Non-contact respiration and heartbeat joint detection method based on FMCW (Frequency Modulated Continuous Wave) radar

Abstract

The invention belongs to the technical field of life information monitoring, and particularly relates to an FMCW radar-based non-contact breathing and heartbeat combined detection method, which comprises the following steps of: identifying the chest of a target to be detected by utilizing a target detection algorithm to obtain a target area; controlling a steering engine to enable the FMCW radar to move to a corresponding target area, transmitting a linear frequency modulation signal, receiving a reflection signal, and performing frequency mixing to obtain an intermediate frequency signal; performing distance fast Fourier transform on the sampling data of the intermediate frequency signal to obtain a distance feature, and determining a distance unit corresponding to the target according to an extreme value of the distance feature; a phase value of a distance unit where a target is located is extracted to perform phase unwrapping and phase difference operation, then a phase difference signal is subjected to signal decomposition by using a variational mode decomposition (VMD) algorithm, and a respiration and heartbeat signal is processed by using an independent component analysis (ICA) algorithm to obtain a respiration signal and a heartbeat signal; and S4, estimating the respiratory frequency and the heart rate through the frequency spectrums of the respiratory signal and the heartbeat signal. According to the invention, the signal processing efficiency and accuracy are improved.

Description

Technical Field

[0001] The invention belongs to the technical field of life information monitoring, and in particular relates to a non-contact breathing and heartbeat joint detection method based on FMCW radar. Background Art

[0002] Although contact vital signs monitoring systems play an important role in medical treatment and health maintenance, they also have some shortcomings. Contact monitoring technology requires direct contact with the skin, which makes the user experience poor and may cause discomfort to the user, especially when worn for a long time, which may cause problems such as skin airtightness, sweat stains and redness. In addition, contact monitoring equipment may be restricted by user activities when used. For example, the use of wearing and binding during sleep monitoring may change the user's sleeping habits and affect the accuracy of the monitoring results. Contact monitoring equipment may not be applicable in certain specific scenarios, such as in medical environments or public safety occasions where non-contact monitoring is required. When a patient suffers from a skin disease that causes skin ulcers, or suffers from an infectious skin disease, the wearing of a contact vital signs monitoring instrument may cause health and safety problems. In summary, although contact vital signs monitoring systems are still indispensable in some medical occasions, the development of non-contact monitoring technology is also inevitable. Therefore, it is of great significance to provide more comfortable, convenient and accurate monitoring without feeling monitoring, without wearing and binding, without direct contact with the skin, without changing sleeping habits, and without restraint. Summary of the invention

[0003] Based on the above-mentioned shortcomings and deficiencies in the prior art, one of the objects of the present invention is to at least solve one or more of the above-mentioned problems in the prior art. In other words, one of the objects of the present invention is to provide a non-contact breathing and heartbeat joint detection method based on FMCW radar that meets one or more of the above-mentioned needs.

[0004] In order to achieve the above-mentioned object of the invention, the present invention adopts the following technical solutions:

[0005] A non-contact breathing and heartbeat joint detection method based on FMCW radar comprises the following steps:

[0006] S1. Using the target detection algorithm to identify the chest of the target to be detected, and obtain the target area;

[0007] S2, control the servo to make the FMCW radar move to the corresponding target area and transmit linear frequency modulation signals, receive reflected signals, and mix to obtain intermediate frequency signals;

[0008] S3. Perform a fast Fourier transform on the sampled data of the intermediate-frequency signal to obtain range features, and then determine the range cell corresponding to the target based on the extreme values of the range features; extract the phase values of the range cell where the target is located for phase unwrapping and phase difference operations, and then use the variational mode decomposition (VMD) algorithm to decompose the phase difference signal, and use the independent component analysis (ICA) algorithm to process the respiratory and heartbeat signals to obtain the respiratory signal and the heartbeat signal;

[0009] S4. Estimate the respiratory rate and heart rate through the spectra of the respiratory signal and the heartbeat signal.

