A SE and SSD based ultra-wideband radar life detection algorithm

Abstract

The application discloses a kind of based on SE and SSD's ultra-wideband radar life detection algorithm, belong to radar life detection technical field.The method first adopts channel signal subtraction method, direct reduction average method, linear trend suppression, automatic control gain, slow time filtering and so on to the noise signal in original radar echo signal Clutter and pre-processing;Then by calculating the time delay of the SE value peak of radar acceptance pulse along the slow time direction is selected to locate target.Based on SE value selection and recombination these adjacent distance gate signal, more vital sign information can be obtained than single frame signal, and observation time can be reduced.Radar echo signal is decomposed and reconstructed using SSD, according to the energy percentage of the obtained SSCs in the breathing and heartbeat frequency band, the signal is reconstructed, and the frequency of breathing and heartbeat is calculated using FFT.The radar life detection method proposed in the application can accurately locate the target and extract life information, effectively improve the signal-to-noise ratio, and has wide research value and application prospect.

Description

Technical Field

[0001] This invention relates to the field of ultra-wideband radar life detection technology, and in particular to an ultra-wideband radar life detection algorithm based on SE and SSD. Background Technology

[0002] Ultra-wideband radar detection technology is primarily designed for life-saving operations, thus requiring extremely high accuracy, stability, and anti-interference capabilities. Currently, the life detection technologies used worldwide mainly include light wave detection, infrared detection, and acoustic wave detection. While these technologies can detect life signals, they suffer from poor penetration, low accuracy, and weak anti-interference capabilities, making them difficult to apply effectively in complex, harsh, and variable rescue environments and having little impact on accelerating life-saving efforts. Ultra-wideband radar detection technology, as an emerging means of detecting vital signs, essentially emits electromagnetic waves, thus possessing extremely strong penetration capabilities. It can penetrate non-metallic obstacles and reflect off the human body surface. It also boasts advantages such as high range resolution, strong penetration, and strong anti-interference capabilities, and is not easily affected by external environmental factors such as weather, temperature, and light. Therefore, it is an ideal method for detecting vital signs with promising application prospects. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes an ultra-wideband radar life detection algorithm based on SE and SSD.

[0004] The technical solution adopted in this invention is an ultra-wideband radar life detection algorithm based on SE and SSD, and the overall process is as follows: Figure 1 As shown, it includes the following 9 steps.

[0005] Step 1: Establish a mathematical model of vital signs signals. The specific steps are as follows.

[0006] Step 1.1: Assuming that in the detection scenario, when the radar system and the human body reference distance d0 are constant, the distance d(t) between the two can be considered to change with the movement of the chest cavity. For the surrounding static environment, the pulses emitted at different times have the same reflection and refraction law. Therefore, the pulse response is the sum of the human target response and the response of the surrounding environment, which can be expressed as:

[0007]

[0008] Where δ(τ) is the impulse function; a v Let a be the reflection coefficient of the human body to radar waves. i Let τ be the reflection coefficient of the radar wave for the i-th static environmental factor; v and τ i These are the corresponding time delays for the human body and the i-th environmental factor, respectively.

[0009] Step 1.2: Ignoring noise except for minor human movements (i.e., breathing and heartbeat), the measured d(t) should be a constant. According to the basic principle of pulse-type ultra-wideband radar life detection, the instantaneous distance between the radar system's receiving antenna and the human chest can be expressed as:

[0010] d(t)=d0+Δd=d0+A r sin(2πf r t)+A h sin(2πf h t) (2)

[0011] Where d0 is the reference distance of the straight path between the radar antenna front end and the human chest cavity, which can be considered as the positioning distance of the trapped person; A r and A h These represent the changes in chest cavity amplitude caused by breathing and heartbeat in the trapped person, respectively. r The amplitude is approximately 1-2 cm, and the amplitude of chest vibration caused by heartbeat is A. h It is approximately 0.5 mm, because the amplitude change in the chest cavity caused by heartbeat is relatively small. It is generally believed that Δd is mainly caused by respiratory signals; f r f is the breathing rate of the trapped person. h t represents the heart rate; t represents the slow time, which is the actual radar test time.

[0012] Step 1.3: Based on the relationship between the speed, distance, and time of radar waves, the time delay can be expressed as follows.

[0013]

[0014] Where v is the propagation speed of electromagnetic waves (v = 3 × 10⁻⁶) 8 m / s), τ r τ h These represent the time delays in echo signals caused by chest cavity movements resulting from respiratory and heartbeat signals, respectively.

[0015] Step 1.4: Assuming the ultra-wideband pulse radar propagation signal is p(τ), * represents convolution operation, and n(t,τ) is random environmental noise, then the echo signal received by the receiving antenna is...

[0016]

[0017] Step 1.5: Discretize the radar echo signal to obtain the radar echo matrix, where τ=mδ T m = 0, 1, 2…M-1 is the fast time of the echo signal, T f The sampling interval represents the fast time direction; t = nT sn = 0, 1, 2…N-1 is the slow time of the echo signal, T s This indicates the sampling interval in the slow time direction.

[0018]

[0019] Step 2: Preprocess the raw radar echo to remove background clutter and further improve the signal-to-noise ratio. The specific steps are as follows.

[0020] Step 2.1: Subtraction of radar echo signals. The simplest and most direct method to eliminate background clutter in radar echo signals is to subtract each line of radar echo signal from the previous line. During this process, constant values ​​are eliminated. For the radar echo moment R... M×N Assume the fast time series is Y j j = 1, 2, 3, ..., N, where N is the number of sampling points during the scanning time, and the result is obtained by subtracting the channel signals.

