UWB-GPR joint de-noising method based on layered dynamic collaborative optimization algorithm

Abstract

The invention discloses a UWB-GPR joint de-noising method based on a layered dynamic collaborative optimization algorithm, and belongs to the crossing field of signal processing and geophysical exploration. Aiming at the problems of modal aliasing phenomenon and poor fitness in the traditional denoising method, the method comprises the following steps of: preprocessing underground reflected waves to extract an effective signal interval; using a hierarchical dynamic collaborative optimization algorithm (HDCO) to globally optimize a dynamic noise gain factor and count the number of integrated iterations for the target function; performing multi-scale decomposition on the echo signal by using a multi-scale noise injection mode decomposition algorithm (MSNIMD), and screening effective intrinsic oscillation mode components in combination with self-adaptive multi-scale spectrum entropy; a nonlinear coefficient shrinkage rule is adopted for reserved components, and secondary filtering of noise is achieved; and re-synthesizing to obtain a reflection waveform. According to the method, the problem of noise residue caused by fixed parameters of a traditional algorithm is solved, and the problem of time-frequency characteristic annihilation caused by coupling of noise in the signal and an effective signal is solved.

Description

Technical Field

[0001] The present invention belongs to the cross - field of signal processing and geophysical exploration, and relates to a UWB - GPR joint denoising method based on a hierarchical dynamic collaborative optimization algorithm. Background Technique

[0002] Ultra - wideband ground - penetrating radar technology (UWB - GPR) is widely used in fields such as engineering investigation, archaeological exploration, and highway quality inspection due to its high resolution, fast detection speed, and strong anti - interference ability. When the ground - penetrating radar detects underground target bodies, due to the influence of system and environmental factors, the received echo not only contains target reflection signals, antenna - to - antenna coupling signals, ground clutter, but also contains a lot of noise information, showing the characteristics of non - stationary and multi - scale noise superposition, resulting in a decline in signal quality, thus seriously affecting the recognition and detection of target bodies.

[0003] Traditional ground - penetrating radar signal denoising methods, such as empirical mode decomposition, are prone to mode mixing when decomposing noisy signals, resulting in the inability to effectively separate the effective signal from the noise. And the simple wavelet threshold denoising method depends on artificial experience to select the decomposition layer number and threshold rule, and has poor adaptability to non - stationary signals. For the parameters in the existing traditional algorithms, they need to be set manually, and it is difficult to adapt to complex noise environments. The above problems will lead to a significant decline in the accuracy of signal reconstruction in low - signal - to - noise ratio scenarios for traditional methods, seriously affecting the recognition accuracy of underground target bodies.

[0004] The present invention proposes a UWB - GPR joint denoising method based on a hierarchical dynamic collaborative optimization algorithm to solve the problems of noise residue and signal distortion caused by fixed parameters in traditional mode decomposition; overcome the dependence of traditional denoising methods on artificial experience; and achieve the denoising processing of ultra - wideband ground - penetrating radar signals in non - stationary noise environments. Summary of the Invention

[0005] The object of the present invention is to solve the problems of noise residue and signal distortion caused by fixed parameters in traditional mode decomposition; overcome the dependence of traditional denoising methods on artificial experience; and achieve the denoising processing of ultra - wideband ground - penetrating radar signals in non - stationary noise environments.

[0006] To achieve the above object, the present invention provides the following solution:

[0007] The present invention provides a UWB - GPR joint denoising method based on a hierarchical dynamic collaborative optimization algorithm, including the following steps:

[0008] S1: Collect the underground reflection waveforms received by UWB - GPR and the antenna, perform a simple pre - processing algorithm on them to eliminate the interference of near - field coupling, and extract the effective signal interval;

[0009] S2: Using the Hierarchical Dynamic Cooperative Optimization algorithm (HDCO), globally optimize the dynamic noise gain factor and the statistical integration iteration times for the objective function to generate an optimal combination of decomposition parameters;

[0010] S3: Use the Multi-Scale Noise Injection Modal Decomposition algorithm (MSNIMD) to perform multi-scale decomposition on the UWB-GPR signal, combine the adaptive multi-scale spectral entropy to screen out the effective intrinsic oscillation mode components, and eliminate the high-frequency noise-dominated modes;

[0011] S4: Apply the non-linear coefficient shrinkage rule to the remaining components to achieve secondary filtering of the residual noise in the frequency domain;

[0012] S5: Re-synthesize to obtain a high-fidelity reflected waveform of the underground target and evaluate the algorithm performance.

