Human body target positioning method of singular spectrum analysis method based on tensor decomposition
By employing a singular spectrum analysis method based on tensor decomposition, the problem of the significant impact of respiratory signal frequency estimation on human target localization was solved. EEMD-supervised tensor decomposition was used to process non-stationary signals, improving the signal-to-noise ratio and ensuring the accuracy of human target localization and detection performance in complex environments.
Patent Information
- Application Number
- CN202411113331.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-14
- Publication Date
- 2024-11-29
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In existing human target localization methods based on multi-channel singular spectrum analysis, the estimation of respiratory signal frequency has a significant impact on the human target localization algorithm, resulting in an unsatisfactory or decreased signal-to-noise ratio (SNCR).
A singular spectrum analysis method based on tensor decomposition is adopted. By preprocessing, embedding tensors, decomposing and reconstructing UWB bioradar echo signals, the tensor decomposition process is supervised by ensemble empirical mode decomposition (EEMD) to replace the traditional singular value decomposition (SVD) for processing non-stationary signals.
It improves the signal-to-noise ratio (SNCR), ensuring the accuracy of human target localization, effectively suppressing clutter interference, and improving detection accuracy in complex environments.
Smart Images

Figure CN119044958A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of human body positioning in emergency search and rescue, and in particular to a human body target positioning method based on a singular spectrum analysis method of tensor decomposition. Background Art
[0002] Bio-radar is a new type of radar that uses radar as a technical means to conduct non-contact detection, positioning and imaging of living organisms. Bio-radar emits electromagnetic waves, which can penetrate non-metallic media of a certain thickness, such as brick walls, prefabricated panels, clothing, etc. The reflected waves are modulated by the physiological activities of the living organism, such as breathing, heartbeat, pulse, body movement, etc., and a series of signal processing is performed on the echo signal to extract the physiological parameters, movement trajectory and other life information of the living organism. Bio-radar technology has the technical advantages of non-contact, all-day, and all-weather detection.
[0003] There are many methods for locating human targets based on bio-radar, some are based on energy, and some are based on parameterized bispectral estimation. In 2015, our research group studied a human target positioning method based on Multichannel Singular Spectrum Analysis (MSSA). Compared with the traditional singular value decomposition method and adaptive spectral line enhancement method, this method has a better ability to improve the signal-to-noise ratio.
[0004] However, when performing singular spectrum analysis, subspace selection is still an open problem, and the selection of reference signals is of great significance. In the previous work of our research group, a period delay of the breathing signal was selected as the reference signal, but this depends on the accurate estimation of the signal frequency. When the frequency estimation is inaccurate, the signal-to-noise ratio (SNCR) of the obtained result is not ideal. When the frequency estimation error is large, the SNCR is even less than 0. Therefore, when the breathing signal is selected as the reference signal, the estimation of the breathing signal frequency has a greater impact on the positioning algorithm of the human target.
[0005] Therefore, it is necessary to propose a method that does not require a preset reference signal to realize the subspace selection and reconstruction process according to the properties of the signal itself, improve the SNCR, and ensure the accuracy of human target positioning. Summary of the invention
[0006] The present invention aims to provide a human target positioning method based on a singular spectrum analysis method of tensor decomposition, so as to solve the problem that the estimation of the breathing signal frequency in the human target positioning method based on a multi-channel singular spectrum analysis method has a great influence on the positioning algorithm of the human target.
[0007] In order to achieve the above object, the present invention provides the following technical solutions:
[0008] A human target positioning method based on a singular spectrum analysis method of tensor decomposition, the human target positioning method based on a singular spectrum analysis method of tensor decomposition comprises the following steps:
[0009] S1. Ultra-wideband (UWB) bio-radar echo signal acquisition and preprocessing;
[0010] S2, embed the preprocessed one-dimensional time series into a tensor;
[0011] S3, performing tensor decomposition;
[0012] S4, group reconstruction;
[0013] S5. Hankel matrix.