[0010] As a preferred solution, in the step S2, the chirp signal transmitted by the FMCW radar is expressed as:

[0011]

[0012] where A T is the amplitude of the signal, f0 is the starting frequency of the chirp signal, B is the bandwidth, T0 is the duration of the chirp signal, is the phase noise, and t is the time;

[0013] The echo signal received by the quadrature receiver from the target to be measured is expressed as:

[0014]

[0015] where A R is the amplitude of the echo signal, t d = 2S(t) / c, c is the speed of light, the distance from the chest to the radar S(t) = d0 + x(t), d0 is the distance from the target to the radar, and x(t) is the displacement caused by chest movement;

[0016] The received signal and the transmitted signal are mixed through two quadrature I / Q channels, and then the intermediate-frequency signal is obtained through a low-pass filter and is expressed as:

[0017]

[0018] where, is the phase noise, l d = 2B(d0 + x(t)) / cT0, μ b (t) = 4π(d0 + x(t)) / λ0, λ0 is the wavelength corresponding to the frequency f0, and j is the imaginary unit.

[0019] As a preferred solution, in the step S2, the frequency and phase of the intermediate-frequency signal are extracted for determining the distance between the target to be measured and the radar and the chest displacement of the target to be measured.

[0020] As a preferred solution, in step S3, the intermediate frequency signal is first preprocessed by filtering and noise reduction.

[0021] As a preferred solution, in step S3, the intermediate frequency signal is preprocessed by filtering and noise reduction using a band - pass filter, and the lower and upper limit frequencies of the band - pass filter are set to retain the respiration and heartbeat signals.

[0022] As a preferred solution, step S4 specifically includes:

[0023] The respiration signal and the heartbeat signal are respectively subjected to frequency detection. Using the fast Fourier transform method, the spectrum of the signal is calculated, and by analyzing the frequencies corresponding to the spectrum peaks, the respiration rate and heart rate are obtained.

[0024] As a preferred solution, in step S2, the FMCW radar uses 60GH Z FMCW millimeter - wave radar.

[0025] As a preferred solution, in step S2, the distance between the FMCW radar and the target to be measured is not greater than 2m.

[0026] As a preferred solution, in step S1, the target detection algorithm is the YOLOv5s algorithm.

[0027] Compared with the prior art, the beneficial effects of the present invention are:

[0028] Based on the respiration and heartbeat signal separation algorithm of the FMCW radar, the present invention not only improves the efficiency and accuracy of signal processing, but also provides important technical support for the further development of non - contact vital sign monitoring technology; the present invention proposes a method combining variational mode decomposition (VMD) - based vital sign detection and independent component analysis (ICA). The VMD algorithm can effectively separate the respiration and heartbeat signals, and combined with the ICA algorithm, it further improves the accuracy and stability of signal separation. Description of the Drawings

[0029] Figure 1 It is the network architecture diagram of YOLOv5s of the embodiment of the present invention;

[0030] Figure 2 It is the chest recognition flowchart of the target to be measured of the embodiment of the present invention;

[0031] Figure 3 It is the schematic diagram of the radar sensor of the embodiment of the present invention;

[0032] Figure 4Schematic diagram of radar signal transmission and reception according to an embodiment of the present invention;

[0033] Figure 5 Schematic diagram of the intermediate frequency signal according to an embodiment of the present invention;

[0034] Figure 6 Flow chart of intermediate frequency signal processing according to an embodiment of the present invention; Detailed implementation manners

[0035] To more clearly illustrate the embodiments of the present invention, the following will describe the specific implementation manners of the present invention with reference to the accompanying drawings. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings, and other implementation manners can also be obtained.

[0036] The present invention aims to study and improve the traditional breathing and heartbeat signal separation technology using millimeter radar waves. The developed breathing and heartbeat signal separation algorithm based on FMCW radar not only improves the efficiency and accuracy of signal processing, but also provides important technical support for the further development of non-contact vital sign monitoring technology. The present invention proposes a method combining variational mode decomposition (VMD) - based vital sign detection and independent component analysis (ICA). The VMD algorithm can effectively separate breathing and heartbeat signals. Combining with the ICA algorithm further improves the accuracy and stability of signal separation.

