[0084] Where: w L is the center frequency of the Lth-order mode function, is the quadratic penalty factor, represents the convolution calculation; δ(t) is the impulse function;

[0085] In order to make the residual signal f r The energy of (t) is minimized, and a new constraint J2 needs to be defined to minimize the spectrum overlap between the residual signal and the energy of the L-order mode:

[0086]

[0087] Among them, β L (t) is the impulse response of the selected filter, the purpose is to better constrain the signal, and its frequency response The expression is:

[0088]

[0089] The first two constraints J1 and J2 cannot distinguish the L-order mode from the L-1-order mode. A new constraint needs to be defined to separate them. Refer to the generation step of the second constraint. The frequency response of the filter used The expression is:

[0090]

[0091] Therefore, the expression of constraint J3 is established as:

[0092]

[0093] Where: β i (t) is the impulse response of the filter selected for the third constraint.

[0094] After establishing the three constraints, in order to reconstruct the signal by adding the L-order mode function, L-1 mode and the residual signal, it is necessary to add the last constraint, which is expressed as:

[0095]

[0096] When the L-1 order mode is known, solving the L order mode can be transformed into a constrained minimization problem:

[0097]

[0098] Where: α is a parameter used to balance the three constraints J1, J2, and J3.

[0099] Step 2-2, use the WOA algorithm to determine the optimal parameter α of SVMD. Figure 3 As shown, the preprocessed signal is input, the population size of WOA, the maximum number of iterations, the upper and lower bounds of the search space are initialized, and the value range of the balance parameter α is determined.

[0100] Initialize the parameters set in the whale optimization algorithm, calculate the fitness value of each IMF decomposed by SVMD, and select the individual whale position corresponding to the minimum fitness value as the optimal solution;

[0101] Iterate within the set range of iteration times. When the random number p in WOA is less than 0.5, go directly to the next step. Otherwise, perform spiral update on the individual position of the whale. The spiral update formula is as follows:

[0102]

[0103] Where: p is the probability of predation mechanism, which is a random number between [0,1]; A is the coefficient variable; X * (t) is the position vector of the optimal solution; b is a constant used to define the shape of the logarithmic spiral; D * is the distance between the current search individual and the current optimal solution; l is a random number with a uniform distribution in the range [-1,1].

[0104] By changing the probability p, parameter α and coefficient variable A, the whale will update its position in different ways and move towards a better position.

[0105] Calculate the value of the coefficient variable A. If |A| < 1, perform encirclement contraction. Since the encirclement contraction mechanism and spiral update mechanism of WOA are performed simultaneously, the execution probability of the two is equal, both 0.5, so the encirclement contraction mathematical formula is the same as the spiral update formula above. If |A| > 1, do not perform encirclement contraction but perform random search. The random search formula is as follows:

[0106] D r =|C·X r (txt)

[0107] X(t+1)=X r (t)-AD r

[0108] Where: X r (t) is the position vector of a randomly selected individual whale, D r is the distance between the currently selected individual whale and its prey, and C is the coefficient variable.

[0109] Determine whether the iteration is completed. When the number of iterations T<T maxWhen , the fitness of each whale is calculated. When the fitness function value is better than the previous optimal solution, the optimal solution is updated. Otherwise, the iteration is terminated and the optimal parameter α is output.

[0110] Step 2-3, fitness calculation using the maximum mutual information coefficient:

[0111] The maximum mutual information coefficient (MIC) is used to measure the degree of association between two variables X and Y. The definition of mutual information I(X; Y) is as follows:

[0112]

[0113] Where: p(·) is the joint density probability. The stronger the correlation between X and Y, the larger the value of I(X; Y).

[0114] When solving the joint density probability p(·), the two variables X and Y are divided into a grid diagram with i rows and j columns, and the mutual information value between X and Y is calculated. The maximum mutual information value obtained is normalized. The calculation formula of MIC is:

[0115]

[0116] Where: B is the upper limit of the grid graph.

[0117] In order to prevent modal aliasing, the MIC between the decomposed IMFs should be minimized and the MIC between the IMFs and the original signal should be maximized. Therefore, the fitness f based on the maximum mutual information coefficient is fitness The calculation formula is as follows:

[0118]

[0119] Where: x(t) is the input signal; α is the balance parameter in SVMD.

