Ground penetrating radar parameter optimization and general survey method for highway subgrade condition detection

Abstract

The present application relates to the ground penetrating radar parameter optimization and general survey method for highway subgrade condition detection. The method includes four core modules: one is the ground penetrating radar parameter optimization configuration, through static basic configuration and dynamic adaptive adjustment, combined with real-time sensing and PID control, realizing the automatic correction of parameters when the medium changes; the second is the rapid general survey execution, combining traffic flow, road surface vision, historical disease data, optimizing the route through Dijkstra algorithm, and determining the anomaly through multi-feature fusion; the third is the optimization imaging processing, proposing the multi-layer medium correction time delay imaging and complex disease classification imaging, improving the deep disease recognition rate; the fourth is the maintenance decision linkage, quantifying the disease grade and priority, automatically outputting the maintenance scheme, and shortening the decision cycle. The present application solves the problems of poor dynamic adaptation, difficult complex disease recognition and disconnected decision of the prior art, has been verified in multiple road sections, and is suitable for highway subgrade disease whole chain monitoring and maintenance.

Description

Technical Field

[0001] This invention relates to the field of non-destructive testing technology for highways, and in particular to a method for optimizing ground-penetrating radar parameters and conducting surveys for detecting the condition of highway subgrades. Background Technology

[0002] Ground-penetrating radar (GPR) has become a core technology for roadbed detection due to its advantages of non-destructive, rapid, and continuous detection. However, existing technologies still have some shortcomings in engineering practice, making it difficult to meet the requirements of efficient and high-precision roadbed detection. For example, in existing technologies, the core parameters of GPR (center frequency, sampling window, etc.) are mostly set once before detection, without considering the dynamic fluctuations of roadbed medium properties during the detection process. Most of these technologies only target fixed medium parameters (preset dielectric constant). The design parameters do not cover the field detection process. It is difficult to adapt to dynamic environmental changes due to variations in moisture content and temperature.

[0003] In addition, traditional roadbed surveys mostly rely solely on radar reflection data to identify anomalies, without combining key data such as road surface visual characteristics and traffic flow. On the one hand, there is a strong correlation between surface defects on the road surface (such as cracks wider than 5mm and settlement greater than 2cm) and internal defects in the roadbed. For example, road surface settlement often corresponds to roadbed voids, but current technology has not established a spatial mapping relationship between the two. This makes it easy for the strong reflections of metal manhole covers and guardrail foundations to be misjudged as voids, resulting in a generally high misjudgment rate. On the other hand, multi-lane detection does not combine real-time traffic flow to adjust the route. During the morning rush hour, the traffic flow density is high, and traditional fixed-lane sequential detection is prone to causing traffic congestion and reducing efficiency.

[0004] In addition, existing imaging algorithms also have certain shortcomings. First, for media with three or more layers, such as air layer, asphalt surface layer, base layer and subgrade layer, a simplified two-layer media time delay model is still used, which does not take into account the difference in electromagnetic wave attenuation of each layer of media, which can easily lead to blurred imaging of deep defects and low recognition rate. Second, for complex defects with abnormal water content and coexistence of cavities, the traditional weighting factor is based on only a single energy feature, which makes it difficult to distinguish the reflection differences of different defects and the misjudgment rate is high. Summary of the Invention

[0005] In view of the above problems, the present invention provides a ground-penetrating radar application method for the entire process of highway subgrade condition detection, including coverage parameter configuration, survey execution, imaging analysis, and maintenance decision-making.

[0006] The specific technical solution is as follows:

[0007] Ground-penetrating radar parameter optimization and survey methods for highway subgrade condition detection include the following steps:

[0008] S1, Ground Penetrating Radar (GPR) parameter optimization configuration, including static basic parameter configuration and dynamic parameter adaptive adjustment; the static basic parameter configuration is based on the roadbed layering and the relative permittivity of the medium, and the center frequency is quantized and selected according to a preset correlation formula; the sampling time window base value is calculated based on the maximum detection depth and medium parameters, and the actual value is 1.3 times the base value; the sampling interval is set so that the sampling rate is more than 4 times the highest frequency of the reflected wave; the dynamic parameter adaptive adjustment is achieved by acquiring the dynamic parameters of the medium through the vehicle-mounted real-time sensing system, and the sensing data is transmitted to the control unit in real time via the CAN bus; based on the dynamic parameters of the medium, the center frequency and sampling time window are adjusted in real time through a PID control algorithm to control the penetration depth deviation to be less than 3%;

[0009] S2, Rapid Survey Execution, includes multi-source data preprocessing, intelligent detection, and multi-feature fusion anomaly determination. Multi-source data preprocessing involves collecting traffic flow data, historical road damage data, and road surface visual data for the detected road sections, establishing a historical damage-radar feature association database, dividing traffic flow time periods, and simultaneously calibrating the spatiotemporal coordinates of radar, sensor, and visual data. Intelligent detection, based on traffic flow time periods, plans the lane detection sequence using the Dijkstra algorithm, prioritizing lanes with low traffic density. A vehicle-mounted ground-penetrating radar system continuously scans at a constant speed of 60-65 km / h, simultaneously recording radar data, road surface images, and sensor data. Multi-feature fusion anomaly determination involves extracting radar features and road surface image features in 10m units, calculating anomaly probabilities using a logistic regression model, and classifying confirmed anomalies, suspected anomalies, and normal road sections according to probability thresholds.

[0010] S3, optimized imaging processing, including multi-layer media imaging and complex disease classification; the multi-layer media imaging involves establishing a multi-layer media model of the roadbed, calculating the electromagnetic wave propagation delay using a modified time delay formula, solving for refraction points through a layer-by-layer one-dimensional search, and performing inverse attenuation weighted enhancement on deep scattering data to generate multi-layer media imaging results; the complex disease classification involves extracting the energy, waveform entropy, and in-phase axis curvature of the imaging grid to form feature vectors, inputting them into a pre-trained lightweight model, and outputting type determination results for cavities, voids, and abnormal water content.