[0021] Y j '=Y j+1 -Y j (6)

[0022] Step 2.2: Direct subtraction of average method, take the average value of radar echo data along the scanning time direction, that is, DC component; DC component is generated by stationary objects. Since the distance between the stationary object and the detection device is fixed, while the human body's breathing signal and heartbeat signal are periodically changing, the DC component can be removed by subtracting this average value from the original data. The result of eliminating DC component is:

[0023]

[0024] in, It can be approximated as the DC component in the background clutter.

[0025] Step 2.3: Linear Trend Suppression. In practical applications, radar echo data exhibits a linear trend that changes with slow time. The main influencing factors are time jitter and drift of the triggering unit in the radar system. To eliminate the linear trend, the Linear Trend Suppression (LTS) method is used to estimate the linear trend of each slow time slice and subtract it from the original slow time slice. The formula is as follows.

[0026] W = Ω T -X(X T X) -1 X T Ω T (8)

[0027] Where X = [x1, x2], x1 = [1, 2, ..., N] T x2 = [1,1,…,1] T .

[0028] Step 2.4: Automatic gain control. The amplitude of the radar echo signal caused by the slight movement of the target is mainly related to the lateral distance of the human chest cavity and the relative distance between the human target and the radar. In the actual detection environment, the multipath effect will also interfere with the target echo. Automatic gain control (AGC) is used to enhance the weak vital signs signal in the slow time direction, and the corresponding gain coefficient is calculated according to the energy in the selected time window 2λ+1, so as to achieve adaptive control. Take the echo signal r(τ,n1) of the n1th frame as an example.

[0029]

[0030] r E (τ,t)=g mask (τ,t)×r(τ,t) (10)

[0031] Among them, g mask (τ i ,t) represents the gain coefficient, while r E (t,τ) is the signal after AGC processing.

[0032] Step 2.5: Since the frequency range of human respiration is between 0.1-0.8Hz and the frequency range of heartbeat is between 0.8-2.5Hz, a bandpass filter is used to filter out high-frequency noise signals. The filter is selected as a Butterworth filter, and its square amplitude function is defined as follows.

[0033]

[0034] Step 3: Use the sample entropy method to obtain the location information of the human target in the radar echo matrix. The specific steps are as follows.

[0035] Suppose a fast time sequence m1, and a slow time sequence of length N X[m1,i]={y(1),y(2),…,y(i)}, i=1,2,3,…,N,Y∈W M×N Y forms a vector sequence of dimension q according to its index. q (1),…,Y q (N-q+1), where Y q (i) = {y(i), y(i+1), ..., y(i+q-1)}, 1≤i≤N-q+1, these vectors represent q consecutive values ​​of y starting from the i-th one.

[0036] Define vector Y q(i) and Y q The distance between (j) is d[Y] q (i),Y q [j] represents the absolute value of the maximum difference between the corresponding elements of the two pairs.

[0037] d[Y q (i),Y q [j]=max k＝0,…,m-1 (|x(i+k)-x(j+k)|) (12)

[0038] For a given Y q (i), Statistics Y q (i) and Y q The number of j (1≤j≤Nq, j≠i) whose distance between (j) is less than or equal to r, denoted as B. i For 1≤i≤Nq, define .

[0039]

[0040] Definition B (q) (r) is.

[0041]

[0042] Increase the dimension to q+1 and calculate Y. q+1 (i) and Y q+1 The number of j (1≤j≤Nq, j≠i) whose distance between (j) is less than or equal to r, denoted as A. i For 1≤i≤Nq, define .

[0043]

[0044] Define A (q) (r) is.

[0045]

[0046] Thus, B (q) (r) is the probability of two sequences matching at point q with similarity tolerance r, while A (q) (r) is the probability that two sequences match at point q+1, and the sample entropy is defined as follows.

[0047]

[0048] When q is a finite value, it can be estimated using the following formula.

[0049]

[0050] The value of SampEn ranges from [0,1]. The size of SampEn represents the complexity of the slow time series X[m1,i]. The smaller the value of SampEn, the more regular the changes in the time series; conversely, the larger the value, the more random the changes in the time series.

[0051] In radar echo matrix W M×N In the data, the slow-time directional changes of the human target's location range are relatively regular, resulting in a low SE value for the human target region. Therefore, the location P of the human target can be determined by finding the minimum SE value. pos This allows us to obtain the distance information of the target.

[0052]

[0053] Where v = 3 × 10 8 m / s and T f This indicates the sampling interval in the fast time direction.

[0054] Assume the lateral distance D of the human chest cavity tho Then, based on the sampling interval T in the fast time direction... f The number of points occupied by the distance from the human chest cavity in the received pulse can be calculated, i.e., P. tho =2D tho / vT f Where v = 3 × 10 8 m / s; Let ψ represent the vital signs signal matrix, which can then be expressed as .

[0055]

[0056] Step 4: Range estimation, selecting signals at the human body location and recombining the signals.

[0057] Step 5: Perform bandpass filtering on the combined signal.

[0058] Step 6: Perform SSD decomposition on the preprocessed signal. The specific operation steps are as follows.

