Ground penetrating radar to-be-measured map adaptive optimization and generation method suitable for YOLO model

Abstract

The invention provides a ground penetrating radar to-be-measured map adaptive optimization and generation method suitable for a YOLO model. The method comprises the steps of obtaining ground penetrating radar data; null drift removal processing is carried out on the collected ground penetrating radar data; according to the mean value of the amplitude intensity of each sampling point in the radar data subjected to zero drift removal, calculating the change trend of the radar wave amplitude by adopting a median filtering method; according to the radar data subjected to zero drift removal, carrying out statistics on a ground penetrating radar early-stage signal amplitude feature extreme value; determining an amplitude compensation initial sampling point, and performing adaptive amplitude compensation based on a radar wave amplitude change trend; performing spectral analysis and digital filtering on the ground penetrating radar data after amplitude compensation by using fast Fourier transform, and eliminating direct current interference in the ground penetrating radar data after digital filtering; and according to the ground penetrating radar data after DC interference elimination, generating a ground penetrating radar to-be-measured map sequence adaptive to the YOLO model. The problems of poor to-be-measured map quality, low signal-to-noise ratio and slow generation speed of the existing ground penetrating radar are solved.

Description

Technical Field

[0001] This invention relates to the technical field of ground-penetrating radar (GPR) map optimization, and more particularly to an adaptive optimization and generation method for GPR maps to be measured. Background Technology

[0002] Ground-penetrating radar (GPR) boasts advantages such as high precision, high efficiency, continuous non-destructive imaging, real-time imaging, and intuitive results, making it a primary technology for long-distance, engineered, and highly efficient non-destructive detection of hidden road defects. However, data processing and identification of hidden road defects still rely heavily on human experience, leading to strong subjectivity, a lack of unified standards, and significant time and labor costs. Especially with the widespread application of multi-channel systems, the sheer volume of data has increased dramatically, making it difficult for human experience to guarantee timely data processing and accurate defect identification, easily resulting in missed diagnoses and misdiagnoses. Converting GPR data into a series of GPR maps and applying the YOLO deep learning algorithm for intelligent identification can significantly improve the speed, accuracy, and precision of diagnosing hidden road defects, while significantly reducing labor and testing costs. It also provides technical support for targeted, minimally invasive treatment of hidden road defects, enabling rapid elimination of road safety hazards, prevention of major road disasters, improvement of the safety and resilience of road infrastructure, and protection of public safety and property.

[0003] The ground-penetrating radar (GPR) map characteristics of hidden road defects are not only related to the type, shape, size, and depth of the defects, but also affected by the heterogeneity of the surrounding road structural layer materials, resulting in significant noise, chaotic waveforms, and low signal-to-noise ratio in the radar data. Therefore, before converting GPR data into GPR maps, different signal processing algorithms are needed to process and optimize the data. Patent application number 202511022032.6 discloses a method for generating three-dimensional GPR maps. It employs preprocessing methods such as zero-bias removal, zero-point adjustment, gain adjustment, digital filtering, and background elimination to remove inherent low-frequency DC components, enhance echo signals from deep defects, eliminate interference waves and lateral oscillation stripes at fixed positions, and highlight abnormal targets. However, this method still relies on manual experience to process and optimize GPR data; gain adjustment requires manual setting, and the frequency band of digital filtering follows a fixed rule. There are numerous ground-penetrating radar (GPR) manufacturers, resulting in diverse data formats, varying acquisition parameters (such as antenna frequency, time window, number of sampling points, and channel spacing), and different application scenarios (such as municipal roads, highways, national and provincial trunk roads, and rural roads). Processing and optimizing GPR data based on human experience is not only highly subjective and lacks unified standards, but is also time-consuming and labor-intensive, making it difficult to meet the high-speed and automated requirements of AI-based diagnostics of GPR maps for processing massive amounts of raw GPR data. Summary of the Invention

[0004] To address the technical problems of poor image quality, low signal-to-noise ratio, and low generation efficiency caused by the strong subjectivity in current empirical methods for optimizing and generating ground-penetrating radar (GPR) maps, this invention proposes an adaptive optimization and generation method for GPR maps suitable for the YOLO model. By statistically analyzing the amplitude and spectral distribution range of GPR data, the method adaptively adjusts the amplitude compensation parameters and digital filtering parameters to optimize the quality of GPR maps and generate a sequence of GPR maps suitable for intelligent recognition by the YOLO model. This overcomes the problems of poor image quality, low signal-to-noise ratio, and slow generation speed in current GPR map generation methods.

