A method for suppressing ringing noise in ground-penetrating radar under non-uniform environments

Abstract

This invention belongs to the field of ground-penetrating radar (GPR) technology, and particularly relates to a method for suppressing ringing noise in GPR under non-uniform environments. The specific process is as follows: Step 1: Establish a non-uniform environment model and simulate ringing noise in the model; Step 2: Construct a background ringing noise matrix to suppress the ringing noise, obtaining echo data B after removing the ringing noise; Step 3: For the echo data matrix B obtained in Step 2, use singular value decomposition (SVD) to reduce the intensity of non-uniform clutter, obtaining echo data matrix C; Step 4: For the echo data matrix C, use symmetric filtering to suppress non-uniform clutter, obtaining echo data matrix S; Step 5: Using the peak signal-to-noise ratio (PSNR) between the echo data before processing and after processing in Steps 1-4 as the objective function, select parameters using a Bayesian optimization algorithm to obtain the parameter values ​​of the echo data matrix at the maximum PNR, ultimately achieving GPR ringing noise suppression.

Description

Technical Field

[0001] This invention belongs to the field of ground penetrating radar technology, and particularly relates to a method for suppressing ringing noise in ground penetrating radar under non-uniform environments. Background Technology

[0002] Ground penetrating radar (GPR) is a non-destructive testing technology that has seen rapid development in recent years. Its principle involves transmitting high-frequency electromagnetic waves into the ground via a transmitting antenna. When these waves propagate to areas with significant differences in underground material properties, they are reflected and refracted. The receiving antenna receives these reflected echoes, processes the signals, and forms an echo signal image, thus enabling the detection of underground targets. Due to its advantages such as wide detection range of target materials, high resolution, real-time imaging, and non-destructive testing, GPR has been widely applied in fields such as highway quality inspection, municipal pipeline inspection, geological and hydrological monitoring, building damage detection, and exploration. However, due to the physical structure and electrical characteristics of the transmitting antenna itself, ringing can occur. When the antenna transmits a pulse signal, not all the energy is released; some energy flows back and forth between the antenna tip and the input feed line. Each flow generates a secondary pulse, causing the antenna to transmit multiple gradually decaying pulses, resulting in ringing. Because of impedance mismatch between the antenna and the ground, these multiple gradually decaying pulses form reflected echoes on the ground, appearing as horizontal stripes covering the entire echo image. Generally, ringing noise masks the hyperbolic characteristics of the target echo, affecting the quality of ground-penetrating radar (GPR) echo images. GPR echo signals are also affected by non-uniform clutter caused by variations in soil conditions and surface roughness, making target echo signal detection more difficult.

[0003] To suppress non-uniform clutter and ringing noise in ground-penetrating radar (GPR), researchers have proposed several traditional methods, including mean cancellation, median cancellation, singular value decomposition (SVD), discrete wavelet transform (DWT), wavenumber domain filtering, and deep learning-based methods. Mean cancellation and median cancellation are simple in principle and easy to implement, but when used for ringing noise suppression, they tend to generate new horizontal stripes at the target echo apex and are less effective at suppressing non-uniform clutter. Singular value decomposition is also ineffective at suppressing ringing noise, and parameter selection relies on human experience. Discrete wavelet transform suffers from difficulties in selecting wavelet basis functions, as different basis functions can significantly affect the processing results. Wavenumber domain filtering can target the ringing noise based on its frequency, but for non-uniform clutter, a fixed wavenumber domain filter cannot effectively suppress it, requiring more complex adaptive filtering methods. Deep learning-based methods can adaptively achieve optimal noise suppression through a large amount of training data, and their noise suppression effect is good. However, they require a large amount of data for training, as well as a lot of computing resources and time. Furthermore, the generalization performance of the model is poor, resulting in poor noise suppression performance in new scenarios. Summary of the Invention

[0004] To address the aforementioned problems, this invention proposes a method for suppressing ringing noise in ground-penetrating radar under non-uniform environments. Compared to traditional methods, this invention solves the problems of difficulty in model and parameter selection and poor generalization performance under different scenarios, and demonstrates better suppression effects on non-uniform clutter and ringing noise.

