Non-contact multi-target heart rate variability detection method based on adaptive variational mode decomposition

Abstract

The present application relates to a kind of non-contact multi-target heart rate variability (HRV) detection method based on adaptive variational mode decomposition (VMD).The method first uses radar technology to detect the human target in the measured scene to determine the accurate distance window of each target.Then, the signal phase is extracted from the distance window by the differential and cross-multiplication algorithm, and then the thoracic displacement signal of each person to be tested is obtained.The present application optimizes the VMD parameters using the improved GOA algorithm, proposes a new adaptive function HIE (combining permutation entropy, mutual information and energy loss rate) as the objective function of GOA, and realizes the complete adaptive VMD decomposition of the thoracic displacement signal of each person to be tested.Through this decomposition, the heartbeat signal and the breathing signal are extracted, so that the heart rate variability characteristics of each person to be tested can be accurately detected.The present method provides a high-precision, non-contact HRV detection method, which is suitable for multi-target scene.

Description

Technical Field

[0001] The present invention belongs to the technical fields of millimeter wave radar sensors, heart rate variability detection, signal processing, etc., and specifically relates to a non-contact multi-target heart rate variability detection method based on adaptive variational mode decomposition. Background Art

[0002] Heart Rate Variability (HRV) refers to the variation in the difference in the inter-beat interval (IBI) cycles of consecutive heartbeats. It is an important physiological indicator that reflects the ability of the autonomic nervous system to regulate heart rhythm. As a non-invasive physiological parameter, HRV is widely used to assess overall heart health and the ability of the autonomic nervous system (ANS) responsible for regulating heart activity. Traditional HRV detection methods usually rely on electrocardiograms (ECGs) or heart rate monitoring devices. These devices have problems such as inconvenient placement and uncomfortable wearing. They are not suitable for special groups such as infants, patients with skin burns, and patients with sleep breathing disorders, and have significant limitations.

[0003] In recent years, the rapid development of non-contact HRV detection technologies has provided a new approach to addressing these issues. These technologies can acquire physiological signals without direct contact with the subject, not only avoiding user discomfort but also being applicable to specific populations. Methods such as those based on voice signals, thermal imaging, RGB images, and lasers have significant drawbacks, including susceptibility to environmental factors such as lighting, privacy concerns, and the inability to measure over extended periods.

[0004] Radio frequency (RF)-based methods can effectively address these issues because RF signals are unaffected by environmental factors like light and temperature and can penetrate non-metallic media. They extract respiratory and heartbeat information by transmitting electromagnetic waves and receiving reflected signals from the human chest, which vary depending on the tiny movements caused by breathing and heartbeat.

[0005] Patent publication number CN116236188A proposes a heart rate determination method. This method uses differential calculations and similarity determination to determine multiple predicted heart rates. However, HRV estimation in this method requires accurate timing of each heartbeat, which is much more complex than simply measuring heart rate. Factors such as breathing, trunk movement, and environmental noise can interfere with heartbeat signals, making accurate HRV estimation extremely challenging. Therefore, most existing heart rate detection methods cannot meet the high-precision requirements of HRV estimation.

[0006] Patent publication number CN118000698A proposes a heart rate variability monitoring device and method based on millimeter wave radar. It includes a signal preprocessing module and a heart rate variability algorithm processing module. The signal preprocessing is completed by eliminating static clutter from the signal, separating the random movement of the human body, and removing the information segments damaged by movement. In the heart rate variability algorithm processing module, the chest phase composite signal is decomposed, and the parameters are selected through the sparrow search algorithm, and finally the heartbeat signal is extracted. However, the high-order harmonic components of breathing will inevitably affect the heartbeat signal. At the same time, the high-order harmonic components of breathing and the heartbeat signal have similar waveforms, which may directly lead to the wrong selection of the heartbeat signal, which will lead to a decrease in the accuracy of HRV detection.

[0007] The patent with publication number CN114732390A proposes a non-contact heart rate variability monitoring method based on frequency modulated continuous wave radar. The location of the heart of the monitored object is extracted through FFT; the motion information of the heart of the monitored object is extracted using phase correlation, and the acceleration signal is calculated; the obtained acceleration signal is smoothed, and the position of the segmentation point between each heartbeat is estimated by the peak detection method; an acceleration template signal of a single heartbeat is generated and the acceleration signal is precisely segmented to obtain the heartbeat interval of each heartbeat, and the heart rate variability index is calculated. However, the use of an acceleration template has obvious limitations. A single template cannot handle a variety of heartbeat waveforms, and the disadvantages are obvious under long-term monitoring. In addition, this method is only applicable to a single target. In the case of multiple targets, the motion signal of the heart extracted by this method is even weaker, making it difficult to perform correct HRV estimation.