[0037] The present invention first uses the YOLOv5 algorithm to identify the human chest. When the FMCW millimeter-wave radar is controlled by a servo to be in front of the human body, the reflection area of the human chest target area for the radar signal is relatively large. Then, data is collected by the FMCW radar, and the breathing and heartbeat frequencies are jointly detected. The millimeter-wave radar emits a linearly frequency-modulated continuous wave, receives the human reflection signal, and mixes it to obtain a difference frequency signal. The breathing and heartbeat movements cause changes in the reflection signal, which are reflected in the difference frequency signal. Filter and denoise the difference frequency signal to improve the accuracy of subsequent processing. After initializing the parameters, use the VDM-ICA joint algorithm to further separate the breathing and heartbeat signals. The VDM algorithm decomposes the preprocessed signal into multiple sub-signals, which contain components related to breathing and heartbeat but are mixed. Prepare the signal after VDM decomposition as the input, and ICA separates the signals mainly caused by heartbeat and breathing changes based on the independence assumption. Finally, use the spectral analysis method to calculate the spectra of the separated heartbeat and breathing signals respectively to determine their frequencies, and achieve more accurate breathing and heartbeat frequency detection.

[0038] Specifically, the non-contact breathing and heartbeat joint detection method based on FMCW radar according to the embodiments of the present invention specifically includes the following processes:

[0039] 1. Chest recognition and positioning based on the YOLOv5s algorithm;

[0040] When the human body is directly in front of the FMCW millimeter-wave radar, the reflection area of target regions such as the human chest for the radar signal is relatively large. The millimeter-wave signal emitted by the radar is vertically incident on the human body surface, and the proportion of the reflected signal returning along the original path to the radar receiver is relatively high. This makes the intensity of the received reflected signal strong, which can reduce errors and is beneficial to signal detection and subsequent processing.

[0041] YOLOv5s is a model architecture in the YOLOv5 series. It has a good balance between accuracy and speed and is suitable for target detection tasks. As Figure 1 shown, the network model structure of YOLOv5s consists of four parts: Input (input end), Backbone (backbone network), Neck (neck network), and Prediction (prediction end).

[0042] The work of the Input end includes preprocessing the input image and unifying its size for training and inference in the detection model.

[0043] In the first layer of the Backbone (backbone network), the Focus structure is proposed. It first performs a block slicing operation on the feature map and then connects the results and sends them to the subsequent modules. Its function is to improve computing power while not losing information. Two CSP structures are designed in YOLOv5. Among them, CSP1_X is applied to the Backbone, and CSP2_X is applied to the Neck. Some modifications are made to the FPN+PAN structure of the Neck part to enhance the network's feature fusion ability.

[0044] The loss function and non-maximum suppression (NMS) constitute the prediction end. The loss function of YOLOv5 consists of three parts, and its calculation formula:

[0045] Loss = λ1L loc +λ2L obj +λ3L cls

[0046] λ1, λ2, λ3 are loss term weight coefficients, and L loc is the bounding box loss, L obj is the confidence loss, and L cls is the class loss.

[0047] Adopt images of the human chest area under various conditions, including people wearing different clothes (such as T-shirts, shirts, dresses, etc.), different postures (standing, sitting, bending, etc.), and different lighting conditions (strong light, weak light, natural light, etc.). Use annotation tools to annotate the chest area in the images. Input the training set data into the YOLOv5s model for training. During the training process, the model calculates the loss function based on the annotated chest area and the predicted results.

[0048] Such as Figure 2 As shown, deploy the trained YOLOv5 model into OpenMv, and obtain real-time images through the camera of OpenMv. Transmit the preprocessed images to the loaded YOLOv5s model for inference. The model will output information such as the bounding box coordinates (including the upper left and lower right coordinates or the center coordinates and width and height) of each predicted chest area, and the class confidence.