[0059] Step 6.1: Construct a new trajectory matrix. Set the embedding dimension and data length of the vital sign signal vector ζ(t) to M and N respectively, and construct an M-row N-column matrix X. The i-th row of matrix X is represented by ζ. i (ζ(i),…,ζ(N),ζ(1),…,ζ(i-1)), i=1,…M, matrix X is.

[0060]

[0061] By moving the bottom right element of matrix X to the top left, we obtain a new trajectory function X(N×K), where K = N - M + 1.

[0062] Step 6.2: Calculate the residual component V at the j-th iteration by adaptively selecting the size M of the embedding dimension. j Power Spectral Density (PSD) of (n), residual component.

[0063]

[0064] And estimate the frequency f corresponding to the maximum peak value of PSD. max .

[0065] In the first iteration, if f max / f s <10 -3 (f s The sampling frequency is 10. -3 (where M is the threshold), and M is set to N / 3, the margin signal can be regarded as a trend component.

[0066] When the number of iterations is greater than 1, the embedding dimension M is set to 1.2 × (f s / f max The period of the separated component is 1.2 times, which is beneficial to improving the identification capability of SSA and ensuring that SSD can give full play to the role of adaptive selection spectrum filter.

[0067] Step 6.3: Reconstruct the singular components after j iterations. When the iteration number j = 1, detect a large trend term from the signal, obtain the matrix through the first left and right eigenvectors, and bisect the matrix diagonally to obtain the signal component g. (1) (n).

[0068] When the iteration number j>1, the time scale component g (j) (n) Frequency components are concentrated in the interval [f max -Δf,f max +Δf], where Δf represents the margin signal v (j) (n) Half of the main peak bandwidth, the left eigenvector in the frequency band [f max -Δf,f max The indices of the principal vectors with prominent spectral peaks in [+Δf] are established as a subset I. j =(I j ={i1,…,i p Then, these principal components are used to construct a matrix, and finally, the corresponding SSC signal component g is obtained by applying the diagonal averaging method to the matrix. (j) (n).

[0069] Step 6.4: Set the stopping condition for signal decomposition iteration. Each iteration of decomposition yields a new singular spectral signal component g.(j) When (n), a new residual signal will also be obtained.

[0070] v (j+1) (n)=v (j) (n)-g (j) (n) (23)

[0071] After j iterations, the formula for calculating the normalized mean square error (NMSE) between the residual signal and the original signal is as follows.

[0072]

[0073] The iterative decomposition ends when the normalized mean square error (NMSE) is less than a given threshold (th = 1%); otherwise, the residual signal v is... (j) (n) is considered a new input signal, and the decomposition process continues until the condition is met and the iteration stops. Assuming there are m singular spectral components, the decomposition result of SSD is...

[0074]

[0075] Step 7: Calculate the percentage of respiratory and heart rate signals.

[0076] Given that the human respiratory frequency band is 0.1Hz to 0.8Hz and the heart rate frequency band is 0.8Hz to 2.8Hz, assuming that the human echo signal is decomposed into m SSCs by SSD, the energy percentage of respiration and heart rate is calculated for each SSC in the frequency domain, as shown in the following formula.

[0077]

[0078]

[0079] Where E(j) is the frequency domain energy of the j-th SSC; E r (j) and E h (j) represent the energy within the respiratory and heartbeat frequency bands of the j-th SSC; δ r and δ h These are the energy ratio thresholds for judging respiration and heartbeat, respectively.

[0080] Step 8: Reconstruct the SSC components that meet the requirements into respiratory and heartbeat signals.

[0081] The SSC components that meet the requirements are reconstructed into respiratory and heartbeat signals.

[0082] S r (t)=∑SSC(t) (28)

[0083] S h(t)=∑SSC(t) (29)

[0084] Step 9: Perform spectral analysis on the reconstructed signal using Fast Fourier Transform (FFT).

[0085] The beneficial effects of the above technical solution are as follows: This invention provides an ultra-wideband radar life detection algorithm based on SE and SSD; the method locates the target by calculating the time delay corresponding to the peak value of the radar received pulse along the slow time direction for echo selection. Since the human body has displacement changes relative to the radar during breathing, the life signal is distributed on the selected echo and its adjacent range gates. Based on the SE value selection and recombination of the signals on these adjacent range gates, more vital sign information can be obtained than a single frame signal, and the observation time can be reduced; SSD is used to decompose and reconstruct the radar echo signal, and the signal is reconstructed according to the energy percentage of SSCs obtained from the decomposition in the breathing and heartbeat frequency bands, and FFT is used to calculate the breathing and heartbeat frequencies. Attached Figure Description

[0086] Figure 1 This is a flowchart of an ultra-wideband radar life detection algorithm based on SE and SSD according to the present invention.

[0087] Figure 2 This is a diagram illustrating the vital signs signal model of the present invention.

[0088] Figure 3 This is a flowchart of the distance detection and echo selection algorithm of the present invention.

[0089] Figure 4 This is the result of the preprocessing algorithm of the present invention.

[0090] Among them, (a) the original radar signal, the result of RPS background clutter removal, the result of linear trend suppression, and the result obtained after processing by the AGC algorithm; (b) the result of LTS processing; and (c) the comparison results before and after slow time filtering.

[0091] Figure 5 The SE algorithm of this invention provides the positioning distance curve.

[0092] Among them, (a) positioning results at 0.35m; (b) positioning results at 1.0m; and (c) positioning results at 1.5m.

[0093] Figure 6 This is an SSD decomposition diagram of the life signal of the present invention.