[0005] The technical solution for implementing the present invention is as follows:

[0006] A method for suppressing ringing noise in ground-penetrating radar under non-uniform environments, the specific process of which is as follows:

[0007] Step 1: Establish a non-uniform environment model and simulate ringing noise in the environment model;

[0008] Step 2: Construct a background ringing noise matrix to suppress the ringing noise, and obtain echo data B after removing the ringing noise;

[0009] Step 3: For the echo data matrix B obtained in Step 2, the intensity of non-uniform clutter is reduced by using the singular value decomposition method to obtain the echo data matrix C;

[0010] Step 4: For the echo data matrix C, use symmetric filtering to suppress non-uniform clutter and obtain the echo data matrix S;

[0011] Step 5: Using the peak signal-to-noise ratio (PSNR) between the echo data before processing and after processing in steps 1-4 as the objective function, the parameters are selected using a Bayesian optimization algorithm to obtain the parameter values ​​of the echo data matrix at the maximum PNR, thus ultimately achieving ground-penetrating radar ringing noise suppression.

[0012] Optionally, the specific method for simulating ringing noise described in this invention is as follows: In a non-uniform environment model, the receiving antenna and the transmitting antenna are placed at the same horizontal position, and an iron plate is placed at a set distance from the transmitting antenna to reflect the pulse signal emitted by the transmitting antenna. The signal received by the receiving antenna simulates ringing noise.

[0013] Optionally, the specific process of step 2 in this invention is as follows:

[0014] First, take a horizontal window of length l, traverse the echo data matrix through the sliding window, and trim and weight the echo data in the window to initially remove the target echo signal and obtain the background ringing noise matrix.

[0015] Secondly, the generated background ringing noise matrix is ​​iteratively optimized until all target echo signals are removed, leaving only the background ringing noise matrix;

[0016] Finally, the background ringing noise matrix is ​​subtracted from the original signal echo data to obtain the echo data after removing the ringing noise.

[0017] Optionally, the trimming described in this invention involves: removing the minimum and maximum values ​​within the window according to a trimming ratio α, and averaging the remaining values ​​within the window to obtain the average value A of the window after trimming by ratio α. α ;

[0018] The weighting is as follows: A custom parameter s is set, and the weighting is calculated based on the remaining values ​​in the window and A. α The difference is normalized with weights, and then a weighted sum is performed to obtain the value A of the background ringing noise at the center of the window. rc .

[0019] Optionally, the weighting weights described in this invention are:

[0020]

[0021] Where s is a user-defined parameter, d i The magnitude 'a' of the i-th element in the window. i With A α The difference;

[0022] Then apply the weight W i Normalization is performed to obtain normalized weights.

[0023] Optionally, in step 3 of the present invention, the method of using singular value decomposition to reduce the intensity of non-uniform clutter is as follows: for strip-shaped non-uniform clutter generated by rough surfaces, the first one or several larger singular values ​​are selected for removal; for non-uniform clutter generated by underground sand and gravel, smaller singular values ​​or singular values ​​close to 0 are selected for removal.

[0024] Optionally, the specific process of step 4 in this invention is as follows: traverse all elements in matrix C using a sliding window, and obtain the position J with the strongest symmetry in each row of matrix C. m Calculate the strongest position J in each row. m The symmetric weight matrix α(i) is calculated, and the symmetric weight matrix β[i,j] at each element in the data matrix C is calculated. The influence factors γ and μ of the two weights are set, and the synthetic symmetric filtering weight matrix w[i,j] is calculated. The matrix w[i,j] is used to implement symmetric filtering of matrix C to suppress non-uniform clutter.

[0025] Optionally, the specific process of step 4 in this invention is as follows:

[0026] Calculate J for each row m The symmetric weight matrix α(i) at point i is obtained and normalized.

[0027]

[0028] Where J0(i) represents the column number at the position with the strongest symmetry in the i-th row;

[0029] Calculate the symmetric weight matrix β[i,j] at each element in the data matrix C, and then normalize it.