[0120] Step 3, obtain the initial reconstructed signal. Calculate the energy E(i) of each IMF component and the energy E(i) of each IMF in the heartbeat signal frequency range of 0.8 to 2 Hz. h (i) Since the energy of the IMF before and after decomposition is almost equal to the energy of the original signal, the component whose energy ratio reaches a certain threshold is selected for signal reconstruction, that is:

[0121]

[0122] Where: ε is the energy ratio threshold for determining whether it is a heartbeat signal. The higher the ε value, the higher the proportion of the heartbeat component in the IMF. When the threshold ε is 0.9, the accuracy of the decomposition can be ensured.

[0123] Select IMF with ε greater than or equal to 0.9 for reconstruction and obtain the initial reconstructed signal S h1 , which can be expressed as:

[0124]

[0125] Through the method of Step 2, we can determine that the optimal parameter α is 1855, and the number of decomposition layers after iteration is 4. Therefore, Figure 4 As shown in Figure 1, a total of 4 IMF components are obtained after SVMD decomposition. It can be seen that during the initial decomposition, only the center frequencies of IMF2 and IMF3 fall within the heartbeat interval [0.8, 2]. Among IMF2, IMF3, and IMF4, only the energy ratio of IMF2 (0.9813) is greater than the threshold, so it is used as the initial reconstructed signal. Figure 5 As shown, compared with the original signal, the first reconstructed signal can ensure that the detected heartbeat signal is within the range of [0.8, 2], and filter out a large amount of noise, and the interference to the heartbeat signal is partially suppressed.

[0126] Step 4, obtain the secondary reconstructed signal. In order to ensure the correlation between the reconstructed signal and the original signal, the initial reconstructed signal S h1 Perform secondary decomposition and calculate the relationship between each IMF and S after decomposition h1 Correlation coefficient:

[0127]

[0128] Where ui and f represent the decomposed modal component and the original signal, respectively; E(·) and D(·) represent the expectation and variance of the random variable, respectively; r represents the correlation coefficient between IMF and the original signal, and its relationship with the correlation is shown in Table 1.

[0129] Table 1 Relationship between correlation coefficient and correlation

[0130]

[0131] The correlation coefficient threshold σ is set to 0.3, the correlation coefficient between the IMF and the original signal is calculated, and the IMF with the largest correlation coefficient is selected as the target IMF to obtain the secondary reconstructed signal S h2 :

[0132]

[0133] For the initial reconstructed signal S h1 After secondary decomposition, the time domain and frequency domain diagrams of the decomposed IMF are as follows Figure 6 As shown, the IMF with the largest correlation coefficient is selected for secondary reconstruction. Figure 6The correlation coefficients between the middle IMF and the initial reconstructed signal are 0.5974, 0.6519, 0.6248, and 0.3360, respectively. IMF2 with the largest correlation coefficient is selected to perform secondary reconstruction on the initial reconstructed signal.

[0134] Step 5: Solve the heart rate signal. Perform envelope demodulation on the secondary reconstructed signal and perform fast Fourier transform on the secondary reconstructed signal to obtain the heart rate frequency f H , then the heart rate HR is: HR = f H ×60(times / min).

[0135] The corresponding spectrum diagrams before and after the heartbeat signal decomposition are as follows Figure 7 As shown in the figure, the center frequencies of the original signal, the first decomposition signal, and the second decomposition signal are 1.00Hz, 1.11Hz, and 1.21Hz respectively. The heart rate collected by the ECG module is 73 beats / minute, which is converted to a center frequency of about 1.22Hz. The accuracy of the second decomposition signal is improved by 17% and 8% respectively compared with the original signal and the first decomposition signal, and the measured heart rate is closer to the true value.