[0013] Furthermore, in S1, collecting the underground reflected waveform and performing preprocessing specifically includes:

[0014] Detect the underground target through UWB-GPR and the antenna. The detection range is 2×0.5 meters, the detection depth is 0.5 meters, and the detection target is an underground ant nest. Obtain the A-scan and B-scan waveforms of the underground reflection, and select a certain A-scan waveform for the experiment; Since there is a lot of interference information such as direct waves in the echo signal obtained during the detection by UWB-GPR, in order to verify the algorithm of the present invention, first perform a series of preprocessing on the echo signal to eliminate the interference of near-field coupling, and extract the effective signal interval.

[0015] Furthermore, in S2, using the HDCO algorithm to generate the optimal parameter combination specifically includes:

[0016] Using the Hierarchical Dynamic Cooperative Optimization algorithm (HDCO), globally optimize the dynamic noise gain factor (DNG) and the statistical integration iteration times (SIC) for the objective function to generate an optimal combination of decomposition parameters; By setting the parameter ranges of the dynamic noise gain factor and the statistical integration iteration times, randomly generate the positions of the hierarchical agents, calculate the fitness with the goal of minimizing the multi-scale spectral entropy, obtain the distances and positions of the dominant agent Alpha, the secondary agent Beta, and the auxiliary agent Delta, and perform boundary correction and rounding.

[0017] Furthermore, in S3, using the Multi-Scale Noise Injection Modal Decomposition algorithm (MSNIMD) to perform multi-scale decomposition on the signal specifically includes:

[0018] Perform MSNIMD decomposition on the echo signal to generate multi-scale EOM components, separate noise and effective signals; according to the optimized DNG and SIC parameters obtained, inject adaptive multi-scale noise into the echo signal to form a noisy signal; then extract local maximum and minimum points for multi-scale adaptive oscillation decomposition (MSAOD) decomposition, calculate the mean curve, extract the first EOM, and repeat the screening until the set stop criterion is met; then perform ensemble averaging calculation on the energy of the EOM to obtain its ensemble; finally, perform screening to distinguish between noise-dominated and signal-dominated EOM components.

[0019] Further, in S4, adopt the non-linear coefficient shrinkage rule for the retained components to achieve secondary filtering of residual noise in the frequency domain. The specific method is as follows:

[0020] Process according to the improved non-linear coefficient shrinkage rule to suppress high-frequency noise and retain signal edge features; first perform three-layer processing on the underground echo signal, then determine the suppression parameters, and then process according to the improved non-linear coefficient shrinkage rule, and finally re-synthesize the processed underground echo.

[0021] Further, in S5, re-synthesize to obtain a high-fidelity underground target reflection waveform and evaluate the algorithm performance. The specific method is as follows:

[0022] Re-synthesize the echo after algorithm processing, and evaluate the effective signal index (ESI), error signal energy (ESE), and timing correlation index (TCI) to verify the practicability of the algorithm.

[0023] The present invention has achieved the following beneficial technical effects compared with the prior art:

[0024] 1. The hierarchical dynamic collaborative optimization algorithm provided by the present invention can dynamically match the core parameters of the multi-scale noise injection modal decomposition algorithm to achieve adaptive decomposition of the noisy echo signal. The parameters in traditional decomposition algorithms need to be manually set and are difficult to adapt to complex noise environments. The present invention can avoid the decomposition instability problem caused by manual parameter setting, improve the self-adaptability of decomposition, and solve the problems of noise residue and signal distortion caused by fixed parameters in traditional algorithms.

[0025] 2. The improved MSNIMD decomposition combined with the non-linear coefficient shrinkage rule noise reduction algorithm provided by the present invention can screen effective components according to adaptive multi-scale spectral entropy and perform secondary elimination of residual noise, greatly improving the high fidelity of the underground target reflection waveform, avoiding the mode mixing phenomenon existing in traditional decomposition algorithms, and effectively solving the problem of video feature annihilation caused by the coupling of strong background noise and weak effective signals in ground penetrating radar signals. Description of the Drawings

[0026] Figure 1 It is the architecture diagram of the UWB-GPR joint denoising method based on the MSNIMD decomposition and the nonlinear coefficient shrinkage rule processing of the hierarchical dynamic collaborative optimization algorithm provided by the embodiments of the present invention.