[0014] Furthermore, in step S1, the human micro-motion model caused by human life activities, namely breathing and heartbeat, relative to the radar transceiver antenna is:
[0015] d(t)=d0+m(t)=d0+m b sin(2πf b t)+m h sin(2πf h t)
[0016] Where d0 indicates the distance between the human target position and the transmitting and receiving antenna, m b , m h are the amplitudes of breathing and heartbeat, respectively, b , f h are the frequencies of breathing and heartbeat respectively. The two-dimensional echo signal collected by the radar can be expressed as:
[0017]
[0018] p(t) is the normalized transmission waveform signal, A i is the amplitude of the i-th multipath component, τ i is the signal delay caused by the i-th multipath component, A is the amplitude of body motion, τ d (t) is the signal delay caused by d(t), and its expression is:
[0019]
[0020] Where c is the speed of light, sampling r(t,τ) yields R[m,n]=r(τ=mT f ,t=nT s ), T f ,T s The sampling intervals for the fast and slow time dimensions, respectively;
[0021] UWB bio-radar system uses time series τ = mT in fast time dimension f (m=1,2,…M) samples the received waveform, and in the slow time dimension at discrete moments t=nT s (n=1,2,…N) sampling pulse waveform, the sampled radar echo signal is stored as a two-dimensional array R of size M×N.
[0022] Furthermore, in step S1, each signal in the UWB bio-radar echo signal is smoothed and filtered, mainly to remove the high-frequency interference introduced on the signal, and its mathematical expression is:
[0023]
[0024] Where l = 1, 2, ..., L, L = ML f +1, L f is the filter order, the optimal filter order L f =85.
[0025] Furthermore, in step S1, the distance accumulation of the smoothed and filtered echo signal is performed on the matrix along the distance direction, which can be expressed as:
[0026]
[0027] Where L is the accumulation window width along the distance direction, k (k = 1, 2, 3, ... K) is the coordinate of the distance dimension after distance accumulation, and K is the number of sampling points in the distance direction after accumulation. symbol Indicates rounding down.
[0028] Furthermore, in step S2, the point signal x is segmented into matrices with l as non-overlapping time windows The size of the matrix is The relationship between the matrix and the time series is as follows
[0029]
[0030] in Then convert the matrix into a tensor, that is, After adding windows, each layer of the tensor is formed, which is mathematically expressed as:
[0031]
[0032] l1-o is the overlap size of adjacent windows, and l1 is the window size.
[0033] Furthermore, in step S3, the tensor is subjected to canonical or parallel factor analysis (CANDECOMP / PARAFAC), and the basic formula of CP decomposition is:
[0034]
[0035] where x ijk is a third-order tensor The (i,j,k)th element of , R is the order of the model.
[0036] Furthermore, in step S3, the subgradient method is used to solve g ijk , the iteration equation is
[0037]
[0038] in
[0039]
[0040] μ is the iteration step size.
