Ground penetrating radar data fusion optimization method for tunnel detection

Abstract

The invention discloses a ground penetrating radar data fusion optimization method for tunnel detection. The method comprises the following steps: step 1, selecting an antenna frequency; step 2, cutting data; step 3, preprocessing the data; step 4, multi-frequency data space-time standardization; step 5, multi-frequency data fusion; step 6, Gaussian blur processing; and 7, image screening: selecting an optimized image with the best robustness. The method disclosed by the invention is specially used for tunnel detection radar data, interferences of metal davits, expansion joints and the like can be effectively removed, the fused multi-frequency data naturally transits near the boundary, the boundary and double images basically disappear, and the problems in practical application of tunnel engineering detection are effectively solved. According to the method, contradictory unification of large detection depth and high resolution is realized on the same image, targets such as steel bars and layers in the image are more easily and automatically identified by an artificial intelligence algorithm, and a technical foundation is laid for subsequent AI interpretation.

Description

Technical Field

[0001] The present invention belongs to the field of tunnel non-destructive testing, and in particular relates to a ground penetrating radar data fusion optimization method for tunnel testing in this field. Background Art

[0002] Ground-penetrating radar (GPR) is a safe and convenient nondestructive testing device widely used in tunnel engineering quality inspections. In recent years, numerous universities and research institutions at home and abroad have developed vehicle-mounted, non-contact GPR systems. These systems, primarily composed of a radar host, a multi-frequency antenna, and a telescopic bracket, can simultaneously collect data from multiple survey lines. Processing yields information such as lining thickness, cavities behind the lining, cracks in the secondary lining concrete, and voids in the secondary lining concrete. Currently, GPR still faces several challenges in practical detection. Multi-frequency data corresponds to multiple images, requiring comprehensive consideration of the detection results from multiple antennas for the same inspection location, which can lead to difficulty identifying and accurately locating defects. Furthermore, the multi-frequency, multi-line measurement method leads to a surge in data volume, exponentially increasing the workload for data processing personnel. This wastes human resources and negatively impacts tunnel inspection work.

[0003] GPR data fusion processing is a way to solve the above problems. In 2014, the team from Central South University used a deterministic deconvolution interpolation method to fuse the data collected by 100MHz and 400MHz antennas. The information carried by the echo data can be displayed on a single graph, but there are many multiple waves after fusion, making it difficult to identify defects. The following year, the team used S transform to fuse multi-frequency data, improving the image resolution. In 2018, Xu Xianlei and others from China University of Mining and Technology sorted out the process of ground penetrating radar multi-frequency data fusion processing. This method is closer to practical applications, but there are some "ghosting" of the target after fusion, which is not conducive to interpretation. In 2020, the Tongji University team performed weighted fusion of multi-frequency data in the time domain and frequency domain. Although the boundary line of the fused image is obvious, the type and location of the defect near the boundary line are difficult to determine. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a ground penetrating radar data fusion optimization method for tunnel detection.

[0005] The present invention adopts the following technical solutions:

[0006] A ground penetrating radar data fusion optimization method for tunnel detection, the improvement of which comprises the following steps:

[0007] Step 1, select the antenna frequency:

[0008] Select a high-frequency antenna with a center frequency of 800MHz-1500MHz and a medium-frequency antenna with a center frequency of 300MHz-400MHz to work together;

[0009] Step 2, data trimming:

[0010] Cut the original data into segments according to the data record and the marking position on the original data;

[0011] Step 3, data preprocessing;

[0012] Perform data preprocessing on the cropped data;

[0013] Step 4: Spatiotemporal normalization of multi-frequency data:

[0014] Unify the data of different frequencies into the same coordinate system with the distance and horizontal directions as the coordinate axes;

[0015] Step 5: Multi-frequency data fusion:

[0016] The medium frequency and high frequency data processed in step 4 are fused using the data fusion method;

[0017] Step 6: Gaussian blur processing;

[0018] Step 7, image screening:

[0019] Select the optimized image with the best robustness.

[0020] Furthermore, the data preprocessing operations in step 3 include zero bias correction, digital filtering, gain adjustment, zero point position adjustment and background elimination processing.

[0021] Furthermore, in step 4, horizontally, the maximum number of scanned channels in the data of different frequencies is used as a standard to interpolate the channel numbers of the data of other frequencies so that the channel numbers of all the data in the horizontal direction are consistent, thereby completing the horizontal standardization.