[0011] Furthermore, it also includes step S4, maintenance decision linkage, which includes disease level assessment and priority ranking and report generation; the disease level assessment is based on disease size, reflection intensity and highway grade, dividing the disease into three levels: minor, moderate and severe, and matching the corresponding maintenance cycle; the priority ranking and report generation is to rank the maintenance priority according to the risk-cost score, automatically generate a decision report containing disease parameters, maintenance plan and estimated cost, and push it to the highway maintenance system.

[0012] Further, in step S1, the subgrade layering includes a subgrade layer, a base course, a cement concrete surface layer, and an asphalt concrete surface layer; the preset correlation formula for selecting the center frequency is: In the formula, The center frequency of the antenna. The relative permittivity of the medium, For radar vertical resolution, the center frequency of the roadbed layer is selected as 100-400MHz, the center frequency of the base layer is selected as 400-900MHz, the center frequency of the cement concrete surface layer is selected as 900-1200MHz, and the center frequency of the asphalt concrete surface layer is selected as greater than 1200MHz.

[0013] Furthermore, in step S1, the real-time sensing system includes a high-frequency capacitive moisture content sensor, an infrared temperature sensor, and an inertial measurement unit; the dynamic parameters of the medium include roadbed moisture content, temperature, and antenna tilt data.

[0014] Furthermore, in step S2, the spatiotemporal coordinates are calibrated by time synchronization through a GPS timing module, and the mapping relationship between road surface image pixels and actual mileage is established by a checkerboard calibration method.

[0015] Furthermore, in step S2, the vehicle-mounted ground-penetrating radar system includes a ground-penetrating radar host, an air-coupled antenna with a lateral width of 3.75m or greater, a wheel encoder, and a high-definition camera; the pulse repetition frequency of continuous scanning is 200kHz or greater, and the operating frequency band 3dB range is 100-500MHz.

[0016] Further, in step S2, the anomaly probability formula of the logistic regression model is: ,in, , where is the radar feature weight; , where is the image feature weight; This represents the comprehensive value of radar characteristics. The image feature synthesis value; when A value greater than 0.8 is considered an anomaly. A value between 0.5 and 0.8 is considered a suspected abnormality. A value less than 0.5 is considered normal.

[0017] Further, in step S3, the lightweight model is a pre-trained MobileNet-V2 model, and its feature vector is... ,in For energy, For waveform entropy, The curvature of the in-phase axis.

[0018] Further, in step S4, the criteria for determining the severity of the disease are: the maximum side length of the disease is less than 0.3m or the reflection intensity is less than 0.2; the criteria for determining the moderate severity of the disease are: the maximum side length of the disease is between 0.3m and 0.8m or the reflection intensity is between 0.2 and 0.5; and the criteria for determining the severity of the disease are: the maximum side length of the disease is greater than 0.8m or the reflection intensity is greater than 0.5.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] (1) The present invention provides a ground-penetrating radar parameter optimization and survey method and device for detecting roadbed conditions. The method includes a ground-penetrating radar parameter optimization configuration module. Through static basic configuration and dynamic adaptive adjustment, combined with real-time sensing and PID control, the parameters are automatically corrected when the medium changes, breaking through the limitations of traditional static parameters and improving the detection accuracy compared with traditional technology.

[0021] (2) The present invention provides a ground-penetrating radar parameter optimization and survey method and device for detecting roadbed conditions. The method includes a rapid survey execution module that integrates traffic flow, road surface visual data, and historical disease data. It optimizes the route through the Dijkstra algorithm, determines anomalies through multi-feature fusion, and reduces the misjudgment rate compared with traditional technologies.

[0022] (3) The present invention provides a ground-penetrating radar parameter optimization and survey method and device for detecting roadbed conditions. The method includes an optimized imaging processing module, proposes multi-layer medium-corrected time-delay imaging and complex disease classification imaging, and improves the deep disease identification rate and type identification rate compared with traditional technology.

[0023] (4) The present invention provides a ground-penetrating radar parameter optimization and survey method and device for detecting the condition of highway subgrade. The method also includes a maintenance decision linkage module, which quantifies the disease level and priority, and outputs maintenance plan, thereby shortening the decision cycle. Attached Figure Description

[0024] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0025] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0026] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0027] Example 1

[0028] Ground-penetrating radar parameter optimization and survey methods for highway subgrade condition detection (reference) Figure 1 This includes the following steps:

[0029] S1, Ground Penetrating Radar Parameter Optimization Configuration: including static basic parameter configuration and dynamic parameter adaptive adjustment; the static basic parameter configuration is based on the roadbed layering and the relative permittivity of the medium. The center frequency is selected by quantification according to a preset correlation formula; the basic value of the sampling time window is calculated based on the maximum detection depth and medium parameters, and the actual value is 1.3 times the basic value; the sampling interval is set so that the sampling rate is more than 4 times the highest frequency of the reflected wave; the dynamic parameter adaptive adjustment is to obtain the medium dynamic parameters through the vehicle-mounted real-time sensing system, and the sensing data is transmitted to the control unit (radar host) in real time via CAN bus; based on the medium dynamic parameters, the center frequency and sampling time window are adjusted in real time through PID control algorithm to control the penetration depth deviation to be less than 3%, and when the sensor data is abnormal, the sliding mean is used to replace or switch to the average parameters of the previous road segment;

[0030] S2, rapid survey execution, includes multi-source data preprocessing, intelligent detection, and multi-feature fusion anomaly determination. Multi-source data preprocessing involves collecting traffic flow data (time-segmented vehicle flow density), historical damage data (drilling reports and repair records from the past 3 years), and road surface visual data (vehicle-mounted cameras) for the detected road sections. A historical damage-radar feature association database is established, traffic flow time periods are divided, and the spatiotemporal coordinates of radar, sensor, and visual data are simultaneously calibrated. Intelligent detection, based on traffic flow time periods, uses the Dijkstra algorithm to plan the lane detection sequence, prioritizing lanes with low traffic density. A vehicle-mounted ground-penetrating radar system continuously scans at a constant speed of 60-65 km / h, simultaneously recording radar data, road surface images, and sensor data. Multi-feature fusion anomaly determination involves extracting radar features (average energy E) in 10m units. 10 The probability of anomalies is calculated by using a logistic regression model, and road sections are classified into confirmed anomalies, suspected anomalies, and normal road sections according to probability thresholds.