[0136] In order to verify the effectiveness of the vital sign detection method based on WOA-SVMD, four methods, namely band-pass filtering, empirical mode decomposition (EMD), complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and the vital sign detection method based on WOA-SVMD, are used to decompose and reconstruct the heartbeat signal of the same signal, and the heartbeat frequency collected by the ECG device is used as the reference signal. The heartbeat signals processed by the four methods are shown in Figure 2. Figure 8 As shown. Figure 8 It can be seen that the bandpass filter method contains a lot of clutter and cannot separate the respiratory harmonics from the heartbeat signal; the noise added by CEEMDAN will produce redundant components and affect the separation of the heartbeat signal; the signal waveform decomposed by EMD is prone to modal aliasing problems; and the heartbeat signal extracted by the vital signs detection method based on WOA-SVMD is relatively pure, the signal curve is smooth, and the noise in the frequency domain of the heartbeat signal is small, indicating that the vital signs detection method based on WOA-SVMD has better noise suppression performance than the other three methods, which is conducive to the detection of weak signals.

[0137] The vital signs signals of 7 volunteers were selected and processed by the above four methods respectively. The results are shown in Table 2 below, where experimental numbers 1 to 7 are the heart rate data of 7 volunteers, and the last is the RMSE result. It can be seen that the maximum RMSE of bandpass filtering is 12.58, while that of EMD and CEEMDAN are 3.59 and 5.69 respectively. The RMSE of this vital sign detection method based on WOA-SVMD is only 1.25, which is much smaller than other comparison methods, and a clean heartbeat signal can be extracted.

[0138] Table 2 Root mean square error of four methods

[0139]

[0140]

[0141] Aiming at the problem that heartbeat signals are easily affected by respiratory harmonics and other noises, this life feature detection algorithm based on WOA-SVMD selects the maximum mutual information coefficient as the fitness function, decomposes the original FMCW radar signal through WOA-SVMD, and then introduces the energy ratio and correlation coefficient as two indicators to reconstruct the heartbeat signal twice, and finally selects the appropriate heartbeat signal. Experimental results show that this life feature detection algorithm based on WOA-SVMD can better suppress the respiratory harmonics and noise in the heartbeat signal, and improve the detection accuracy of the heart rate.

Claims

1. A vital sign detection method based on WOA-SVMD, characterized in that: The specific steps include: Step 1, separation of respiratory signal and clutter heartbeat signal: pre-process the raw data collected by the radar to obtain the respiratory rate and clutter heartbeat signal; Step 2, decomposing the clutter heartbeat signal: Based on the optimal parameter α obtained by WOA, the clutter heartbeat signal is decomposed by SVMD to obtain a series of intrinsic mode functions IMF; Step 3, obtain the initial reconstructed signal: calculate the energy E(i) of each IMF and the energy E(i) of each IMF in the heartbeat signal frequency range of 0.8 to 2 Hz h (i), according to E(i) and E h (i) Calculate the energy ratio and select the IMF whose energy ratio reaches the set energy ratio threshold ε for reconstruction to ensure the accuracy of decomposition, and obtain the initial reconstructed signal S h1 ; Step 4, obtain the secondary reconstruction signal: the initial reconstruction signal S h1 Perform secondary decomposition, calculate the correlation coefficient between each IMF after decomposition and the initial reconstructed signal, reconstruct the IMF whose correlation coefficient is greater than the set correlation coefficient threshold σ, and obtain the secondary reconstructed signal S h2 ; Step 5, heart rate solution: The secondary reconstructed signal is envelope demodulated and then fast Fourier transformed to obtain the heart rate f H , then the heart rate HR is: HR = f H ×60.

2. The WOA-SVMD-based vital sign detection method according to claim 1, characterized in that: Step 1 is as follows: Step 1-1, introduce MTI for processing, and perform subtraction operation on the signals with the same amplitude and phase; Step 1-2, after determining the distance unit where the person is located according to the radar, the extracted phase is unwrapped; Step 1-3, perform phase differentiation on the unwrapped signal, and use IIR to separate the differentiated signal to obtain the respiratory frequency and clutter heartbeat signal.

3. The method for detecting vital signs based on WOA-SVMD according to claim 1, characterized in that: Step 2 The specific steps are as follows: Step 2-1, decompose the clutter heartbeat signal using SVMD; Step 2-2, use the WOA algorithm to determine the optimal balance parameter α of SVMD.