[0049] According to the center coordinates (x c , y c ) output by the model, taking the upper left corner of the image as the origin (0, 0), with the positive x-axis direction horizontally to the right and the positive y-axis direction vertically downward to establish a coordinate system, we can determine the specific position of the chest in the image. OpenMV and the servo pan-tilt are connected and controlled through PWM wave signals. According to the center coordinates of the chest area, the angle that the servo pan-tilt needs to rotate can be calculated to make the FMCW millimeter-wave radar emit signals directly at the human chest.

[0050] 2. Data acquisition based on FMCW radar;

[0051] Currently, there are mainly three types of radar systems used for non-contact vital sign detection, namely continuous wave (CW) radar, ultra-wideband (UWB) radar, and frequency-modulated continuous wave (FMCW) radar. Among them, the transmission frequency of the FMCW radar is linearly modulated with time. It can detect the chest cavity displacement using the Doppler effect while detecting the target distance. By extracting the frequency and phase of the intermediate frequency signal, accurate distance and displacement information of the measured object can be obtained. Compared with microwave radar, millimeter-wave radar has a wider bandwidth and shorter wavelength, so it has higher range resolution and sensitivity, making it suitable for non-contact vital sign detection. The FMCW millimeter-wave radar has excellent performance in both distance measurement and micro-displacement measurement. Therefore, a 60GH Z FMCW millimeter-wave radar is used for non-contact vital sign detection.

[0052] The radar sensor detects targets by transmitting frequency-modulated continuous wave signals (FMCW signals). The characteristic of continuous waves is that they change continuously over a period of time. When such signals hit a person's body, the body will cause minute vibrations due to breathing and heartbeat, resulting in the reflected wave carrying signals with the human body vibrations. Secondly, the received reflected signal is mixed with a part of the transmitted signal. This process generates an intermediate frequency signal, which is determined by the signal difference after comparing the transmitted signal and the reflected signal. This frequency difference is actually the distance between the target and the radar sensor and the phase difference caused by the weak vibrations of the human body. During this process, the intermediate frequency signal is processed into a complex form, and then the radar sensor performs corresponding algorithm analysis on these complex signals to extract the signals related to breathing and heartbeat.

[0053] As Figure 3 shown, the R60ABD1 radar module is based on a one-transmit and three-receive antenna form: The wide-beam radar module is mainly suitable for the top-mounted installation mode. By algorithm control within a certain angle range, it accurately scans the action tomography of the whole human body; it realizes the function of collecting breathing and heart rate in different postures when the human body is static and dynamic. As Figure 4 shown, the radar emits millimeter-wave signals in the 60G frequency band. After the signals contact the measured target and are reflected, the transmitted signal is demodulated. Then, through demodulation, amplification, ADC and other processes, the initial data is obtained. Next, the initial data is processed by the VMD-ICA algorithm in the MCU unit to calculate the amplitude, frequency, and phase information of the signal, and finally, the measurement and evaluation of the heart rate and breathing frequency data of the measured target are realized.

[0054] In the embodiment of the present invention, the R60ABD1 millimeter-wave radar sensor is adopted. In this system, when measuring multiple data in various complex environments, it is necessary to ensure that the anti-interference ability of the sensor itself is qualified. The R60ABD1 millimeter-wave radar sensor works without being affected by environments such as temperature, humidity, noise, airflow, and light, and can measure various data more stably and accurately. At the same time, the detection distance of the R60ABD1 millimeter-wave radar sensor is wide, generally between 1.5m and 2m, which can better meet the non-contact requirements of the system and is more convenient to respond to the human body vibration signals at different distances. Secondly, the R60ABD1 millimeter-wave radar has the advantage of low power consumption, and the output power is within 0.5W, which can significantly extend the service life of the device and reduce the maintenance cost.

[0055] In the actual operation process, after testing, the distance between the radar sensor and the body of the tested person should be kept at about 1.5m, that is, the effective measurement range of the radar sensor is 1.5m. After building the module, as Figure 5 shown, when the radar transmitted signal meets the measured target and returns to the signal receiving module, the time delay generated during this process is called Δtd (same as t in the following formula d ), after mixing the transmitted signal and the echo signal by the mixing module, an intermediate-frequency signal with an unchanged frequency can be obtained.