[0022] Furthermore, in step 4, the data is interpolated in the range direction, and the intermediate frequency radar data is set as R m (i,j), where the sampling point of each data is i=1,2,...,i m , the data is located in channel j = 1, 2, ..., j m , the time window is T m ; High frequency radar data is R h (p,q), where the sampling point of each data is p=1,2,...,p h , the data is located in channel q=1,2,...,q h , the time window is T h ;

[0023] Due to Tm ≥T h , interpolate the intermediate frequency data in the distance upward, and the number of sampling points of each intermediate frequency data after interpolation is:

[0024]

[0025] At this time, the sampling point of each channel of the intermediate frequency radar data is k, k=1,2,...,i m_new , the interpolation position value is:

[0026]

[0027] In the above formula, w1 is the weight, 0≤w1≤1.

[0028] Furthermore, the data fusion methods in step 5 include Bayesian reasoning method, principal component analysis method, DS reasoning method, Laplace pyramid method, LP fusion method and two-dimensional wavelet transform method.

[0029] Furthermore, the two-dimensional Gaussian function in step 6 is:

[0030]

[0031] In the above formula, x, y are the coordinate positions of the pixel points on the two-dimensional image, r is the blur radius, r 2 =x 2 +y 2 , σ is the standard deviation of the normal distribution, G(x, y) is the blur kernel, and different σ values ​​are selected to obtain multiple blur kernels, which are convolved with the fused image to obtain multiple optimized images.

[0032] Furthermore, in step 7, the parameters of multiple optimized images are calculated, including information entropy, average gradient, root mean square error, and peak signal-to-noise ratio, and the variation curves of these parameters when the standard deviation of the Gaussian function takes different values ​​are obtained, and the optimized image with the best robustness is selected.

[0033] The beneficial effects of the present invention are:

[0034] The method disclosed in this paper is specifically designed for tunnel inspection radar data, effectively removing interference from metal davits, expansion joints, and other sources. The fused multi-frequency data transitions naturally near demarcation lines, virtually eliminating demarcation lines and ghosting, effectively resolving challenges encountered in tunnel engineering inspection applications. This method achieves the contradictory balance of deep detection and high resolution within the same image, making it easier for artificial intelligence algorithms to automatically identify objects such as rebar and layers within the image, laying the technical foundation for subsequent AI interpretation. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic flow diagram of the method of the present invention;

[0036] Figure 2 It is a schematic diagram of the tunnel structure;

[0037] Figure 3 It is the raw data image of 400MHz ground penetrating radar;

[0038] Figure 4 It is the raw data image of 900MHz ground penetrating radar;

[0039] Figure 5 This is the cropped 400MHz data image;

[0040] Figure 6 This is the cropped 900MHz data image;

[0041] Figure 7 It is the pre-processed 400MHz data image;

[0042] Figure 8 It is the pre-processed 900MHz data image;

[0043] Figure 9 This is the 400MHz data image after time and space normalization;

[0044] Figure 10 This is the 900MHz data image after time and space normalization;

[0045] Figure 11 is the fused data image;

[0046] Figure 12 It is a graph of information entropy that changes with the standard deviation of the Gaussian function;

[0047] Figure 13 is a graph of the average gradient as the standard deviation of the Gaussian function changes;

[0048] Figure 14 This is the optimized image when σ is 1.5. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0050] Example 1: This example discloses a ground penetrating radar data fusion optimization method for tunnel detection, such as Figure 1 As shown, the following steps are included:

[0051] Step 1, select the antenna frequency:

[0052] Ground-penetrating radar (GPR) scans along the surface of the secondary lining. The transmitting antenna (Tx) emits electromagnetic waves into the lining. These waves pass through the secondary lining, waterproofing layer, primary support, and surrounding rock. When they encounter locations with varying relative permittivity, they are reflected and picked up by the receiving antenna (Rx). Locations with varying relative permittivity, such as voids, inter-layer variations, loose compaction, water-rich areas, and rebar, are reflected in the radar image.

[0053] The thickness of the initial support shotcrete is generally 5-30cm, and the thickness of the secondary lining concrete is 30-60cm. Taking into account the thickness of the surrounding rock, the GPR detection depth of the tunnel inspection should be at least 200cm, and the distance resolution should be better than 5cm.