[0041] Furthermore, regarding the reference tensor f ijk , the EEMD method is used to add the intrinsic mode functions falling within the respiratory frequency band as the reference signal, and embedding and other operations are performed on it to make it tensorized, and the grouping and reconstruction after the tensor decomposition are supervised. The specific steps of the EEMD are as follows:
[0042] 1) Add white noise of a certain intensity to the original signal sequence;
[0043] 2) Perform EMD decomposition on the noisy signal to obtain a series of IMF components;
[0044] 3) Add white noise of different intensities to the original signal sequence, repeat steps 1) and 2) for a total of M times;
[0045] 4) The IMF components obtained by M-times EMD decomposition are averaged to obtain the final IMF of the original signal sequence, which is expressed as:
[0046]
[0047] Further, in step S5, let Represents the one-dimensional data signal obtained after mapping, which can be obtained by diagonal averaging. The mathematical expression is
[0048]
[0049] Compared with the prior art, the present invention has the following beneficial effects:
[0050] The present invention selects tensor decomposition to replace the singular value decomposition (SVD) process in singular spectrum analysis. The method is more suitable for processing non-stationary signals. In subspace selection and reconstruction, the ensemble empirical mode method (EEMD) is used to supervise the tensor decomposition process. The traditional singular spectrum analysis (SSA) method does not take the inherent non-stationary characteristics of the signal into consideration and cannot correctly group and reconstruct the signal. The tensor decomposition replaces the singular value decomposition (SVD) process in the traditional singular spectrum analysis (SSA) to process such signals well. In addition, EEMD plays a role in supervising the reconstruction process in tensor decomposition, and also avoids the decrease of SNCR caused by incorrect frequency estimation of the respiratory signal. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] Figure 1 The flowchart of the method for locating human targets based on the singular spectrum analysis method of tensor decomposition is shown;
[0052] Figure 2 Build tensor diagrams for time series;
[0053] Figure 3 It is a schematic diagram of CP decomposition;
[0054] Figure 4 It is the original echo signal of UWB bio-radar;
[0055] Figure 5 These are the result diagrams after processing by the two algorithms; (a) radar signal after processing by TSSA algorithm; (b) range-spectrum diagram after processing by TSSA algorithm; (c) radar signal after processing by MSSA algorithm; (d) range-spectrum diagram after processing by MSSA algorithm. DETAILED DESCRIPTION
[0056] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments:
[0057] Table 1 Abbreviations
[0058]
[0059] like Figure 1 As shown, the human target positioning method based on the singular spectrum analysis method of tensor decomposition comprises the following steps:
[0060] S1. Ultra-wideband (UWB) bio-radar echo signal acquisition and preprocessing:
[0061] The process of UWB bio-radar capturing human life signals is mathematically described: the human micro-motion model caused by human life activities, i.e. breathing and heartbeat, relative to the radar transceiver antenna is:
[0062] d(t)=d0+m(t)=d0+m b sin(2πf b t)+m h sin(2πf h t) (1)
[0063] Where d0 indicates the distance between the human target position and the transmitting and receiving antenna, m b and m h are the amplitudes of breathing and heartbeat, respectively, b and f h are the frequencies of breathing and heartbeat respectively. The two-dimensional echo signal collected by the radar can be expressed as:
[0064]
[0065] p(t) is the normalized transmission waveform signal, A i is the amplitude of the i-th multipath component, τ i is the signal delay caused by the i-th multipath component, A is the amplitude of body motion, τ d (t) is the signal delay caused by d(t).
[0066]
[0067] Where c is the speed of light. Sampling r(t,τ) yields R[m,n]=r(τ=mT f ,t=nT s ), T f and T s are the sampling intervals for the fast and slow time dimensions, respectively.
[0068] UWB bio-radar system uses time series τ = mT in fast time dimension f (m=1,2,…M) samples the received waveform, and in the slow time dimension at discrete moments t=nT s (n=1,2,…N) sampling pulse waveform. The sampled radar echo signal is stored as a two-dimensional array R of size M×N.
[0069] In actual applications, due to the complex detection environment and the instability of the radar system, the data collected by the UWB bio-radar system contains not only the signals reflected by the living organisms, but also a variety of noise and clutter interference. In order to better achieve the detection purpose, the radar echo signal needs to be pre-processed.
[0070] First, each signal in the UWB bio-radar echo signal is smoothed and filtered to remove the high-frequency interference introduced by the signal. The mathematical expression is:
[0071]
[0072] Where l = 1, 2, ..., L, L = ML f +1. L f is the filter order, and the optimal filter order L is determined by experiment f =85.
[0073] In a specific embodiment of the present invention, the sampling frequency of the UWB bio-radar system in the distance direction is 8192, which means that if the radar echo signal is directly processed, the computer's computing speed will be very high, which is not convenient for post-disaster rescue to use portable computers for lightweight data processing. Therefore, it is necessary to perform distance accumulation operations in the distance direction to reduce the amount of calculation. In addition, accumulation in the distance direction is equivalent to performing windowed smoothing filtering in the distance direction, which can increase the signal-to-noise ratio of the channel signal.
