Method for detecting global moisture content of highway base layer based on three-dimensional ground penetrating radar

Abstract

The invention discloses an expressway base layer global moisture content detection method based on a three-dimensional ground penetrating radar, which comprises the following steps: constructing a three-dimensional multiphase random medium electromagnetic simulation model by considering aggregate grading and a pore structure; based on the model, simulating the propagation process of radar waves in a base layer under different moisture content conditions to obtain simulated radar reflection wave data; the A-scan waveform is converted into an A-scan waveform; extracting a reflected wave mean value, a standard deviation and a peak value coefficient, so as to establish a reflected wave intensity-dielectric constant-moisture content three-variable regression model; based on a line tracking theory, inversing an actually measured dielectric constant through actually measured reflected wave two-way travel time and horizon thickness; calculating actually measured moisture content data by utilizing the actually measured dielectric constant obtained by inversion; and constructing a complete random field model representing the spatial variability of the moisture content, and carrying out constraint and calibration on the random field model by taking the actually measured moisture content as a sample point through a Kriging interpolation method, so as to generate a base-layer global moisture content visual distribution diagram.

Description

Technical Field

[0001] This invention belongs to the field of non-destructive testing technology for road structure performance, specifically relating to a method for detecting the moisture content of the entire base layer of a highway based on three-dimensional ground-penetrating radar. Background Technology

[0002] Semi-rigid base courses are an important component of highway structural systems. Their internal moisture content directly affects the material's elastic modulus, flexural strength, and volumetric stability, making it a key indicator for assessing the health of the base course. When the moisture content deviates from the optimum moisture content range, the base course is more prone to structural defects such as strength degradation, voids, loosening, subsidence, and water damage under vehicle loads and environmental cycles, seriously affecting road safety and service life.

[0003] Currently, the moisture content detection of highway base courses mainly employs discrete destructive testing methods such as drying, microwave methods, and core drilling. These methods have a limited number of measurement points, low detection efficiency, and difficulty in reflecting the continuous spatial variation of base course moisture content. Base course materials exhibit typical multiphase random heterogeneous characteristics; the random spatial distribution of aggregate phase, matrix phase, and pores causes significant scattering noise and dielectric disturbances, leading to an unstable mapping relationship between radar reflection wave characteristics and moisture content. Existing forward modeling methods using the finite-difference time-domain method often employ layered homogeneous models or random medium models for moisture content calculation, failing to consider the multiphase composition and spatially random distribution of the actual base course material, thus limiting the model's applicability in practical engineering environments. Furthermore, as a parameter with significant spatial variability, the continuous distribution of moisture content is difficult to reliably represent through interpolation with a limited number of measurement points. Summary of the Invention

[0004] Purpose of the invention: To address the problems of low engineering applicability of simulation models, complex radar signal noise, and difficulty in achieving continuous characterization of moisture content across the entire highway base course in moisture content detection, this invention proposes a method for detecting the full-area moisture content of highway base courses based on three-dimensional ground-penetrating radar. This method overcomes the limitations of traditional detection methods, such as discreteness, inefficiency, and poor model applicability, and achieves non-destructive, continuous, and high-precision automated detection of moisture content in highway base courses, providing reliable data support for road structure health assessment and intelligent maintenance.

[0005] Technical Solution: This invention proposes a method for detecting the overall moisture content of highway base courses based on three-dimensional ground-penetrating radar, comprising the following steps:

[0006] By considering aggregate gradation and pore structure, a three-dimensional multiphase random medium electromagnetic simulation model is constructed; the three-dimensional multiphase random medium electromagnetic simulation model includes an air layer, an asphalt concrete layer and a base layer from top to bottom.

[0007] Based on the constructed three-dimensional multiphase random medium electromagnetic simulation model, the propagation process of radar waves in the base layer under different moisture contents was simulated, and simulated radar reflection wave data under different moisture contents were obtained.