[0027] Figure 2 It is the algorithm flowchart of the improved algorithm for optimizing the MSNIMD decomposition parameters based on the hierarchical dynamic collaborative optimization algorithm provided by the embodiments of the present invention. Specific embodiments

[0028] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the protection scope of the present invention.

[0029] The core of the present invention is to provide a UWB-GPR joint denoising method based on the hierarchical dynamic collaborative optimization algorithm to solve the problems of echo noise interference of existing ultra-wideband ground penetrating radar and the inadaptability of fixed parameter decomposition.

[0030] In order to enable those skilled in the art to better understand the solution of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0031] S1: By collecting the underground reflection waveforms received by the UWB-GPR and the antenna, perform a simple preprocessing algorithm on them to eliminate the interference of near-field coupling and extract the effective signal interval.

[0032] Detect the underground target through the UWB-GPR and the antenna. The detection range is 2×0.5 meters, the detection depth is 0.5 meters, and the detection target is an underground ant nest. Obtain the A-scan and B-scan waveforms of the underground reflection, and select a certain A-scan waveform for the experiment; since there is a lot of interference information such as direct waves in the echo signal obtained during the detection by the UWB-GPR, in order to verify the algorithm of the present invention, first perform a series of preprocessing on the echo signal to eliminate the interference of near-field coupling and extract the effective signal interval.

[0033] S2: Use the hierarchical dynamic collaborative optimization algorithm (HDCO) to globally optimize the dynamic noise gain factor and the statistical integration iteration times for the objective function, and generate the optimal decomposition parameter combination.

[0034] Using the Hierarchical Dynamic Cooperative Optimization algorithm (HDCO), globally optimize the dynamic noise gain factor (DNG) and the statistical integration iteration count (SIC) for the objective function to generate an optimal combination of decomposition parameters; by setting the parameter ranges of the dynamic noise gain factor and the statistical integration iteration count, randomly generate the positions of the hierarchical agents, calculate the fitness with the goal of minimizing the multi-scale spectral entropy, obtain the distances and positions of the dominant agent Alpha, the secondary agent Beta, and the auxiliary agent Delta, and perform boundary correction and rounding.

[0035] S21: Set the parameter ranges of the dynamic noise gain factor and the statistical integration iteration count, and randomly generate the positions of the hierarchical agents: Positions i,j = 1b j + rand()·(ub j - 1b j ), where i is the number, j = 1 corresponds to the dynamic noise gain factor, and j = 2 corresponds to the statistical integration iteration count.

[0036] S22: Then, perform fitness calculation to minimize the multi-scale spectral entropy: where K is the number of eigen-oscillation modes (EOM), SpEn is the multi-scale spectral entropy, and B m (r) is the matching probability of the signal sequence under the embedding dimension m and tolerance.

[0037] S23: Update the positions of the agents, with the parameter α decreasing linearly: where t is the current iteration count and T max is the maximum iteration count. The coefficient vectors and are respectively: where and are random vectors within [0, 1].

[0038] S24: Calculate the distances of the dominant agent, secondary agent, and auxiliary agent and update their positions: where and are the position vectors of the three-layer agents respectively.

[0039] S25: Perform boundary correction and rounding: SIC optimized = round(x SIC ).

[0040] S3: Use the Multiscale Noise Injection Mode Decomposition Algorithm (MSNIMD) to perform multiscale decomposition on the UWB-GPR signal, combine the adaptive multiscale spectral entropy to screen out the effective intrinsic oscillation mode components, and eliminate the high-frequency noise-dominated modes.

[0041] Perform MSNIMD decomposition on the echo signal to generate multiscale EOM components, separate the noise and the effective signal; according to the optimized DNG and SIC parameters obtained, inject adaptive multiscale noise into the echo signal to form a noisy signal; then extract the local maximum and minimum points for Multiscale Adaptive Oscillation Decomposition (MSAOD) decomposition, calculate the mean curve, extract the first EOM, and repeat the screening until the set stop criterion is met; then perform ensemble averaging on the energy of the EOM to obtain its ensemble; finally, perform screening to distinguish between the noise-dominated and signal-dominated EOM components.

[0042] S31: Perform MSNIMD decomposition to separate the noise and the effective signal. First, inject adaptive noise to construct a noisy signal: σ x = std(x), where k is the noise instance number, k = 1, 2,..., SIC. The noisy signal is: x( k ) = x + DNG·ε( k ).