[0074] The distance accumulation of the smoothed and filtered echo signal along the distance matrix can be expressed as
[0075]
[0076] Where L is the accumulation window width along the distance direction, k (k = 1, 2, 3, ... K) is the coordinate of the distance dimension after distance accumulation, and K is the number of sampling points in the distance direction after accumulation. symbol Indicates rounding down. Based on experimental evaluation and experience, K = 200 is generally taken to ensure that the signal-to-noise ratio is enhanced without losing information.
[0077] S2, embed the preprocessed one-dimensional time series into a tensor;
[0078] Singular spectrum analysis based on tensor decomposition requires embedding the one-dimensional time series into a tensor and then performing tensor decomposition.
[0079] like Figure 2 As shown, x is segmented into matrices with l as non-overlapping time windows The size of the matrix is The relationship between the matrix and the time series is as follows:
[0080]
[0081] in Then the matrix is converted into a tensor. The conversion process is as follows Figure 2 As shown, the matrix After adding windows, each layer of the tensor is formed, which is mathematically expressed as:
[0082]
[0083] l1-o is the overlap size of adjacent windows, and l1 is the window size.
[0084] S3, performing tensor decomposition;
[0085] There are two main methods for decomposing tensors: TUCKER decomposition (also known as high-order singular value decomposition) and canonical or parallel factor analysis (CANDECOMP / PARAFAC). The present invention adopts the CP decomposition method, which was independently proposed by Carroll and Harshman in 1970.
[0086] The basic formula for CP decomposition is:
[0087]
[0088] where x ijk is a third-order tensor The (i,j,k)th element of , R is the order of the model.
[0089] like Figure 3 As shown, CP decomposition is to decompose a third-order tensor into the sum of tensors of rank 1. The decomposition form is:
[0090]
[0091] where R is a positive integer and The element-by-element form is,
[0092]
[0093] There are 6 possible forms of horizontal CP decomposition:
[0094] X (I×JK) =[X ∷1 ,...,X ∷K ]=A(C⊙B) T (11)
[0095] X (J×KI) =[X 1∷ ,...,X I∷ ]=B(A⊙C) T (12)
[0096] X (K×IJ) =[X :1: ,…,X :J: ]=C(B⊙A) T (13)
[0097]
[0098] Where X⊙Y represents the Khatri-Rao product of the m×n matrix X and the l×n matrix Y, and
[0099]
[0100] Similar to the horizontal CP form, there are also 6 forms of vertical expansion:
[0101]
[0102]
[0103] S4, group reconstruction;
[0104] The tensor decomposition method used requires group reconstruction, so it will include a weighted process. The present invention uses ensemble empirical mode decomposition (EEMD) to supervise the reconstruction process. The mathematical representation of this process is
[0105]
[0106] where ||·|| represents the Frobenius norm, R is the order, and a ir ,b jr ,c kr are the tensor factors, G = {g ijk} is a superdiagonal tensor. Since the objective function (26) is convex, different optimization methods can be used to minimize it. The subgradient method is an extension of the traditional descent method and can be used to deal with non-differentiable convex functions, so it has a wider range of problems than traditional methods. The subgradient method is applied to solve g ijk The iterative equation is
[0107]
[0108] in
[0109]
[0110] μ is the iteration step size.
[0111] About the reference tensor f ijk ,This paper adopts the EEMD method to add the intrinsic mode functions falling within the respiratory frequency band as the reference signal, and performs operations such as embedding on it to make it tensorized, and supervises the grouping and reconstruction after tensor decomposition.