[0030]

[0031] Optionally, step 4 of this invention calculates the synthesized symmetric filtering weighting matrix w[i,j], and uses the matrix w[i,j] to perform symmetric filtering on matrix C to suppress non-uniform clutter; the specific process is as follows:

[0032] Calculate the synthetic symmetric filter weighting matrix w[i,j] and perform symmetric filtering to suppress non-uniform clutter, expressed as:

[0033] w[i,j]=e γ·α[i] ·e μ·β[i,j]

[0034] s[i,j]=c[i,j]·w[i,j]

[0035] Where γ and μ are the influence factors of the symmetric weight matrices α[i] and β[i,j], and s[i,j] is the signal after symmetric filtering to suppress non-uniform clutter.

[0036] Optionally, in step 5 of the present invention, parameters such as sliding window length l, trimming ratio α, weighting intensity s, and symmetric weight matrix influence factors γ and μ in the symmetric filtering process are selected using a Bayesian optimization algorithm.

[0037] Beneficial effects:

[0038] First, based on the characteristics of ringing noise, non-uniform clutter, and target echo, the present invention sets a horizontal sliding window in the received signal data matrix, and performs trimming and weighting processing on the signal within the window. By continuously iterating and updating the signal within the window, the invention achieves the suppression effect of non-uniform clutter and ringing noise.

[0039] Secondly, this invention first uses singular value decomposition to reduce the intensity of non-uniform clutter during non-uniform clutter suppression, thereby improving the suppression effect. For the selection of some key parameters, this invention uses a Bayesian optimization algorithm for automatic selection, solving the problem of difficult parameter selection. Simulation experiments have verified that, under a non-uniform environment model, this invention has good suppression effects on both non-uniform clutter and ringing noise. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a flowchart of the present invention;

[0042] Figure 2 This is a simulation model diagram of a non-uniform environment established by the method of this invention;

[0043] Figure 3 These are simulation results of the method of the present invention under non-uniform environment; Figure 3 (a) is the B-scan image obtained from gprMax simulation. Figure 3 (b) Echo data after adding ringing noise. Figure 3 (c) is the added ringing noise;

[0044] Figure 4 This is a diagram showing the ringing noise suppression results of the method of the present invention; Figure 4 (a) is the background ringing noise constructed using the method in step 2 of this invention. Figure 4 (b) is the result after ringing noise suppression;

[0045] Figure 5 This is a diagram showing the non-uniform clutter suppression results of the method of the present invention; Figure 5(a) is the result after suppressing ringing noise and then performing singular value decomposition. Figure 5 (b) is in Figure 5 (a) Based on the singular value decomposition process, symmetric filtering is performed to suppress non-uniform clutter.

[0046] Specific implementation process

[0047] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0048] It should be noted that, in the absence of conflict, the following embodiments and features can be combined with each other; and, based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0049] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0050] This application provides an embodiment of a method for suppressing ringing noise in ground-penetrating radar under non-uniform environments, such as... Figure 1 As shown, the specific process is as follows:

[0051] Step 1: Establish a non-uniform environment model and simulate ringing noise in the environment model;

[0052] The non-uniform environment model in this step was constructed using the Peplinski soil model in the gprMax simulation software. This model can simulate the non-uniform environment in real soil, taking into account the influence of soil components such as sand, clay, and moisture on the overall dielectric constant of the soil. Different non-uniform environment models can be established by controlling the proportion of sand and clay and the range of volumetric water content.

[0053] Ringing noise occurs because when an antenna transmits a pulse signal, not all the energy is released; some energy flows back and forth between the antenna tip and the input feed line. Each flow generates a secondary pulse, causing the antenna to transmit multiple gradually decaying pulses. Due to impedance mismatch between the antenna and the ground, these progressively decaying pulses are reflected by the ground, forming multiple ground reflection waves, thus creating ringing noise. In this step, we use gprMax simulation software to simulate ringing noise. A metal plate is placed 5cm away from the transmitting antenna to reflect the pulse signal emitted by the transmitting antenna. The receiving antenna is at the same horizontal position as the transmitting antenna, 10cm apart. The signal received by the receiving antenna is used to simulate ringing noise.