[0008] In summary, traditional HRV detection methods, such as electrocardiograms and heart rate monitors, have problems such as inconvenient placement and discomfort, and are not suitable for special groups such as infants, patients with skin burns, and patients with sleep breathing disorders. Some non-contact HRV detection methods, such as those based on voice signals, thermal imaging, RGB images, and lasers, are easily affected by environmental factors such as light, can leak privacy, and cannot measure for a long time. Existing HRV detection methods based on millimeter-wave radar still have difficulties in signal extraction and interference in multi-target scenarios. Summary of the Invention

[0009] In response to the above problems, the present invention provides a method for continuous HRV monitoring that is more accurate and applicable to multi-target scenarios.

[0010] In summary, to overcome the interference of respiratory harmonics, human trunk motion, and environmental noise on heartbeat signal extraction in multi-target scenarios, this paper proposes a non-contact multi-target heart rate variability detection method based on adaptive variational mode decomposition. Two important innovations are made in the signal processing method:

[0011] (1) A heartbeat signal extraction method named Health-VMD is proposed. This method uses an improved locust optimization algorithm (GOA) to optimize the variational mode decomposition (VMD) parameters, and can fully adaptively extract heartbeat signals from human chest displacement signals in multi-target scenarios for HRV estimation.

[0012] (2) An objective function (HIE) that is consistent with the characteristics of vital sign signals is designed for Health-VMD. This objective function consists of three indicators: permutation entropy, mutual information, and energy loss rate. It can accurately extract heartbeat signals in complex environments such as multi-target environments and effectively suppress interference such as respiratory high-order harmonics and human trunk motion.

[0013] This solution can achieve high-accuracy HRV detection for multiple people, while being contactless and highly applicable. These advantages give this solution significant application potential and practical value in the field of HRV detection and analysis.

[0014] The method first uses radar technology to perform multi-target detection on human targets in the measured scene to determine the precise distance window of each target. Subsequently, the signal phase is extracted from the distance window through differentiation and cross-multiplication algorithm to obtain the chest displacement signal of each subject. The present invention uses an improved GOA algorithm to optimize the VMD parameters and proposes a new adaptation function HIE (combining permutation entropy, mutual information and energy loss rate) as the objective function of GOA to achieve fully adaptive VMD decomposition of the chest displacement signal of each subject. Through this decomposition, the heartbeat signal and the breathing signal are extracted respectively, so that the heart rate variability characteristics of each subject can be accurately detected. This method provides a high-precision, non-contact HRV detection method suitable for multi-target scenarios.

[0015] The technical solution specifically adopted by the present invention to solve the technical problem is:

[0016] A non-contact multi-target heart rate variability detection method based on adaptive variational mode decomposition is proposed. After multi-target detection is performed on the original radar data signal to obtain the precise distance window of each human target in the measured scene, the phase of the signal within the distance window is extracted through differentiation and cross-multiplication algorithm, thereby obtaining the chest displacement signal of each subject. Through the VMD method with adaptive parameters and using the adaptation function HIE for vital sign signal extraction as the objective function of GOA, a fully adaptive VMD decomposition of the chest displacement signal of each subject is achieved, and the heartbeat signal and respiratory signal of each subject are obtained. The heart rate variability characteristics of each subject are then detected from the heartbeat signal.

[0017] Furthermore, the signal processing flow of the multi-target detection includes distance FFT, angle FFT, static clutter elimination, minimum unit mean constant false alarm rate detection and DBSCAN clustering.

[0018] Furthermore, before extracting the phase of the signal within the range window through the differentiation and cross-multiplication algorithm, a least squares-based circle fitting method is used to estimate the DC component, and then the DC component is eliminated from the signal to achieve the purpose of DC offset correction.

[0019] Furthermore, the adaptive parameter VMD method optimizes the modal number and modal bandwidth control parameters of VMD through an improved GOA optimization algorithm; the improved GOA optimization algorithm introduces a nonlinear decreasing coefficient on the basis of GOA, and this coefficient changes with the increase in the number of algorithm iterations, thereby enhancing the global search capability and local search capability in the early and late stages of the iteration, thereby improving the convergence accuracy of the algorithm; and introduces the Levy flight strategy to improve the GOA algorithm, thereby preventing the algorithm from falling into local optimality by promoting population diversity.

[0020] Further, the mathematical model of the improved GOA optimization algorithm is as follows:

[0021]

[0022] Among them, Y i represents the position of the i-th locust in the locust swarm, and the superscript d represents the improved algorithm; d ij =|x i -x j | is the distance between the i-th individual and the j-th individual, N is the number of individuals in the population, and z is the function that defines the interaction force between individuals:

[0023]

[0024] Among them, p represents the strength of attraction, q is the range of attraction scale;

[0025] ub d and lb d are the upper and lower bounds of the D-dimensional search space, respectively. Represents the optimal individual in the current population, and parameter c is the improved nonlinear decreasing coefficient, which is calculated by the following formula;

[0026]

[0027] In the formula, l represents the current number of iterations of the algorithm, and L represents the maximum number of iterations;

[0028] The Levy flight strategy is used to improve the GOA algorithm, which can prevent the algorithm from falling into local optimality by promoting population diversity.