[0005] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0006] An adaptive optimization and generation method for ground-penetrating radar (GPR) target maps applicable to the YOLO model, comprising the following steps:

[0007] Step 1: Use ground-penetrating radar to collect data on hidden road defects and obtain ground-penetrating radar data;

[0008] Step 2: Perform zero-drift removal processing on the collected ground-penetrating radar data to obtain zero-drift removed radar data;

[0009] Step 3: Based on the mean amplitude intensity of each sampling point in the zero-drift radar data, the median filtering method is used to calculate the trend of radar wave amplitude variation;

[0010] Step 4: Statistically analyze the extreme values ​​of the early signal amplitude characteristics of ground-penetrating radar based on the radar data after zero-drift removal;

[0011] Step 5: Determine the starting sampling point for amplitude compensation based on the extreme values ​​of the early radar signal amplitude characteristics, and perform adaptive amplitude compensation based on the trend of radar wave amplitude variation;

[0012] Step 6: Use Fast Fourier Transform to perform spectrum analysis on the amplitude-compensated ground penetrating radar data to determine the filtering parameters, perform digital filtering based on the filtering parameters, and eliminate DC interference in the digitally filtered ground penetrating radar data.

[0013] Step 7: Generate a sequence of ground-penetrating radar (GPR) test maps adapted to the YOLO model based on the GPR data after DC interference cancellation.

[0014] Furthermore, in step 1, the ground-penetrating radar (GPR) uses a shielded ground-coupled GPR antenna combined with a profiling method to collect multi-channel waveform data of hidden road defects at equal intervals, thereby acquiring two-dimensional GPR data. Where m = 1, 2, ..., M represents the sampling point number, M is the number of sampling points, and n = 1, 2, ..., N represents the channel number. This is the number of the Dao.

[0015] Furthermore, the collected ground-penetrating radar data undergoes zero-drift removal processing, including:

[0016] Zero drift removal is performed by subtracting the mean value of each data point from the sampled data.

[0017] ,

[0018] in, For radar data with zero drift removed, k is any sampling point.

[0019] Furthermore, the median filtering method is used to calculate the variation trend of radar wave amplitude, including: calculating the variation trend of ground-penetrating radar wave amplitude with sampling time.

[0020]

[0021] in, The mean absolute value of the amplitude at the m-th sampling point. This represents the trend of change at the m-th sampling point, where m = 1, 2, ..., M represents the sampling point number. It is a median function. +1 represents the length of the filtering window.

[0022] Furthermore, step 4 includes:

[0023] Within the 0-1 / 4 time window, all sampling points and channels in the zero-drift-free radar data are traversed, and the early signal amplitude characteristics are statistically analyzed and extracted: the maximum value of the early signal amplitude. and minimum value and the maximum value and minimum value Number of sampling points and .

[0024] Furthermore, the starting sampling point for amplitude compensation is determined based on the extreme values ​​of the early radar signal amplitude characteristics, including:

[0025] Sampling points based on the maximum amplitude of the early signal and minimum sampling point The maximum value in the range is used as the starting sampling point for amplitude compensation. .

[0026] Furthermore, the method for adaptive amplitude compensation based on the trend of radar wave amplitude variation is as follows:

[0027]

[0028] in, The amplitude trend values ​​from sampling point 1 to M. This is the data after amplitude compensation.

[0029] Furthermore, the fast Fourier transform (FFT) was used to perform spectral analysis on the amplitude-compensated ground-penetrating radar (GPR) data to determine the filtering parameters. This included: using the FFT to analyze the amplitude-compensated GPR data... Perform spectrum analysis to obtain its main frequency. Determine the high cutoff frequency of the digital filter. and low cutoff frequency .

[0030] Furthermore, digital filtering is performed based on the filtering parameters, including: applying a Chebyshev filter to the amplitude-compensated data. Perform bandpass filtering and retain and The signals between them are used to obtain digitally filtered radar data. .

[0031] Furthermore, the aforementioned zero-drift removal method is used to eliminate radar data after digital filtering. DC interference in the system.

[0032] The beneficial effects of this invention are as follows:

[0033] Existing ground-penetrating radar (GPR) map optimization methods rely on manual experience, resulting in high subjectivity, numerous steps, and slow speed. This invention flexibly adjusts amplitude compensation and digital filtering parameters based on the amplitude and spectral characteristics of the raw GPR data, automatically removing zero drift and background noise, eliminating background noise and signal interference in the GPR data, improving the signal-to-noise ratio, and automatically generating a sequence of GPR maps to be measured.

[0034] This method requires no human intervention throughout the optimization and generation of ground-penetrating radar (GPR) maps. It not only completely eliminates the influence of subjective human factors and personal experience, but also significantly improves the optimization and generation speed of GPR maps. By quantitatively calculating the temporal distribution characteristics of amplitude in the original GPR data, this method adaptively adjusts the starting position and parameter changes of amplitude compensation, reducing the probability of errors caused by insufficient human experience in amplitude compensation.

[0035] This method calculates the spectral characteristics of ground-penetrating radar data and adaptively adjusts digital filtering parameters, thus avoiding errors caused by human subjective experience.