[0031] S3, optimized imaging processing, including multi-layer media imaging and complex disease classification; the multi-layer media imaging involves establishing a multi-layer media model of the roadbed, calculating the electromagnetic wave propagation delay using a modified time delay formula, solving for refraction points through a layer-by-layer one-dimensional search, and performing inverse attenuation weighted enhancement on deep scattering data to generate multi-layer media imaging results; the complex disease classification involves extracting the energy, waveform entropy, and in-phase axis curvature of the imaging grid to form feature vectors, inputting them into a pre-trained lightweight model, and outputting type determination results for cavities, voids, and abnormal water content.

[0032] The sampling window is actually set at 1.3 times the base value to account for spatial fluctuations in the dielectric constant and moisture content of the subgrade medium. A 30% redundancy is reserved to avoid signal interruption in deeper layers due to changes in geological parameters. This ensures the sampling window fully covers the electromagnetic wave's round-trip propagation time, improving signal integrity and reducing the rate of missed detection of deep defects. A sampling rate more than four times the highest frequency of the reflected wave follows the Nyquist sampling theorem. The highest frequency of the reflected wave is 1.5 times the center frequency. A four-times sampling rate effectively avoids spurious frequency interference, reduces signal distortion, and improves radar image resolution—doubling the resolution compared to a two-times sampling rate. Controlling the penetration depth deviation to less than 3% ensures accurate depth positioning in subgrade defect detection. A deviation exceeding 3% leads to a defect depth identification error >5cm, affecting maintenance work. A depth positioning error ≤2cm meets the accuracy requirements for subgrade detection. The use of vehicle-mounted ground-penetrating radar systems for continuous scanning at a constant speed of 60-65 km / h meets the requirements for uninterrupted traffic detection on high-grade highways. This speed range aligns with the minimum driving speed requirements for expressways and matches the ground-penetrating radar pulse repetition frequency (≥200kHz). The detection time per kilometer is less than 1 minute, and the daily detection range reaches 200 km, doubling the detection efficiency compared to 30 km / h, while maintaining good data continuity. Extracting radar features and road surface image features in 10m units combines the spatial distribution patterns of roadbed defects. The 10m unit balances detection accuracy and efficiency, avoiding data redundancy caused by excessively small units and missed defects caused by excessively large units. This improves the efficiency of abnormal road section identification, reduces the false positive rate, and caters to both large-scale surveys and precise identification needs.

[0033] Furthermore, as one implementation method, it also includes step S4, maintenance decision linkage, which includes disease level assessment and priority ranking and report generation; the disease level assessment is based on disease size, reflection intensity and highway grade, dividing the disease into three levels: minor, moderate and severe, and matching the corresponding maintenance cycle; the priority ranking and report generation is to rank the maintenance priority according to the risk-cost score, automatically generate a decision report containing disease parameters, maintenance plan and estimated cost, and push it to the highway maintenance system.

[0034] Further, as a specific implementation, in step S1, the subgrade layering includes a subgrade layer, a base course, a cement concrete surface layer, and an asphalt concrete surface layer; the preset correlation formula for selecting the center frequency is: In the formula, The center frequency of the antenna. The relative permittivity of the medium, For radar vertical resolution, the center frequency of the roadbed layer is selected as 100-400MHz, the center frequency of the base layer is selected as 400-900MHz, the center frequency of the cement concrete surface layer is selected as 900-1200MHz, and the center frequency of the asphalt concrete surface layer is selected as greater than 1200MHz.

[0035] Furthermore, as a specific implementation method, in step S1, the formula for calculating the basic value of the sampling window is as follows: In the formula, The speed of electromagnetic waves in a medium is expressed in m / ns. The maximum detection depth is expressed in meters (m).

[0036] Furthermore, as a specific implementation, in step S1, the sampling interval The unit is ns, and the sampling rate is = To ensure that the sampling rate is the highest frequency of the reflected wave ( 4 times that of ).

[0037] Furthermore, as a specific implementation, in step S1, the real-time sensing system includes a high-frequency capacitive moisture content sensor (measurement range 0-40%, accuracy ±1%), an infrared temperature sensor (measurement range -20-80℃, accuracy ±0.5℃), and an inertial measurement unit; the dynamic parameters of the medium include roadbed moisture content, temperature, and antenna tilt data.

[0038] Furthermore, as a specific implementation, in step S1, the dynamic parameter adaptive adjustment also includes a fault tolerance mechanism. When the sensor data fluctuation is greater than 5%, the average value of the five adjacent sensor points is automatically used as the replacement. When any sensor fails, the radar system switches to the average medium parameter of the first 100m road segment to calculate the static basic parameters and marks the sensor failure segment in the data file.

[0039] Furthermore, as a specific implementation, in step S2, the hardware parameters of the vehicle-mounted ground-penetrating radar system meet the following requirements: the instantaneous transmit power of the ground-penetrating radar host is greater than or equal to 6W, the dynamic range is greater than or equal to 150dB, and the A / D conversion bit depth is greater than or equal to 12; the lateral width of the air-coupled antenna is greater than or equal to 3.75m, which is suitable for full-width coverage of standard lanes; the wheel encoder resolution is less than or equal to 1mm, and the position calibration error is less than 0.5m.

[0040] Electromagnetic wave attenuation coefficient The acquisition method is as follows: Prioritize the use of moisture content sensor data from the real-time sensing system, and then... calculate, The moisture content is used; when no sensor data is available, the corresponding filler material is retrieved from the preset roadbed filler material parameter database. By default, the database includes mainstream fillers such as silty clay, sandy soil, and gravelly soil at different moisture contents. value.