[0112] Empirical Mode Decomposition (EMD) is a method for processing non-stationary nonlinear signals. Compared with other time-frequency analysis methods such as short-time Fourier transform and wavelet transform, EMD has many advantages. For example, EMD does not need to set basis functions, and can decompose the signal into multiple intrinsic mode functions (IMF) based on the characteristics of the signal itself. Therefore, it has the advantages of multi-resolution and adaptive signal decomposition. However, the conventional EMD method has the problem of mode aliasing, which makes the physical meaning of each IMF unclear. As an improved EMD method, EEMD can effectively solve the problem of mode aliasing.
[0113] The following is a detailed description of EEMD:
[0114] After the original signal passes through EMD, a series of IMFs can be obtained. Each IMF must meet two conditions: (1) the number of extreme values and the number of zero points of the component signal must be equal or differ by at most one; (2) at any data position, the average value of the envelope of the local maximum value and the envelope of the local minimum value is 0.
[0115] EMD is achieved through a screening process, and its specific operations are as follows: 1) Find all the maximum and minimum points of the signal, use the cubic spline function to connect all the maximum (minimum) points into an upper (lower) envelope, and average the upper and lower envelopes to obtain the average envelope m1(t); 2) Subtract the average envelope m1(t) from the original signal x(t) to obtain h(t); 3) Use h(t) as the signal to be decomposed and repeat steps 1) and 2) many times until the envelope is symmetrical under certain conditions. The final h is the first IMF component, C1(t); 4) Subtract C1(t) from the original signal to obtain a signal sequence r1(t) = x(t)-C1(t) with high-frequency components removed; 5) Repeat the above process to obtain the second IMF component, and continue to repeat the above process until the final difference sequence is monotonic. At this point, the original signal can be represented by the sum of the IMF component and the residual:
[0116]
[0117] EEMD is proposed to solve the problem of mode aliasing in EMD decomposition. The EEMD method adds different noises to the original signal sequence, and performs EMD decomposition on the noisy signal, repeats it many times, and averages all the IMFs obtained by decomposition to obtain the final IMF. The specific steps are as follows:
[0118] 1) Add white noise of a certain intensity to the original signal sequence;
[0119] 2) Perform EMD decomposition on the noisy signal to obtain a series of IMF components;
[0120] 3) Add white noise of different intensities to the original signal sequence, repeat steps 1) and 2) for a total of M times;
[0121] 4) Average the IMF components obtained by M-times EMD decomposition to obtain the final IMF of the original signal sequence.
[0122]
[0123] Since the white noise added each time is evenly distributed on the time-frequency scale, the white noise constructs a uniform time-frequency space scale. For each EMD decomposition, the different frequency scales of the signal will be automatically projected onto the frequency scale corresponding to the uniform time-frequency space. Since the artificially added white noise is unrelated for each EMD, the IMF components obtained by the M-time EMD decomposition are averaged to offset the influence of the added white noise.
[0124] S5. Hankel matrix.
[0125] Usually, the Y matrix obtained by grouping is not a Hankel matrix. In order to map Y to a one-dimensional signal, it needs to be Hankel matrixed. Usually, the time series is obtained by averaging the anti-diagonal values.
[0126] make It represents the one-dimensional data signal obtained after mapping, which can be obtained by diagonal averaging. The mathematical expression is:
[0127]
[0128] The present invention selects tensor decomposition to replace the singular value decomposition (SVD) process in singular spectrum analysis. The method is more suitable for processing non-stationary signals. In subspace selection and reconstruction, the ensemble empirical mode method (EEMD) is used to supervise the tensor decomposition process. The traditional singular spectrum analysis (SSA) method does not take the inherent non-stationary characteristics of the signal into consideration and cannot correctly group and reconstruct the signal. The tensor decomposition replaces the singular value decomposition (SVD) process in the traditional singular spectrum analysis (SSA) to process such signals well. In addition, EEMD plays a role in supervising the reconstruction process in tensor decomposition, and also avoids the decrease of SNCR caused by incorrect frequency estimation of the respiratory signal.