[0008] Simulated radar reflection wave data under different moisture contents are converted into A-scan waveforms under different moisture contents.

[0009] The mean, standard deviation, and peak value of reflected waves are extracted from A-scan waveforms under different moisture contents. These mean, standard deviation, and peak value are used as characterization indicators of reflected wave intensity. Using the dielectric constant and reflected wave intensity characteristic parameters corresponding to different moisture contents as sample data, a mapping relationship between reflected wave intensity, dielectric constant, and moisture content is established through multiple regression analysis, thereby constructing a three-variable regression model of reflected wave intensity-dielectric constant-moisture content.

[0010] Measured radar reflection wave data are obtained. Based on line tracking theory, the dielectric constant of the measured data is inverted by the measured two-way travel time of the reflected wave and the layer thickness. The dielectric constant of the measured data is substituted into the three-variable regression model of reflected wave intensity-dielectric constant-water content to obtain the measured water content data at the corresponding measuring point.

[0011] A fully random field model is established based on the covariance matrix decomposition method. In the fully random field model, the water content parameter at any point in space is regarded as a random variable that follows a preset probability distribution. The correlation between different spatial locations is described by the autocorrelation function and the autocorrelation distance, thereby characterizing the spatial variability of the water content parameter.

[0012] Using the measured moisture content data as known constraints, the Kriging interpolation method is used to constrain and calibrate the distribution of the completely random field model, so that the completely random field model matches the measured moisture content data while maintaining consistent statistical laws, thereby obtaining a visual distribution map of moisture content across the entire grassroots area.

[0013] Furthermore, the aforementioned construction of a three-dimensional multiphase random medium electromagnetic simulation model by considering aggregate gradation and pore structure specifically includes:

[0014] A pavement structure layer model is established from top to bottom, including an air layer, an asphalt concrete layer, and a base course. The air layer is used to simulate the positional relationship between the radar antenna and the asphalt concrete layer. The asphalt concrete layer is characterized using a random medium model. The base course is constructed by reconstructing the three-dimensional geometry of the aggregate particles and randomly placing the reconstructed aggregate particles according to the aggregate gradation range to form an aggregate phase. A vertical force is applied to compact the aggregate phase, causing the aggregate particles to form irregularly distributed unfilled spaces during the contact and rearrangement process. These unfilled spaces serve as a porous phase and together with the aggregate phase, constitute the multiphase random porous structure of the base course.

[0015] Determine the radar antenna frequency;

[0016] The model mesh size is determined based on the minimum wavelength of radar wave propagation, the dimensions of the road structure layer model, and the pixels of the base layer profile.

[0017] The sampling window is determined based on the propagation time of radar waves within the road surface structure layer model;

[0018] Determine the dielectric constants of different structural layers in the pavement structural layer model;

[0019] Based on the established road surface structure layer model, the determined radar antenna frequency, model mesh size, sampling time window, and the dielectric constants corresponding to different structural layers, a three-dimensional multiphase random medium electromagnetic simulation model is constructed.

[0020] Furthermore, different moisture content conditions are achieved by adjusting the air-to-water ratio in the aggregate phase pores of the base layer.

[0021] Furthermore, the specific operation of converting simulated radar reflection wave data under different moisture content conditions into A-scan waveforms under different moisture content conditions includes:

[0022] Simulated radar reflection wave data under different moisture content conditions are input into the denoising model to obtain denoised simulated radar reflection wave data.

[0023] The denoised simulated radar reflection wave data is converted into A-scan waveforms;

[0024] The denoising model consists of a first SCB module, a self-attention deep autoencoder, and a second SCB module connected in sequence.

[0025] In the first SCB module, the input simulated radar reflection wave data is first processed by a 1×1 convolutional layer for dimensionality reduction. The tensor output by the convolution is uniformly divided into two groups, X1 and X2, and then processed by SwinTransformer and residual convolution to obtain outputs Y1 and Y2 respectively. The outputs Y1 and Y2 are concatenated together and further processed by 1×1 convolution. Finally, residual addition is performed with the input simulated radar reflection wave data to obtain the feature tensor Z.