[0036] Compared with existing technologies, this technology automatically optimizes the quality of ground-penetrating radar (GPR) maps by quantitatively calculating the amplitude attenuation characteristics and spectral characteristics of ground-penetrating radar (GPR) data, adaptively adjusting amplitude compensation parameters and digital filtering parameters, and overcoming the problems of high subjectivity and low efficiency in the current empirical methods for optimizing and generating GPR maps. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art 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.

[0038] Figure 1 This is step 1) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 1, which obtains the original ground-penetrating radar data profile of road structure layer cracks at a main frequency of 400MHz.

[0039] Figure 2 This is step 1) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 1, which obtains the original ground-penetrating radar data profile of road structure layer cracks at a main frequency of 200MHz.

[0040] Figure 3 Step 2) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 1 is to obtain the profile of the road structure layer crack after automatic zero-drift removal of the 400MHz main frequency ground-penetrating radar data.

[0041] Figure 4 Step 2) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 1 is to obtain the profile of the road structure layer crack crack after automatic zero-drift removal of the 200MHz main frequency ground-penetrating radar data.

[0042] Figure 5 This is step 3) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 1, which obtains the amplitude mean and amplitude variation trend diagrams after automatic zero-drift removal of 400MHz ground-penetrating radar data of road structure layer cracks;

[0043] Figure 6 This is step 3) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 1, which obtains the amplitude mean and amplitude variation trend diagrams after automatic zero-drift removal of ground-penetrating radar data with a main frequency of 200MHz for road structure layer cracks;

[0044] Figure 7 Step 5) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 2 is to obtain the profile of the road structure layer crack after adaptive amplitude compensation of ground-penetrating radar data with a main frequency of 400MHz.

[0045] Figure 8 Step 5) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 2 is to obtain the profile of the road structure layer crack after adaptive amplitude compensation of ground-penetrating radar data with a main frequency of 200MHz.

[0046] Figure 9 Step 6) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 2 is to obtain the profile of the road structure layer crack after adaptive digital filtering of ground-penetrating radar data with a main frequency of 400MHz.

[0047] Figure 10 Step 6) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 2 is to obtain the profile of the road structure layer crack after adaptive digital filtering of ground-penetrating radar data with a main frequency of 200MHz.

[0048] Figure 11 Step 6) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 2 is to obtain the cross-sectional image of the road structure layer crack after automatic background removal of the 400MHz main frequency ground-penetrating radar data.

[0049] Figure 12 Step 6) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 2 is to obtain the cross-sectional image of the road structure layer crack after automatic background removal of the 200MHz main frequency ground-penetrating radar data.

[0050] Figure 13 The steps 1), 2), 5), and 6) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to the YOLO model used in Example 2 are: the original waveform of ground-penetrating radar single channel data with a main frequency of 400MHz for road structure layer cracks, the waveform with automatic zero drift removal, the waveform with adaptive amplitude compensation, the waveform with adaptive digital filtering, and the waveform with automatic background removal.

[0051] Figure 14 The steps 1), 2), 5), 6), and 7) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to the YOLO model used in Example 2 are: the original waveform of ground-penetrating radar single channel data with a main frequency of 200MHz for road structure layer cracks, the waveform with automatic zero drift removal, the waveform with adaptive amplitude compensation, the waveform with adaptive digital filtering, and the waveform with automatic background removal.

[0052] Figure 15This is step 1) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 3, which obtains the original ground-penetrating radar data profile of road structure layer uneven settlement disease at the main frequency of 400MHz.

[0053] Figure 16 This is step 1) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 3, which obtains the original ground-penetrating radar data profile of the road structure layer uneven settlement disease at the main frequency of 200MHz.

[0054] Figure 17 This is step 2) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 3, which obtains the profile of the 400MHz ground-penetrating radar data of uneven settlement disease in road structure layer after automatic zero-drift removal.

[0055] Figure 18 This is step 2) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 3, which obtains the profile of the 200MHz ground-penetrating radar data of uneven settlement disease in road structure layer after automatic zero-drift removal.

[0056] Figure 19 Step 5) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 3 is to obtain the profile of the road structure layer uneven settlement disease ground-penetrating radar data with adaptive amplitude compensation at the main frequency of 400MHz.

[0057] Figure 20 Step 5) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 3 is to obtain the profile of the road structure layer uneven settlement disease ground-penetrating radar data with adaptive amplitude compensation at the main frequency of 200MHz.

[0058] Figure 21 Step 6) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 3 is to obtain the profile of the road structure layer uneven settlement disease ground-penetrating radar data with a main frequency of 400MHz after adaptive digital filtering.

[0059] Figure 22 Step 6) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 3 is to obtain the profile of the 200MHz main frequency ground-penetrating radar data of uneven settlement disease of road structure layer after adaptive digital filtering.

[0060] Figure 23Step 6) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 3 is to obtain the profile of the 400MHz ground-penetrating radar data of uneven settlement disease of road structure layer after automatic background removal.