[0094] Figure 7 This is a diagram showing the energy proportion of the SSC component in the respiratory and cardiac frequency bands of this invention.

[0095] Figure 8This is a diagram showing the reconstructed respiratory and heartbeat signals from the present invention.

[0096] Figure 9 This is a spectrum diagram of the respiratory and heartbeat signals of the present invention.

[0097] Among them, (a) is the spectrum of respiratory signals; and (b) is the spectrum of heartbeat signals. Detailed Implementation

[0098] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0099] In actual radar detection, radar echo signals mainly include the vital signs of the target human body, as well as noise interference and other clutter signals from the surrounding environment. The main purpose of life detection is to extract vital sign information from complex echo signals based on the characteristics of human vital signs such as breathing and heartbeat. The overall algorithm flow is as follows: Figure 1 As shown, in the preprocessing stage, background clutter is first simply eliminated by subtracting the channel signal; the DC component in the background clutter is eliminated by direct subtraction averaging; the linear trend suppression method is used to eliminate the linear trend in the radar echo data; then, multipath effect is reduced by automatic gain control, and weak respiratory and heartbeat signals are enhanced from the slow time direction to improve the signal-to-noise ratio of human vital signs; finally, high-frequency noise in the echo signal is filtered out by Butterworth low-pass filter; then, the echo selection is performed to locate the target by calculating the time delay corresponding to the peak value of the radar received pulse along the slow time direction; finally, the SSD algorithm is used to decompose and reconstruct the radar echo signal, and the signal is reconstructed based on the energy percentage of SSCs in the respiratory and heartbeat frequency bands obtained from the decomposition, and the frequencies of respiratory and heartbeat are calculated by FFT.

[0100] The vital sign signal model in this embodiment is as follows: Figure 2 As shown, the specific steps are as follows.

[0101] Assuming that in the detection scenario, when the radar system and the human body reference distance d0 are constant, the distance d(t) between them can be considered to change with the movement of the chest cavity; and for the surrounding static environment, the pulses emitted at different times have the same reflection and refraction law, so the pulse response is the sum of the human target response and the response of the surrounding environment, which can be expressed as.

[0102]

[0103] Where δ(τ) is the impulse function; a v Let a be the reflection coefficient of the human body to radar waves. i Let τ be the reflection coefficient of the radar wave for the i-th static environmental factor; v and τ iThese are the corresponding time delays for the human body and the i-th environmental factor, respectively.

[0104] Ignoring noise and excluding micro-movements of the human body (i.e., breathing and heartbeat), the measured d(t) should be a constant. According to the basic principle of pulse-type ultra-wideband radar life detection, the instantaneous distance between the radar system's receiving antenna and the human chest can be expressed as:

[0105] d(t)=d0+Δd=d0+A r sin(2πf r t)+A h sin(2πf h t) (2)

[0106] Where d0 is the reference distance of the straight path between the radar antenna front end and the human chest cavity, which can be considered as the positioning distance of the trapped person; A r and A h These represent the changes in chest cavity amplitude caused by breathing and heartbeat in the trapped person, respectively. r The amplitude is approximately 1-2 cm, and the amplitude of chest vibration caused by heartbeat is A. h It is approximately 0.5 mm, because the amplitude change in the chest cavity caused by heartbeat is relatively small. It is generally believed that Δd is mainly caused by respiratory signals; f r f is the breathing rate of the trapped person. h t represents the heart rate; t represents the slow time, which is the actual radar test time.

[0107] Based on the relationship between the speed, distance, and time of radar waves, the time delay can be expressed as:

[0108]

[0109] Where v is the propagation speed of electromagnetic waves (v = 3 × 10⁻⁶) 8 m / s), τ r τ h These represent the time delays in echo signals caused by chest cavity movements resulting from respiratory and heartbeat signals, respectively.

[0110] Assuming the propagation signal of the ultra-wideband pulse radar is p(τ), * represents convolution operation, and n(t,τ) is random environmental noise, then the echo signal received by the receiving antenna is...

[0111]

[0112] The radar echo matrix is ​​obtained by discretizing the radar echo signal.

[0113]

[0114] Where τ=mδ Tm = 0, 1, 2…M-1 is the fast time of the echo signal, T f The sampling interval represents the fast time direction; t = nT s n = 0, 1, 2…N-1 is the slow time of the echo signal, T s This indicates the sampling interval in the slow time direction.

[0115] The result of the preprocessing algorithm in this embodiment is as follows: Figure 4 As shown, the specific steps are as follows.

[0116] Channel signal subtraction method: The simplest and most direct way to eliminate background clutter in radar echo signals is to subtract each line of radar echo signal from the previous line. The constant value maintained during this process will be eliminated. For radar echo moment R... M×N Assume the fast time series is Y j j = 1, 2, 3, ..., N, where N is the number of sampling points during the scanning time, and the result is obtained by subtracting the channel signals.

[0117] Y j '=Y j+1 -Y j (6)

[0118] Direct subtraction averaging method: Take the average value of the radar echo data along the scanning time direction, which is the DC component; the DC component is generated by stationary objects. Since the distance between the stationary object and the detection device is fixed, while the human body's breathing and heartbeat signals change periodically, the DC component can be removed by subtracting this average value from the original data. The result of eliminating the DC component is:

[0119]

[0120] in, It can be approximated as the DC component in the background clutter.