[0054] Step 2: Construct a background ringing noise matrix to suppress ringing noise and obtain echo data after removing ringing noise;

[0055] Ringing noise appears as large-amplitude horizontal stripes in radar echo signal images, almost completely masking the target echo signal and non-uniform clutter. First, based on the characteristics of the horizontal distribution of ringing noise, a sliding window is used to traverse the echo data matrix, and the echo data within the window is trimmed and weighted to initially remove the target echo signal, obtaining the background ringing noise matrix. Second, the generated background ringing noise matrix is ​​iteratively optimized until all target echo signals are removed, leaving only the background ringing noise matrix. Finally, this matrix is ​​subtracted from the original signal echo data to obtain the echo data after removing the ringing noise. The detailed steps are as follows:

[0056] In the radar echo signal matrix, a horizontal window of length l is taken starting from the first row. The window length can be changed according to different radar echo data, and the number of elements in the window is denoted as L. The elements in the window are arranged in ascending order, and then the minimum and maximum values ​​in the window are removed according to a set ratio α. Since the amplitude variation of ringing noise within the window is very small, it can be assumed that the removed extreme values ​​are generated by the target echo signal and non-uniform clutter. The average value of the remaining values ​​in the window is obtained by averaging the values ​​in the window after trimming by ratio α, denoted as A. α Represented as:

[0057]

[0058] Among them, a i It is the amplitude of the i-th element, N = αL.

[0059] The obtained mean A α It can be used as a reference value for the ringing noise in the center of the window, but due to the randomness of the α value setting, some extreme values ​​may not be removed within the window, thus affecting the average value. Therefore, α cannot be used as a reference value. αThis is equivalent to the true value of the background ringing noise. The solution is to compare the remaining values ​​within the window with A. α The difference is normalized with weights, and then a weighted sum is performed to obtain the value A of the background ringing noise at the center of the window. rc , where r and c represent the row and column of the window's center position. The weight W of the i-th element within the window. i Represented as:

[0060]

[0061] Where s is a user-defined parameter, representing the value based on a. i With A α The weighted intensity is set by the difference. d i The magnitude 'a' of the i-th element in the window. i With A α The difference is expressed as: Then, the weight W is... i Normalization is performed to obtain normalized weights. Reuse The amplitude values ​​within the window are weighted and summed to obtain the value A of the background ringing noise at the center of the window. rc The above process can be represented as follows:

[0062]

[0063] Slide the horizontal window along each row until the entire echo data matrix has been traversed, initially obtaining the background ringing noise matrix A. Repeat the same process for A, iterating a certain number of times until all target echoes and non-uniform clutter are removed, resulting in a matrix containing only background ringing noise. Then, subtract the background ringing noise matrix from the original echo data matrix to achieve the effect of suppressing ringing noise.

[0064] Step 3: For the echo data matrix B obtained in Step 2, the intensity of non-uniform clutter is reduced by using the singular value decomposition method to obtain the echo data matrix C;

[0065] Singular value decomposition (SVD) utilizes the properties of singular values ​​to decompose the echo data matrix after removing ringing noise into complementary subspaces of the target echo signal and the non-uniform clutter subspace, thereby increasing the signal-to-noise ratio. Let B be the echo data matrix after removing ringing noise in step 2, and its singular value decomposition is as follows:

[0066] B = U × S × V T

[0067] Where U and V are orthogonal matrices, and S is a singular value matrix, S = diag(σ1, σ2, ..., σ r ), r = rank(B). B T The eigenvalue of B is λ.i And it is sorted in descending order. These are called singular values ​​of matrix B.

[0068] Matrix B can be represented using singular values ​​as follows:

[0069]

[0070] B = B1 + B2 + ... + B k

[0071] Among them, B i It is a matrix with the same dimensions as B, called the eigenimage of B.