[0129] A standard ruins model of the China Earthquake Emergency Rescue Center was used as the experimental scene to collect experimental data without human targets under the ruins. Because the data is collected in a real environment, the echo must contain non-static clutter, random noise, and system jitter interference. The original echo data is as follows: Figure 4 shown.
[0130] The ideal human target is established based on the characteristic analysis of the breathing and body motion response of the buried human body based on IR-UWB. The amplitude of human breathing is changed to obtain signal data with different SNCR input strengths. The human target positioning method based on TSSA and the human target positioning method based on multichannel singular spectrum analysis (MSSA) are used to compare the input SNCR of the two methods. The results are shown in Table 2. The respiratory frequency involved in the human target positioning method based on MSSA is estimated using the fast Fourier transform algorithm (FFT).
[0131] Table 2 SNCR results after two algorithms processing
[0132]
[0133]
[0134] From the results in the table, it can be seen that the proposed method for detecting human targets based on tensor decomposition is more effective than the multi-channel singular spectrum analysis method. Under different input SNCR conditions, the output SNCR has been improved to varying degrees. This is because the traditional SSA method does not take into account the inherent non-stationary characteristics of the signal, so it is impossible to correctly group and reconstruct the signal. Tensor decomposition replaces the SVD process in the traditional SSA to process such signals well. In addition, EEMD plays a role in supervising the reconstruction process in tensor decomposition, and also avoids the decrease in SNCR caused by incorrect estimation of the respiratory signal frequency.
[0135] Take one set of data and process it using two algorithms respectively. The processing results are as follows: Figure 5 This result can more intuitively show that the TSSA-based human target positioning algorithm has better noise and clutter suppression capabilities than the MSSA algorithm. Figure 5 (a) The non-human target area is relatively Figure 5 The non-human target area in (c) is much purer. Figure 5 (b) and Figure 5 (d) Two result graphs also show that Figure 5 (d) shows more clutter components, which means that the processing algorithm is prone to inaccurate human target positioning.
[0136] Through experimental evaluation, the proposed detection method is conducive to improving the accuracy of human target detection under ruins. The improved human target positioning method can effectively suppress clutter interference and ensure the accuracy of human target positioning.
[0137] The above is only an embodiment of the present invention, and the common knowledge such as the known specific technical solutions and / or characteristics in the solution is not described in detail here. It should be pointed out that for those skilled in the art, without departing from the technical solution of the present invention, several modifications and improvements can be made, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.
Claims
1. A human target positioning method based on a singular spectrum analysis method of tensor decomposition, characterized in that: The following steps are involved: S1. Ultra-wideband (UWB) bio-radar echo signal acquisition and preprocessing; S2, embed the preprocessed one-dimensional time series into a tensor; S3, performing tensor decomposition; S4, group reconstruction; S5. Hankel matrix.
2. The human body target positioning method based on the singular spectrum analysis method of tensor decomposition according to claim 1, characterized in that: In step S1, the human micro-motion model caused by human life activities, i.e., breathing and heartbeat, relative to the radar transceiver antenna is: d(t)=d0+m(t)=d0+m b sin(2πf b t)+m h sin(2πf h t) Where d0 indicates the distance between the human target position and the transmitting and receiving antenna, m b , m h are the amplitudes of breathing and heartbeat, respectively, b , f h are the frequencies of breathing and heartbeat respectively. The two-dimensional echo signal collected by the radar can be expressed as: p(t) is the normalized transmission waveform signal, A i is the amplitude of the i-th multipath component, τ i is the signal delay caused by the i-th multipath component, A is the amplitude of body motion, τ d (t) is the signal delay caused by d(t), and its expression is: Where c is the speed of light, sampling r(t,τ) yields R[m,n]=r(τ=mT f ,t=nT s ), T f ,T s The sampling intervals for the fast and slow time dimensions, respectively; UWB bio-radar system uses time series τ = mT in fast time dimension f (m=1,2,…M) samples the received waveform, and in the slow time dimension at discrete moments t=nT s (n=1,2,…N) sampling pulse waveform, the sampled radar echo signal is stored as a two-dimensional array R of size M×N.