[0043] S32: Extract the local extreme points, extract the maximum points M (k) and the minimum points N (k) of x (k) , and generate a curve: Calculate the mean curve as:

[0044] S33: Extract the first EOM: Repeat the screening until the stop criterion is met.

[0045] S34: Ensemble averaging: Obtain the ensemble {EOM1, EOM2,... EOM K}.

[0046] S35: Perform screening to distinguish between the noise-dominated and signal-dominated EOM components. First, calculate for each component: where m is 2, r = 0.15·std(EOM i ), then set the condition θ = μ - 0.5σ, screen the EOM that meets the condition: selectedEOM = {EOM i | SpEn i > θ}.

[0047] S4: Apply the non-linear coefficient shrinkage rule to the retained components to achieve secondary filtering of the residual noise in the frequency domain.

[0048] Process according to the improved non-linear coefficient shrinkage rule to suppress high-frequency noise and retain the signal edge features; first perform three-layer processing on the underground echo signal, then determine the suppression parameters, and then process according to the improved non-linear coefficient shrinkage rule, and finally re-synthesize the processed underground echo.

[0049] S41: Perform three-layer processing: [cA3, cD3, cD2, cD1] = wavedec(EOM i , 3,'sym,6'), where cA3 is the third-layer approximation coefficient and cD i are the detail coefficients of each layer.

[0050] S42: Determine the suppression parameters, and the formula is where σ is the estimated noise standard deviation and N is the signal length.

[0051] S43: Perform improved non-linear threshold processing: where α and γ are the attenuation rate of controlling small coefficients and the shrinkage intensity of adjusting large coefficients respectively.

[0052] S44: Perform signal synthesis processing, and the formula is:

[0053] S5: Re-synthesize to obtain the reflection waveform of the underground target with high fidelity and evaluate the algorithm performance.

[0054] S51: Superimpose the signals to obtain the final signal:

[0055] S52: Evaluate the Effective Signal Index (ESI):

[0056] S53: Evaluate the Error Signal Energy (ESE):

[0057] S54: Evaluate the Temporal Correlation Index (TCI):

[0058] The above has introduced in detail the UWB-GPR joint denoising method based on the hierarchical dynamic collaborative optimization algorithm provided by the present invention. Each embodiment in the specification is described in a progressive manner, and the key point of each embodiment is to illustrate the differences from other embodiments. The same and similar parts among the embodiments can be referred to each other.

[0059] The content not described in detail in the specification of the present invention belongs to the prior art well-known to those skilled in the art.

Claims

1. A UWB-GPR joint denoising method based on a hierarchical dynamic collaborative optimization algorithm, characterized in that: The following steps are involved: S1: By collecting the underground reflected waveforms received by UWB-GPR and antenna, a simple pre-processing algorithm is applied to eliminate the interference of near-field coupling and extract the effective signal interval; S2: Using the hierarchical dynamic collaborative optimization algorithm (HDCO) to globally optimize the dynamic noise gain factor and the number of statistical integration iterations for the objective function, the optimal decomposition parameter combination is generated; S3: The multi-scale noise injection modal decomposition algorithm (MSNIMD) is used to perform multi-scale decomposition of the signal, and the adaptive multi-scale spectral entropy is combined to screen the effective intrinsic oscillation modal components and eliminate the high-frequency noise-dominated modes; S4: A nonlinear coefficient shrinkage rule is applied to the retained components to achieve secondary filtering of residual noise in the frequency domain; S5: Resynthesize the high-fidelity underground target reflection waveform and evaluate the algorithm performance.

2. The UWB-GPR joint denoising method based on hierarchical dynamic collaborative optimization algorithm according to claim 1 is characterized in that: In S1: The underground target object is detected by UWB-GPR and antenna, with a detection range of 2×0.5 meters and a detection depth of 0.5 meters. The detection target object is an underground ant nest, and the A-scan and B-scan waveforms reflected underground are obtained, and one of the A-scan waveforms is selected for the experiment; since the echo signal obtained by UWB-GPR during detection contains a lot of interference information such as direct waves, in order to verify the algorithm of the present invention, a series of preprocessing is first performed on the echo signal to eliminate the interference of near-field coupling, and the effective signal interval is extracted.