[0072] In this invention, based on the distribution characteristics of non-uniform clutter and its energy intensity relative to the target echo, singular values ​​corresponding to the non-uniform clutter can be flexibly selected for removal, thereby reducing the energy of the non-uniform clutter and highlighting the characteristics of the target echo signal. For strip-shaped non-uniform clutter generated by rough surfaces, since its clutter energy is relatively large compared to the target echo, the first or several larger singular values ​​should be selected for removal; for non-uniform clutter generated by underground sand and gravel, since its clutter energy is relatively small compared to the target echo, smaller singular values ​​or singular values ​​close to 0 should be selected for removal.

[0073] Step 4: For the echo data matrix C, use symmetric filtering to suppress non-uniform clutter and obtain the echo data matrix S;

[0074] In ground-penetrating radar echo images, the target echo exhibits a distinct hyperbolic characteristic, with symmetry on both sides of the hyperbola's apex. Based on the symmetry difference between the target echo and non-uniform clutter, symmetrical filtering can be used to suppress the non-uniform clutter. The detailed steps are as follows:

[0075] First, let the data matrix after singular value decomposition in step 3 be C, where each element is represented as c[i,j]. A sliding window of length K and width M is used to traverse all elements in matrix C. Within the sliding window, the position with the strongest symmetry in each row of matrix C is obtained by multiplying the values ​​of symmetrical elements and taking the maximum value; this position is highly likely to be the location of the target echo hyperbola. The above method is expressed as follows:

[0076]

[0077] Among them, J m It is the position with the strongest symmetry in each row of matrix C.

[0078] Secondly, calculate J for each row. m The symmetric weight matrix α(i) at point i is expressed as:

[0079]

[0080] After obtaining α[i], normalize it to eliminate the influence of excessive weight at the vertex of the echo hyperbola on the entire symmetric filtering process.

[0081] Next, calculate the symmetric weight matrix β[i,j] at each element in the data matrix C, expressed as:

[0082]

[0083] Corresponding to α[i], β[i,j] also needs to be normalized. Furthermore, during the calculation of the symmetric weight matrix, since the data matrix C may contain elements with a value of 0, the denominator of the symmetric weight matrix may be 0, leading to calculation errors. To address this, the average of the valid values ​​in the two rows immediately above and below the location of the calculation error is taken as the value at the location of the error.

[0084] Finally, the weighted matrix w[i,j] of the synthesized symmetric filter is calculated, and symmetric filtering is performed to suppress non-uniform clutter, as shown below:

[0085] w[i,j]=e γ·α[i] ·e μ·β[i,j]

[0086] s[i,j]=c[i,j]·w[i,j]

[0087] Where γ and μ are the influence factors of the symmetric weight matrices α[i] and β[i,j], and s[i,j] is the signal after symmetric filtering to suppress non-uniform clutter.

[0088] Step 5: Using the peak signal-to-noise ratio (PSNR) between the echo data before processing and after processing in steps 1-4 as the objective function, select parameters using a Bayesian optimization algorithm to obtain the parameter value that maximizes the PSNR, ultimately achieving suppression of ringing noise in ground-penetrating radar.

[0089] Bayesian optimization is an algorithm used to optimize an objective function by constructing a Gaussian process to select parameters and gradually approximate the optimal value of the objective function. In the process of constructing a background ringing noise matrix to suppress ringing noise, some important parameters, such as the sliding window length *l*, the signal trimming ratio *α* within the window, the weighting intensity *s*, and the influence factors *γ* and *μ* of the symmetric weight matrix in the symmetric filtering process, can be automatically selected using Bayesian optimization. Its objective function is the peak signal-to-noise ratio (PSNR) between the processed data and the unprocessed data in both processes. The parameter values ​​that maximize the PSNR are obtained through Bayesian optimization.

[0090] This invention proposes a method for suppressing ringing noise in ground-penetrating radar under non-uniform environments. The method suppresses ringing noise by constructing a background ringing noise matrix. After reducing the energy of non-uniform clutter using singular value decomposition, symmetric filtering is then applied to further suppress the non-uniform clutter. To address the difficulty of manually selecting parameters, a Bayesian optimization algorithm is used to obtain the parameter values ​​that maximize the peak signal-to-noise ratio (PSNR) as the objective function. Finally, the suppression effectiveness of the above method for ringing noise and non-uniform clutter is verified using simulation software.