[0061] Figure 24 Step 6) of the adaptive optimization and generation method for ground-penetrating radar test spectrum applicable to YOLO model used in Example 3 is to obtain the profile of the 200MHz ground-penetrating radar data of uneven settlement disease of road structure layer after automatic background removal.

[0062] Figure 25 This is a flowchart of the method of the present invention. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] Example 1

[0065] An adaptive optimization and generation method for ground-penetrating radar target maps applicable to the YOLO model, such as Figure 25 As shown, the steps are as follows:

[0066] Step 1): Use ground-penetrating radar to collect data on hidden road defects and obtain ground-penetrating radar data.

[0067] In this embodiment, appropriate antenna types and acquisition parameters are selected based on different road scenarios and their hidden defects to obtain high-quality ground-penetrating radar (GPR) data. The GPR acquires road hidden defect data at equal intervals using a shielded ground-coupled GPR antenna and a profiling method. The GPR antenna moves equidistantly along the survey line, acquiring single-channel waveforms (time-intensity signals at a single point) at different locations. Multiple waveforms are combined to form two-dimensional radar data. Where m = 1, 2, ..., M represents the sampling point number, M is the number of sampling points, and n = 1, 2, ..., N represents the channel number. This is the number of the Dao.

[0068] Shielded ground-coupled ground-penetrating radar (GPR) antennas possess advantages such as strong anti-electromagnetic interference capability, good environmental adaptability, wide main frequency distribution range, large penetration depth, and high spatial resolution. They are sensitive to subtle electrical differences in structures and are widely used in the detection of hidden road defects. To maximize the penetration of electromagnetic wave energy into the underground medium, shielded GPR antennas must operate as close to the ground surface as possible. In this case, for shielded GPR antennas with small transmit / receive offsets, the airborne and ground-borne direct waves cannot be separated but mix and superimpose to form the early GPR signal. In GPR data, the early radar signal is a crucial indicator for locating the ground surface.

[0069] Ground-penetrating radar (GPR) transmits high-frequency electromagnetic waves underground via a shielded ground-coupled transmitting antenna. When these waves propagate through the underground medium, they encounter electrical anomalies or irregularities, generating reflected or scattered waves. Some of these reflected waves are received by the receiving antenna at the surface and converted into a discrete numerical sequence that varies over time, recording the changes in radar signal strength, phase, and other parameters. The change in radar signal over time at a single transmitting point is called a single-channel waveform. The GPR transmitting and receiving antennas move along the survey line at a fixed offset, acquiring multiple channels of waveforms at different locations along the line, forming two-dimensional GPR data. A two-dimensional radar data volume is a typical two-dimensional array, where rows represent sampling points, columns represent channel numbers, the horizontal axis represents the position information on the survey line, and the vertical axis represents the two-way travel time (or converted depth information) of the radar wave received at each sampling point.

[0070] Step 2): Perform zero-drift removal processing on the collected mine detection data to obtain zero-drift removed radar data.

[0071] For shielded ground-coupled ground-penetrating radar (GPR) antennas with a center frequency ≤400MHz, the amplitude in the GPR data may sometimes be entirely positive or entirely negative, or exhibit asymmetry in the positive and negative half-cycles. This is because the raw GPR data contains DC drift, causing the mean amplitude of the raw GPR data to deviate from zero. Zero drift can lead to deviations in subsequent amplitude analysis and compensation, making it impossible to accurately identify weak reflection signals from defects. This offset must be eliminated first to ensure symmetry in the positive and negative waveforms of the signal, thereby quickly correcting the data baseline offset and providing clean basic data for subsequent amplitude characteristic statistics and deep signal enhancement.

[0072] In this embodiment of the application, step S2 includes:

[0073] Zero drift removal is performed by subtracting the mean value of each data point from the sampled data. The expression is as follows:

[0074]

[0075] Where m=1,2,…,M represents the sampling point number; n=1,2,…,N represents the channel number.

[0076] Electromagnetic waves propagating in underground media are affected by factors such as geometric diffusion, absorption attenuation, scattering loss, and transmission loss. This results in stronger amplitudes of early radar signals from near the surface and weaker amplitudes of radar reflection signals from deeper underground layers in ground-penetrating radar data. Therefore, to highlight the radar wave response characteristics of deep targets, amplitude compensation is needed for the weak radar reflection signals from deep layers, as described in steps S3-S5.

[0077] Step 3): Based on the mean amplitude intensity of each sampling point in the zero-drift radar data, the median filtering method is used to calculate the trend of radar wave amplitude intensity with sampling time (or exploration depth).

[0078] In this embodiment, the average amplitude of each sampling point of the ground penetrating radar data is used. The median filtering method was used to calculate the variation trend of ground-penetrating radar wave amplitude with sampling time (or exploration depth). The calculation formula is:

[0079]

[0080]

[0081] in, The mean absolute value of the amplitude at the m-th sampling point. This represents the trend of change at the m-th sampling point, where m = 1, 2, ..., M represents the sampling point number. It is a median function. +1 represents the length of the filter window, which is usually an odd number ≥3.