[0041] In step S1, the ground-penetrating radar parameter optimization configuration implements a two-tiered parameter management system of static basic configuration and dynamic adaptive adjustment to ensure parameter adaptability under different environments. The static basic configuration is based on the roadbed layer structure and inherent media properties, quantifying core parameters such as center frequency, sampling window, and sampling rate. Parameter selection principles prioritize meeting detection depth requirements, while also considering resolution. For example, for roadbed detection depths of 0.8-2.2m, a 100-400MHz antenna is selected; for base layer detection depths of 0.3-0.8m, a 400-900MHz antenna is selected, ensuring effective coverage of each structural layer. Dynamic adaptive adjustment addresses fluctuations in on-site media properties by automatically adjusting parameters through real-time sensing, parameter calculation, and fault-tolerant correction. An incremental PID control algorithm is employed, using the penetration depth deviation (target depth - measured depth) as feedback to adjust the center frequency and sampling window in real time. PID parameters, such as the proportional coefficient, are determined through on-site debugging. =2.5, integral coefficient =0.8, differential coefficient =0.3), the specific adjustment logic is as follows:

[0042] Frequency adjustment: ,in,

[0043] , For penetration depth deviation, This is the static fundamental frequency.

[0044] Time window adjustment: In the formula, This is the real-time attenuation coefficient. , For static base time windows, This is the baseline attenuation coefficient.

[0045] Furthermore, as a specific implementation method, in step S2, the traffic flow time period is divided into low peak (less than 50 vehicles / km), medium peak (50-150 vehicles / km), and high peak (more than 150 vehicles / km). Based on the traffic flow data, the detection lane sequence is dynamically planned using the Dijkstra algorithm, prioritizing the detection of lanes with a traffic flow density of less than 50 vehicles / km during peak hours.

[0046] Furthermore, as a specific implementation method, the spatiotemporal coordinates are calibrated by time synchronization through a GPS timing module, and the mapping relationship between road surface image pixels and actual mileage is established by a checkerboard calibration method.

[0047] Furthermore, as a specific implementation, the vehicle-mounted ground-penetrating radar system includes a ground-penetrating radar main unit, an air-coupled antenna with a lateral width of 3.75m or greater, a wheel encoder, and a high-definition camera; it also includes a detection vehicle. Multi-vehicle road sections are detected sequentially from the inside to the outside, and time-space synchronization of radar data, road surface images, and sensor data is achieved through GPS timing.

[0048] Furthermore, as a specific implementation method, in step S2, the anomaly probability formula of the logistic regression model is: ,in, , where is the radar feature weight; , where is the image feature weight; This is the comprehensive value of radar characteristics. The image feature synthesis value; when A value greater than 0.8 is considered an anomaly. A value between 0.5 and 0.8 is considered a suspected abnormality. A value less than 0.5 is considered normal.

[0049] In step S2, the rapid census execution implements multi-source data preprocessing, intelligent route planning, high-precision detection, and fusion judgment, adapting to scenarios without traffic interruption. Multi-source data time synchronization is achieved through a GPS timing module, and spatial synchronization is achieved through a checkerboard calibration method.

[0050] Time synchronization: One synchronization pulse is generated every second to trigger the radar host, camera, and sensor to record data simultaneously, ensuring that the timestamp deviation is less than 1μs;

[0051] Spatial synchronization: A 1m×1m checkerboard calibration plate is laid on the road surface to establish the mapping relationship between image pixels and actual mileage (e.g., 100 pixels = 1m) to ensure that the spatial deviation between road surface features and radar data is less than 0.1m.

[0052] Route planning and detection execution are based on Dijkstra's shortest path algorithm, using lanes as nodes and traffic flow density as path weights to plan the optimal detection sequence.

[0053] Node definition: For a two-way six-lane road segment, the nodes are the inner lane 1, inner lane 2, outer lane 1, and outer lane 2;

[0054] Weighting: The weight is 1 when the traffic density is less than 50 vehicles / km, 2 when it is between 50 and 150 vehicles / km, and 3 when it is greater than 150 vehicles / km.

[0055] Path finding: Starting from the node with the smallest weight, find the path with the smallest total weight (e.g., during peak hours, the outer two lanes with a weight of 1 are detected first, while the inner one lane with a weight of 3 is detected later).

[0056] In the multi-feature fusion anomaly detection, a logistic regression model is used to fuse radar features and image features for detection. The specific steps are as follows:

[0057] Feature extraction (using 10m as the basic unit):

[0058] Radar characteristics: average energy (Integration interval = sampling time window), waveform entropy ( , For the first The amplitude probability of each sampling point ), continuity of the same phase axis =Number of consecutive tracks / Total number of tracks;

[0059] Image features: apparent area of ​​disease (Area of ​​cracks + settlement zone, unit: m²) 2 ), Maximum side length of the disease (When the crack width is greater than 5mm or the settlement is greater than 2cm) ,otherwise );

[0060] Model Training: A logistic regression model was trained using 1000 sets of historical data (500 normal sets and 500 abnormal sets) to obtain the weights. (Radar feature weights) (Image feature weights), model formula:

[0061] ,

[0062] In the formula, for , for .

[0063] Judgment criteria: A value greater than 0.8 indicates an anomaly; 0.5 ≤ A value ≤0.8 is suspected to be abnormal; A value less than 0.5 is normal.

[0064] Furthermore, as a specific implementation, in step S3, the lightweight model is a pre-trained MobileNet-V2 model, and the feature vector is... ,in For energy, For waveform entropy, Curvature of the in-phase axis. Complex disease classification imaging: energy extraction for each imaging grid. Waveform entropy Curvature of the same phase axis Forming eigenvectors Input the pre-trained MobileNet-V2 lightweight model and output the probabilities of voids, delamination, and abnormal moisture content. Determine the disease type based on the principle of maximizing probability.

