Millimeter wave radar signal motion noise removal method based on total variation

Abstract

The invention belongs to the field of non-contact vital sign detection, and particularly relates to a millimeter wave radar signal motion noise removal method based on total variation, and the method specifically comprises the steps: S1, obtaining thoracic cavity signals containing motion noise through a millimeter wave radar, rearranging the thoracic cavity signals, and converting the thoracic cavity signals into a gray-scale map; s2, after a chest image containing noise is obtained, second-order gradient information of the image in different directions is calculated; s3, according to gradient information in different directions, sparse characteristics of noise and a bottom layer image structure are obtained, and transverse and longitudinal second-order regularization constraint terms are respectively designed by introducing Lp pseudo norms; meanwhile, a global sparse constraint term is designed based on the prior characteristics of the noise; and S4, restoring the denoised image into a one-dimensional thoracic cavity signal. According to the method, the sparse characteristics of the motion noise and the underlying image structure are fully utilized, and the sparse characteristics are better described by introducing the Lp pseudo norm, so that the motion noise is effectively removed, and the detail characteristics of the thoracic cavity signal are better reserved.

Description

Technical Field

[0001] The present invention belongs to the technical field of non-contact vital sign detection, and specifically is a method for removing motion noise of millimeter wave radar signals based on total variation. Background Art

[0002] With the increase in population, the conditions of patients in hospitals vary. When encountering patients with skin diseases or large-area burns, it is not appropriate to use contact devices for heartbeat detection. Therefore, it is necessary to study the detection of heartbeat signals in non-contact conditions. When using millimeter-wave radar to collect human chest signals, the human body cannot remain completely still, and there will always be unconscious or imperceptible movements. Therefore, the collected signals often contain motion noise. The presence of motion noise reduces the quality of chest signal data and seriously affects their subsequent applications in medical treatment, rescue and other fields. The existing motion noise removal algorithms can be roughly divided into two categories: filtering-based methods and signal decomposition-based methods, but they often have certain defects. Among them, filtering-based methods usually use various filters to remove motion noise, such as Chebyshev filters, wavelet decomposition, and wavelet-Chebyshev combination filters. Signal decomposition-based methods mainly extract useful and effective components by finding a set of modal components and their respective center frequencies, such as empirical mode decomposition and variational mode decomposition. However, these two methods cannot achieve good results when dealing with non-periodic irregular noise. Therefore, we propose a millimeter-wave radar signal motion noise removal method based on total variation to solve the problem of motion noise interference. Summary of the invention

[0003] 1. Technical issues to be resolved

[0004] In view of the shortcomings of the prior art, the present invention provides a method for removing motion noise of millimeter-wave radar signals based on total variation, which solves the problems raised in the above background technology.

[0005] (II) Technical solution

[0006] In order to achieve the above-mentioned purpose, the present invention specifically adopts the following technical solutions:

[0007] A method for removing motion noise of millimeter wave radar signals based on total variation specifically comprises the following steps:

[0008] S1, after obtaining the millimeter-wave radar acquisition data matrix from the millimeter-wave radar echo signal data set, static clutter filtering, unwrapping and differential processing are performed to obtain the phase difference signal; the phase difference signal matrix is ​​taken out in turn according to the chest signal period, and these taken out row vectors are put into each row of the chest signal grayscale matrix in turn. Considering the periodicity of the signal, the newly constructed chest signal grayscale matrix should be relatively smooth when converted into a grayscale image, and at the same time have certain sparse characteristics in structure.

[0009] S2, after obtaining the chest signal image containing motion noise, the second-order gradient information of the image in different directions is calculated respectively to obtain the second-order gradient in the horizontal direction, the second-order gradient in the vertical direction and the second-order mixed gradient. The calculation formula of the gradient information in different directions of the image is:

[0010]

[0011] Where: I(x,y) represents the pixel value of the image, G xx (x, y) represents the second-order gradient information in the horizontal direction of the image, G yy (x, y) represents the second-order gradient information in the vertical direction of the image, G xy (x,y) represents the second-order mixed gradient information of the image, and (x,y) represents the row and column position index of the image pixel.

[0012] S3, based on the second-order gradient information in different directions, obtains the sparse characteristics of noise and underlying image structure, and designs second-order regularization constraints in different directions by introducing the Lp pseudo-norm; at the same time, a global sparse constraint is designed based on the prior characteristics of motion noise, and its related expression is:

[0013] R0=||YZ||1

[0014]

[0015]

[0016]

[0017] Among them: Y represents the ideal image without noise, Z represents the image polluted by noise, L x =[-1,1] represents the convolution kernel of the horizontal difference, L y =[-1,1] T represents the convolution kernel of the longitudinal difference, R0 represents the global sparse constraint term, R xx represents the horizontal second-order regularization constraint, R yy represents the vertical second-order regularization constraint, R xy represents the mixed second-order regularization constraint term.