[0029] The Lévy flight strategy is introduced to update the individual positions of locusts. In each iteration, a random 1 / 3 of the locusts are allowed to update their positions using the Lévy flight strategy. The formula is as follows:

[0030]

[0031] Where size(D) represents the dimension of the problem to be solved (which can be understood as the number of variables in the optimization problem. For example, if the optimization variables are the number of modes and the modal bandwidth control parameter, the dimension of the problem is 2). Represents dot product, i represents the current iteration number, which is used to control the flight distance, and β represents the Levy index, which allows a larger search range in the early stage of the iteration and a smaller search range in the later stage.

[0032] Furthermore, the product of permutation entropy, mutual information, and energy loss ratio is used as the objective function to achieve parameter optimization through the GOA algorithm, as shown in the following formula:

[0033] HIE=H X ·H Y ·I(X;Y)·e

[0034] Among them, HIE represents the objective function, H X represents the permutation entropy of the respiratory component, H Y represents the permutation entropy of the heartbeat component, I(X; Y) represents the mutual information between the respiratory component and the heartbeat component, and e represents the energy loss ratio.

[0035] Furthermore, the permutation entropy formulas of the respiratory component and the heartbeat component are as follows:

[0036]

[0037] Where H represents the permutation entropy, N is the total number of permutations, and p i represents the probability of occurrence of the i-th element;

[0038] The formula for the mutual information between the respiratory component and the heartbeat component is as follows:

[0039]

[0040] Where I(X;Y) represents the mutual information between the respiratory signal X and the heartbeat signal Y, p(x,y) represents the joint probability distribution function, and p(x) and p(y) represent the marginal probability distribution functions of X and Y, respectively.

[0041] The formula for energy loss ratio is as follows:

[0042]

[0043] Where f(m) represents the human chest displacement signal, u k represents the kth modal component, Σu k Represents the reconstructed signal.

[0044] Furthermore, the VMD decomposition process is:

[0045] First, the GOA parameters are initialized and several random parameter combinations are generated. Then, VMD decomposition is performed on each parameter combination to find respiratory and heartbeat signal components whose frequencies are within the respiratory frequency range and the heartbeat frequency range. The HIE is calculated and the minimum HIE and the corresponding parameter combination are recorded. The mutual information between the heartbeat component and the respiratory component is calculated respectively, and the component with the minimum mutual information is selected as the heartbeat signal component. Then, 2 / 3 of the locusts update their positions according to the mathematical model of the improved GOA optimization algorithm, and 1 / 3 of the locusts perform Lévy flight. Subsequently, the new minimum HIE and the corresponding parameter combination are calculated. The parameter combination is iteratively updated according to the conditions until the end criterion is met. Finally, the optimal parameter combination is output as the global optimal solution.

[0046] And, an electronic device includes a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the non-contact multi-target heart rate variability detection method based on adaptive variational mode decomposition as described above are implemented.

[0047] A non-transitory computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the non-contact multi-target heart rate variability detection method based on adaptive variational mode decomposition as described above.

[0048] Compared to existing technologies, the non-contact, multi-target HRV detection solution based on millimeter-wave radar proposed in this invention and its preferred embodiment offers significant advantages. By utilizing millimeter-wave radar to collect signals and applying advanced processing algorithms, this solution can simultaneously monitor the HRV of multiple targets, significantly facilitating continuous HRV monitoring in multi-target environments.

[0049] Compared with methods that perform HRV detection by calculating acceleration signals and using differential enhancement algorithms, the present invention uses multi-target recognition to distinguish multiple targets in dimensions such as distance and angle, thereby enabling simultaneous detection of multiple people.

[0050] Compared to heart rate extraction methods using VMD, the adaptive parameter VMD algorithm proposed in this paper uses an improved GOA optimization algorithm for parameter optimization. This algorithm can fully adaptively extract heartbeat signals from chest displacement signals. It also proposes an objective function (HIE) that is more suitable for heartbeat signal extraction. This method can extract more accurate HRV information and is less susceptible to the influence of human trunk motion, environmental noise interference, and respiratory higher harmonic components, demonstrating greater robustness and versatility.

[0051] Its main innovations include at least:

[0052] 1. A HRV monitoring method combined with multi-target detection is proposed to achieve simultaneous detection of heart rate variability (HRV) of multiple people. In addition, more accurate heart rate information can be obtained through HRV.

[0053] 2. A VMD decomposition method with adaptive parameters is proposed, and the VMD parameters are adaptively optimized through an improved GOA optimization algorithm to adapt to different application scenarios and the physiological signal characteristics of different individuals.