3. The human body target positioning method based on the singular spectrum analysis method of tensor decomposition according to claim 1, characterized in that: In step S1, each signal in the UWB bio-radar echo signal is smoothed and filtered, mainly to remove the high-frequency interference introduced on the signal. The mathematical expression is: Where l = 1, 2, ..., L, L = ML f +1, L f is the filter order, the optimal filter order L f =85.
4. The method for locating a human target based on a singular spectrum analysis method of tensor decomposition according to claim 3, characterized in that: In step S1, the distance accumulation of the smoothed and filtered echo signal along the distance direction is performed on the matrix, which can be expressed as: Where L is the accumulation window width along the distance direction, k (k = 1, 2, 3, ... K) is the coordinate of the distance dimension after distance accumulation, and K is the number of sampling points in the distance direction after accumulation. symbol Indicates rounding down.
5. The method for locating a human target based on a singular spectrum analysis method of tensor decomposition according to claim 1, characterized in that: In step S2, the point signal x is segmented into matrices with l as non-overlapping time windows The size of the matrix is The relationship between the matrix and the time series is as follows in Then convert the matrix into a tensor, that is, After adding windows, each layer of the tensor is formed, which is mathematically expressed as: l1-o is the overlap size of adjacent windows, and l1 is the window size.
6. The method for locating a human target based on a singular spectrum analysis method of tensor decomposition according to claim 1, characterized in that: In step S3, the tensor is subjected to canonical or parallel factor analysis (CANDECOMP / PARAFAC), and the basic formula of CP decomposition is: where x ijk is a third-order tensor The (i,j,k)th element of , R is the order of the model.
7. The human body target positioning method based on the singular spectrum analysis method of tensor decomposition according to claim 1 is characterized in that: In step S4, Ensemble Empirical Mode Decomposition (EEMD) is used to supervise the reconstruction process, and the mathematical representation of the process is: where ||·|| represents the Frobenius norm, R is the order, and a ir ,b jr ,c kr are the tensor factors, G = {g ijk } is a superdiagonal tensor.
8. The method for locating a human target based on a singular spectrum analysis method of tensor decomposition according to claim 7, characterized in that: In step S3, the subgradient method is used to solve for g ijk , the iteration equation is in μ is the iteration step size.
9. The method for locating a human target based on a singular spectrum analysis method of tensor decomposition according to claim 8, characterized in that: About the reference tensor f ijk , the EEMD method is used to add the intrinsic mode functions falling within the respiratory frequency band as the reference signal, and embed it to make it tensorized, and supervise the grouping and reconstruction after tensor decomposition. The specific steps of the EEMD are as follows: 1) Add white noise of a certain intensity to the original signal sequence; 2) Perform EMD decomposition on the noisy signal to obtain a series of IMF components; 3) Add white noise of different intensities to the original signal sequence, repeat steps 1) and 2) for a total of M times; 4) The IMF components obtained by M-times EMD decomposition are averaged to obtain the final IMF of the original signal sequence, which is expressed as:
10. The method for locating a human target based on a singular spectrum analysis method of tensor decomposition according to claim 1, characterized in that: In step S5, let Represents the one-dimensional data signal obtained after mapping, which can be obtained by diagonal averaging. The mathematical expression is
Citation Information
Patent Citations
Separating and identifying method for multiple human body objects in distance direction of UWB impact biological radar
CN106019254A
Strong reflection clutter elimination method based on double-source IR-UWB (Ultra-Wideband) bioradar
CN106093868A
Human and animal non-contact detection distinguishing method based on IR-UWB biological radar signals
CN110879388A
Life information detection method and system based on singular spectrum decomposition algorithm
CN115016033A
Human body vital sign detection method and system based on ultra wide band radar
CN118452868A