[0018] In particular, in the field of image processing, the Lp norm is generally defined as The Lp pseudo-norm is defined as

[0019] S4, combining the second-order regularization constraints in different directions in S3 with the global sparse constraints, obtains the complete total variation model expression for removing random motion noise:

[0020]

[0021] Where: μ represents the coefficient of the second-order sparse regularization term.

[0022] S5, using the alternating multiplier method to iteratively solve the total variation model for removing noise in S4, including the following steps:

[0023] S501, converting the constrained problem into an unconstrained problem by introducing intermediate variables, and obtaining the augmented Lagrangian function of the problem.

[0024] The augmented Lagrangian function expression of the unconstrained problem is:

[0025]

[0026] Where: A0 = YZ, A1 = L x *L x *Y, A2 = L y *L y *Y, A3 = L x *L y *Y, Λ0, Λ1, Λ2 and Λ3 represent Lagrange multipliers, γ0, γ1, γ2 and γ3 represent penalty coefficients, max represents the maximum value, and min represents the minimum value.

[0027] S502, solving the sub-problems of each variable in the augmented Lagrangian function one by one, and obtaining its iterative formula respectively.

[0028] (1) Based on the above augmented Lagrangian function, we can get the subproblem of Y as follows:

[0029]

[0030]

[0031]

[0032]

[0033] Through the convolution theorem and fast Fourier transform, the iterative formula of Y is obtained as follows:

[0034]

[0035]

[0036]

[0037] Where: represents the Fourier transform operator, represents the inverse Fourier transform operator, express Conjugate mapping, · represents dot product, / represents dot division.

[0038] (2) Based on the above augmented Lagrangian function, we can get the sub-problems of intermediate variables A0, A1, A2 and A3 respectively:

[0039]

[0040]

[0041]

[0042]

[0043] Through the contraction theorem, the iterative formulas of the intermediate variables A0, A1, A2 and A3 are obtained as follows:

[0044]

[0045]

[0046]

[0047]

[0048] Where:

[0049] (3) Finally, using the gradient ascent method, the iterative formulas for the Lagrange multipliers Λ0, Λ1, Λ2, and Λ3 are obtained as follows:

[0050]

[0051]

[0052]

[0053]

[0054] Among them, λ is the learning rate.

[0055] S503, after multiple iterations of each variable, until the relevant conditions or the maximum number of iterations are met, the optimal solution of Y can be finally obtained, that is, the chest signal image after the motion noise is removed.

[0056] (III) Beneficial effects

[0057] Compared with the prior art, the present invention provides a method for removing motion noise of millimeter wave radar signals based on total variation, which has the following beneficial effects:

[0058] This method converts the one-dimensional physiological signal collected by millimeter-wave radar into a two-dimensional image signal, and then uses the sparse characteristics of noise and the underlying image structure to denoise it. It not only effectively removes noise, but also better preserves the detailed characteristics of the chest signal, providing a new idea for physiological signal denoising.

[0059] The present invention uses the second-order gradient information of the chest signal image when solving the proposed total variation model for removing motion noise. The second-order gradient information of the image can better highlight the noise and has a certain sparsity. Compared with other norms, the Lp pseudo-norm has better absorption and characterization capabilities, so the Lp norm is used in the second-order sparse regularization constraint term, and at the same time, the L1 norm is used in the global sparse constraint term to ensure the convergence of the model. This significantly reduces the noise originally distributed throughout the frequency domain, and the signal is concentrated within the ideal chest signal frequency range. The motion noise is basically filtered out, and it has better application prospects in processing vital sign signals collected by millimeter-wave radar. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a flowchart of a method for removing motion noise of millimeter wave radar signals based on total variation of the present invention;

[0061] Figure 2 This is a time domain diagram of the signal before denoising of the present invention;

[0062] Figure 3 This is a signal spectrum diagram before denoising of the present invention;

[0063] Figure 4 A time domain diagram of a signal after denoising using the motion noise removal method of the present invention;

[0064] Figure 5 This is a spectrum diagram of a signal after denoising using the motion noise removal method of the present invention. DETAILED DESCRIPTION

[0065] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0066] Example

[0067] like Figure 1 As shown, a method for removing motion noise of millimeter wave radar signals based on total variation is proposed in one embodiment of the present invention, comprising the following steps:

[0068] S1, after obtaining the millimeter-wave radar acquisition data matrix from the millimeter-wave radar echo signal data set, static clutter filtering, unwrapping and differential processing are performed to obtain the phase difference signal; the phase difference signal matrix is ​​taken out in turn according to the chest signal period, and these taken out row vectors are put into each row of the chest signal grayscale matrix in turn. Considering the periodicity of the signal, the newly constructed chest signal grayscale matrix should be relatively smooth when converted into a grayscale image, and at the same time have certain sparse characteristics in structure.