[0054] 3. A new objective function HIE is proposed, which is designed specifically for extracting human vital signs signals. Using it in the VMD parameter optimization algorithm can obtain more accurate heartbeat signals.

[0055] This method uses an adaptive parameter optimization algorithm and a newly proposed objective function (HIE) to achieve continuous HRV monitoring of multiple targets. It solves the problem that heartbeat signals are easily affected by human trunk motion and higher harmonic components of respiratory signals, and achieves continuous HRV monitoring of multiple targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0057] Figure 1 Schematic diagram of the overall solution of an embodiment of the present invention;

[0058] Figure 2Flowchart of the improved VMD algorithm provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0059] To make the features and advantages of this patent more clearly understood, the following embodiments are specifically described in detail as follows:

[0060] It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meaning as commonly understood by those skilled in the art to which this application belongs.

[0061] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0062] First, the specific implementation process and steps of the non-contact multi-target HRV detection method based on millimeter wave radar provided by the embodiment of the present invention are as follows: Figure 1 As shown:

[0063] Step 1: Collecting radar raw data signals about multiple targets;

[0064] Step 2: Perform multi-target detection on the radar raw data signal. This mainly involves identifying multiple targets and obtaining the range and azimuth of each target. This includes range FFT, angle FFT, static clutter elimination, minimum unit mean constant false alarm rate (SOCA-CFAR) detection, and DBSCAN clustering. After this step, the radar signal detection window corresponding to each target to be detected in the detection interval can be determined.

[0065] Step 3: After obtaining the precise range window of the target from the previous step, the phase of the signal within the range window is further extracted using the Differentiate and Crossmultiply (DACM) algorithm, thereby obtaining the human chest displacement signal. To eliminate the influence of the DC components of the I and Q channels on the phase extraction, a DC offset compensation operation is performed.

[0066] Step 4: Utilize the proposed VMD method with adaptive parameters, preferably optimize the VMD parameters through the improved GOA optimization algorithm, and use the proposed adaptation function HIE for vital sign signal extraction as the objective function of GOA, so as to perform a fully adaptive VMD decomposition of the human chest displacement signal to obtain the heartbeat signal, and finally perform HRV estimation.

[0067] The following is a detailed introduction to the multi-target HRV detection method that the present invention focuses on:

[0068] HRV detection can generally be divided into three main steps: target recognition, phase extraction, and heartbeat signal extraction. Target recognition primarily involves identifying multiple targets, determining the distance and angle of each target, and determining the detection window for each target. The specific signal processing process includes range FFT, angle FFT, static clutter removal, minimum unit mean constant false alarm rate (SOCA-CFAR) detection, and DBSCAN clustering. The target detection window is obtained flexibly, combining both range and azimuth information for precise spatial positioning, or using only a single piece of information (such as range or azimuth) to simplify the process or reduce costs.

[0069] After obtaining the detection window of the target, this embodiment further extracts the phase information of the signal within the detection window through DACM phase extraction. In order to eliminate the influence of the DC components of the I and Q channels on the phase extraction, a DC offset compensation operation is required.

[0070] The present invention utilizes a VMD method with adaptive parameters for heartbeat signal extraction, optimizing VMD parameters through an improved GOA optimization algorithm. Furthermore, an adaptive function (HIE) tailored to the characteristics of vital sign signals is proposed as the GOA objective function to improve the accuracy of heartbeat signal extraction.

[0071] (1) Target Identification: Target detection is crucial before vital sign detection. For example, determining the detection window by distance requires first determining the distance bin in which the target is located. A fast Fourier transform (FFT), i.e., a distance FFT, is then performed on each chirp in a fast time to locate the target.

[0072]

[0073] Where X(i,m) is the corresponding distance bin information, i is the index of the distance bin, and s IF (n,m) represents the digitized channel information for the nth ADC sample and the mth chirp. After the range FFT, the range bins with targets have higher energy than the reflected range bins without targets. In indoor scenes, various static objects (such as walls, tables, and metal objects) strongly reflect electromagnetic waves. These static objects may generate higher range bin energy than human targets, thus affecting human target detection. Therefore, eliminating static target clutter interference is essential.