[0056] The FMCW radar transmits a linearly frequency-modulated signal. The quadrature receiver captures the echo signal reflected by the target and performs quadrature mixing with the transmitted signal to obtain an intermediate-frequency signal. Extracting the frequency and phase of the intermediate-frequency signal can determine the distance between the target and the radar and the chest displacement of the target. The transmitted linearly frequency-modulated signal can be expressed as:

[0057]

[0058] In the formula: A T is the amplitude of the signal; f0 is the starting frequency of the linearly frequency-modulated signal; B is the bandwidth; T0 is the duration of the linearly frequency-modulated signal; is the phase noise, and the received signal can be considered as the delay of the transmitted signal; d0 is the distance from the target to the radar, x(t) is the displacement caused by chest movement, then the distance from the chest to the radar is S(t) = d0 + x(t); the delay time is t d = 2S(t) / c, where c is the speed of light; then the received signal can be expressed as:

[0059]

[0060] In the formula: A R is the amplitude of the received signal. The received signal and the transmitted signal are mixed through two orthogonal I / Q channels, and then the intermediate-frequency signal y(t) is obtained through a low-pass filter, which can be expressed as:

[0061]

[0062] In the formula: is the phase noise; l d = 2B(d0 + x(t)) / cT0, μ b (t) = 4π(d0 + x(t)) / λ0, λ0 is the wavelength corresponding to the frequency f0, and j is the imaginary unit;

[0063] For the displacement amplitude of the heartbeat being 0.1 - 0.5 mm, the displacement amplitude of breathing being 1 - 12 mm, the heartbeat frequency being 0.8 - 2 H Z , and the breathing frequency being 0.1 - 0.5 H Z , in the phase , both the chest displacement x(t) and the wavelength λ c are in the millimeter range, so a small chest displacement will also cause a phase Obvious changes can be achieved by continuously transmitting a chirp signal and extracting the phase of the intermediate-frequency signal to sample the chest displacement x(t), and then estimating the respiration rate and heart rate.

[0064] 3. Signal preprocessing;

[0065] Perform preprocessing operations such as filtering and noise reduction on the collected intermediate-frequency signal to remove some clutter interference and high-frequency noise to improve the accuracy of subsequent processing.

[0066] Apply a band-pass filter that allows signals within a certain frequency range to pass through. This frequency range can be set according to the frequency ranges of the respiration and heartbeat signals. For respiration and heartbeat signal detection, the frequency of the heartbeat signal is generally around 1 - 2 Hz, and the frequency of the respiration signal is lower (about 0.2 - 0.4 Hz). Set the lower limit frequency of the band-pass filter to 0.1 Hz (to remove lower-frequency interference) and the upper limit frequency to 3 Hz (to filter out high-frequency noise), which can effectively retain the respiration and heartbeat signals.

[0067] 4. VMD-ICA joint algorithm (VDM algorithm for signal decomposition - ICA algorithm for separating heartbeat and respiration signals);

[0068] The FMCW radar vital sign detection process based on the VMD-ICA joint algorithm is as Figure 6 shown, mainly including four steps: target positioning, phase extraction, signal separation, and frequency estimation. First, perform a range fast Fourier transform on the sampled data of the intermediate-frequency signal to obtain range characteristics. Then determine the range bin corresponding to the target according to the extreme values of the range characteristics; extract the phase values of the range bin where the target is located for phase unwrapping and phase difference operations. After that, use the VMD algorithm to decompose the phase difference signal, and use ICA to further process the respiration and heartbeat signals to improve the detection accuracy of the heartbeat signal, obtain the respiration signal and the heartbeat signal, and the respiration rate and heart rate can be preliminarily estimated through the spectra of these two signals.

[0069] 4.1. Signal decomposition method based on variational mode decomposition (VMD);

[0070] Initialize the parameters of the VDM algorithm to determine relevant parameters such as the number of decomposition layers and the initial mode functions. Decompose the preprocessed signal into multiple sub-signal components through the VDM algorithm. These components represent signal components with different frequency characteristics, which contain information related to respiration and heartbeat but are in a mixed state.

