Non-contact respiration detection method

Abstract

The invention discloses a non-contact respiration detection method. First, an access point (AP), a base station (STA) and a capture device are arranged in a target area, and channel state information is obtained by capturing a beamforming feedback frame (BFI) periodically sent from the STA to the AP. And then, decompressing and extracting amplitude signals and time information in the BFI frame, performing phase-space reconstruction on the extracted representative subcarrier signals of the antenna combination, and extracting and evaluating the optimal antenna combination through dynamic characteristics. Thirdly, performing Hampel filtering on the subcarrier amplitude signals of the optimal antenna combination to remove abnormal values, performing phase-space reconstruction and dynamic feature extraction on all subcarriers, and selecting the optimal subcarrier by using a K-means clustering algorithm; and then, Savitzky-Golay filtering and band-pass filtering are carried out on the selected optimal subcarrier signal, and linear interpolation is carried out through principal component analysis (PCA) dimension reduction to obtain an equally-spaced sampling signal. And finally, performing fast Fourier transform (FFT) on the processed signal, and estimating the respiratory rate through an innovative peak detection algorithm. The invention provides a non-contact respiration detection method by utilizing a phase space reconstruction and dynamic feature extraction technology, and compared with a traditional method, the method can better capture nonlinear dynamic features of signals and more comprehensively evaluate dynamic features of each subcarrier, so that the subcarrier most suitable for respiration detection is selected, and the detection accuracy is improved. The breathing signal and the interference signal are effectively distinguished, and the accuracy and robustness of non-contact breathing detection based on the Wi-Fi signal are remarkably improved.

Description

Technical Field

[0001] The invention belongs to the field of radio technology, and in particular relates to a non-contact breathing detection method. Background Art

[0002] With the rapid development of the Internet of Things and smart home technologies, non-contact vital signs monitoring methods have received increasing attention. Among them, respiratory rate, as one of the important physiological parameters of the human body, is of great significance for health monitoring, sleep quality assessment, and early diagnosis of certain diseases. Traditional respiratory monitoring methods usually require wearing special equipment, such as breathing belts or nasal tubes. Although these methods are highly accurate, they can cause discomfort to users, especially during long-term monitoring or sleep monitoring, affecting the user's normal activities and sleep quality.

[0003] In recent years, non-contact respiration detection methods based on wireless signals have become a research hotspot. These methods mainly use the slight changes in wireless signals caused by human respiratory movements during propagation to estimate respiration rate. Among them, respiration detection methods based on Wi-Fi signals have attracted much attention due to their high popularity and low cost. Such methods usually use channel state information (CSI) or beamforming feedback information (BFI) in Wi-Fi signals to perform respiration detection.

[0004] In Wi-Fi signal-based respiration detection, since different subcarriers have different wavelengths, their sensitivity to respiration signals is different, so selecting the right subcarrier is crucial to improving detection accuracy. At present, common subcarrier selection methods include variance-based selection methods, short-term breathing-to-noise ratio (BNR)-based selection methods, and machine learning-based selection methods. The variance-based selection method assumes that subcarriers with larger variances contain more respiration information, so the subcarrier with the largest variance is selected for respiration detection. However, this method may be affected by environmental noise, resulting in unstable selection results. The selection method based on short-term breathing-to-noise ratio (BNR) selects the best subcarrier by calculating the ratio of the respiration signal strength of each subcarrier to the total signal strength. Although this method takes the signal-to-noise ratio into account, it may ignore other important characteristics of the signal. Some researchers have also tried to use machine learning algorithms to automatically select the best subcarrier. Although such methods can learn the complex features carried by the subcarrier signal, they often require a large amount of training data and may have overfitting problems.