[0121] Linear Trend Suppression: In practical applications, radar echo data exhibits a linear trend that changes with slow time. The main influencing factors are time jitter and drift of the triggering unit in the radar system. To eliminate the linear trend, the Linear Trend Suppression (LTS) method is used. The linear trend of each slow time slice is estimated and subtracted from the original slow time slice. The formula is as follows.

[0122] W = Ω T -X(X T X) -1 X T Ω T (8)

[0123] Where X = [x1, x2], x1 = [1, 2, ..., N] T x2 = [1,1,…,1] T .

[0124] Automatic gain control: The amplitude of the radar echo signal caused by the slight movement of the target is mainly related to the lateral distance of the human chest cavity and the relative distance between the human target and the radar. The multipath effect in the actual detection environment will also interfere with the target echo. Automatic gain control (AGC) is used to enhance the weak vital signs signal in the slow time direction, and the corresponding gain coefficient is calculated according to the energy in the selected time window 2λ+1, thereby realizing adaptive control. Take the echo signal r(τ,n1) of the n1th frame as an example.

[0125]

[0126] r E (τ,t)=g mask (τ,t)×r(τ,t) (10)

[0127] Among them, g mask (τ i ,t) represents the gain coefficient, while r E (t,τ) is the signal after AGC processing.

[0128] Slow-time filtering: Since the frequency range of human respiration is between 0.1-0.8Hz and the frequency range of heartbeat is between 0.8-2.5Hz, a bandpass filter is used to filter out high-frequency noise signals. The filter is chosen as a Butterworth filter, and its squared amplitude function is defined as follows.

[0129]

[0130] The distance detection and echo selection algorithm flow of this embodiment is as follows: Figure 3 As shown, the specific steps are as follows.

[0131] Suppose a fast time sequence m1, and a slow time sequence of length N X[m1,i]={y(1),y(2),…,y(i)}, i=1,2,3,…,N,Y∈W M×N Y forms a vector sequence of dimension q according to its index. q (1),…,Y q (N-q+1), where Y q (i) = {y(i), y(i+1), ..., y(i+q-1)}, 1≤i≤N-q+1, these vectors represent q consecutive values ​​of y starting from the i-th one.

[0132] Define vector Y q(i) and Y q The distance between (j) is d[Y] q (i),Y q [j] represents the absolute value of the maximum difference between the corresponding elements of the two pairs.

[0133] d[Y q (i),Y q [j]=max k＝0,...,m-1 (|x(i+k)-x(j+k)|) (12)

[0134] For a given Y q (i), Statistics Y q (i) and Y q The number of j (1≤j≤Nq, j≠i) whose distance between (j) is less than or equal to r, denoted as B. i For 1≤i≤Nq, define .

[0135]

[0136] Definition B (q) (r) is.

[0137]

[0138] Increase the dimension to q+1 and calculate Y. q+1 (i) and Y q+1 The number of j (1≤j≤Nq, j≠i) whose distance between (j) is less than or equal to r, denoted as A. i For 1≤i≤Nq, define .

[0139]

[0140] Define A (q) (r) is.

[0141]

[0142] Thus, B (q) (r) is the probability of two sequences matching at point q with similarity tolerance r, while A (q) (r) is the probability that two sequences match at point q+1, and the sample entropy is defined as follows.

[0143]

[0144] When q is a finite value, it can be estimated using the following formula.

[0145]

[0146] The value of SampEn ranges from [0,1]. The size of SampEn represents the complexity of the slow time series X[m1,i]. The smaller the value of SampEn, the more regular the changes in the time series; conversely, the larger the value, the more random the changes in the time series.

[0147] In radar echo matrix W M×N In the data, the slow-time directional changes of the human target's location range are relatively regular, resulting in a low SE value for the human target region. Therefore, the location P of the human target can be determined by finding the minimum SE value. pos This allows us to obtain the distance information of the target.

[0148]

[0149] Where v = 3 × 10 8 m / s and T f This indicates the sampling interval in the fast time direction.

[0150] Assume the lateral distance D of the human chest cavity tho Then, based on the sampling interval T in the fast time direction... f The number of points occupied by the distance from the human chest cavity in the received pulse can be calculated, i.e., P. tho =2D tho / vT f Where v = 3 × 10 8 m / s. Let ψ represent the vital signs signal matrix, which can then be expressed as...

[0151]

[0152] The SE algorithm positioning distance curve in this embodiment is as follows: Figure 5 As shown, in the slow-time direction of radar echo reception, the signal complexity of the human body location range differs from that of the non-human body location range. Furthermore, human breathing and heartbeat signals exhibit more regular and predictable changes than background noise. Therefore, the PE value is higher in the non-human body location range and lower in the human body location range. As shown in the figure, the estimated distances for 0.35, 1.0, and 1.5 m are 0.3457 m, 1.0163 m, and 1.5062 m, respectively, with corresponding measurement errors of 0.0043 m, 0.0163 m, and 0.0062 m.

[0153] The SSD decomposition diagram of the vital signals in this embodiment is as follows: Figure 6 As shown, the specific steps are as follows.

[0154] Constructing a new trajectory matrix: Given that the embedding dimension and data length of the vital sign signal vector ζ(t) are M and N respectively, construct an M x N matrix X, where the i-th row of matrix X represents ζ. i(ζ(i),…,ζ(N),ζ(1),…,ζ(i-1)), i=1,…M, matrix X is.