[0074] Since the frequency and phase of the static target echo signal remain unchanged in the received signals at different times of the same antenna, while the echo signal of the target chest will change, the background clutter cancellation algorithm is used to eliminate the static clutter. After the echo signal vectors at different times are added and averaged, the static target echo amplitude remains the same as before, while the dynamic target echo is vector cancelled and becomes smaller. Subtracting the averaged signal from the echo signal at other times, the static target echo is subtracted, while the dynamic target echo is less affected. In this way, the static clutter signal can be effectively eliminated. The implementation process is as follows: first, the average value of the intermediate frequency signals at different times of each receiving antenna is obtained, and a reference intermediate frequency signal is obtained; then, all intermediate frequency signals are subtracted from the reference intermediate frequency signal, so that the static clutter elimination is completed, which can be represented as:

[0075]

[0076]

[0077] In the formula, M is the number of slow time dimension sampling points. After static clutter elimination, the target in the Range-FFT spectrum needs to be detected to obtain the distance bin range of the target. Since the distance bin without a reflecting target only contains noise, the energy of the distance bin with a reflecting target is greater than that of the distance bin without a reflecting target. However, since different targets have different distances, according to the principle of radar, the reflection energy of the near target is large, and the reflection energy of the far target is small, so it is not possible to set a predetermined energy threshold for target detection. Therefore, the CFAR algorithm is used to detect the target. The CFAR algorithm is a method that can adaptively adjust the threshold according to the noise level of the interference, so that the false alarm probability remains at a constant value. Since there are multiple targets in the scene, in order to avoid the masking effect and reduce the miss detection rate, the SOCA-CFAR algorithm is used to detect the target in the embodiment.

[0078] Since the human target has a certain volume, it contains multiple reflection points, and the head, chest and limbs of the human body can all produce echo signals. After CFAR detection, there will be multiple distance bins corresponding to the same target. Therefore, a clustering algorithm is needed to identify the number of targets and divide multiple distance bins of the same target together. DBSCAN is a density-based clustering algorithm that can divide data points into different clusters according to their density distribution, without the need to specify the number of clustering clusters in advance, and can identify and exclude noise data from the clustering clusters. Considering the uncertainty of the number of targets in the actual application scenario, the DBSCAN algorithm is used to cluster without knowing the number of clusters.

[0079] Unlike other parts of the human body, the distance between the chest and the radar changes slowly due to breathing and heartbeat movements, resulting in significant phase variations. Therefore, the chest's range bins have a large phase variance. Within each clustering result, the range bin with the largest phase variance within the range of range bins is selected for further analysis, corresponding to the range bin where chest motion is most pronounced.

[0080] (2) Phase extraction: After completing human target detection, in order to restore the displacement of the human chest, it is necessary to extract phase information from the target position. However, due to the inevitable defects of circuit components, the extracted phase signal will contain a DC component, causing the phase to deviate from the true value and produce phase distortion. is the target chest distance bin information at distance bin index j, and the extracted phase can be expressed as:

[0081]

[0082] Where DC I 、DC Q The real and imaginary parts of the DC bias are used respectively. The DC component is estimated by the circle fitting method based on least squares (LS), and then eliminated from the signal to achieve the purpose of DC bias correction. It can be expressed as:

[0083]

[0084]

[0085] After DC offset compensation, the phase information of the complex signal needs to be extracted. When measuring displacements greater than half a wavelength, FMCW radars face phase ambiguity caused by phase jumps. This is especially true for millimeter-wave FMCW radars (4 mm wavelength at 77 GHz), where chest movement caused by breathing and heartbeat ranges from 0.2 to 12 mm, making the phase ambiguity problem particularly severe. Generally, a phase unwrapping algorithm is used to calibrate the phase to solve this problem. However, this algorithm greatly increases computational complexity, and if noise is present, automatic calibration may fail. To address this problem, an extended DACM algorithm is used to address the phase discontinuity problem, which can be expressed as:

[0086]

[0087] Then, the human chest displacement signal can be expressed as:

[0088]

[0089] (3) Heartbeat signal extraction: After obtaining the human thoracic displacement signal, the heartbeat signal needs to be extracted from it. The human thoracic displacement signal is mainly composed of respiratory signal and heartbeat signal. The thoracic movement range caused by breathing is 1-12 mm, and the frequency is 0.1-0.5 Hz. The thoracic movement range caused by heartbeat is 0.2-0.5 mm, and the frequency is 0.8-2.0 Hz. These two can be regarded as quasi-periodic signals, and the human respiratory and heartbeat signals have different characteristics in amplitude and frequency. Therefore, the VMD decomposition method can be used to decompose the human thoracic displacement signal. However, the selection of parameters in the VMD decomposition method will greatly affect the decomposition result. Therefore, the adaptive parameter VMD decomposition method is proposed in this scheme, which can completely adaptively decompose the vital sign signal. The improved GOA is used for parameter optimization, and a function HIE suitable for the characteristics of vital sign signal is proposed as the target function of GOA. Thus, the extraction of heartbeat signal is realized, and the HRV is estimated.

[0090] The core idea of VMD is to build and solve a variational problem. Assuming that the human thoracic displacement signal f(m) is decomposed into K components, ensuring that the decomposition sequence is a modal component with a center frequency and a limited bandwidth, and the sum of the estimated bandwidths of each mode is minimized, the constraint variational model of VMD is as follows:

[0091]

[0092] where u k ={u1,u2,...,u K} is each modal function, ω k ={ω1,ω2,...,ω K} is the center frequency of each mode.