[0005] Although the above-mentioned several types of subcarrier selection methods have improved the effect of subcarrier selection to a certain extent, there are still some limitations. Therefore, there is an urgent need for a new subcarrier selection method that can make full use of the dynamic characteristics of the signal, improve the accuracy and robustness of the selection, and maintain low computational complexity and strong environmental adaptability. In response to this problem, the present invention proposes a non-contact breathing detection method. First, the subcarrier amplitude signal and time information are obtained from the beamforming feedback information (BFI) captured in the Wi-Fi system, and then the antenna combination and subcarrier are evaluated by using phase space reconstruction technology and dynamic feature extraction methods, and the k subcarrier signals with the best perception performance are selected to weight and synthesize a new subcarrier signal for breathing detection. Then, a series of filtering and dimensionality reduction processing are performed on the selected optimal breathing detection signal, and finally the estimated breathing rate of the target is obtained by fast Fourier transform (FFT) and innovative peak detection algorithm. Through phase space reconstruction and dynamic feature extraction, this method can more comprehensively evaluate the dynamic characteristics of each subcarrier, thereby selecting the subcarrier most suitable for breathing detection and improving the accuracy of detection. Compared with the traditional variance or BNR-based methods, this method can capture the nonlinear dynamic characteristics of the signal and better distinguish the respiratory signal from the interference signal. Summary of the invention

[0006] The purpose of the present invention is to provide a contactless breathing detection method. The method first obtains the subcarrier amplitude signal and time information through the beamforming feedback information (BFI) in the Wi-Fi system, and then uses the phase space reconstruction technology and the dynamic feature extraction method to evaluate the antenna combination and the subcarrier, and selects the k subcarrier signals with the best perception performance to synthesize a new subcarrier signal according to the feature score weight. Then, a series of filtering and dimensionality reduction processing is performed on the selected optimal breathing detection signal, and finally the fast Fourier transform (FFT) and the innovative peak detection algorithm are used to accurately estimate the breathing rate of the subject. This method can make full use of the dynamic characteristics of the Wi-Fi signal, select the subcarrier signal with the best perception performance for breathing detection, and improve the accuracy and robustness of breathing detection based on Wi-Fi signals.

[0007] A non-contact breathing detection method according to the present invention comprises the following steps:

[0008] Step 1: Place a N AP The access point (AP) with N antennas STA The antenna base station (Station, STA) and a capture device, assuming N AP ≥N STAA subject breathes quietly near the AP and STA. The STA is associated with the AP. The AP periodically sends data frames to the STA. The capture device captures the beamforming feedback frames (BFI) periodically sent from the STA to the AP for target breathing detection.

[0009] Step 2: Decompress and extract the amplitude signal and time information in the BFI frame to obtain N AP× N STA Channel state information for all subcarriers of the antenna combination;

[0010] Step 3: Extract the N AP× N STA The representative subcarrier signal of the antenna combination is reconstructed in phase space (PSR), and the optimal antenna combination is evaluated by dynamic feature extraction, which specifically includes the following steps:

[0011] Step 3 (a) select the first n representative subcarriers of each antenna combination according to the subcarrier energy. Subcarriers with high energy usually contain more information and are more representative of channel characteristics.

[0012] Step 3 (ii): Selecting the appropriate delay time is the key to phase space reconstruction. The average mutual information (AMI) method is used to calculate the optimal delay time τ for each representative subcarrier. The calculation formula of the average mutual information (AMI) method is:

[0013]

[0014] Where P(X(t),X(t+τ)) is the joint probability distribution of X(t) and X(t+τ), P(X(t)) and P(X(t+τ)) are the marginal probability distributions of X(t) and X(t+τ), respectively. The τ that minimizes I(τ) is selected as the optimal delay time.

[0015] Step 3 (iii): The embedding dimension defines the dimension of the reconstructed phase space. The CAO method is used to determine the optimal embedding dimension m for each representative subcarrier. The specific steps of the CAO method are as follows:

[0016] (1) Calculate E(d):

[0017]

[0018] Among them, y(i,d) is the d-dimensional reconstruction vector, and n(i,d) is the time index of the nearest neighbor of y(i,d);

[0019] (2) Calculate E * (d)

[0020]

[0021] (3) When E * (d) When it no longer changes significantly, the corresponding d is the optimal embedding dimension m.