[0155]

[0156] By moving the bottom right element of matrix X to the top left, we obtain a new trajectory function X(N×K), where K = N - M + 1.

[0157] By adaptively selecting the size M of the embedding dimension, the residual component V is calculated at the j-th iteration. j Power Spectral Density (PSD) of (n), residual component.

[0158]

[0159] And estimate the frequency f corresponding to the maximum peak value of PSD. max .

[0160] In the first iteration, if f max / f s <10 -3 (f s The sampling frequency is 10. -3 (where M is the threshold), and M is set to N / 3, the margin signal can be regarded as a trend component.

[0161] When the number of iterations is greater than 1, the embedding dimension M is set to 1.2 × (f s / f max The period of the separated component is 1.2 times, which is beneficial to improving the identification capability of SSA and ensuring that SSD can give full play to the role of adaptive selection spectrum filter.

[0162] Reconstructing the singular components after j iterations: When the iteration number j=1, a large trend term is detected from the signal. The matrix is ​​obtained through the first left and right eigenvectors. The signal component g is obtained by diagonally bisecting this matrix. (1) (n).

[0163] When the iteration number j>1, the time scale component g (j) (n) Frequency components are concentrated in the interval [f max -Δf,f max +Δf]. Where Δf represents the margin signal v. (j) (n) Half of the main peak bandwidth, the left eigenvector in the frequency band [f max -Δf,f max The indices of the principal vectors with prominent spectral peaks in [+Δf] are established as a subset I. j =(I j ={i1,…,ip Then, these principal components are used to construct a matrix, and finally, the corresponding SSC signal component g is obtained by applying the diagonal averaging method to the matrix. (j) (n).

[0164] Signal decomposition iteration stopping condition setting: Each iteration of decomposition obtains a new singular spectrum signal component g. (j) When (n), a new residual signal will also be obtained.

[0165] v (j+1) (n)=v (j) (n)-g (j) (n) (23)

[0166] After j iterations, the formula for calculating the normalized mean square error (NMSE) between the residual signal and the original signal is as follows.

[0167]

[0168] The iterative decomposition ends when the normalized mean square error (NMSE) is less than a given threshold (th = 1%); otherwise, the residual signal v is... (j) (n) is considered a new input signal, and the decomposition process continues until the condition is met and the iteration stops. Assuming there are m singular spectral components, the decomposition result of SSD is...

[0169]

[0170] In this embodiment, the energy proportion of the SSC component within the respiratory and cardiac frequency bands is as follows: Figure 7 As shown in the figure, SSC9, SSC10, and SSC11 all account for more than 90% of the energy in the respiratory frequency band, so these three SSC components can be used to reconstruct the respiratory signal. For the heartbeat signal, only SSC6 accounts for more than 90% of the energy in the heartbeat frequency band, while the other SSCs account for a relatively small percentage of the energy in the heartbeat frequency band, indicating that they contribute less to the reconstruction of the heartbeat signal. Therefore, SSC6 can be used to reconstruct the heartbeat signal.

[0171] The reconstructed respiratory and heart rate signals of this embodiment are as follows: Figure 8 As shown, the specific steps are as follows.

[0172] Given that the human respiratory frequency band is 0.1Hz to 0.8Hz and the heart rate frequency band is 0.8Hz to 2.8Hz, assuming that the human echo signal is decomposed into m SSCs by SSD, the energy percentage of respiration and heart rate is calculated for each SSC in the frequency domain, as shown in the following formula.

[0173]

[0174]

[0175] Where E(j) is the frequency domain energy of the j-th SSC; E r (j) and E h (j) represent the energy within the respiratory and heartbeat frequency bands of the j-th SSC; δ r and δ h These are the energy ratio thresholds for judging respiration and heartbeat, respectively; the SSC components that meet the requirements are reconstructed into respiration and heartbeat signals.

[0176] S r (t)=∑SSC(t) (28)

[0177] S h (t)=∑SSC(t) (29)

[0178] After reconstructing the respiratory and heartbeat signals, frequency domain analysis was performed using the FFT method. The spectra of the respiratory and heartbeat signals in this embodiment are as follows: Figure 9 As shown.

[0179] The performance of the EEMD algorithm and the SSD algorithm was compared using the same set of radar echo signals. The comparison results are shown in Table 1, where F r F is the respiratory rate. h Heart rate, SNR r SNF represents the signal-to-noise ratio of respiratory signals. h This indicates the ratio of heartbeat signals.

[0180] Table 1

[0181]

[0182] The proposed ultra-wideband radar life detection algorithm based on SE and SSD has significantly improved performance compared to some existing life detection methods. Compared with EEMD, it improves the accuracy and real-time performance of signal separation and effectively enhances the signal-to-noise ratio of respiratory and heartbeat signals.

[0183] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these are merely illustrative examples, and various changes or modifications can be made to these embodiments without departing from the principles and essence of the present invention. The scope of the present invention is defined only by the appended claims.