[0026] In the self-attention deep autoencoder, the feature tensor output by the first SCB module is denoised and restored through a self-attention enhanced encoder-decoder structure.

[0027] The second SCB module is the inverse operation of the first SCB module, and the second SCB module outputs the denoised simulated radar reflection wave data.

[0028] Furthermore, the specific operations for extracting the mean, standard deviation, and peak value of reflected waves from A-scan waveforms under different moisture content conditions include:

[0029] The A-scan waveforms under different moisture contents were decomposed by EMD to obtain the components, including: basic mode components and residual components.

[0030] Perform Hilbert transform on each component to obtain the corresponding Hilbert spectrum;

[0031] Hilbert spectra of different components were synthesized to obtain Hilbert spectra of A-scan waveforms under different moisture contents;

[0032] Based on the Hilbert spectrum of the obtained A-scan waveform, statistical analysis is performed on the amplitude of the Hilbert spectrum within a preset time window corresponding to the base layer reflection interface. The mean, standard deviation, and peak value of the amplitude within the time window are calculated respectively. These are used to characterize the overall energy level, amplitude dispersion, and distribution morphology of the reflected wave, thereby obtaining the mean, standard deviation, and peak value of the reflected wave under different moisture content conditions.

[0033] Furthermore, the acquisition of measured radar reflected wave data, based on line tracking theory, involves inverting the dielectric constant through the measured two-way travel time of the reflected wave and the layer thickness. Specific operations include:

[0034] According to the following formula, based on line tracing theory, the dielectric constant is inverted by measuring the two-way travel time of the reflected wave and the layer thickness:

[0035]

[0036]

[0037] in, Let be the thickness of the i-th layer. For the two-way travel time of this layer, For the speed of transmission, At the speed of light, Let be the dielectric constant of the i-th layer.

[0038] Furthermore, the aforementioned method for establishing a completely random field model based on covariance matrix decomposition specifically includes:

[0039] A spatial discrete model of the grassroots test section is constructed, the grassroots test section is divided into several spatial discrete units for random field modeling, and the spatial coordinates of the center point of each spatial discrete unit are output.

[0040] The mean of the moisture content parameter follows a log-normal distribution. and standard deviation Convert to the mean of the corresponding normal distribution and standard deviation The calculation formula is as follows:

[0041]

[0042]

[0043] The coordinates of the center points of each model unit are substituted into the exponential autocorrelation function to calculate the covariance matrix. The covariance matrix Perform Cholesky decomposition This yields the lower triangular matrix L;

[0044] The exponential autocorrelation function is:

[0045] ;

[0046] In the formula, l is the autocorrelation distance, and h is the relative distance between two sample points;

[0047] If a column vector S is formed by randomly generating n independent random samples that follow a standard normal distribution, then the standard normal random field P can be expressed as: ;

[0048] A mathematical transformation is performed on the standard normal random field P to obtain a field with a mean of . The sum and variance are Normally distributed random fields: ;

[0049] Taking the logarithm of a normally distributed random field yields a completely random field model.

[0050] Furthermore, the Kriging interpolation is calculated using the following formula:

[0051]

[0052]

[0053] in, This represents the value of the variation function between positions i and j. The weighting coefficients are assigned to the moisture content sample data at position i; m is the number of known moisture content sample points. The value of the variation function between position i and the predicted point. For Lagrange daily numbers; for The actual measured moisture content at the location; For prediction points The estimated value.

[0054] Furthermore, the base course is a cement-stabilized crushed stone base course; the cement-stabilized crushed stone base course is a semi-rigid road base course structure formed by mixing, spreading and compacting graded crushed stone as the main aggregate and cement as the binding material.