[0022] Step 3 (4): Calculate the average values ​​of τ and m of all representative subcarriers as new τ and m, and reconstruct the phase space of each representative subcarrier according to the determined delay time τ and embedding dimension m to obtain the reconstructed trajectory. The time series of each subcarrier will be mapped into the m-dimensional phase space. The formula for phase space reconstruction is:

[0023] X(t)=[x(t),x(t+τ),x(t+2τ),…,x(t+(m-1)τ)]

[0024] Among them, X(t) is the original time series, τ is the delay time, and m is the embedding dimension. The phase space reconstruction process is as follows:

[0025] For the time series X(t), t=1,2,…,N, construct an m-dimensional vector:

[0026] X(t)=[x(t),x(t+τ),x(t+2τ),…,x(t+(m-1)τ)]

[0027] Among them, t=1,2,…,N-(m-1)τ.

[0028] In this way, we obtain a trajectory containing N-(m-1)τ m-dimensional vectors, each of which represents a point in the reconstructed phase space.

[0029] Step 3 (V) Analyze the reconstructed phase space and evaluate the dynamic characteristics of each representative subcarrier signal, including periodicity, stability, recurrence quantification analysis (RQA) characteristics and geometric characteristics:

[0030] (1) Periodicity: Perform fast Fourier transform on the phase space trajectory and calculate the average amplitude of the spectrum as a periodicity indicator;

[0031]

[0032] Among them, FFT represents fast Fourier transform, and |·| represents the modulus of a complex number.

[0033] (2) Stability: The standard deviation of the phase space trajectory is calculated as a stability indicator;

[0034]

[0035] where μ is the mean value of X(t).

[0036] (3) Recursive quantitative analysis (RQA) characteristics: including recurrence rate (RR), determinism (DET) and average diagonal line length (L);

[0037] ①Construct a recursive graph R:

[0038] First, choose a suitable threshold ∈, and then for each pair of points i and j, calculate:

[0039] R(i,j)=Θ(∈-||X(i)-X(j)||)

[0040] where Θ is the Heaviside step function, which is 1 if the argument is nonnegative and 0 otherwise.

[0041] ② Calculate the recurrence rate (RR):

[0042]

[0043] ③Calculation certainty (DET):

[0044] First find all the diagonal line segments and record their lengths.

[0045]

[0046] where P(l) is the number of diagonals of length l, l min is the minimum diagonal length (usually 2).

[0047] ④Calculate the average diagonal length (L):

[0048]

[0049] (4) Geometric characteristics: including the average distance from the trajectory to the center of mass and the volume occupied by the trajectory.

[0050]

[0051] Volume = ∏(max(X i )-min(X i ))

[0052] Where C is the center of mass of the trajectory, X i is the i-th dimension coordinate.

[0053] Step 3 (six): standardize the extracted dynamic features, calculate the feature scores, and select the antenna combination with the highest score as the optimal antenna combination according to the feature scores.

[0054] Step 4: Perform Hampel filtering on the subcarrier amplitude signal of the optimal antenna combination to remove outliers;

[0055] Step 5: Perform phase space reconstruction and dynamic feature extraction on all subcarriers of the optimal antenna combination, and use the K-means clustering algorithm to select the best subcarrier to obtain the optimal breathing detection signal, which specifically includes the following steps:

[0056] Step 5 (i), performing phase space reconstruction and dynamic feature extraction on all subcarriers of the optimal antenna combination according to the delay time τ and embedding dimension m of the representative subcarrier of the optimal antenna combination obtained in step 3;

[0057] Step 5 (ii), standardizing the extracted dynamic features and calculating feature scores;

[0058] Standardization process:

[0059] Feature score calculation: score feature =w1Z1+w2Z2+…+w n Z n

[0060] Among them, w i is the weight of each feature, Z i is the standardized eigenvalue.