[0071] The variational mode decomposition algorithm assumes that any signal is composed of a series of sub-signals with specific center frequencies and finite bandwidths, that is, the intrinsic mode function (IMF). The IMF is defined as an amplitude-modulated and frequency-modulated signal and can be expressed as

[0072]

[0073] Where: A k (t) and are the amplitude and phase of the k-th mode u k (t) respectively. The amplitude A k (t) and the instantaneous phase change relatively little, and the mode u k (t) has a central frequency ω k . The mode u k (t) and its central frequency ω k can be solved by a variational problem, and the solution process needs to satisfy two conditions:

[0074] (1) The sum of the modal bandwidths is minimized; The calculation method of the modal bandwidth in the original VDM formula may not accurately capture the subtle frequency changes in the signal. When dealing with respiratory and heartbeat signals, the frequency changes of these signals are relatively subtle and vulnerable to noise interference. When calculating the modal bandwidth, an adaptive weight is introduced for each frequency component. According to the local characteristics of the signal, such as the amplitude change rate of the signal or the energy distribution of the spectrum, the weight value is dynamically adjusted. So that the decomposed modes can more accurately reflect the true frequency components of the signal. The bandwidth calculation formula is

[0075]

[0076] Where, is the partial derivative symbol; δ(t) is the unit impulse function; ω(t) is the adaptive weight function, which can be calculated according to the local characteristics of the signal;

[0077] (2) The sum of the modes is equal to the original signal. Assume that the original signal f(t) is decomposed into k (positive integer) modes. According to the above conditions, the constrained variational expression is

[0078]

[0079] Where: δ(t) is the unit impulse function; * represents convolution. Introduce the Lagrange multiplier λ and the penalty factor α, change the constrained problem into an unconstrained problem, and transform the constrained minimization problem into finding the saddle point of the augmented Lagrangian function L, and its expression is:

[0080]

[0081] Where d(t) = <λ(t), f(t) - ∑ k u k (t)>, the modal component in variational mode decomposition (VMD), which represents the k-th intrinsic mode function (IMF) after the original signal f(t) is decomposed. It represents the square of the norm of the error between the original signal f(t) and the signal ∑ k reconstructed by the modal component u k u k (t).

[0082] u k (t) and ω k can be obtained by iterative alternating direction method of multipliers, and the iterative formula is as follows:

[0083]

[0084]

[0085]

[0086] In the formula: and respectively represent the Fourier transforms of f(t), λ(t) and u k (t), the superscript n is the number of iterations, the subscripts i, k are the modal orders; τ is the update rate of the Lagrange multiplier. The convergence condition of the iteration is

[0087]

[0088] In the formula: z r and z a are the relative error and the absolute error respectively. When both formulas are satisfied, the iteration stops.

[0089] 4.2. ICA-based respiratory and heartbeat separation algorithm;

[0090] The ICA algorithm is based on the independence assumption of signals. By finding the inverse matrix of the appropriate mixing matrix, the multi-dimensional mixed signal is separated into independent source signals. In this application scenario, it is expected to separate the change signals mainly caused by heartbeat and the change signals mainly caused by respiration. According to the characteristics of the separated signals, the signal components corresponding to heartbeat and respiration are identified respectively. The heartbeat signal has a relatively high frequency and a relatively small amplitude change, while the respiration signal has a relatively low frequency and a relatively large amplitude change.

[0091] The signal separation process of the embodiment of the present invention specifically includes:

[0092] 1. Centering and whitening: Then the signal is centered and whitened to ensure that the signal has a zero mean and the covariance matrix is the identity matrix;

[0093] 2. ICA separation: Apply the ICA algorithm to estimate the mixing matrix and the independent component s(t);

[0094] 3. Selection and reconstruction: Select the components representing heartbeat and respiration from the independent components separated by ICA, usually based on frequency characteristics and known physiological knowledge.

[0095] 4. Reconstruct the signals: Use the selected independent components to reconstruct the clean heartbeat and respiration signals.