[0093] To solve the above constraint optimization problem, a quadratic penalty term α and a Lagrange multiplier λ are used to convert the constraint variational problem into an unconstrained variational problem, and the augmented Lagrangian

[0094]

[0095] The saddle point of the augmented Lagrangian is found by iterating the sub-optimization sequence through the ADMM method. Specifically, for all ω≥0, update

[0096]

[0097] where l is the iteration number, are the frequency domain expressions of f(m), u k (m), λ(m) respectively, and then ω k is updated:

[0098]

[0099] For all ω ≥ 0, update:

[0100]

[0101] Where γ represents the noise tolerance. Continue the above iteration until the convergence condition is met:

[0102]

[0103] Here, ε represents the convergence criterion tolerance. Parameter settings are crucial. A mode number K that is too large may reduce the actual information contained in each mode; conversely, a mode that is too small may result in insufficient decomposition and an inability to accurately capture important features in the signal. A mode bandwidth control parameter (or quadratic penalty term) α that is too large may cause the signal to lack detailed information; conversely, a mode that is too small may contain more noise. The settings of these two parameters can significantly impact the decomposition results. Compared to these two parameters, γ and ε have less impact on the decomposition results and are typically set to their default values ​​in the original VMD algorithm. Therefore, finding the optimal parameter combination that matches the signal is key to the VMD method.

[0104] Locusts often hunt and migrate in large gatherings. While larval locusts move slowly and over a small area, adults can move rapidly over large spaces. The GOA algorithm is a new swarm intelligence optimization algorithm derived by simulating the swarm characteristics of locusts. Compared with differential evolution (DE), particle swarm optimization (PSO), and gray wolf optimization (GWO), the GOA algorithm has the advantages of stronger global search capabilities and faster convergence. Its mathematical model is shown as follows:

[0105] Y i =S i +G i +A i (15)

[0106] Among them, Y i represents the position of the i-th locust in the locust swarm, S i is the mutual influence between the i-th locust and other individuals in the population, G i is the gravity on the i-th locust, A i is the wind force experienced by the i-th locust.

[0107] S i The expression is as follows:

[0108]

[0109] Among them, dij = |x i -x j | is the distance between the ith individual and the jth individual, is the unit vector from the ith individual to the jth individual, N is the number of individuals in the population, and z is a function that defines the force of interaction between individuals:

[0110]

[0111] where p represents the strength of attraction and q is the scale of the attraction range. In this paper, p = 0.5 and q = 1.5 are taken, and the positions of individuals in the population are limited to the range [1, 4] to ensure that the individuals in the population interact with each other.

[0112] To solve the problem that the original locust quickly reaches the comfort zone but the population does not converge to the specified point, the original GOA is modified by introducing an improved nonlinear decreasing coefficient, which changes with the number of iterations of the algorithm, thereby enhancing the global search ability in the early and late stages of iteration.

[0113]

[0114] where ub d and lb d are the upper and lower bounds of the D-dimensional search space, is the best individual in the current population, and the parameter c is the improved nonlinear decreasing coefficient, calculated as follows.

[0115]

[0116] where l represents the current iteration number of the algorithm, and L represents the maximum iteration number. The use of a nonlinear weight coefficient makes the value of the inertia weight change less in the early and late stages of iteration, enabling the algorithm to enhance the global search ability in the early stages of iteration and have good local development ability in the late stages of iteration, thereby improving the convergence accuracy of the algorithm.

[0117] Due to the rapid decrease in species diversity, the GOA algorithm is improved by using the Levy flight strategy to prevent the algorithm from converging to the local optimum too early. The algorithm is improved by promoting population diversity to avoid falling into local optima. Levy flight is a non-Gaussian random process with step length following Levy distribution. The characteristic of Levy flight is to randomly walk with a small step length and occasionally jump in a large step length to change direction, effectively increasing the diversity of individual positions in the population and avoiding falling into local optima.

[0118] The definition of Levy distribution is as follows:

[0119] Levy(y) ~ |y-β (20)

[0120] In the formula, β represents a Levy index, and y represents a step length of the Levy flight. The step length of the Levy flight is simulated by using a Mantegna algorithm, and the calculation formula is as follows:

[0121] y = μ / |ν| 1 / β (21)

[0122] In the formula, β = 1.5, μ and ν are subjected to Gaussian distribution, wherein σ ν = 1, and Γ(·) represents a gamma function.

[0123] The Levy flight is introduced to update the position of locust individuals, and in each iteration, random 1 / 3 locusts adopt the Levy flight strategy to update the position, and the formula is as follows:

[0124]

[0125] Wherein size(D) represents the dimension of the problem, represents a dot product, i represents a current iteration number, and d is used to control the flight distance, so that a large range of search movement can be performed in the early iteration, and a small range of search movement can be performed in the later iteration.