[0055] Beneficial Effects: This invention simulates the electromagnetic wave propagation process under different moisture contents by constructing a three-dimensional multiphase random medium electromagnetic simulation model that realistically reflects the gradation composition and pore structure of the base course. It utilizes a deep denoising model to suppress material heterogeneity scattering and multiple reflection interference, extracting reflected wave intensity characteristics such as mean, standard deviation, and peak value from the simulation waveform, and establishing a three-variable regression model of reflected wave intensity, dielectric constant, and moisture content. By acquiring measured radar echo signals using three-dimensional ground-penetrating radar, the dielectric constant of the road base course under real conditions is obtained based on the two-way travel time of the reflected wave and the layer thickness, thus correcting the moisture content prediction results. A completely random field model of the spatial variability of moisture content is constructed, and Kriging interpolation and field core drilling results are integrated to calibrate the model parameters, achieving continuous visualization of the base course moisture content across the entire domain. Compared with existing technologies, this invention has the following advantages:

[0056] (1) This invention constructs an aggregate phase by randomly placing aggregate particles of different sizes according to the aggregate gradation range, and applies vertical compaction only to the cement-stabilized crushed stone base layer, so that the aggregate particles undergo spatial rearrangement during the compaction process, thereby forming an irregularly distributed pore structure between the aggregate particles; at the same time, in the electromagnetic simulation, the aggregate phase and the pore phase are respectively assigned corresponding dielectric parameters, thereby constructing a three-dimensional electromagnetic simulation model of a multiphase random medium considering aggregate gradation and pore structure, so as to realize the true expression of the internal microstructure of the material;

[0057] (2) The proposed Swin-Conv Block and self-attention deep autoencoder (ADAE) joint denoising model achieves the coordinated expression of fine-grained local features and global features in radar signals through dual feature extraction of residual convolution and self-attention mechanism, which significantly suppresses material heterogeneous scattering noise and multiple reflection interference.

[0058] (3) The fully random field-Kriging interpolation fusion modeling method introduced in this invention realizes the continuous expression of the moisture content of highway base course in the spatial domain. By describing the spatial variability through the random field model and constraining and calibrating the random field model by combining the measured moisture content data with Kriging interpolation, the full-domain three-dimensional visualization reconstruction of moisture content is realized, which makes up for the limitation of existing methods that can only obtain discrete point data;

[0059] (4) The moisture content detection system constructed by this invention integrates data acquisition, noise preprocessing, moisture content detection and visualization analysis modules, forming a complete automated process from three-dimensional radar data input to the output of moisture content map of the entire base layer, which significantly improves detection efficiency and data processing automation, and provides efficient and intelligent technical support for health monitoring and smart maintenance of highway base layers;

[0060] (5) The present invention can realize non-destructive, continuous and high-precision automatic detection of moisture content of highway base course, providing reliable data support for road base course health monitoring and intelligent maintenance. Attached Figure Description

[0061] Figure 1 This is a flowchart of a method for detecting the overall moisture content of highway base course based on three-dimensional ground-penetrating radar, provided in an embodiment of the present invention.

[0062] Figure 2 This refers to the three-dimensional reconstructed aggregate phase profile in an embodiment of the present invention;

[0063] Figure 3 This refers to the three-dimensional multiphase random medium electromagnetic simulation model constructed based on the finite-difference time-domain method in the embodiments of the present invention;

[0064] Figure 4 This is a diagram of the denoising model architecture in an embodiment of the present invention;

[0065] Figure 5 These are radar A-scan waveforms before and after noise reduction in an embodiment of the present invention.

[0066] Figure 6 These are the original radar signal and the basic mode components and residual components after EMD decomposition in the embodiments of the present invention.

[0067] Figure 7 This is a visualization distribution map of the overall moisture content of the base layer generated in an embodiment of the present invention. Detailed Implementation

[0068] The technical solution of the present invention will now be further described in conjunction with the accompanying drawings and embodiments.