[0096] Before the ICA algorithm, pre-whiten the observed signals to transform the covariance matrix of the signals into an identity matrix, reducing the impact of noise on the algorithm. At the same time, introduce a regularization term into the objective function of ICA to constrain the norm or smoothness of the solution and prevent overfitting to noise. In the objective function,

[0097]

[0098] where λ is the regularization parameter, and ‖W‖ 2 is the regularization term. By adjusting the value of λ, the effects of signal separation and noise suppression are balanced. This method can improve the robustness of the ICA algorithm to noise to a certain extent, enabling accurate separation of respiration and heartbeat signals even in the presence of noise.

[0099] The algorithm for separating respiration and heartbeat signals based on independent component analysis (ICA) is a powerful statistical method for extracting independent signal sources from multi-channel biological signal recordings. This method is widely applied to complex physiological signals such as respiration measurement, aiming to separate the pure heartbeat and respiration signals for further analysis and diagnosis. The core idea of the ICA algorithm is to decompose the multivariate signal (observed signal) into additive sub-components (independent components), which are assumed to be statistically independent non-Gaussian signal sources.

[0100] For respiration and heartbeat signals, the ICA model can be expressed as:

[0101] x(t) = A(t)·s(t) + n(t) (3 - 11)

[0102] where x(t) represents the observed signal vector, A(t) is the time-varying mixing matrix, s(t) is the independent signal source vector (including heartbeat and respiration signals), and n(t) is the noise vector. In practical applications, respiration and heartbeat signals are often non-stationary, and their statistical characteristics change with time. Introduce a time-varying mixing matrix to describe the non-stationary characteristics of the signals, and adopt the sliding window technique to estimate the mixing matrix within each window and update the matrix parameters as the window slides to track the non-stationarity of the signals. This method can better capture the changes in respiration and heartbeat signals at different time periods and improve the separation effect.

[0103] The mathematical basis of the ICA algorithm in the embodiments of the present invention:

[0104] Fast ICA is a widely used ICA algorithm. Based on the principle of maximum non-Gaussianity, it finds independent components by iteratively optimizing the non-Gaussianity of independent components. Its basic update rule is as follows:

[0105] w + = E{x(t)g(w T x(t))}-E{g′(w T x(t))}w (3-12) where w is the weight vector used to extract a single independent component, g(·) is a non-linear function, often chosen as the logarithmic hyperbolic tangent function or the exponential function, g′(·) is the derivative of the non-linear function, and E{} represents the expected value. Next, normalize the weight vector:

[0106]

[0107] This process is repeated until the weight vector converges. By repeating this process for all independent components, an independent component matrix can be constructed, and then this matrix is used to analyze and reconstruct the original signal, so as to achieve the purpose of separating respiratory and heartbeat signals. The ICA algorithm provides a powerful tool for separating and analyzing heartbeat and respiratory signals. The accurate separation of these signals is crucial for clinical diagnosis and biomedical research. The success of this algorithm lies in its ability to extract statistically independent components from multi-channel and complex physiological signals without prior knowledge of the signal source or the mixing process.

[0108] 5. Frequency Detection

[0109] Perform frequency detection on the separated heartbeat and respiratory signals respectively. Use the fast Fourier transform FFT method to calculate the spectrum of the signal, so as to determine the heartbeat frequency and respiratory frequency. By analyzing the frequency corresponding to the spectrum peak, the frequency values of respiration and heartbeat can be obtained, realizing the detection of respiratory and heartbeat frequencies.

[0110] FFT is an algorithm for efficiently calculating the discrete Fourier transform DFT. For the discrete-time signal x(n) (n = 0, 1, 2, …, N - 1), the calculation formula of DFT is:

[0111]

[0112] where X(k) is the spectrum of the signal. The FFT algorithm utilizes properties such as the symmetry and periodicity of the DFT, greatly reducing the computational amount.

[0113] Calculate the frequency axis corresponding to the spectrum after FFT. For the case where the sampling frequency is f s , and the signal length is N, the frequency resolution frequency axis f(k) = kΔf (k = 0, 1, …, N - 1).