[0061] Step 5 (iii), using the K-means clustering algorithm to divide the subcarriers into k categories, selecting the subcarrier with the highest feature score in each category, and taking the selected k subcarriers as the best subcarriers;

[0062] Step 5 (iv), weighting the k subcarriers according to the characteristic scores to synthesize a new subcarrier signal.

[0063] Weighted synthesis formula:

[0064] Among them, S i is the i-th selected subcarrier signal, w i is its corresponding feature score.

[0065] Step 6: Perform Savitzky-Golay filtering on the selected optimal breathing detection signal to smooth the data, and extract the breathing signal frequency band through bandpass filtering;

[0066] Step 7: Perform principal component analysis (PCA) on the filtered signal to reduce its dimension, and perform linear interpolation to obtain the signal S sampled at equal intervals. interp(t) ;

[0067] Step 8: Perform fast Fourier transform (FFT) on the interpolated signal and estimate the respiratory rate through an innovative peak detection algorithm, which specifically includes the following steps:

[0068] Step 8 (a), performing a fast Fourier transform (FFT) on the interpolated signal obtained in step 7;

[0069] Step 8 (ii) perform wavelet denoising on the FFT output signal, and find the peak in the denoised signal. The peak height is required to be no less than 10% of the maximum signal value, the peak spacing is no less than 5% of the frequency axis length, and the protrusion is no less than 30% of the maximum signal value. The peak detection conditions are:

[0070] Height condition: peakheight>0.1×max(|X(f)|)

[0071] Spacing condition: peakdistance>0.05×length(f)

[0072] Prominence condition: peakprominence>0.3×max(|X(f)|)

[0073] Step 8 (iii), screening effective peaks in the frequency range of 0.1 Hz to 0.5 Hz, and calculating the score of each peak;

[0074]

[0075] Step 8 (iv), sort the peaks according to the scores, select the top three peaks with the highest scores, calculate the weighted average frequency of these three peaks, the weight is the FFT amplitude of each peak, and convert the weighted average frequency into the number of breaths per minute, which is the estimated respiratory rate (times / minute).

[0076] Weighted average frequency calculation:

[0077]

[0078] Among them, f i is the frequency of the ith peak, |X(f i )| is the corresponding FFT amplitude.

[0079] Beneficial Effects

[0080] The present invention proposes a contactless breathing detection method based on the preferential selection of subcarrier signals. First, the subcarrier amplitude signal and time information are obtained from the beamforming feedback information (BFI) captured in the Wi-Fi system, and then the antenna combination and subcarrier are evaluated using phase space reconstruction technology and dynamic feature extraction methods, and the k subcarrier signals with the best perception performance are selected to weight and synthesize the optimal breathing detection signal. Next, a series of filtering and dimensionality reduction processes are performed on the selected optimal breathing detection signal, and finally the estimated breathing rate of the target is obtained through fast Fourier transform (FFT) and innovative peak detection algorithm. Through phase space reconstruction and dynamic feature extraction, this method can more comprehensively evaluate the dynamic characteristics of each subcarrier, thereby selecting the subcarrier that is most suitable for breathing detection and improving the accuracy and robustness of breathing detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 It is a flow chart of the present invention;

[0082] Figure 2 It is a schematic diagram of a specific implementation of the human breathing detection method of the present invention. Specific implementation plan

[0083] The present invention will be further described below in conjunction with the accompanying drawings:

[0084] like Figure 1 A non-contact breathing detection method is shown, which specifically includes the following steps:

[0085] Step 1: Place a N AP The access point (AP) with N antennas STA The antenna base station (Station, STA) and a capture device, assuming N AP ≥N STA A subject breathes quietly near the AP and STA. The STA is associated with the AP. The AP periodically sends data frames to the STA. The capture device captures the beamforming feedback frames (BFI) periodically sent from the STA to the AP for target breathing detection.