[0065] In step S3, the optimized imaging processing proposes an optimized imaging algorithm for multi-layered media and complex defects, achieving accurate imaging and classification. Multi-layered media imaging is adapted to media with three or more layers. Taking a four-layer model (air layer, asphalt surface layer, base layer, and subgrade layer) as an example, parameters for each layer are defined, including an air layer thickness of 0.3m (antenna height), a relative permittivity of 1.0, and an electromagnetic wave attenuation coefficient. The electromagnetic field strength is 0.01 dB / m, the electromagnetic wave velocity is 0.3 m / ns; the asphalt surface layer thickness is 0.18 m, the relative permittivity of the medium is 6.0, and the electromagnetic wave attenuation coefficient is [missing information]. The electromagnetic induction coefficient is 0.3 dB / m, the electromagnetic wave velocity is 0.122 m / ns; the thickness of the substrate is 0.6 m, the relative permittivity of the medium is 10.0, and the electromagnetic wave attenuation coefficient is [missing information]. The electromagnetic wave velocity is 0.095 m / ns; the roadbed thickness is 1.4 m; the relative permittivity of the medium is 8.0; and the electromagnetic wave attenuation coefficient is [missing information]. The electromagnetic wave velocity is 0.106 m / ns, with a wavelength of 0.6 dB / m. Traditional time delay formulas do not consider layered attenuation; this invention proposes a modified formula:

[0066] ,

[0067] In the formula: The total two-way propagation delay (ns) of electromagnetic waves in multilayer media. The medium layer number (starting from 1, 1 is the air layer, 2 is the asphalt surface layer, 3 is the base layer, ..., n is the bottommost subgrade layer); For the first Layer and First The x-coordinate (m) of the refraction point at the interface; (Horizontal coordinates of the transmitting antenna on the ground, in meters); (Horizontal coordinates of the imaging grid, in meters); for The propagation speed of electromagnetic waves in a layered medium (unit: m / ns). , The speed of light in a vacuum. , For the first The relative permittivity of the dielectric layer, For the first The electromagnetic wave velocity of the layered medium; For the first Electromagnetic wave attenuation coefficient (dB / m) of the layered medium. , For the first The moisture content of the medium in the layer; For the first The thickness of the dielectric layer; For the first The attenuation compensation factor of the layer medium (used to correct the time delay calculation deviation caused by the energy attenuation of electromagnetic waves propagating in deep media).

[0068] The method for finding the refraction point employs a layer-by-layer one-dimensional search:

[0069] 1. Initial range: Refraction point of the first layer (air layer - asphalt layer) The search scope is ;

[0070] 2. Iterative solution: The golden section method is used to find the solution that corrects the delay. smallest , and then Based on this, determine the refraction point of the second layer (asphalt layer - base course). Search scope Iterate until the th layer;

[0071] Error verification: The delay calculation error of the four-layer model is less than 2%, which is a reduction compared to the traditional two-layer model.

[0072] Deep signal enhancement and imaging: For scattering data of media with 3 layers or less, inverse attenuation weighting is used (the... Weight of layer medium The signal amplitude of the deep layer was enhanced, increasing from 0.8mV to 1.5mV, resulting in a greater than 40% improvement in imaging clarity. The final imaging results were output in depth-lateral coordinates, with a resolution of 15cm laterally and 6cm in depth.

[0073] In the construction of multi-feature vectors for complex disease classification imaging, three types of features are extracted to form vectors for each imaging grid (size 15cm×6cm). .energy The calculation method for (mV) is the mean amplitude of the scattered data within the grid. For differences in disease characteristics, E is greater than 15 when there are cavities, and 10 ≤ mV when there are voids. ≤15, when the moisture content is abnormal <10; Waveform entropy The calculation method is as follows In the formula For the first The amplitude probability of each sampling point The number of sampling points represents the difference in cavitation characteristics among disease sites. Greater than 0.8, 0.5≤ when empty ≤0.8, when the moisture content is abnormal Less than 0.5; Curvature of the in-phase axis The method is used to quantify the curvature of the in-phase axis at the same reflection interface (such as the subgrade layer interface or the boundary of the defect) in the radar profile. The calculation process strictly follows the logic of in-phase axis extraction, curve fitting, and curvature solution. The specific steps are as follows:

[0074] 1. Establish a two-dimensional rectangular coordinate system with the lateral direction of the radar profile as the x-axis (unit: m, corresponding to highway mileage or lateral offset) and the depth direction as the z-axis (unit: m, downward is the positive direction); select continuous and clear reflection phase axes (such as peak phase axes or valley phase axes) in the radar profile to ensure that the extracted phase axes have no obvious discontinuities (the number of discontinuities is less than or equal to 3, and the length of a single discontinuity is less than 0.5m, which can be completed by linear interpolation); along the extracted phase axes, select sampling points uniformly at lateral intervals of 0.1m, with the number of sampling points in each calculation unit (corresponding to the lateral range of the imaging grid) being greater than or equal to 10, to ensure the accuracy of curve fitting;

[0075] 2. Extracted sampling point data of the in-phase axis , … Preprocessing is performed; The criteria remove outliers that deviate from the overall trend, and calculate all sampling points. mean of values and standard deviation If a certain sampling point satisfy If a point is found to be out of the loop, it is replaced by the linear interpolation result of two adjacent sampling points. The preprocessed in-phase axis data is smoothed using the moving average method (window size = 5 sampling points) to eliminate local fluctuations caused by random noise and preserve the overall curvature trend.

[0076] 3. A quadratic polynomial fitting method is used to model the curve of the smoothed in-phase axis data. The fitting function is in the form of: In the formula, represents the lateral coordinate of the sampling point on the same phase axis, in meters; These are the depth coordinates of the sampling points on the same phase axis, in meters. , , The fitting coefficients are obtained using the least squares method.

[0077] 4. The fitting coefficients are solved using the least squares method. The objective function is to minimize the sum of squared fitting errors for all sampling points, i.e.:

[0078] ,

[0079] Through the , , Taking the partial derivatives of each equation and setting them to zero, we obtain the system of equations:

[0080] ,

[0081] Substitute the preprocessed sampling points Solving the system of equations yields the fitting coefficients. , , Complete the modeling of the in-phase axis curve;

[0082] 5. Based on the curvature formula of a plane curve in differential geometry, and combined with the results of quadratic polynomial fitting, the in-phase axis at any transverse position... Curvature of the in-phase axis at the location (Unit: m) -1 The calculation is as follows: Taking the derivative of the fitted function, we obtain the slope of the in-phase axis: ; Continuing to differentiate, we get Then, substituting into the formula for the curvature of a plane curve, we get:

[0083] ,

[0084] 6. Obtain the in-phase axis curvature corresponding to all sampling points within the calculation unit (imaging grid). The average value is taken as the final result of the curvature of the in-phase axis of this unit, that is:

[0085] ,

[0086] In the formula, To calculate the number of in-phase axis sampling points within the calculation unit, Greater than or equal to 10.