[0082] Due to wavefront diffusion and absorption attenuation, the amplitude of the original radar data decreases exponentially with increasing detection distance. This makes it impossible to accurately identify the reflected signals of deep road defects in the original radar data. Therefore, it is necessary to extract the overall attenuation trend through median filtering to provide a basis for subsequent deep signal compensation. By quantifying the depth attenuation law of radar waves, we can avoid over- or under-compensation of subsequent amplitude compensation and improve the accuracy of compensation.

[0083] Step 4): Statistically analyze the extreme values ​​of the early signal amplitude characteristics of ground-penetrating radar based on the radar data after zero drift removal.

[0084] The early signal is a mixture of "airborne direct wave" and "ground direct wave" electromagnetic waves emitted by the ground-penetrating radar antenna (it cannot be separated due to the small antenna transmit / receive offset). Its duration is approximately one signal cycle, and it serves as a crucial indicator and basis for locating the ground surface and measuring amplitude compensation. The early signal typically lies within a 0-1 / 4 time window; statistical analysis within this range can accurately capture the amplitude characteristics of the early signal.

[0085] In this embodiment, within a time window range of 0-1 / 4, all sampling points and channels are traversed to statistically analyze and extract the early signal amplitude characteristics: the maximum value of the early signal amplitude. and minimum value and the maximum value and minimum value Number of sampling points and .

[0086] Step 5): Determine the starting sampling point for amplitude compensation based on the extreme values ​​of the early radar signal amplitude characteristics, and perform adaptive amplitude compensation based on the trend of radar wave amplitude change.

[0087] In this embodiment of the application, firstly, the sampling point is based on the maximum value of the early signal amplitude. and minimum sampling point The maximum value in the range is used as the starting sampling point for amplitude compensation. .

[0088] Furthermore, adaptive amplitude compensation is performed based on the trend of radar wave amplitude variation, including:

[0089]

[0090] in, The data is after amplitude compensation. This represents the result of median filtering, which is the complete sequence of amplitude trend values ​​for all sampling points (1 to M).

[0091] The core of amplitude compensation is to enhance weak signals in deep layers, while early signals are strong signals near the surface. Their extreme values ​​can serve as the basis for judging the starting point of compensation, providing an accurate "starting point benchmark" for subsequent amplitude compensation and avoiding compensation deviations caused by manual judgment of the starting point.

[0092] Traditional methods involve manually setting compensation parameters, which can easily lead to insufficient compensation for deep layers or overcompensation for shallow layers. This step adaptively calculates the compensation coefficient based on the data's own trends, accurately matching the degree of signal attenuation in deep layers. The reflected signal intensity of deep defects is significantly enhanced, avoiding missed detections due to weak signals, while maintaining the authenticity of shallow signals.

[0093] Step 6): Use Fast Fourier Transform to perform spectrum analysis on the amplitude-compensated ground penetrating radar data to determine the filtering parameters, perform bandpass digital filtering based on the filtering parameters, and eliminate DC interference in the bandpass digitally filtered ground penetrating radar data.

[0094] Ground-penetrating radar (GPR) data contains not only effective reflected waves from hidden road defects, but also scattered waves caused by the heterogeneity of the background medium, environmental noise, and system clutter. This results in a chaotic waveform in the GPR data, with blurred and indistinct reflected waves, necessitating filtering. The raw GPR data also contains significant background noise, resulting in a chaotic waveform and blurred, indistinct reflected waves in the radar profile, requiring further filtering. By utilizing the frequency differences between effective and interfering waves, system clutter and background noise can be eliminated, while retaining the geologically significant effective waves, thereby highlighting the effective signal and improving the signal-to-noise ratio.

[0095] In this embodiment of the application, Fast Fourier Transform is used to process the amplitude-compensated ground-penetrating radar data. Perform spectrum analysis to obtain its main frequency. Determine the high cutoff frequency of the digital filter. and low cutoff frequency .

[0096] Furthermore, a Chebyshev filter is used for bandpass filtering to preserve... and The signals between them are filtered to remove interference from other frequencies, resulting in digitally filtered radar data. .

[0097] Ground-penetrating radar (GPR) early signals exhibit strong amplitude, stable waveform, and clear phase, with a duration approximately equal to one signal cycle. The lower the center frequency of the ground-coupled antenna, the longer the early signal duration and the greater the depth-dependent influence range. The early GPR signal couples and superimposes with reflected waves from very shallow road surface defects, severely interfering with the identification of these hidden defects. Therefore, it is necessary to eliminate interference from the early GPR signal and other horizontally in-phase axes, i.e., background removal. This invention also employs the method of subtracting the mean value of each channel from all sampling points for DC interference removal, expressed as:

[0098]

[0099] in, The data is ground-penetrating radar data after DC interference removal. m=1,2,…,M represents the sampling point number, and n=1,2,…,N represents the channel number.