[0086] Step 2: Decompress and extract the amplitude signal and time information in the BFI frame to obtain N AP× N STA Channel state information for all subcarriers of the antenna combination;

[0087] Step 3: Extract the N AP× N STAThe representative subcarrier signal of the antenna combination is reconstructed in phase space (PSR), and the optimal antenna combination (TX, RX) is evaluated by dynamic feature extraction, which specifically includes the following steps:

[0088] Step 3 (a) select the first n representative subcarriers of each antenna combination according to the subcarrier energy. Subcarriers with high energy usually contain more information and are more representative of channel characteristics.

[0089] Step 3 (ii): Selecting a suitable delay time is the key to phase space reconstruction. The average mutual information (AMI) method is used to calculate the optimal delay time τ for each representative subcarrier. The calculation formula of the average mutual information (AMI) method is:

[0090]

[0091] Where P(X(t),X(t+τ)) is the joint probability distribution of X(t) and X(t+τ), P(X(t)) and P(X(t+τ)) are the marginal probability distributions of X(t) and X(t+τ), respectively. The τ that minimizes I(τ) is selected as the optimal delay time.

[0092] Step 3 (iii): The embedding dimension defines the dimension of the reconstructed phase space. The CAO method is used to determine the optimal embedding dimension m for each representative subcarrier. The specific steps of the CAO method are as follows:

[0093] (1) Calculate E(d):

[0094]

[0095] Among them, y(i,d) is the d-dimensional reconstruction vector, and n(i,d) is the time index of the nearest neighbor of y(i,d);

[0096] (2) Calculate E * (d)

[0097]

[0098] (3)When E * (d) When it no longer changes significantly, the corresponding d is the optimal embedding dimension m.

[0099] Step 3 (4): Calculate the average values ​​of τ and m of all representative subcarriers as new τ and m, and reconstruct the phase space of each representative subcarrier according to the determined delay time τ and embedding dimension m to obtain the reconstructed trajectory. The time series of each subcarrier will be mapped into the m-dimensional phase space. The formula for phase space reconstruction is:

[0100] X(t)=[x(t),x(t+τ),x(t+2τ),…,x(t+(m-1)τ)]

[0101] Among them, X(t) is the original time series, τ is the delay time, and m is the embedding dimension. The phase space reconstruction process is as follows:

[0102] For the time series X(t), t=1,2,…,N, construct an m-dimensional vector:

[0103] X(t)=[x(t),x(t+τ),x(t+2τ),…,x(t+(m-1)τ)]

[0104] Among them, t=1,2,…,N-(m-1)τ.

[0105] In this way, we obtain a trajectory containing N-(m-1)τ m-dimensional vectors, each of which represents a point in the reconstructed phase space.

[0106] Step 3 (V) Analyze the reconstructed phase space and evaluate the dynamic characteristics of each representative subcarrier signal, including periodicity, stability, recurrence quantification analysis (RQA) characteristics and geometric characteristics:

[0107] (1) Periodicity: Perform fast Fourier transform on the phase space trajectory and calculate the average amplitude of the spectrum as a periodicity indicator;

[0108]

[0109] Among them, FFT represents fast Fourier transform, and |·| represents the modulus of a complex number.

[0110] (2) Stability: The standard deviation of the phase space trajectory is calculated as a stability indicator;

[0111]

[0112] where μ is the mean value of X(t).

[0113] (3) Recursive quantitative analysis (RQA) characteristics: including recurrence rate (RR), determinism (DET) and average diagonal line length (L);

[0114] ①Construct a recursive graph R:

[0115] First, choose a suitable threshold ∈, and then for each pair of points i and j, calculate:

[0116] R(i,j)=Θ(∈-||X(i)-X(j)||)

[0117] where Θ is the Heaviside step function, which is 1 if the argument is nonnegative and 0 otherwise.

[0118] ② Calculate the recurrence rate (RR):

[0119]

[0120] ③Calculation certainty (DET):

[0121] First find all the diagonal line segments and record their lengths.

[0122]

[0123] where P(l) is the number of diagonals of length l, l min is the minimum diagonal length (usually 2).