Claims

1. A SE and SSD based ultra-wideband radar life detection method, characterized by, The method comprises the following 9 steps: Step 1: a vital sign signal mathematical model is established; Step 2: the original radar echo is preprocessed to remove background clutter, and the signal-to-noise ratio is further improved; Step 3: the sample entropy method is used to obtain the position information of the human body target in the radar echo matrix; Step 4: range estimation, selecting the signal at the position of the human body and recombining the signal; Step 5: the combined signal is processed by band-pass filtering, and then SSD decomposition is performed to calculate the percentage of the respiratory signal and the heartbeat signal; According to the human respiratory frequency band of 0.1Hz-0.8Hz and the heartbeat frequency band of 0.8Hz-2.8Hz, it is assumed that the human body echo signal is decomposed into m SSCs by SSD, and the energy percentage of the respiratory and heartbeat is calculated for each SSC in the frequency domain, as follows: where E(j) is the frequency energy of the jthSSC; E r (j) and E h (j) are the energies in the respiratory and heartbeat bands in the jthSSC, respectively; δ r and δ h are the energy ratio thresholds for judging the respiration and heartbeat, respectively. Step 6: the SSC components meeting the requirements are reconstructed into respiratory and heartbeat signals; Step 7: fast Fourier transform (FFT) is used to perform frequency spectrum analysis on the reconstructed signal.

2. The SE and SSD based ultra-wideband radar life detection method according to claim 1, characterized in that, The process of step 1 is as follows: Step 1.1: it is assumed that when the radar system and the human body reference distance d0 are certain in the detection scene, the distance d(t) between them changes with the movement of the chest; and for the surrounding static environment, the pulses emitted at different times have the same reflection refraction rule, so the pulse response is the sum of the human body target response and the surrounding environment response, which can be represented as: where δ(τ) is an impulse function; a v is the reflection coefficient of the human body to the radar wave, a i is the reflection coefficient of the ith stationary environmental factor to the radar wave; τ v and τ i are the corresponding time delays of the human body and the ith environmental factor, respectively. Step 1.2: ignoring the noise except for the human body micro-motion, i.e. the respiratory and heartbeat movement, the measured d(t) should be a constant value, according to the basic principle of pulse-type ultra-wideband radar life detection, the instantaneous distance between the receiving antenna of the radar system and the human chest can be represented as: d(t) = d0+ Ad = d0+ A r sin(2πf r t) + A h sin(2πf h t) (2) where d0 is the reference distance of the straight path between the front end of the radar antenna and the chest cavity of the human body, which is considered as the positioning distance of the trapped person; A r and A h respectively represent the chest cavity amplitude changes caused by breathing and heartbeat of the trapped person, A r The amplitude of the breathing signal is 1-2 cm, and the amplitude of the chest vibration caused by the heartbeat A h is 0.5 mm, because the relative change of the chest amplitude caused by the heartbeat is small, mainly caused by the breathing signal; f r is the breathing frequency of the trapped person, f h is the heartbeat frequency; t is the slow time, that is, the actual radar test time; Step 1.3: according to the relationship among the speed, distance and time of the radar wave, the time delay can be represented as: where v is the propagation speed of the electromagnetic waves, τ r , τ h are the time delays of the echo signals caused by the chest movements due to the respiration and heartbeat signals, respectively. Step 1.4: assuming that the ultra-wideband pulse radar propagation signal is p(τ), * represents convolution operation, and n(t,τ) is random environmental noise, then the echo signal received by the receiving antenna is: Step 1.5: the radar echo matrix is obtained by discretely processing the radar echo signal: where τ = mδ T , m = 0, 1, 2...M-1 is the echo signal fast time, T f represents the sampling interval in the fast time direction; t = nT s , n = 0, 1, 2...N-1 is the echo signal slow time, T s represents the sampling interval in the slow time direction.

3. The SE and SSD based ultra-wideband radar life detection method of claim 1, wherein, The process of step 2 is as follows: Step 2.1: The simplest and direct method to eliminate the background clutter in the radar echo signal is to subtract each row of the radar echo signal from the previous row of the radar echo signal, and the constant quantity will be eliminated in this process; let the sampling point number in the fast time direction be M, and for the radar echo matrix R M×N , assuming that the fast time sequence is Y j , j = 1, 2, 3, …, N, N is the sampling point number of the scanning time, and the result of the channel signal subtraction method is: Y j = Y j+1 - Y j (6) Step 2.2: direct average method, the average value of the radar echo data along the scanning time direction is taken, i.e. the direct current component; the direct current component is generated by the stationary object, and the distance between the stationary object and the detection device is fixed and unchanged, while the respiratory signal and the heartbeat signal of the human body change periodically, so the direct current component can be removed by subtracting the average value from the original data, and the result of eliminating the direct current component is: wherein Approximately regarded as the direct current component in the background clutter; Step 2.3: linear trend suppression, in actual application, the radar echo data has a linear trend changing with the slow time, and the main influencing factors are the time jitter and drift of the trigger unit in the radar system; in order to eliminate the linear trend, the linear trend suppression (LTS) method is used to estimate the linear trend of each slow time slice and subtract it from the original slow time slice, and the formula is: W = Ω T - X (X T X) -1 X T Ω T (8) where X = [x1, x2], x1 = [1, 2,..., N] T , x2 = [1, 1,..., 1] T ; Step 2.4: Automatic control gain, the amplitude of the radar echo signal caused by the target micro-motion is mainly related to the transverse distance of the human chest and the relative distance between the human target and the radar, and the multipath effect in the actual detection environment will also interfere with the target echo; an automatic gain control (AGC) is used to enhance the weak vital sign signal in the slow time direction, and the corresponding gain coefficient is calculated according to the energy in the selected time window 2λ+1, so as to realize adaptive control, taking the nth frame of echo signal r(τ,n1) as an example: r E (τ,t) = g mask (τ,t) x r(τ,t) (10) where g mask (τ i ,t) represents a gain coefficient, while r E (t,τ) is the signal after AGC processing; Step 2.5: Since the frequency range of human respiration is between 0.1-0.8Hz, and the frequency range of heartbeat is between 0.8-2.5Hz, a band-pass filter is used to filter out high-frequency noise signals, and the Butterworth filter is selected, whose square amplitude function is defined as:

4. The SE and SSD based ultra-wideband radar life detection method of claim 1, wherein, The process of step 3 is as follows: Assume fast time m1, for a slow time sequence X[m1, i] = {y(1), y(2), …, y(i)} of length N and sampling points M, i = 1, 2, 3, …, N, Y ∈ W M×N , according to the serial number to form a group of vector sequences with dimension q, Y q (1), …, Y q (N-q+1), where Y q (i) = {y(i), y(i+1), …, y(i+q-1)}, 1 ≤ i ≤ N-q+1, these vectors represent the values of q consecutive y starting from the i-th. The distance d[Y q (i), Y q (j)] between the vectors Y q (i) and Y q (j) is the absolute value of the maximum difference of the corresponding elements: d[Y q (i),Y q (j)] = max k＝0,…,m-1 (|x(i+k)-x(j+k)|) (12) For a given Y q (i), count Y q (i) and Y q (j) between which the distance is less than or equal to r, 1≤j≤N-q, j≠i, and record it as B i For 1≤i≤N-q, define: Definition B (q) (r) is: increasing the dimension to q + 1, statistics Y q+1 (i) and Y q+1 the number of j, 1≤j≤N-q, j≠i, whose distance between (i) and Y i For 1≤i≤N-q, define: Definition A (q) (r) is: Thus, B (q) (r) is the probability of matching q points of the two sequences under similarity r, while A (q) (r) is the probability of matching q+1 points of the two sequences, and the sample entropy is defined as: When q is a finite value, it can be estimated by the following formula: The value range of SampEn is [0, 1], and the size of SampEn represents the complexity of the slow-time sequence X[m1,i], the smaller the value of SampEn, the more regular the time series changes, and vice versa, the more random the time series changes; In the radar echo matrix W M×N The slow time direction data change of the human target position range is relatively regular, so the SE value corresponding to the human target area is low, and therefore the position P of the human target can be determined by finding the minimum value of the SE value pos , thereby obtaining the distance information of the target: where v = 3 x 10 8 m / s and T f denotes the sampling interval in the fast time direction; Assume the lateral distance D of the human chest tho ; then according to the sampling interval T in the fast time direction f , the number of points occupied by the distance of the human chest in the received pulse can be calculated, that is, P tho = 2D tho / vT f , where v = 3 x 10 8 m / s; the vital sign signal matrix is represented by ψ, which can be calculated from the radar echo matrix:

5. The SE and SSD based ultra-wideband radar life detection method of claim 1, wherein, The process of step 6 is as follows: Step 6.1: Constructing a new trajectory matrix, set the embedding dimension and data length of the vital signs signal vector ζ(t) as M and N respectively, construct an M by N matrix X, the i-th row of the matrix X is expressed as ζ i (ζ(i),…,ζ(N),ζ(1),…,ζ(i-1)),i=1,…M, the matrix X is: The right lower corner elements of the X matrix are moved to the upper left corner to obtain a new trajectory function X, with a dimension of N×K, where K=N-M+1; Step 6.2: Compute the residual component V at the jth iteration by adaptively selecting the size M of the embedding dimension j (n) of the Power Spectral Density (PSD) of the residual component: and estimate the frequency f corresponding to the maximum peak of the PSD max ; In the first iteration, if f max / f s <10 -3 , f s is the sampling frequency, 10 -3 is the threshold value, M is set as N / 3, and the residual signal can be regarded as a trend component; When the iteration number is greater than 1, set the embedding dimension M = 1.2 x (f s / f max ) is 1.2 times the separated component period, which is conducive to improving the SSA identification ability and ensuring that the SSD fully plays the role of adaptive selection of spectral filters; Step 6.3: Reconstructing the singular principal components after j iterations, when the iteration number j = 1, a large trend item is detected from the signal, the matrix is obtained by the first left and right eigenvectors, and the signal component g is obtained by diagonal partitioning the matrix (1) (n); When the iteration number j>1, the time scale component g (j) (n) Frequency components are concentrated in the interval [f max -Δf,f max +Δf], where Δf represents the margin signal v (j) (n) Half of the main peak bandwidth, the left eigenvector in the frequency band [f max -Δf,f max The indices of the principal vectors with prominent spectral peaks in [+Δf] are established as a subset I. j =(I j ={i1,…,i p Then, these principal components are used to construct a matrix, and finally, the corresponding SSC signal component g is obtained by applying the diagonal averaging method to the matrix. (j) (n); Step 6.4: Setting the iteration stop condition for signal decomposition, obtaining new singular spectrum signal components g at each iteration (j) (n) also a new residual signal: v (j+1) (n) = v (j) (n) - g (j) (n) (23) After j iterations, the normalized mean squared error NMSE (Normalized Mean Squared Error, NMSE) of the residual signal and the original signal is calculated as follows: The iteration decomposition ends when the normalized mean square error (NMSE) of the angelica is less than a given threshold value, th = 1%; otherwise, the residual signal v (j) (n) Continue the decomposition process as a new input signal until the condition is met to stop iteration, assuming that there are m singular spectral components, and the decomposition result of the SSD is:

Images