[0054] According to the distance resolution r, the average relative dielectric constant ε of the tunnel lining r2 According to formula (1), the radar transmitting pulse center frequency f can be preliminarily determined c If the center frequency selected in this way cannot meet the maximum detection depth, the center frequency should be lowered to give priority to meeting the detection depth.

[0055]

[0056] Center frequency f c The unit is megahertz (MHz). Considering the tunnel lining and surrounding rock as a whole, which is mainly composed of reinforced concrete and stone, ε is taken here. r2 =10. Based on the minimum initial support thickness, r = 0.05m, the resulting center frequency is greater than 474MHz. However, the maximum detection depth of a GPR with a center frequency of 474MHz is approximately 1m, which is less than the combined thickness of the tunnel lining and surrounding rock. Therefore, the center frequency that meets the maximum detection depth should be appropriately lowered. Considering the antenna frequencies used by mainstream off-the-shelf GPR products in the market, a range of 300MHz-400MHz is selected.

[0057] On the other hand, according to the deepest detection distance h of the radar and the average relative dielectric constant ε of the tunnel lining r2 , formula (2) can limit the radar transmission pulse center frequency f c The maximum value of:

[0058]

[0059] Here h = 2m, f c At the same time, considering the requirements of distance resolution and horizontal resolution, a high-frequency antenna with a center frequency of 800MHz-1500MHz and a medium-frequency antenna with a center frequency of 300MHz-400MHz are selected to work together to meet the requirements of large detection depth and high resolution.

[0060] Step 2, data trimming:

[0061] GPR detects tunnel layers, defects, and other targets, primarily along the vault, sidewall, and invert survey lines. Multiple-frequency radars can be integrated, allowing data from two channels with different frequencies to be captured on a single survey line. Alternatively, separate detection can be employed, with the start and end points calibrated on the survey line before detection. During detection, the data is marked to record the start and end points. During data collection, any locations that cause data discontinuities, such as metal davits or expansion joints, should be marked immediately.

[0062] In order to match radar data of different frequencies to the same detection position, and provide a prerequisite for multi-frequency data fusion at the same detection position, and at the same time eliminate the data corresponding to metal davits and expansion joints at fixed positions on the data that do not need to be fused or even interfere with the fusion, the raw data should be segmented according to the paper data records and the marking positions on the raw data;

[0063] Step 3, data preprocessing;

[0064] The cropped data undergoes data preprocessing, including zero-bias correction, digital filtering, gain adjustment, zero-point position adjustment, and background removal. This processing significantly reduces radar multiple fixed waves, limits the data frequency to its bandwidth, removes some high-frequency clutter, and achieves a consistent zero-time position for all data.

[0065] Step 4: Spatiotemporal normalization of multi-frequency data:

[0066] Fusion processing can only be performed by unifying data of different frequencies in the same coordinate system with distance and horizontal axes. Data cropping ensures that the starting and ending positions of data of different frequencies are aligned.

[0067] In the range direction, the intermediate frequency data window is greater than or equal to the high frequency data window, and the intermediate frequency data needs to be interpolated. Assume that the intermediate frequency radar data is R m (i,j), where the sampling point of each data is i=1,2,...,i m , the data is located in channel j = 1, 2, ..., j m , the time window is T m ; High frequency radar data is R h (p,q), where the sampling point of each data is p=1,2,...,p h , the data is located in channel q=1,2,...,q h , the time window is T h ;

[0068] Due to T m ≥T h , interpolate the intermediate frequency data in the distance upward, and the number of sampling points of each intermediate frequency data after interpolation is:

[0069]

[0070] At this time, the sampling point of each channel of the intermediate frequency radar data is k, k=1,2,...,i m_new , the interpolation position value is:

[0071]

[0072] In the above formula, w1 is the weight, 0≤w1≤1.

[0073] In the horizontal direction, the maximum number of scan channels in the data of different frequencies is used as the standard to interpolate the channel numbers of the data of other frequencies so that the channel numbers of all the data in the horizontal direction are consistent, thus completing the horizontal standardization.

[0074] With j m >q h For example, interpolate the high-frequency data to obtain the channel l where the high-frequency data is located after interpolation, l=1,2,...,j m , the interpolation position value is:

[0075]

[0076] In the above formula, w2 is the weight, 0≤w2≤1.