[0124] ④Calculate the average diagonal length (L):

[0125]

[0126] (4) Geometric characteristics: including the average distance from the trajectory to the center of mass and the volume occupied by the trajectory.

[0127]

[0128] Volume = ∏(max(X i )-min(X i ))

[0129] Where C is the center of mass of the trajectory, X i is the i-th dimension coordinate.

[0130] Step 3 (six): standardize the extracted dynamic features, calculate the feature scores, and select the antenna combination with the highest score as the optimal antenna combination according to the feature scores.

[0131] Step 4: Perform Hampel filtering on the subcarrier amplitude signal of the optimal antenna combination to remove outliers;

[0132] Step 5: Perform phase space reconstruction and dynamic feature extraction on all subcarriers of the optimal antenna combination, and use the K-means clustering algorithm to select the best subcarrier S best , to obtain the optimal breathing detection signal, specifically including the following steps:

[0133] Step 5 (i), performing phase space reconstruction and dynamic feature extraction on all subcarriers of the optimal antenna combination according to the delay time τ and embedding dimension m of the representative subcarrier of the optimal antenna combination obtained in step 3;

[0134] Step 5 (ii), standardizing the extracted dynamic features and calculating feature scores;

[0135] Standardization process:

[0136] Feature score calculation: score feature =w1Z1+w2Z2+…+w n Z n

[0137] Among them, w i is the weight of each feature, Z i is the standardized eigenvalue.

[0138] Step 5 (iii), using the K-means clustering algorithm to divide the subcarriers into k categories, selecting the subcarrier with the highest feature score in each category, and taking the selected k subcarriers as the best subcarriers;

[0139] Step 5 (iv), weighting the k subcarriers according to the characteristic scores to synthesize a new subcarrier signal.

[0140] Weighted synthesis formula:

[0141] Among them, S i is the i-th selected subcarrier signal, w i is its corresponding feature score.

[0142] Step 6: Perform Savitzky-Golay filtering on the selected optimal breathing detection signal to smooth the data, and extract the breathing signal frequency band through bandpass filtering;

[0143] Step 7: Perform principal component analysis (PCA) on the filtered signal to reduce its dimension, and perform linear interpolation to obtain the signal S sampled at equal intervals. interp(t) ;

[0144] Step 8: Perform fast Fourier transform (FFT) on the interpolated signal and estimate the respiratory rate through an innovative peak detection algorithm, which specifically includes the following steps:

[0145] Step 8 (a), performing a fast Fourier transform (FFT) on the interpolated signal obtained in step 7;

[0146] Step 8 (ii) perform wavelet denoising on the FFT output signal, and find the peak in the denoised signal. The peak height is required to be no less than 10% of the maximum signal value, the peak spacing is no less than 5% of the frequency axis length, and the protrusion is no less than 30% of the maximum signal value. The peak detection conditions are:

[0147] Height condition: peakheight>0.1×max(|X(f)|)

[0148] Spacing condition: peakdistance>0.05×length(f)

[0149] Prominence condition: peakprominence>0.3×max(|X(f)|)

[0150] Step 8 (iii), screening effective peaks in the frequency range of 0.1 Hz to 0.5 Hz, and calculating the score of each peak;

[0151]

[0152] Step 8 (iv), sort the peaks according to the scores, select the top three peaks with the highest scores, calculate the weighted average frequency of these three peaks, the weight is the FFT amplitude of each peak, and convert the weighted average frequency into the number of breaths per minute, which is the estimated respiratory rate (times / minute).

[0153] Weighted average frequency calculation:

[0154]

[0155] Among them, f i is the frequency of the ith peak, |X(f i )| is the corresponding FFT amplitude.