[0091] This completes a method for suppressing ringing noise in ground-penetrating radar under non-uniform environments. Example

[0092] To verify the proposed method for suppressing ringing noise in ground-penetrating radar under non-uniform environments, a simulation experiment was designed for analysis. The generation of ringing noise and the construction of the non-uniform environment model are shown in step 1. The non-uniform environment model is 1m × 0.4m in size, with a mesh size of dx = dy = dz = 0.002m. The upper layer is a Peplinski soil model with a thickness of 0.05m, and the lower layer is a dry soil model with a thickness of 0.3m. In the Peplinski soil model, the proportion of sand is 0.7, the proportion of clay is 0.3, and the volumetric water content ranges from 0.001 to 0.1. The target profile is a circle with a radius of 0.02m and a burial depth of 0.15m. The transmitting antenna uses a Ricker wavelet with a frequency of 1.3GHz, the distance between the transmitting and receiving antennas is 0.05m, the antenna translation step size is 0.005m, and the time window length is 6ns. A schematic diagram of the simulation scenario is shown below. Figure 2 As shown.

[0093] Figure 3 (a) is the B-scan image obtained from gprMax simulation. In the original data, from top to bottom, there are strong surface reflection echoes, non-uniform soil model reflection echoes, and hyperbolic characteristic echoes of the target reflection. Figure 3 (b) Echo data after adding ringing noise. Figure 3 (c) is the added ringing noise. Due to the ringing noise, the target echo is completely masked, and it is impossible to distinguish whether there is a target underground from the image.

[0094] Figure 4 (a) is the background ringing noise constructed using the method in step 2 of this invention, and... Figure 3 (c) Compared to the added ringing noise, the background ringing noise constructed by this method essentially replicates the added background ringing noise. Figure 3 Subtracting the background ringing noise from the data in (b) yields the following result: Figure 4 (b) shows the result after ringing noise suppression, in which surface reflection echoes and ringing noise are removed, and the noise suppression effect is good.

[0095] Figure 5 (a) shows the result after suppressing ringing noise and then performing singular value decomposition, where the first two larger singular values ​​have been removed. Figure 4 (b) The non-uniform clutter horizontal stripe with higher energy. Figure 5 (b) is in Figure 5 (a) Based on singular value decomposition, symmetric filtering is performed to suppress non-uniform clutter. Figure 4 (b) Compared to the majority of non-uniform clutter being suppressed, the target echo is more prominent.

[0096] The above description, in conjunction with specific embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such deductions or substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for suppressing ringing noise in ground-penetrating radar under non-uniform environments, characterized in that, The specific process is as follows: Step 1: Establish a non-uniform environment model and simulate ringing noise in the environment model; Step 2: Construct a background ringing noise matrix to suppress the ringing noise, and obtain echo data B after removing the ringing noise; Step 3: For the echo data matrix B obtained in Step 2, the intensity of non-uniform clutter is reduced by using singular value decomposition to obtain the echo data matrix C; Step 4: For the echo data matrix C, use symmetric filtering to suppress non-uniform clutter and obtain the echo data matrix S; Step 5: Using the peak signal-to-noise ratio (PSNR) between the echo data before processing and after processing in steps 1-4 as the objective function, the parameters are selected using a Bayesian optimization algorithm to obtain the parameter values ​​of the echo data matrix at the maximum PNR, thus ultimately achieving suppression of ringing noise in ground-penetrating radar. The specific process of step 4 is as follows: traverse all elements of matrix C using a sliding window, and obtain the position J with the strongest symmetry in each row of matrix C. m Calculate the strongest position J in each row. m Given a symmetric weight matrix α[i], calculate the symmetric weight matrix β[i,j] at each element in the data matrix C; Two influence factors, γ and μ, are set, and a symmetric filtering weighting matrix w[i,j] is calculated. The matrix w[i,j] is used to perform symmetric filtering on matrix C to suppress non-uniform clutter.