[0077] Step 5: Multi-frequency data fusion:

[0078] The intermediate frequency and high frequency data processed in step 4 are fused using a data fusion method. The data fusion methods include but are not limited to Bayesian reasoning, principal component analysis (PCA), DS reasoning, Laplace pyramid method, LP fusion method and two-dimensional wavelet transform (WT).

[0079] Step 6: Gaussian blur processing;

[0080] The two-dimensional Gaussian function is:

[0081]

[0082] In the above formula, x, y are the coordinate positions of the pixel points on the two-dimensional image, r is the blur radius, r 2 =x 2 +y 2 , σ is the standard deviation of the normal distribution, G(x, y) is the blur kernel, and different σ values ​​are selected to obtain multiple blur kernels, which are convolved with the fused image to obtain multiple optimized images.

[0083] Step 7, image screening:

[0084] Calculate the parameters of multiple optimized images, including but not limited to information entropy, average gradient, root mean square error, and peak signal-to-noise ratio, and obtain the variation curves of these parameters when the standard deviation of the Gaussian function takes different values. Select the optimized image with the best robustness.

[0085] The steps of the method of the present invention are described in detail with examples below:

[0086] Step 1: Use medium frequency 400MHz and high frequency 900MHz ground penetrating radar to conduct tests on a tunnel. The tunnel structure is as follows: Figure 2 shown.

[0087] The 400MHz radar window is set to 50ns, and the 900MHz radar window is set to 25ns. The two frequency radars are set with the same track spacing, and the number of scanning points is 512. The 400MHz and 900MHz ground penetrating radars are detected along the same survey line. During the detection, the starting and ending positions of the survey line, metal davits, expansion joints, and other locations that cause discontinuous data collection and data anomalies are marked on the data. The raw data collected by the 400MHz and 900MHz ground penetrating radars are as follows: Figure 3 and Figure 4 shown.

[0088] Step 2: Crop the original data according to the marking position on the original data.

[0089] Select the data before the starting position of the survey line, the data after the ending position, the data of the metal davits, the expansion joints and other locations, and delete them. Then, splice the remaining data in chronological order. The trimmed 400MHz data and 900MHz data are as follows: Figure 5 and Figure 6 Data cropping ensures that the start and end positions of the 400MHz data and the 900MHz data are aligned.

[0090] Step 3: preprocess the data obtained in step 2.

[0091] Preprocessing includes zero bias correction, digital filtering, gain adjustment, zero point adjustment and background elimination. The preprocessed 400MHz data and 900MHz data are as follows: Figure 7 and Figure 8 As shown in the figure, the radar's multiple fixed waves are significantly reduced, the frequency of the data is limited to its bandwidth, some high-frequency clutter is filtered out, and the zero-time position of all data is unified.

[0092] Step 4: Perform spatiotemporal normalization on the data obtained in step 3.

[0093] The 400MHz and 900MHz data must be in the same coordinate system with the range and horizontal axes for fusion processing. Horizontally, the channels of the other frequency data are interpolated using the maximum number of channels among the different frequency data to achieve the same channel number for all horizontal data, completing horizontal standardization. After processing in step 3, the 400MHz data has 1500 channels, while the 900MHz data has 1400 channels. Therefore, the 900MHz data is interpolated. The interpolation position is determined according to equation (5), with a weight of 0.5.

[0094] In the range-up direction, the 400MHz radar window is 50ns, and the 900MHz radar window is 25ns. The intermediate frequency data window is larger than the high frequency data window, requiring interpolation of the intermediate frequency data. The number of radar scan points for both frequencies is 512. Interpolation of the 400MHz data, calculated according to Equation (3), yields 1024 interpolated scan points. The interpolation position is determined according to Equation (4), with a weight of 0.5.

[0095] The 400MHz data and 900MHz data after time and space normalization are as follows: Figure 9 and Figure 10 shown.

[0096] Step 5: Use data fusion method to fuse the intermediate frequency and high frequency data processed in step 4.

[0097] Data fusion methods include but are not limited to Bayesian inference, principal component analysis (PCA), DS inference, Laplace pyramid (LP), and two-dimensional wavelet transform (WT). Here, the two-dimensional wavelet transform method is used, with db2 wavelet as the mother wavelet and a decomposition scale of 4 levels. The fused image is as follows Figure 11 As shown in the figure, the yellow rectangular box marks the position of the dividing line, and the dividing line left after the fusion of different frequency data can be clearly seen.