Claims

1. A non-contact breathing detection method, characterized in that The following steps are involved: Step 1: Place a N AP The access point (AP) with N antennas STA The antenna base station (Station, STA) and a capture device, assuming N AP ≥N STA A subject breathes quietly near the AP and STA. The STA is associated with the AP. The AP periodically sends data frames to the STA. The capture device captures the beamforming feedback frames (BFI) periodically sent from the STA to the AP for target breathing detection. Step 2: Decompress and extract the amplitude signal and time information in the BFI frame to obtain N AP× N STA Channel state information for all subcarriers of the antenna combination; Step 3: Extract the N AP× N STA Perform phase space reconstruction (PSR) on the representative subcarrier signals of the antenna combination, and evaluate the optimal antenna combination (TX, RX) through dynamic feature extraction; Step 4: Perform Hampel filtering on the subcarrier amplitude signal of the optimal antenna combination to remove outliers; Step 5: Perform phase space reconstruction and dynamic feature extraction on all subcarriers of the optimal antenna combination, and use the K-means clustering algorithm to select the best subcarrier S best , get the optimal breathing detection signal; Step 6: Perform Savitzky-Golay filtering on the selected optimal breathing detection signal to smooth the data, and extract the breathing signal frequency band through bandpass filtering; Step 7: Perform principal component analysis (PCA) on the filtered signal to reduce its dimension, and perform linear interpolation to obtain the signal S sampled at equal intervals. interp(t) ; Step 8: Perform Fast Fourier Transform (FFT) on the interpolated signal and estimate the respiratory rate using an innovative peak detection algorithm.

2. A non-contact breathing detection method according to claim 1, characterized in that The step three comprises the following steps: Step 3: Extract the N AP× N STA The representative subcarrier signal of the antenna combination is reconstructed in phase space (PSR), and the optimal antenna combination (TX, RX) is evaluated by dynamic feature extraction, which specifically includes the following steps: Step 3 (a) select the first n representative subcarriers of each antenna combination according to the subcarrier energy. Subcarriers with high energy usually contain more information and are more representative of channel characteristics. Step 3 (ii): Selecting a suitable delay time is the key to phase space reconstruction. The average mutual information (AMI) method is used to calculate the optimal delay time τ for each representative subcarrier. The calculation formula of the average mutual information (AMI) method is: Where P(X(t),X(t+τ)) is the joint probability distribution of X(t) and X(t+τ), P(X(t)) and P(X(t+τ)) are the marginal probability distributions of X(t) and X(t+τ), respectively. The τ that minimizes I(τ) is selected as the optimal delay time. Step 3 (iii): The embedding dimension defines the dimension of the reconstructed phase space. The CAO method is used to determine the optimal embedding dimension m for each representative subcarrier. The specific steps of the CAO method are as follows: (1) Calculate E(d): Among them, y(i,d) is the d-dimensional reconstruction vector, and n(i,d) is the time index of the nearest neighbor of y(i,d); (2) Calculate E * (d) (3) When E * (d) When it no longer changes significantly, the corresponding d is the optimal embedding dimension m. Step 3 (4): Calculate the average values ​​of τ and m of all representative subcarriers as new τ and m, and reconstruct the phase space of each representative subcarrier according to the determined delay time τ and embedding dimension m to obtain the reconstructed trajectory. The time series of each subcarrier will be mapped into the m-dimensional phase space. The formula for phase space reconstruction is: X(t)=[x(t),x(t+τ),x(t+2τ),…,x(t+(m-1)τ)] Among them, X(t) is the original time series, τ is the delay time, and m is the embedding dimension. The phase space reconstruction process is as follows: For the time series X(t), t=1,2,…,N, construct an m-dimensional vector: X(t)=[x(t),x(t+τ),x(t+2τ),…,x(t+(m-1)τ)] Among them, t=1,2,…,N-(m-1)τ. In this way, we obtain a trajectory containing N-(m-1)τ m-dimensional vectors, each of which represents a point in the reconstructed phase space. Step 3 (V) Analyze the reconstructed phase space and evaluate the dynamic characteristics of each representative subcarrier signal, including periodicity, stability, recurrence quantification analysis (RQA) characteristics and geometric characteristics: (1) Periodicity: Perform fast Fourier transform on the phase space trajectory and calculate the average amplitude of the spectrum as a periodicity indicator; Among them, FFT represents fast Fourier transform, and |·| represents the modulus of a complex number. (2) Stability: The standard deviation of the phase space trajectory is calculated as a stability indicator; where μ is the mean value of X(t). (3) Recursive quantitative analysis (RQA) characteristics: including recurrence rate (RR), determinism (DET) and average diagonal line length (L); ①Construct a recursive graph R: First, choose a suitable threshold ∈, and then for each pair of points i and j, calculate: R(i,j)=Θ(∈-||X(i)-X(j)||) where Θ is the Heaviside step function, which is 1 if the argument is nonnegative and 0 otherwise. ② Calculate the recurrence rate (RR): ③Calculation certainty (DET): First find all the diagonal line segments and record their lengths. where P(l) is the number of diagonals of length l, l min is the minimum diagonal length (usually 2). ④Calculate the average diagonal length (L): (4) Geometric characteristics: including the average distance from the trajectory to the center of mass and the volume occupied by the trajectory. Volume = ∏(max(X i )-min(X i )) Where C is the center of mass of the trajectory, X i is the i-th dimension coordinate. Step 3 (six): standardize the extracted dynamic features, calculate the feature scores, and select the antenna combination with the highest score as the optimal antenna combination according to the feature scores.