[0100] Even after filtering, interference from early signal superposition still exists, especially with low-frequency antennas (such as 200MHz) where the early signal duration is long and the interference is more pronounced. A second-stage mean removal process is needed to completely eliminate the background and highlight the defect signals. After processing, the characteristics of the target body's reflected waves (such as the waveform morphology of cracks and subsidence) are fully highlighted without background interference, providing a highly recognizable data foundation for YOLO model identification.

[0101] Step 7): Generate a ground-penetrating radar (GPR) spectrum sequence adapted to the YOLO model based on the GPR data after DC interference cancellation.

[0102] In this embodiment of the application, the ground-penetrating radar data after DC interference removal processing is used. Divide into n rows A two-dimensional matrix, where, The number of sampling points is denoted as r, and each of them is saved as a standardized ground-penetrating radar map with a height and width of m pixels. The file names of the maps are named radarfile_sn_1, radarfile_sn_2, ..., radarfile_sn_n in sequence.

[0103] Example 2

[0104] This embodiment describes the detection of hidden cracks in the structural layer of a section of the Lianyungang-Huoerguosi Expressway in Henan Province. The asphalt surface layer of this expressway is 18cm thick, and the base layer is a 36cm semi-rigid water-stabilized base. Data was collected over 12km, and 302 cracks were detected, establishing a sample library. One hidden crack was randomly selected from the sample library. The following description, in conjunction with the accompanying drawings, further illustrates the content of this embodiment:

[0105] An adaptive optimization and generation method for ground-penetrating radar (GPR) maps suitable for the YOLO model is implemented according to the following steps:

[0106] Step 1) Using a dual-channel ground-penetrating radar ground-coupled shielded antenna with main frequencies of 400MHz and 200MHz, data is collected along the driving direction of the lane to obtain the ground-penetrating radar time-domain reflectometry profile at the main frequency of 400MHz (e.g., Figure 1 (as shown) and the ground-penetrating radar time-domain reflectometry profile with a main frequency of 200MHz (as shown) Figure 2 (As shown). From Figure 1 , 2 It can be seen that in the original radar data, the early signal amplitude near the ground surface is very strong, but as the reception time and detection depth increase, the amplitude decays rapidly, and the deep reflection signal is weak.

[0107] Step 2) Process the collected ground-penetrating radar two-dimensional raw data Figure 1 and Figure 2 Perform zero-drift removal processing to obtain radar data after zero-drift removal. Cross-sectional view as follows Figure 3 and Figure 4 As shown, the corresponding single-channel data is as follows: Figure 13 and 14 As shown. From Figure 3 , 4 As can be seen from Figures 13 and 14, after the "zero drift removal" process, the mean value of the waveform approaches the 0 baseline, and the waveform is symmetrical about the 0 baseline.

[0108] Step 3) Based on ground-penetrating radar data Mean amplitude intensity at each sampling point The median filtering method was used to calculate the variation trend of radar wave amplitude intensity with sampling time (or exploration depth). The average amplitude intensity at main frequencies of 400MHz and 200MHz and amplitude variation trend , respectively Figure 5 and Figure 6 As shown. From Figure 5 , 6 It can be seen that the amplitude intensity of radar data is strongest near the ground surface, but it decays rapidly with increasing reception time and detection depth.

[0109] Step 4) Calculate and extract the maximum value of the early signal amplitude of the ground penetrating radar. Minimum value and the number of sampling points. , ;

[0110] Step 5) with , The maximum value in ( Using this as the starting sampling point for amplitude compensation, adaptive amplitude compensation is performed to obtain... This is used to enhance the amplitude intensity of deep-reflected waves. The radar profiles after adaptive amplitude compensation at main frequencies of 400MHz and 200MHz are shown below. Figure 7 and Figure 8 As shown. From Figure 7 , 8 It can be seen that the deep reflection signal of the radar profile is significantly enhanced after adaptive amplitude compensation, but low-frequency interference exists in some areas.

[0111] Step 6) Apply Fast Fourier Transform to the amplitude-compensated ground-penetrating radar data Perform spectrum analysis to obtain its dominant frequency. Determine the high cutoff of the digital filter and low section The frequency is used to perform bandpass digital filtering to eliminate interference waves and noise in the ground penetrating radar data, resulting in digitally filtered ground penetrating radar data. The radar profiles after adaptive digital filtering at main frequencies of 400MHz and 200MHz are as follows: Figure 9 and Figure 10 As shown. From Figure 9 , 10 As can be seen, digital filtering effectively eliminated low-frequency interference in the radar profile, but significant horizontal interference still exists.