[0114] In the spectrum with the frequency axis calibrated, find the frequency corresponding to the peak. For the respiration and heartbeat signals, the frequency peak of the heartbeat signal is usually around 1 - 2 Hz, and the frequency peak of the respiration signal is generally between 0.1 - 0.5 Hz. The peak position can be determined by comparing the magnitudes of the spectrum amplitudes, and the frequency corresponding to the point with the largest amplitude is taken as the heartbeat or respiration frequency.

[0115] The above description is only a detailed explanation of the preferred embodiments and principles of the present invention. For those of ordinary skill in the art, according to the idea provided by the present invention, there will be changes in the specific implementation manners, and these changes should also be regarded as the protection scope of the present invention.

Claims

1. A non-contact combined respiration and heartbeat detection method based on FMCW radar, characterized in that, It includes the following steps: S1. Use the target detection algorithm to identify the chest of the target to be measured, and obtain the target area; S2. Control the servo to move the FMCW radar to the corresponding target area and emit a chirp signal, receive the reflected signal, and mix it to obtain an intermediate frequency signal; S3. Perform a range fast Fourier transform on the sampled data of the intermediate frequency signal to obtain range features, and then determine the range cell corresponding to the target according to the extreme value of the range features; extract the phase value of the range cell where the target is located for phase unwrapping and phase difference operation, and then use the variational mode decomposition (VMD) algorithm to decompose the phase difference signal, and use the independent component analysis (ICA) algorithm to process the respiration and heartbeat signals to obtain the respiration signal and the heartbeat signal; S4. Estimate the respiration rate and heart rate through the spectra of the respiration signal and the heartbeat signal.

2. The non-contact combined respiration and heartbeat detection method according to claim 1, wherein In the step S2, the chirp signal emitted by the FMCW radar is expressed as: Among them, A T is the amplitude of the signal, f0 is the starting frequency of the chirp signal, B is the bandwidth, T0 is the duration of the chirp signal, is the phase noise, and t is the time; The echo signal reflected by the target to be measured received by the quadrature receiver is expressed as: Among them, A R is the amplitude of the echo signal, t d = 2S(t) / c, where c is the speed of light, the distance from the chest to the radar S(t) = d0 + x(t), d0 is the distance from the target to the radar, and x(t) is the displacement caused by chest movement; The received signal and the transmitted signal are mixed through two orthogonal I / Q channels, and then the intermediate frequency signal is obtained through a low-pass filter and is expressed as: Among them, is the phase noise, l d = 2B(d0 + x(t)) / cT0, μ b (t) = 4π(d0 + x(t)) / λ0, where λ0 is the wavelength corresponding to the frequency f0, and j is the imaginary unit.

3. The non-contact combined respiration and heartbeat detection method according to claim 2, wherein In the step S2, the frequency and phase of the intermediate frequency signal are extracted to determine the distance between the target to be measured and the radar and the chest displacement of the target to be measured.

4. The non-contact combined respiration and heart rate detection method according to claim 1, wherein In the step S3, first perform filtering and noise reduction preprocessing on the intermediate frequency signal.

5. The non-contact combined respiration and heartbeat detection method according to claim 4, characterized in that In the step S3, use a band-pass filter to perform filtering and noise reduction preprocessing on the intermediate frequency signal, and set the lower limit frequency and upper limit frequency of the band-pass filter to retain the respiration and heartbeat signals.

6. The non-contact combined respiration and heartbeat detection method according to claim 1, wherein The step S4 specifically includes: Perform frequency detection on the respiration signal and the heartbeat signal respectively, adopt the fast Fourier transform method, calculate the spectrum of the signal, and obtain the respiration rate and heart rate by analyzing the frequency corresponding to the spectrum peak.

7. The non-contact combined respiration and heartbeat detection method according to claim 1, characterized in that In the step S2, the FMCW radar uses 60GH Z FMCW millimeter-wave radar.

8. The non-contact combined respiration and heartbeat detection method according to claim 1, wherein In the step S2, the distance between the FMCW radar and the target to be measured is not greater than 2m.

9. The non-contact combined respiration and heartbeat detection method according to claim 1, wherein In the step S1, the target detection algorithm is the YOLOv5s algorithm.

Images