3. A non-contact breathing detection method according to claim 1, characterized in that The step five comprises the following steps: Step 5: Perform phase space reconstruction and dynamic feature extraction on all subcarriers of the optimal antenna combination, and use the K-means clustering algorithm to select the best subcarrier S best , to obtain the optimal breathing detection signal, specifically including the following steps: Step 5 (a), performing phase space reconstruction and dynamic feature extraction on all subcarriers of the optimal antenna combination according to the delay time τ and embedding dimension m of the representative subcarrier of the optimal antenna combination obtained in step 3; Step 5 (ii), standardizing the extracted dynamic features and calculating feature scores; Standardization process: Feature score calculation: score feature =w1Z1+w2Z2+…+w n Z n Among them, w i is the weight of each feature, Z i is the standardized eigenvalue. Step 5 (iii), using the K-means clustering algorithm to divide the subcarriers into k categories, selecting the subcarrier with the highest feature score in each category, and taking the selected k subcarriers as the best subcarriers; Step 5 (iv) weighting the k subcarriers according to the characteristic scores to synthesize a new subcarrier signal. Weighted synthesis formula: Among them, S i is the i-th selected subcarrier signal, w i is its corresponding feature score.

4. A non-contact breathing detection method according to claim 1, characterized in that The step eight comprises the following steps: Step 8: Perform fast Fourier transform (FFT) on the interpolated signal and estimate the respiratory rate through an innovative peak detection algorithm, which specifically includes the following steps: Step 8 (a), performing a fast Fourier transform (FFT) on the interpolated signal obtained in step 7; Step 8 (ii), perform wavelet denoising on the FFT output signal, and find the peak in the denoised signal. The peak height is required to be no less than 10% of the maximum signal value, the peak spacing is no less than 5% of the frequency axis length, and the protrusion is no less than 30% of the maximum signal value. The peak detection conditions are: Height condition: peak height>0.1×max(|X(f)|) Spacing condition: peak distance>0.05×length(f) Prominence condition: peak prominence>0.3×max(|X(f)|) Step 8 (iii), screening effective peaks in the frequency range of 0.1 Hz to 0.5 Hz, and calculating the score of each peak; Step 8 (iv), sort the peaks according to the scores, select the top three peaks with the highest scores, calculate the weighted average frequency of these three peaks, the weight is the FFT amplitude of each peak, and convert the weighted average frequency into the number of breaths per minute, which is the estimated respiratory rate (times / minute). Weighted average frequency calculation: Among them, f i is the frequency of the ith peak, |X(f i )| is the corresponding FFT amplitude.

Images