[0112] Ground penetrating radar data after digital filtering DC interference in the middle, to obtain The reflected wave response characteristics of the protruding hidden cracks, and the radar profiles after automatic background removal (elimination of transverse horizontal fixed wave interference) at the main frequencies of 400MHz and 200MHz are respectively as shown in the figures. Figure 11 and Figure 12 As shown; from Figure 11 , 12 As can be seen, after background removal processing, the horizontal fixed interference waves in the radar profile are effectively eliminated, and the crack diffraction waves are clearly visible. The signal-to-noise ratio and resolution of the radar spectrum are significantly improved.

[0113] Figure 13 and Figure 14 These are the original waveforms, automatically zero-drift-removed waveforms, adaptive amplitude-compensated waveforms, adaptive digital-filtered waveforms, and automatically background-removed single-channel radar waveforms at main frequencies of 400MHz and 200MHz, respectively. Figure 13 , 14 The changes in the single-channel waveform after each processing step can be seen more clearly in the image.

[0114] Step 7) Generate a sequence of ground-penetrating radar test maps suitable for intelligent recognition using the YOLO model.

[0115] Other processing methods are the same as in Example 1.

[0116] Example 3

[0117] This embodiment describes the detection of hidden structural layer defects on a section of the Daqing-Guangzhou Expressway in Henan Province. The expressway has an 18cm asphalt surface layer and a 36cm semi-rigid water-stabilized base course. Data was collected over 10km, identifying 40 instances of uneven settlement in the road structural layer. A sample library was established, and one instance of uneven settlement in the road structural layer was randomly selected from this library. The following description, in conjunction with the accompanying drawings, further illustrates this embodiment:

[0118] An adaptive optimization and generation method for ground-penetrating radar (GPR) maps suitable for the YOLO model is implemented according to the following steps:

[0119] Step 1) Using a dual-channel ground-penetrating radar ground-coupled shielded antenna with main frequencies of 400MHz and 200MHz, data is collected along the driving direction of the lane to obtain the ground-penetrating radar time-domain reflectometry profile at the main frequency of 400MHz (e.g., Figure 15 (as shown) and the ground-penetrating radar time-domain reflectometry profile with a main frequency of 200MHz (as shown) Figure 16 (As shown). From Figure 15 , 16 As can be seen from the original ground-penetrating radar data profile, the early signal amplitude near the surface is strong, but the amplitude decays rapidly with the increase of reception time and detection depth, and the deep reflection signal is weak and difficult to identify.

[0120] Step 2) Process the collected ground-penetrating radar two-dimensional raw data Figure 15 and Figure 16 Automatic zero-drift removal is performed to obtain radar data after zero-drift removal. ,like Figure 17 and Figure 18 As shown. From Figure 17 , 18 As can be seen, after the zero-drift removal process, the mean value of the single-channel waveform in the radar profile approaches 0, effectively eliminating signal drift.

[0121] Step 3) Based on ground-penetrating radar data Mean amplitude intensity at each sampling point The median filtering method was used to calculate the variation trend of radar wave amplitude intensity with sampling time (or exploration depth). .

[0122] Step 5) Calculate and extract the maximum value of the early signal amplitude of the ground penetrating radar. Minimum value and the number of sampling points. , ;by , The maximum value in ( Using this as the starting sampling point for amplitude compensation, adaptive amplitude compensation is performed to obtain... This is used to enhance the amplitude intensity of deep-reflected waves. The radar profiles after adaptive amplitude compensation at main frequencies of 400MHz and 200MHz are shown below. Figure 19 and Figure 20 As shown. From Figure 19 , 20 As can be seen, after adaptive amplitude compensation, the deep reflection signal in the radar profile is significantly enhanced, but at the same time, the background noise and interference waves are also enhanced.

[0123] Step 6) Apply Fast Fourier Transform to the amplitude-compensated ground-penetrating radar data Perform spectrum analysis to obtain its dominant frequency. Determine the high cutoff of the digital filter and low section The frequency is used to perform bandpass digital filtering to eliminate interference waves and noise in the ground penetrating radar data, resulting in digitally filtered ground penetrating radar data. The radar profiles after adaptive digital filtering at main frequencies of 400MHz and 200MHz are as follows: Figure 21 and Figure 22 As shown. From Figure 21 , 22 As can be seen from this, after adaptive digital filtering, Figure 19 , 20 Background noise and interference waves were effectively eliminated.

[0124] Ground penetrating radar data after digital filtering DC interference in the middle, to obtain The radar profiles after automatic background removal at main frequencies of 400MHz and 200MHz highlight the reflected wave response characteristics of uneven settlement of the road structure layer. Figure 23 and Figure 24 As shown. From Figure 23 , 24 As can be seen, after automatic background removal processing, the horizontal fixed interference waves such as early signals in the radar profile are significantly eliminated, and the signal-to-noise ratio and resolution of the radar spectrum are significantly improved.