[0069] S2, after obtaining the chest signal image containing motion noise, the second-order gradient information of the image in different directions is calculated respectively to obtain the second-order gradient in the horizontal direction, the second-order gradient in the vertical direction and the second-order mixed gradient. The calculation formula of the gradient information in different directions of the image is:

[0070]

[0071] Where: I(x,y) represents the pixel value of the image, G xx (x, y) represents the second-order gradient information in the horizontal direction of the image, G yy (x, y) represents the second-order gradient information in the vertical direction of the image, G xy (x,y) represents the second-order mixed gradient information of the image, and (x,y) represents the row and column position index of the image pixel.

[0072] S3, based on the second-order gradient information in different directions, obtains the sparse characteristics of noise and underlying image structure, and designs second-order regularization constraints in different directions by introducing the Lp pseudo-norm; at the same time, a global sparse constraint is designed based on the prior characteristics of motion noise, and its related expression is:

[0073] R0=||YZ||1

[0074]

[0075]

[0076]

[0077] Among them: Y represents the ideal image without noise, Z represents the image polluted by noise, L x =[-1,1] represents the convolution kernel of the horizontal difference, L y =[-1,1] T represents the convolution kernel of the longitudinal difference, R0 represents the global sparse constraint term, R xx represents the horizontal second-order regularization constraint, R yy represents the vertical second-order regularization constraint, R xy represents the mixed second-order regularization constraint term.

[0078] In particular, in the field of image processing, the Lp norm is generally defined as The Lp pseudo-norm is defined as

[0079] S4, combining the second-order regularization constraints in different directions in S3 with the global sparse constraints, obtains the complete total variation model expression for removing random motion noise:

[0080]

[0081] Where: μ represents the coefficient of the second-order sparse regularization term.

[0082] S5, using the alternating multiplier method to iteratively solve the total variation model for removing noise in S4, including the following steps:

[0083] S501, converting the constrained problem into an unconstrained problem by introducing intermediate variables, and obtaining the augmented Lagrangian function of the problem.

[0084] The augmented Lagrangian function expression of the unconstrained problem is:

[0085]

[0086] Where: A0 = YZ, A1 = L x *L x *Y, A2 = L y *L y *Y, A3 = L x *L y *Y, Λ0, Λ1, Λ2 and Λ3 represent Lagrange multipliers, γ0, γ1, γ2 and γ3 represent penalty coefficients, max represents the maximum value, and min represents the minimum value.

[0087] S502, solving the sub-problems of each variable in the augmented Lagrangian function one by one, and obtaining its iterative formula respectively.

[0088] (1) Based on the above augmented Lagrangian function, we can get the subproblem of Y as follows:

[0089]

[0090] Through the convolution theorem and fast Fourier transform, the iterative formula of Y is obtained as follows:

[0091]

[0092]

[0093]

[0094] Where: represents the Fourier transform operator, represents the inverse Fourier transform operator, express Conjugate mapping, · represents dot product, / represents dot division.

[0095] (2) Based on the above augmented Lagrangian function, we can get the sub-problems of intermediate variables A0, A1, A2 and A3 respectively:

[0096]

[0097]

[0098]

[0099]

[0100] Through the contraction theorem, the iterative formulas of the intermediate variables A0, A1, A2 and A3 are obtained as follows:

[0101]

[0102]

[0103]

[0104]

[0105] Where:

[0106] (3) Finally, using the gradient ascent method, the iterative formulas for the Lagrange multipliers Λ0, Λ1, Λ2, and Λ3 are obtained as follows:

[0107]

[0108]

[0109]

[0110]

[0111] Among them, λ is the learning rate.

[0112] S503, after multiple iterations of each variable, until the relevant conditions or the maximum number of iterations are met, the optimal solution of Y can be finally obtained, that is, the chest signal image after the motion noise is removed.