[0087] Through extensive experimental data calibration, the curvature of the in-phase shaft corresponding to different roadbed defects was determined. The following scope is used for disease classification and identification:

[0088] Cavity disease: >0.2m -1 The phase axis exhibits a distinctly downward-curving parabolic shape;

[0089] Voiding disease: 0.1m -1 ≤ ≤0.2m -1 The phase axis exhibits a gently curved shape;

[0090] Abnormal moisture content diseases: <0.1m -1 The phase axes are basically straight or slightly curved.

[0091] The lightweight model classification uses a pre-trained MobileNet-V2 model. The specific training and inference steps are as follows:

[0092] 1. Dataset Construction: Collect 5000 sets of disease data, including 1600 sets of voids, 1700 sets of detachment, and 1700 sets of abnormal moisture content, which are divided into training set and test set in a 7:3 ratio;

[0093] 2. Model Training: Freeze the first 10 layers of MobileNet-V2, fine-tune the last 3 layers, use Adam as the optimizer, train for 50 epochs, and achieve a test set accuracy of ≥92%;

[0094] 3. Inference Output: Input feature vector V, output the probability (P) of three types of defects: voids, delamination, and abnormal moisture content. 空 P 脱 P 水 The one with the highest probability is the judgment type.

[0095] The classification results are visualized using color coding and dimension annotation to output an image. In the color mapping, cavities are red, voids are yellow, and abnormal moisture content is blue; the dimension annotation is to mark the three-dimensional dimensions (length × width × depth) of the diseased area.

[0096] Further, as a specific implementation method, in step S4, the definition of highway grade adopts the statutory classification in the "Highway Engineering Technical Standards" (JTG B01—2014), specifically including expressways, Class I highways, Class II highways, Class III highways, and Class IV highways. High-grade highways refer to expressways and Class I highways, while ordinary highways refer to Class II and below. The criteria for determining the minor level of damage are: the maximum side length D of the damage is less than 0.3m or the reflection intensity I is less than 0.2I0 (I0 is the standard cavity reflection intensity). The criteria for determining the medium level of damage are: the maximum side length D of the damage is between 0.3m and 0.8m or the reflection intensity I is between 0.2 and 0.5I0. The criteria for determining the severe level of damage are: the maximum side length D of the damage is greater than 0.8m or the reflection intensity I is greater than 0.5I0. Specifically, the damage level of a high-grade highway is upgraded by one grade, and the damage level of an ordinary highway is upgraded by one grade. The road damage level remains unchanged. Specifically, for minor damage, the maximum side length D of the damage is less than 0.2m or the reflection intensity I is less than 0.15I0 for high-grade highways, and less than 0.3m or less than 0.2I0 for ordinary highways. For medium damage, the maximum side length D of the damage is between 0.2m and 0.3m or between 0.15 and 0.2I0 for high-grade highways, and between 0.3m and 0.8m or between 0.2 and 0.5I0 for ordinary highways. For severe damage, the maximum side length D of the damage is greater than 0.3m or greater than 0.2I0 for high-grade highways, and greater than 0.8m or greater than 0.5I0 for ordinary highways.

[0097] Furthermore, as a specific implementation method, in step S4, the risk-cost scoring formula is priority score = risk coefficient × (1 / maintenance cost coefficient), where the risk coefficient for severe level is 3, medium level is 2, minor level is 1, the risk coefficient for high-grade highways is ×1.5, and the maintenance cost coefficient is set according to the repair method (excavation and backfilling = 3, grouting = 2, monitoring / drainage = 1).

[0098] In the maintenance decision-making linkage in step S4, a linkage mechanism is established for disease characteristics, grade assessment, priority ranking, and report generation, directly connecting with maintenance execution. The disease grade assessment is based on the maximum side length D of the disease, the reflection intensity I (relative to the standard void intensity I0, I0=200mV, measured by a standard laboratory test block), and the highway grade, dividing it into three levels: minor void disease (corresponding to voids) is maintained for a 12-month period, primarily through monitoring, with retesting every 3 months; moderate void disease is maintained for a 6-month period, primarily through localized grouting repair; severe void disease is maintained for a 1-month period, primarily through excavation and backfilling; and minor void disease... The following are the classifications of road types: Minor (18-month maintenance period, primarily monitoring); Medium (9-month maintenance period, primarily pressure grouting repair); Severe (3-month maintenance period, primarily base course repair); Mild (24-month maintenance period, primarily drainage treatment); Medium (12-month maintenance period, primarily soil replacement); Severe (6-month maintenance period, primarily deep drainage). Priority ranking uses a risk-cost comprehensive scoring method, with the formula: Priority Score = Risk Coefficient × (1 / Maintenance Cost Coefficient). The risk coefficient is calculated as follows: Severe = 3, Medium = 2, Mild = 1; High-grade highways × 1.5, Ordinary highways × 1.0. Maintenance cost coefficients are calculated as follows: Excavation and backfilling = 3, Pressure grouting = 2, Monitoring / drainage = 1. In the ranking rules, scores are ranked from highest to lowest; if scores are the same, severe highways are prioritized; if grades are the same, high-grade highways are prioritized. The report generation function automatically generates a roadbed disease maintenance decision report, which includes the following content:

[0099] 1. Detection Overview: Road section station number, detection time, equipment parameters, environmental conditions (temperature, moisture content);

[0100] 2. List of defects: the chainage range, defect type, level, three-dimensional dimensions, and spatial coordinates (mileage + lateral offset) of each abnormal road section.

[0101] 3. Imaging data: Classified and colored radar images (including dimension annotations), and photos of road surface defects (linked to radar data locations);

[0102] 4. Maintenance recommendations: Recommended solutions, estimated costs, and implementation timeframes;

[0103] 5. Data attachments: raw radar data, sensor data, and GPS trajectory data, which can be exported to GIS format.