3. The UWB-GPR joint denoising method based on hierarchical dynamic collaborative optimization algorithm according to claim 1 is characterized in that: In S2: The hierarchical dynamic collaborative optimization algorithm (HDCO) is used to globally optimize the dynamic noise gain factor (DNG) and the statistical integrated iteration number (SIC) for the objective function to generate the optimal decomposition parameter combination; By setting the parameter range of the dynamic noise gain factor and the number of statistical integration iterations, the position of the hierarchical agent is randomly generated, and the fitness calculation is performed with the purpose of minimizing the multi-scale spectral entropy to obtain the distance and position of the dominant agent Alpha, the secondary agent Beta, and the auxiliary agent Delta, and the boundary correction and rounding are performed; First, set the parameter ranges for the dynamic noise gain factor and the number of statistical integration iterations to randomly generate the positions of the layered agents: Positions i,j =1b j +rand()·(ub j -1b j ), where i is the serial number, j=1 corresponds to the dynamic noise gain factor, and j=2 corresponds to the number of statistical integration iterations; then, the fitness calculation is performed to minimize the multi-scale spectral entropy: Where K is the number of intrinsic oscillation modes (EOM), SpEn is the multi-scale spectral entropy, and B m (r) is the matching probability of the signal sequence under the embedding dimension m and tolerance; the position of the agent is updated, and the parameter α decreases linearly: Among them, t is the current iteration number, T max is the maximum number of iterations; the coefficient vector and They are: in, and is a random vector in [0,1]; Calculate the distance to the leading agent, secondary agent, and auxiliary agent and update the position: in, and They are the position vectors of the three-layer agents respectively; finally, boundary correction and rounding are performed: SIC optimized =round(x SIC ).

4. The UWB-GPR joint denoising method based on hierarchical dynamic collaborative optimization algorithm according to claim 1 is characterized in that: In S3: Perform MSNIMD decomposition on the echo signal to generate multi-scale EOM components and separate noise and effective signals; according to the optimized DNG and SIC parameters, inject adaptive multi-scale noise into the echo signal to form a noisy signal; then extract the local maximum and minimum points for multi-scale adaptive oscillation decomposition (MSAOD) decomposition, calculate the mean curve, extract the first EOM, and repeat the screening until the set stop criteria are met; then perform ensemble average calculation on the energy of the EOM to obtain its set; Finally, screening is performed to distinguish between noise-dominated and signal-dominated EOM components; First, perform MSNIMD decomposition to separate noise and valid signals; first inject adaptive noise to construct a noisy signal: σ x = std(x), where k is the noise instance number, k = 1, 2, ..., SIC; the noisy signal is: x( k )=x+DNG·ε( k ); Then extract the local extreme points and extract x( k )'s maximum point M( k ) and the minimum point N( k ), generating the curve: The mean curve is calculated as: Extract the first EOM: Repeat the screening until the stopping criterion is met; calculate the ensemble average: Get the set {EOM1, EOM2, ... EOM K }; Finally, screening is performed to distinguish between noise-dominated and signal-dominated EOM components; first, for each component, calculate: Where m is 2, r = 0.15·std(EOM i ); then set the condition θ=μ-0.5σ, Filter the EOM that meets the conditions: selectedEOM = {EOM i |SpEn i >θ}.

5. The UWB-GPR joint denoising method based on hierarchical dynamic collaborative optimization algorithm according to claim 1 is characterized in that: In S4: The improved nonlinear coefficient shrinkage rule is used to suppress high-frequency noise and retain the edge characteristics of the signal; the underground echo signal is first processed in three layers, the suppression parameters are determined, and then the signal is processed according to the improved nonlinear coefficient shrinkage rule, and finally the processed underground echo is resynthesized; First, the underground echo signal is decomposed into three layers: [cA3, cD3, cD2, cD1] = wavedec (EOM i ,3,'sym,6'), where cA3 is the third-layer approximation coefficient, cD i is the detail coefficient of each layer; then determine the suppression parameter, and the calculation formula Where σ is the noise standard deviation estimate, and N is the signal length; Then, the improved nonlinear coefficient shrinkage rule is used: Among them, α and γ are used to control the decay rate of small coefficients and adjust the contraction strength of large coefficients respectively; Finally, after processing:

6. The UWB-GPR joint noise reduction method based on hierarchical dynamic collaborative optimization algorithm according to claim 1 is characterized in that: In S5: The echoes processed by the algorithm are resynthesized, and the effective signal index (ESI), error signal energy (ESE) and timing correlation index (TCI) are evaluated to verify the practicality of the algorithm. The signals are superimposed to obtain the final signal: The resynthesized signal is evaluated from multiple perspectives to verify the reliability of the algorithm; the effective signal index is evaluated: Error signal energy: Timing correlation index:

Images