[0113] This method converts the one-dimensional physiological signal collected by millimeter-wave radar into a two-dimensional image signal, and then uses the sparse characteristics of noise and the underlying image structure to denoise. It not only effectively removes noise, but also better retains the detailed characteristics of the chest signal, providing a new idea for physiological signal denoising. At the same time, the sparse characteristics of motion noise and the underlying image structure are fully considered. When solving the proposed total variation model for removing motion noise, the second-order gradient information of the chest signal image is used. The second-order gradient information of the image can better highlight the noise and has a certain sparsity. Compared with other norms, the Lp pseudo-norm has better absorption and characterization capabilities. Therefore, the Lp norm is used in the second-order sparse regularization constraint term. At the same time, in order to ensure the convergence of the model, the L1 norm is used in the global sparse constraint term. This significantly reduces the noise originally distributed throughout the frequency domain, and the signal is concentrated in the ideal chest signal frequency range. The motion noise is basically filtered out, thereby improving the quality of the chest signal. It can be helpful for the subsequent application of vital signs signals collected by millimeter-wave radar in rescue, medical and other fields; in Figure 2 , Figure 3 The signal with motion noise is shown in Figure 4 , Figure 5 The figure shows the result of removing motion noise using the proposed method.

[0114] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for removing motion noise from millimeter wave radar signals based on total variation. It is characterized in that The specific steps include: S1: After obtaining the data matrix from the millimeter wave radar echo signal data set, static clutter filtering, unwrapping and differential processing are performed to obtain the phase difference signal matrix, the phase difference signal matrix is ​​rearranged to obtain the chest signal matrix, and the chest signal matrix is ​​used to produce the chest signal sample grayscale image; S2: After obtaining the sample grayscale image of the chest signal containing motion noise, the second-order gradient and second-order mixed gradient information of the horizontal and vertical directions of the image are calculated respectively; S3: The sparse characteristics of motion noise and underlying image structure are obtained based on the second-order gradient information in different directions. The second-order sparse regularization constraints in different directions are designed by introducing the Lp pseudo-norm. At the same time, the global sparse constraints are designed based on the prior characteristics of motion noise. S4: Combine the second-order sparse regularization constraints in different directions in S3 with the global sparse constraints to obtain a complete total variation model for removing motion noise; S5: The total variation model for removing motion noise in S4 is iteratively solved by using the alternating multiplier method, and finally a chest signal image without motion noise can be obtained; the chest signal image is converted into a one-dimensional chest signal.

2. According to the method for removing motion noise of millimeter wave radar signal based on total variation in claim 1, Features: The calculation formula of the gradient information in different directions of the image is: Where: I(x,y) represents the pixel value of the image, G xx (x, y) represents the second-order gradient information in the horizontal direction of the image, G yy (x, y) represents the second-order gradient information in the vertical direction of the image, G xy (x,y) represents the second-order mixed gradient information of the image, and (x,y) represents the row and column position index of the image pixel.

3. According to the method for removing motion noise of millimeter wave radar signal based on total variation in claim 1, Features: The global constraint terms and the second-order sparse regularization constraints in different directions are: R 0 =||Y-Z|| 1 Among them: Y represents the ideal image without noise, Z represents the image polluted by noise, L x =[-1,1] represents the convolution kernel of the horizontal difference, L y =[-1,1] T represents the convolution kernel of the longitudinal difference, R 0 represents the global sparse constraint term, R xx represents the horizontal second-order regularization constraint, R yy represents the vertical second-order regularization constraint, R xy represents the mixed second-order regularization constraint term.

4. According to the method for removing motion noise of millimeter wave radar signal based on total variation in claim 1, Features: The complete total variation model expression for noise removal is: Where: μ represents the coefficient of the second-order sparse constraint regularization term.

5. According to the method for removing motion noise of millimeter wave radar signal based on total variation in claim 1, Features: The solution of the total variation model using the alternating direction multiplier method includes the following steps: S501, introducing intermediate variables to transform the constrained problem into an unconstrained problem, and obtaining the augmented Lagrangian function of the problem; The augmented Lagrangian function of this problem is: Among them: A 0 =YZ,A 1 =L x *L x *Y,A 2 =L y *L y *Y,A 3 =L x *L y *Y,Λ 0 ,Λ 1 ,Λ 2 and Λ 3 represents the Lagrange multiplier, γ 0 , γ 1 , γ 2 and γ 3 Represents the coefficient of the penalty term, max means taking the maximum value, and min means taking the minimum value; S502, solving the sub-problems of each variable in the augmented Lagrangian function one by one, and introducing auxiliary variables to obtain their iterative formulas respectively; S503, after multiple iterations of various variables, the optimal solution of Y is obtained, that is, the chest signal image after noise removal.

Images