[0125] Step 7) Generate a sequence of ground-penetrating radar test maps suitable for intelligent recognition using the YOLO model.

[0126] Other processing methods are the same as in Example 1.

[0127] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive optimization and generation method for ground-penetrating radar (GPR) target maps applicable to the YOLO model, characterized in that, The steps are as follows: Step 1: Use ground-penetrating radar to collect data on hidden road defects and obtain ground-penetrating radar data; Step 2: Perform zero-drift removal processing on the collected ground-penetrating radar data to obtain zero-drift removed radar data; Step 3: Based on the mean amplitude intensity of each sampling point in the zero-drift radar data, the median filtering method is used to calculate the trend of radar wave amplitude variation; Step 4: Statistically analyze the extreme values ​​of the early signal amplitude characteristics of ground-penetrating radar based on the radar data after zero-drift removal; Step 5: Determine the starting sampling point for amplitude compensation based on the extreme values ​​of the early radar signal amplitude characteristics, and perform adaptive amplitude compensation based on the trend of radar wave amplitude variation; Step 6: Use Fast Fourier Transform to perform spectrum analysis on the amplitude-compensated ground penetrating radar data to determine the filtering parameters, perform digital filtering based on the filtering parameters, and eliminate DC interference in the digitally filtered ground penetrating radar data. Step 7: Generate a sequence of ground-penetrating radar (GPR) test maps adapted to the YOLO model based on the GPR data after DC interference cancellation.

2. The adaptive optimization and generation method for ground-penetrating radar target maps applicable to the YOLO model according to claim 1, characterized in that, In step 1, ground-penetrating radar (GPR) uses a shielded ground-coupled GPR antenna combined with a profiling method to collect multi-channel waveform data of hidden road defects at equal intervals, thereby acquiring two-dimensional GPR data. Where m = 1, 2, ..., M represents the sampling point number, M is the number of sampling points, and n = 1, 2, ..., N represents the channel number. This is the number of the Dao.

3. The adaptive optimization and generation method for ground-penetrating radar target maps applicable to the YOLO model according to claim 2, characterized in that, The collected ground-penetrating radar data undergoes zero-drift removal processing, including: Zero drift removal is performed by subtracting the mean value of each data point from the sampled data. ； in, For radar data with zero drift removed, k is any sampling point.

4. The adaptive optimization and generation method for ground-penetrating radar target maps applicable to the YOLO model according to claim 3, characterized in that, The median filtering method is used to calculate the variation trend of radar wave amplitude, including: calculating the variation trend of ground-penetrating radar wave amplitude with sampling time. ； in, The mean absolute value of the amplitude at the m-th sampling point. This represents the trend of change at the m-th sampling point, where m = 1, 2, ..., M represents the sampling point number. It is a median function. +1 represents the length of the filtering window.

5. The adaptive optimization and generation method for ground-penetrating radar target maps applicable to the YOLO model according to any one of claims 1-4, characterized in that, Step 4 includes: Within the 0-1 / 4 time window, all sampling points and channels in the zero-drift-free radar data are traversed, and the early signal amplitude characteristics are statistically analyzed and extracted: the maximum value of the early signal amplitude. and minimum value and the maximum value and minimum value Number of sampling points and .

6. The adaptive optimization and generation method for ground-penetrating radar target maps applicable to the YOLO model according to claim 5, characterized in that, The amplitude compensation starting sampling point is determined based on the extreme values ​​of the early radar signal amplitude characteristics, including: Sampling points based on the maximum amplitude of the early signal and minimum sampling point The maximum value in the range is used as the starting sampling point for amplitude compensation. .

7. The adaptive optimization and generation method for ground-penetrating radar target maps applicable to the YOLO model according to claim 6, characterized in that, The method for adaptive amplitude compensation based on the trend of radar wave amplitude variation is as follows: ； in, The amplitude trend values ​​from sampling point 1 to M. This is the data after amplitude compensation.

8. The adaptive optimization and generation method for ground-penetrating radar target maps applicable to the YOLO model according to claim 7, characterized in that, Spectral analysis of amplitude-compensated ground-penetrating radar (GPR) data using Fast Fourier Transform (FFT) was performed to determine filtering parameters. This included: using FFT to analyze the amplitude-compensated GPR data... Perform spectrum analysis to obtain its main frequency. Determine the high cutoff frequency of the digital filter. and low cutoff frequency .

9. The adaptive optimization and generation method for ground-penetrating radar target maps applicable to the YOLO model according to claim 8, characterized in that, Digital filtering is performed based on filtering parameters, including: applying a Chebyshev filter to the amplitude-compensated data. Perform bandpass filtering and retain and The signals between them are used to obtain digitally filtered radar data. .

10. The adaptive optimization and generation method for ground-penetrating radar target maps applicable to the YOLO model according to claim 9, characterized in that, The zero-drift removal method described above is used to eliminate radar data after digital filtering. DC interference in the system.

Images