2. The method for suppressing ringing noise of ground-penetrating radar in non-uniform environments according to claim 1, characterized in that, The specific method for simulating ringing noise is as follows: In a non-uniform environment model, the receiving antenna and the transmitting antenna are placed at the same horizontal position. An iron plate is placed at a set distance from the transmitting antenna to reflect the pulse signal emitted by the transmitting antenna. The signal received by the receiving antenna simulates ringing noise.

3. The method for suppressing ringing noise of ground-penetrating radar in non-uniform environments according to claim 1, characterized in that, The specific process of step 2 is as follows: First, take a horizontal window of length l, traverse the echo data matrix through the sliding window, and trim and weight the echo data in the window to initially remove the target echo signal and obtain the background ringing noise matrix. Secondly, the generated background ringing noise matrix is ​​iteratively optimized until all target echo signals are removed, leaving only the background ringing noise matrix; Finally, the background ringing noise matrix is ​​subtracted from the original signal echo data to obtain the echo data after removing the ringing noise.

4. The method for suppressing ringing noise of ground-penetrating radar in non-uniform environments according to claim 3, characterized in that, The trimming process involves removing the minimum and maximum values ​​within the window according to a trimming ratio α, and then averaging the remaining values ​​to obtain the average value A of the window after trimming by ratio α. α ; The weighting is as follows: A custom parameter s is set, and the weighting is calculated based on the remaining values ​​in the window and A. α The difference is normalized with weights, and then a weighted sum is performed to obtain the value A of the background ringing noise at the center of the window. rc .

5. The method for suppressing ringing noise of ground-penetrating radar in non-uniform environments according to claim 4, characterized in that, The weights for the weighting are: Where s is a user-defined parameter, d i The magnitude 'a' of the i-th element in the window. i With A α The difference; Then apply the weight W i Normalization is performed to obtain normalized weights.

6. The method for suppressing ringing noise of ground-penetrating radar in non-uniform environments according to claim 1, characterized in that, In step 3, the method of using singular value decomposition to reduce the intensity of non-uniform clutter is as follows: for strip-shaped non-uniform clutter generated by rough surfaces, the first one or several larger singular values ​​are selected for removal; for non-uniform clutter generated by underground sand and gravel, smaller singular values ​​or singular values ​​close to 0 are selected for removal.

7. The method for suppressing ringing noise of ground-penetrating radar in non-uniform environments according to claim 1, characterized in that, Each element of the data matrix C is represented as c[i,j]. A sliding window of length K and width M is used to traverse all elements in the data matrix C. Calculate J for each row m The symmetric weight matrix α[i] at point i is obtained and normalized. Where J0(i) represents the column number at the position with the strongest symmetry in the i-th row; Calculate the symmetric weight matrix β[i,j] at each element in the data matrix C, and then normalize it.

8. The method for suppressing ringing noise of ground-penetrating radar in non-uniform environments according to claim 1, characterized in that, Step 4 calculates the synthesized symmetric filtering weighting matrix w[i,j], and uses this matrix w[i,j] to perform symmetric filtering on matrix C to suppress non-uniform clutter; the specific process is as follows: Calculate the synthetic symmetric filter weighting matrix w[i,j] and perform symmetric filtering to suppress non-uniform clutter, expressed as: w[i,j]=e γ·α[i] ·have been μ·β[i,j] s[i,j]=c[i,j]·w[i,j] Where γ and μ are the influence factors of the symmetric weight matrices α[i] and β[i,j], and s[i,j] is the signal after symmetric filtering to suppress non-uniform clutter.

9. The method for suppressing ringing noise of ground-penetrating radar in non-uniform environments according to claim 1, characterized in that, In step 5, parameters such as sliding window length l, pruning ratio α, weighting intensity s, and symmetric weight matrix influence factors γ and μ in the symmetric filtering process are selected using a Bayesian optimization algorithm.

Images