4. The WOA-SVMD-based vital sign detection method according to claim 3, characterized in that: Step 2-1 is as follows: SVMD decomposes the clutter heartbeat signal into multiple modal functions, updates the modal function by minimizing the difference between the signal and the modal function in each iteration, and extracts a frequency component from the signal until all frequency components in the signal can be captured; Assume that the time domain signal f(t) is decomposed into L-order mode functions u L (t) and the residual part f r (t): f(t)=u L (t)+f r (t) The residual signal f r (t) includes the first L-1 order mode functions and the remaining unprocessed signal f u (t), add the first constraint J1 to ensure the accuracy of the decomposition: Where: w L is the center frequency of the Lth-order mode function; is the quadratic penalty factor; represents the convolution calculation; δ(t) is the impulse function; Define the second constraint J2 to make the residual signal f r (t) has the least spectral overlap with the energy of the L-order mode function: Among them, β L (t) is the impulse response of the filter selected by the second constraint J2, and its frequency response The expression is: Define the third constraint J3 to distinguish the L-order mode function from the L-1-order mode function, and use the frequency response of the filter The expression is: Then the expression of the third constraint J3 is: Where: β i (t) is the impulse response of the filter selected by the third constraint J3; The fourth constraint f(t) is defined so that the L-order mode function, the L-1 mode function and the residual signal can be added to reconstruct the signal. The expression of the fourth constraint f(t) is: When the L-1 order modal function is known, solving the L order modal function is transformed into a constrained minimization problem: Where: α is the balance parameter used to balance constraints J1, J2, and J3.

5. The WOA-SVMD-based vital sign detection method according to claim 3, characterized in that: Step 2-2 is as follows: ① Input the preprocessed signal, initialize the population size, maximum number of iterations, upper and lower bounds of the search space of WOA, and determine the value range of the balance parameter α; ② Initialize the parameters set in the whale optimization algorithm, calculate the fitness value of each IMF decomposed by SVMD, and select the individual whale position corresponding to the minimum fitness value as the optimal solution; ③ Iterate within the set iteration number range. When the random number p in WOA is less than 0.5, go directly to the next step. Otherwise, perform spiral update on the individual position of the whale. The spiral update formula is as follows: Where: p is the probability of predation mechanism, which is a random number between [0,1]; A is the coefficient variable; X * (t) is the position vector of the optimal solution; b is a constant used to define the shape of the logarithmic spiral; D * is the distance between the current search individual and the current optimal solution; l is a random number with a uniform distribution in the range [-1,1]; ④ Calculate the value of the coefficient variable A. If |A| < 1, perform encirclement and contraction. The mathematical formula for encirclement and contraction is the same as the spiral update formula above. If |A| > 1, perform random search. The random search formula is as follows: D r =|C·X r (t)-X(t) X(t+1)=X r (t)-AD r Where: X r (t) is the position vector of a randomly selected individual whale; D r is the distance between the currently selected individual whale and its prey; C is the coefficient variable; ⑤ Determine whether the iteration is completed. When the number of iterations T<T max When , the fitness of each whale is calculated. When the fitness function value is better than the previous optimal solution, the optimal solution is updated. Otherwise, the iteration is terminated and the optimal balance parameter α is output.

6. The WOA-SVMD-based vital sign detection method according to claim 5, characterized in that: The fitness calculation based on MIC in step ② is as follows: The definition of mutual information I(X;Y) is as follows: Where: p(·) is the joint density probability; When solving the joint density probability p(·), the two variables X and Y are divided into a grid diagram with i rows and j columns, and the mutual information value between X and Y is calculated. The maximum mutual information value obtained is normalized. The calculation formula of MIC is: Where: B is the upper limit of the grid graph; Fitness f based on the maximum mutual information coefficient fitness The calculation formula is as follows: Where: x(t) is the input signal; α is the balance parameter in SVMD.

7. The method for detecting vital signs based on WOA-SVMD according to claim 1, characterized in that: In Step 4, the correlation coefficient between each IMF after decomposition and the initial reconstructed signal is calculated using the following formula: Where r represents the correlation coefficient between the IMF and the original signal; ui and f represent the decomposed modal components and the original signal, respectively; E(·) and D(·) represent the expectation and variance of the random variable, respectively.

8. The WOA-SVMD-based vital sign detection method according to claim 1, characterized in that: In Step 3, the energy ratio threshold ε is set to 0.9; in Step 4, the correlation coefficient threshold σ is set to 0.3.

Images