[0104] Example 2

[0105] Based on Example 1, the present invention discloses a ground-penetrating radar parameter optimization and survey device for detecting the condition of highway subgrade, comprising a radar detection unit, a dynamic sensing unit, a multi-source data acquisition unit, a control processing unit, and an execution and interaction unit that are connected in a bidirectional signal connection with the control processing unit.

[0106] The radar detection unit is used to transmit / receive electromagnetic waves and correlate them with road mileage. It includes a ground-penetrating radar main unit, a multi-band air-coupled antenna, and a wheel encoder. The center frequency of the multi-band air-coupled antenna covers 100-1200MHz to adapt to the detection of roadbed, subgrade, and surface layers. The wheel encoder is mounted on the detection vehicle to realize the correlation between radar data and mileage.

[0107] The dynamic sensing unit is used to collect dynamic parameters of the roadbed medium, including a high-frequency capacitive moisture content sensor, an infrared temperature sensor, and an inertial measurement unit.

[0108] The multi-source data acquisition unit is used to simultaneously acquire multi-dimensional data, including a road camera, a GPS timing module, and a traffic flow receiving module. The road camera has a resolution of 3840×2160 and a frame rate of 30fps to extract apparent damage features. The GPS timing module synchronizes radar, sensor, and visual data. The traffic flow receiving module acquires lane-level vehicle flow density via 4G or 5G.

[0109] The control and processing unit is used to run the core algorithm and process data. It includes an embedded controller, a storage module and an algorithm integration module. The algorithm integration module has built-in PID parameter adaptive algorithm, Dijkstra lane planning algorithm, multi-layer medium time delay correction imaging algorithm and MobileNet-V2 disease classification model. It can automatically correct radar parameters, identify abnormal road sections and generate disease classification imaging maps.

[0110] The execution and interaction unit is used to carry the device and output results, including a modified detection vehicle, a display terminal and a maintenance decision output interface. The modified detection vehicle has a range of more than 10 hours and can travel at a constant speed of 60-65 km / h. The display terminal is a touch screen to display radar profiles and imaging results in real time. The maintenance decision output interface is an RS485 or GIS interface to push maintenance decision reports to the highway maintenance system.

[0111] Furthermore, in the radar detection unit, the instantaneous transmit power of the ground-penetrating radar host is greater than or equal to 6W, the dynamic range is greater than or equal to 150dB, and the A / D conversion bit depth is greater than or equal to 12; the lateral width of the multi-band air-coupled antenna is greater than or equal to 3.75m to cover the full width of the standard lane, and it can automatically switch the center frequency according to the roadbed location: roadbed detection switches to 100-400MHz, base layer detection switches to 400-900MHz, cement concrete surface layer detection switches to 900-1200MHz, and asphalt concrete surface layer detection switches to greater than 1200MHz.

[0112] Furthermore, in the dynamic sensing unit, the high-frequency capacitive moisture content sensor is determined by the formula... ( The relative permittivity of the roadbed medium is calculated based on the moisture content. The infrared temperature sensor corrects α by increasing the electromagnetic wave attenuation coefficient by 5% for every 10°C increase in temperature. The inertial measurement unit triggers the control processing unit to perform spatial flip correction on the radar data by outputting the antenna tilt angle in real time, thus eliminating the profile offset caused by the tilt.

[0113] Furthermore, in the multi-source data acquisition unit, the road surface camera extracts road surface cracks using the Canny edge detection algorithm and calculates road surface settlement using the binocular vision principle; the traffic flow receiving module divides traffic flow density into three time periods: low peak, medium peak, and high peak, and transmits the time period data to the control processing unit in real time to support lane planning.

[0114] Furthermore, in the control processing unit, the embedded controller is an NVIDIA Jetson Xavier or an embedded chip with equivalent computing power; the PID parameter adaptive algorithm uses the penetration depth deviation (target depth - measured depth) as feedback quantity and adjusts the center frequency and sampling window.

[0115] Furthermore, in the algorithm integration module of the control processing unit, the Dijkstra lane planning algorithm uses lanes as nodes and traffic flow density as weights to solve for the detection path with the minimum total weight; the multi-layer medium delay correction imaging algorithm calculates the electromagnetic wave propagation delay and solves for the refraction point through a layer-by-layer one-dimensional search.

[0116] Furthermore, in the algorithm integration module of the control processing unit, the MobileNet-V2 disease classification model uses the energy of the imaging grid. Waveform entropy Curvature of the same phase axis For feature vectors It outputs the probability of voids, voids, and abnormal moisture content, and the one with the highest probability is the judgment type, which is adapted to the real-time processing requirements of embedded controllers.

[0117] Furthermore, in the execution and interaction unit, the modified detection vehicle is equipped with hazard lights and a following warning system to ensure the safety of traffic detection without interruption; the maintenance decision output interface pushes maintenance decision reports that include the range, type, level, three-dimensional dimensions, recommended maintenance plan, estimated cost and time limit of the defects, and supports exporting them as GIS format files.

[0118] Furthermore, the dynamic sensing unit also includes a fault-tolerant mechanism: when the deviation of any single sensor data from the mean of the five adjacent sampling points is greater than 5%, the moving average of the five adjacent points is used to replace the fluctuating data; when the sensor has no data output for 10 consecutive seconds, the control processing unit automatically calls the average medium parameters of the previous 100m road section to calculate the radar static parameters and marks the sensing fault section in the data file.

[0119] Furthermore, the radar detection unit, dynamic sensing unit, and multi-source data acquisition unit are connected to the control and processing unit via a CAN bus to ensure real-time data exchange between the units; and all units are powered by the lithium battery of the modified detection vehicle, supporting continuous detection.