[0126] The objective function is a key factor for determining the advantages and disadvantages of the decomposition result in the VMD method. Therefore, the application proposes a function HIE for the characteristics of the breathing and heartbeat signals as the objective function of the GOA, which is an index composed of permutation entropy, mutual information and energy loss ratio. By optimizing these indexes, the optimization algorithm can better adapt to the characteristics of the breathing and heartbeat signals, thereby improving the quality and robustness of signal decomposition.

[0127] Permutation entropy is a physical quantity for describing the degree of disorder of a system, which can be used to measure the complexity or regularity of a signal. When the signal is more regular, the permutation entropy is lower, and vice versa. The breathing and heartbeat signals of the human body can be expressed as a class of sinusoidal signals, so their permutation entropy is generally low, and the formula is as follows:

[0128]

[0129] H represents the permutation entropy, N is the total number of permutations, p i represents the probability of occurrence of the i-th element.

[0130] Mutual information is used to measure the degree of correlation between two signals. A large mutual information indicates a strong correlation or interdependence between the two signals. Conversely, a small mutual information indicates a weak or almost non-existent correlation. Respiration and heartbeat signals are independent signals, and a small mutual information indicates a low correlation and better separation, reducing the impact of respiratory harmonics on the heartbeat signal. The formula is as follows:

[0131]

[0132] Where I(X;Y) represents the mutual information between the respiratory signal X and the heartbeat signal Y, p(x,y) represents the joint probability distribution function, and p(x) and p(y) represent the marginal probability distribution functions of X and Y, respectively.

[0133] Respiratory and heartbeat signals dominate the collected radar signals. The difference between the reconstructed signal and the original signal after decomposition should be as small as possible, otherwise it means that important information in the original signal is lost. Therefore, the energy loss ratio can be used as an indicator to measure the decomposition effect. A smaller energy loss ratio indicates better decomposition quality because it shows that most of the energy of the original signal has been successfully retained after decomposition. The formula for the energy loss ratio is shown below

[0134]

[0135] Where f(m) represents the human chest displacement signal, u k represents the kth modal component, Σu k Represents the reconstructed signal.

[0136] The parameter-adaptive VMD method proposed in this method uses the product of permutation entropy, mutual information, and energy loss ratio as the objective function, and optimizes the parameters through the GOA algorithm. For ease of application, the permutation entropy and mutual information are normalized and converted to a range of 0 to 1. This is shown in the following formula:

[0137] HIE=H X ·H Y ·I(X;Y)·e (26)

[0138] Among them, HIE represents the objective function, H X represents the permutation entropy of the respiratory component, H Y represents the permutation entropy of the heartbeat component, I(X; Y) represents the mutual information between the respiratory component and the heartbeat component, and e represents the energy loss ratio.

[0139] The above model constitutes the theoretical basis of this method. The algorithm flow chart is as follows: Figure 2As shown. The process is as follows: First, initialize the GOA parameters and generate some random parameter combinations, then perform VMD decomposition on each parameter combination, find the respiratory and heartbeat signal components whose frequencies are within the respiratory frequency range and the heartbeat frequency range, calculate HIE and record the minimum HIE and the corresponding parameter combination. However, sometimes the higher harmonic components of breathing are also within the heartbeat frequency range, which leads to the wrong selection of the heartbeat component. In order to select the correct signal component, the mutual information between them and the respiratory component is calculated respectively, and the component with the minimum mutual information is selected as the heartbeat signal component. Then, 2 / 3 of the locusts update their positions according to formula (18), and 1 / 3 of the locusts perform Levy flight. Subsequently, the new minimum HIE and the corresponding parameter combination are calculated. The parameter combination is iteratively updated according to the conditions until the end criteria are met. Finally, the optimal parameter combination is output as the global optimal solution. Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0140] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0141] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0142] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0143] It should be noted that, unless otherwise defined, the technical or scientific terms used in the present invention should have the usual meanings understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0144] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other manner. Any person skilled in the art may utilize the above-disclosed technical content to modify or modify the present invention into equivalent embodiments. However, any simple modifications, equivalent variations, and modifications to the above embodiments that do not depart from the technical content of the present invention and are based on the technical essence of the present invention remain within the scope of protection of the present invention.

[0145] This patent is not limited to the above-mentioned optimal implementation mode. Anyone can derive various other forms of non-contact multi-target heart rate variability detection methods based on adaptive variational mode decomposition under the inspiration of this patent. All equal changes and modifications made according to the scope of the patent application of this invention should be covered by this patent.