[0098] Step 6: Gaussian blur processing.

[0099] The standard deviation σ of the two-dimensional Gaussian function is set to 0.1, 0.5, 1, 1.5, 2, 2.5, 3 to obtain multiple blur kernels, which are convolved with the fused image to obtain multiple optimized images.

[0100] Step 7: Image screening.

[0101] Calculate the information entropy and average gradient of the multiple optimized images obtained in step 6, and draw the image information entropy and average gradient curves corresponding to different Gaussian function standard deviations, as shown below: Figure 12 and Figure 13 Considering the above two parameter curves, as shown in Figure 14As shown in the figure, when σ is 1.5, the upper and lower transitions of the dividing line in the image are natural, the layers and trends below the dividing line can be clearly displayed, and the ghosting of other targets such as shallow steel bars basically disappears.

Claims

1. A ground penetrating radar data fusion optimization method for tunnel detection, characterized in that: The steps include: Step 1, select the antenna frequency: Select a high-frequency antenna with a center frequency of 800MHz-1500MHz and a medium-frequency antenna with a center frequency of 300MHz-400MHz to work together; Step 2, data trimming: Cut the original data into segments according to the data record and the marking position on the original data; Step 3, data preprocessing; Perform data preprocessing on the cropped data; Step 4: Spatiotemporal normalization of multi-frequency data: Unify the data of different frequencies into the same coordinate system with the distance and horizontal directions as the coordinate axes; Step 5: Multi-frequency data fusion: The medium frequency and high frequency data processed in step 4 are fused using the data fusion method; Step 6: Gaussian blur processing; Step 7, image screening: Select the optimized image with the best robustness.

2. The ground penetrating radar data fusion optimization method for tunnel detection according to claim 1, characterized in that: The data preprocessing operations in step 3 include zero bias correction, digital filtering, gain adjustment, zero point position adjustment and background elimination processing.

3. The method for optimizing ground penetrating radar data fusion for tunnel detection according to claim 1, characterized in that: In step 4, horizontally, the maximum number of scanned channels in the different frequency data is used as the standard to interpolate the channel numbers of the other frequency data to make the channel numbers of all data in the horizontal direction consistent, thus completing the horizontal standardization.

4. The method for optimizing ground penetrating radar data fusion for tunnel detection according to claim 1, characterized in that: In step 4, interpolate the data in the range direction, assuming that the intermediate frequency radar data is R m (i,j), where the sampling point of each data is i=1,2,...,i m , the data is located in channel j = 1, 2, ..., j m , the time window is T m ; High frequency radar data is R h (p,q), where the sampling point of each data is p=1,2,...,p h , the data is located in channel q=1,2,...,q h , the time window is T h ; Due to T m ≥T h , interpolate the intermediate frequency data in the distance upward, and the number of sampling points of each intermediate frequency data after interpolation is: At this time, the sampling point of each channel of the intermediate frequency radar data is k, k=1,2,...,i m_new , the interpolation position value is: In the above formula, w1 is the weight, 0≤w1≤1.

5. The ground penetrating radar data fusion optimization method for tunnel detection according to claim 1, characterized in that: The data fusion methods in step 5 include Bayesian inference method, principal component analysis method, DS inference method, Laplace pyramid method, LP fusion method and two-dimensional wavelet transform method.

6. The ground penetrating radar data fusion optimization method for tunnel detection according to claim 1, characterized in that: The two-dimensional Gaussian function in step 6 is: In the above formula, x, y are the coordinate positions of the pixel points on the two-dimensional image, r is the blur radius, r 2 =x 2 +y 2 , σ is the standard deviation of the normal distribution, G(x, y) is the blur kernel, and different σ values ​​are selected to obtain multiple blur kernels, which are convolved with the fused image to obtain multiple optimized images.

7. The method for optimizing ground penetrating radar data fusion for tunnel detection according to claim 6, characterized in that: In step 7, the parameters of multiple optimized images are calculated, including information entropy, average gradient, root mean square error, and peak signal-to-noise ratio. The variation curves of these parameters when the standard deviation of the Gaussian function takes different values ​​are obtained, and the optimized image with the best robustness is selected.

Images