[0120] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0121] The embodiments described above are merely illustrative of implementation methods of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for optimizing and surveying ground-penetrating radar parameters for detecting roadbed conditions, characterized in that, Includes the following steps: S1, Ground Penetrating Radar Parameter Optimization Configuration, including Static Basic Parameter Configuration and Dynamic Parameter Adaptive Adjustment; The static basic parameters are configured based on the roadbed layering and the relative permittivity of the medium, and the center frequency is selected by quantization according to a preset correlation formula; the sampling time window base value is calculated based on the maximum detection depth and medium parameters, and the actual value is 1.3 times the base value; the sampling interval is set so that the sampling rate is more than 4 times the highest frequency of the reflected wave; the dynamic parameter adaptive adjustment is achieved by acquiring the medium dynamic parameters through the vehicle-mounted real-time sensing system, and the sensing data is transmitted to the control unit in real time via the CAN bus; based on the medium dynamic parameters, the center frequency and sampling time window are adjusted in real time through a PID control algorithm to control the penetration depth deviation to be less than 3%; S2, rapid survey execution, includes multi-source data preprocessing, intelligent detection, and multi-feature fusion anomaly determination; the multi-source data preprocessing involves collecting traffic flow data, historical damage data, and road surface visual data of the detection section, establishing a historical damage-radar feature association database, dividing traffic flow time periods, and simultaneously calibrating the spatiotemporal coordinates of radar, sensor, and visual data; the intelligent detection involves planning the lane detection sequence based on traffic flow time periods using the Dijkstra algorithm, prioritizing the detection of lanes with low traffic flow density. The vehicle-mounted ground-penetrating radar system continuously scans at a constant speed of 60-65 km / h, simultaneously recording radar data, road images, and sensor data; the multi-feature fusion anomaly determination is to extract radar features and road image features in 10m units, calculate the anomaly probability through a logistic regression model, and classify confirmed anomalies, suspected anomalies, and normal road sections according to the probability threshold. S3, optimized imaging processing, including multi-layer media imaging and complex disease classification; the multi-layer media imaging involves establishing a multi-layer media model of the roadbed, calculating the electromagnetic wave propagation delay using a modified time delay formula, solving for refraction points through a layer-by-layer one-dimensional search, performing inverse attenuation weighted enhancement on deep scattering data, and generating multi-layer media imaging results; the complex disease classification involves extracting the energy, waveform entropy, and in-phase axis curvature of the imaging grid to form feature vectors, inputting them into a pre-trained lightweight model, and outputting type determination results for cavities, voids, and abnormal water content. The preset correlation formula for selecting the center frequency is as follows: In the formula, The center frequency of the antenna. The relative permittivity of the medium, This refers to the radar's vertical resolution. The corrected delay formula is as follows: ， In the formula: This is the total two-way propagation delay of electromagnetic waves in a multilayer medium; This refers to the dielectric layer number; For the first Layer and First The x-coordinate of the refraction point at the interface; for The propagation speed of electromagnetic waves in a layered medium. , The speed of light in a vacuum. , For the first The relative permittivity of the dielectric layer; For the first Electromagnetic wave attenuation coefficient of the layered medium, , For the first The moisture content of the medium in the layer; For the first The thickness of the dielectric layer; For the first Attenuation compensation factor of the layer medium.

2. The method for optimizing and surveying ground-penetrating radar parameters for detecting roadbed conditions according to claim 1, characterized in that, It also includes step S4, maintenance decision linkage, which includes disease level assessment and priority ranking and report generation; the disease level assessment is based on disease size, reflection intensity and highway grade, dividing the disease into three levels: minor, moderate and severe, and matching the corresponding maintenance cycle; the priority ranking and report generation is to sort the maintenance priorities according to the risk-cost score, automatically generate a decision report containing disease parameters, maintenance plan and estimated cost, and push it to the highway maintenance system.

3. The method for optimizing and surveying ground-penetrating radar parameters for detecting roadbed conditions according to claim 1, characterized in that, In step S1, the roadbed layering includes a subgrade layer, a base course, a cement concrete surface course, and an asphalt concrete surface course; wherein the center frequency of the subgrade layer is selected as 100-400MHz, the center frequency of the base course is selected as 400-900MHz, the center frequency of the cement concrete surface course is selected as 900-1200MHz, and the center frequency of the asphalt concrete surface course is selected as greater than 1200MHz.

4. The method for optimizing and surveying ground-penetrating radar parameters for detecting roadbed conditions according to claim 1, characterized in that, In step S1, the real-time sensing system includes a high-frequency capacitive moisture content sensor, an infrared temperature sensor, and an inertial measurement unit; the dynamic parameters of the medium include the roadbed moisture content, temperature, and antenna tilt data.

5. The method for optimizing and surveying ground-penetrating radar parameters for detecting roadbed conditions according to claim 1, characterized in that, In step S2, the spatiotemporal coordinates are calibrated by time synchronization through a GPS timing module, and the mapping relationship between road surface image pixels and actual mileage is established by a checkerboard calibration method.

6. The method for optimizing and surveying ground-penetrating radar parameters for detecting roadbed conditions according to claim 1, characterized in that, In step S2, the vehicle-mounted ground-penetrating radar system includes a ground-penetrating radar host, an air-coupled antenna with a lateral width of 3.75m or greater, a wheel encoder, and a high-definition camera.

7. The method for optimizing and surveying ground-penetrating radar parameters for detecting roadbed conditions according to claim 1, characterized in that, In step S2, the anomaly probability formula of the logistic regression model is: ,in, , where is the radar feature weight; , where is the image feature weight; This represents the comprehensive value of radar characteristics. The image feature synthesis value; when A value greater than 0.8 is considered an anomaly. A value between 0.5 and 0.8 is considered a suspected abnormality. A value less than 0.5 is considered normal.

8. The method for optimizing and surveying ground-penetrating radar parameters for detecting roadbed conditions according to claim 1, characterized in that, In step S3, the lightweight model is a pre-trained MobileNet-V2 model, and its feature vector is... ,in For energy, For waveform entropy, The curvature of the in-phase axis.

9. The method for optimizing and surveying ground-penetrating radar parameters for detecting roadbed conditions according to claim 2, characterized in that, In step S4, the criteria for determining the severity of the disease are: the maximum side length of the disease is less than 0.3m or the reflection intensity is less than 0.2; the criteria for determining the moderate severity of the disease are: the maximum side length of the disease is between 0.3m and 0.8m or the reflection intensity is between 0.2 and 0.5; and the criteria for determining the severity of the disease are: the maximum side length of the disease is greater than 0.8m or the reflection intensity is greater than 0.5.

Images