Claims

1. A non-contact multi-target heart rate variability detection method based on adaptive variational mode decomposition, characterized by: After performing multi-target detection on the radar raw data signal to obtain the precise range window of each human target in the measured scene, the phase of the signal within the range window is extracted through differentiation and cross-multiplication algorithms, thereby obtaining the chest displacement signal of each subject. Through the adaptive parameter VMD method and using the adaptation function HIE for vital sign signal extraction as the objective function of GOA, a fully adaptive VMD decomposition of the chest displacement signal of each subject is achieved, obtaining each subject's heartbeat signal and respiratory signal, and then detecting the heart rate variability characteristics of each subject from the heartbeat signal. The adaptive parameter VMD method optimizes the modal number and modal bandwidth control parameters of VMD through an improved GOA optimization algorithm. The improved GOA optimization algorithm introduces a nonlinear decreasing coefficient on the basis of GOA, and this coefficient changes with the increase of the number of algorithm iterations, thereby enhancing the global search capability and the local search capability in the early and late stages of the iteration, respectively, thereby improving the convergence accuracy of the algorithm. The Levy flight strategy is introduced to improve the GOA algorithm, which prevents the algorithm from falling into local optimality by promoting population diversity. The mathematical model of the improved GOA optimization algorithm is as follows: in, Indicates the number of locusts in a swarm The position of the locust, the superscript d represents the improved algorithm; For the Individuals and The distance between individuals, is the number of individuals in the population, is the function that defines the interaction force between individuals: in, Indicates the strength of attraction, is the attractiveness scale range; and They are The upper and lower bounds of the dimensional search space, Represents the optimal individual in the current population, parameter is the improved nonlinear decreasing coefficient, which is calculated by the following formula; Where, Indicates the current iteration number of the algorithm, Indicates the maximum number of iterations; The Levy flight strategy is used to improve the GOA algorithm, which can prevent the algorithm from falling into local optimality by promoting population diversity. Introduce Lévy flight to update the individual position of locusts. In each iteration, let the random The locust uses the Levy flight strategy to update its position. The formula is as follows: in represents the dimension of the problem to be solved, represents dot product, Indicates the current iteration number, which is used to control the flight distance. Representing the Levy index allows for a larger search range in the early stages of iteration and a smaller search range in the later stages.

2. The non-contact multi-target heart rate variability detection method based on adaptive variational mode decomposition according to claim 1, characterized in that: The signal processing flow of the multi-target detection includes distance FFT, angle FFT, static clutter elimination, minimum unit mean constant false alarm rate detection and DBSCAN clustering.

3. The non-contact multi-target heart rate variability detection method based on adaptive variational mode decomposition according to claim 1, characterized in that: Before extracting the phase of the signal within the range window through differentiation and cross-multiplication algorithm, the DC component is estimated by least squares based circle fitting method, and then the DC component is eliminated in the signal to achieve the purpose of DC offset correction.

4. The non-contact multi-target heart rate variability detection method based on adaptive variational mode decomposition according to claim 1, characterized in that: The product of permutation entropy, mutual information, and energy loss ratio is used as the objective function, and parameter optimization is achieved through the GOA algorithm, as shown in the following formula: Where HIE represents the objective function, represents the permutation entropy of the respiratory component, represents the permutation entropy of the heartbeat components, represents the mutual information between the respiratory component and the heartbeat component, Represents the energy loss ratio.

5. The non-contact multi-target heart rate variability detection method based on adaptive variational mode decomposition according to claim 4, characterized in that: The permutation entropy formula of the respiratory component and the heartbeat component is as follows: Where, represents the permutation entropy, is the total number of permutations, Indicates the The probability of occurrence of an element; The formula for the mutual information between the respiratory component and the heartbeat component is as follows: Where, Respiratory signal and heartbeat signal The mutual information between represents the joint probability distribution function, and Respectively and The marginal probability distribution function of The formula for energy loss ratio is as follows: in Represents the human chest displacement signal, Indicates the modal components, Represents the reconstructed signal.

6. The non-contact multi-target heart rate variability detection method based on adaptive variational mode decomposition according to claim 1, characterized in that: The process of VMD decomposition is: First, the GOA parameters are initialized and several random parameter combinations are generated. Then, VMD decomposition is performed on each parameter combination to find the respiratory and heartbeat signal components whose frequencies are within the respiratory and heartbeat frequency ranges. The HIE is calculated and the minimum HIE and the corresponding parameter combination are recorded. Calculate the mutual information between the heartbeat component and the respiratory component respectively, and select the component with the smallest mutual information as the heartbeat signal component; then, The locust updates its position according to the mathematical model of the improved GOA optimization algorithm. of locusts performing Levy flights; Subsequently, the new minimum HIE and the corresponding parameter combination are calculated; the parameter combination is iteratively updated according to the conditions until the end criteria are met; finally, the optimal parameter combination is output as the global optimal solution.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the non-contact multi-target heart rate variability detection method based on adaptive variational mode decomposition as described in any one of claims 1 to 5 are implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the non-contact multi-target heart rate variability detection method based on adaptive variational mode decomposition as described in any one of claims 1 to 5 are implemented.

Images