[0069] Three-dimensional ground-penetrating radar (GPR) possesses advantages such as high precision, non-destructive operation, continuous detection, and wide coverage, making it an important technological means to meet the ever-increasing demand for road structure inspection. GPR transmits high-frequency narrowband electromagnetic waves and receives reflected signals. Based on parameters such as the energy, amplitude, and time delay of the incident and reflected waves, it obtains the distribution of dielectric parameters within the pavement structure layer, indirectly reflecting moisture content information. Based on this, this invention proposes a method for detecting the full-area moisture content of highway base courses based on three-dimensional GPR, including:

[0070] S1. A three-dimensional multiphase random medium electromagnetic simulation model considering aggregate gradation and pore structure is constructed based on the finite-difference time-domain method. The propagation process of radar waves in the base layer is simulated under different moisture contents to obtain simulated radar reflection wave data close to the actual base layer conditions. In this step, point cloud coordinate data of the real aggregates are first obtained using laser scanning technology. Then, in PFC3D software, spheres of different sizes are used to fit the aggregate surface shape to achieve three-dimensional reconstruction of the aggregate phase contour, such as... Figure 2 As shown;

[0071] Then, a three-dimensional multiphase random medium electromagnetic simulation model based on the finite-difference time-domain method simultaneously describes the spatial distribution of aggregate phase, matrix phase, and pore phase. This model is constructed using gprMax software, and the parameters set for this simulation model include:

[0072] Model Dimensions: A pavement structure layer model with dimensions of 2.0m × 0.6m × 0.35m was established, comprising, from top to bottom, a 0.06m air layer, a 0.18m asphalt concrete layer, and a 0.36m cement-stabilized crushed stone base course, as shown below. Figure 3 As shown; the air layer positions the antenna 0.06m above the asphalt surface layer, consistent with the actual detection height of the ground-coupled ground-penetrating radar; the asphalt concrete layer uses a random medium model; the cement-stabilized crushed stone base course refers to a semi-rigid road base structure formed by mixing, spreading, and compacting graded crushed stone as the main aggregate and cement as the binder. Its cement content, gradation composition, and structural form meet the general technical requirements for base course structures in highway engineering. The cement-stabilized crushed stone base course reconstructs the three-dimensional geometry of the aggregate particles and randomly places the reconstructed aggregate particles according to the aggregate gradation range to form an aggregate phase. After each grade of aggregate is placed, a vertical force is applied to compact it, causing the aggregate particles to form irregularly distributed unfilled spaces during contact and rearrangement. These unfilled spaces serve as the pore phase, which, together with the aggregate phase, constitutes the multiphase random pore structure of the cement-stabilized crushed stone base course. Antenna frequency: Ricker wavelet is selected. The formula for generating radar waves is as follows:

[0073]

[0074] in, It is the center frequency of the Ricker wavelet. It is the propagation time of electromagnetic waves;

[0075] Antenna frequencies of 600MHz, 800MHz, and 900MHz are commonly used intermediate frequencies, while 1.0GHz, 1.2GHz, and 1.5GHz are commonly used high-frequency antennas.

[0076] Mesh settings: The model mesh size is determined by the minimum wavelength of electromagnetic wave propagation, the model dimensions, and the pixel count of the aggregate model profile. The calculation formula is as follows:

[0077]

[0078]

[0079] Where λ is the minimum wavelength of the Ricker wavelet; c is the speed of light; and f is the highest frequency of the electromagnetic wave. The equivalent dielectric constant of the structural layer; and These are the mesh sizes calculated based on the model and the material profile images, respectively. and These are the length and depth of the model, respectively. and , respectively, represent the pixels in the length and width directions of the cross-sectional image of the heterogeneous material; 𝑥 represents the length direction of the model, and 𝑦 represents the depth direction of the model.

[0080] Time window size: The sampling time window is determined by the propagation time of the electromagnetic wave inside the road surface, and the calculation formula is as follows:

[0081]

[0082] Where h is the depth of the model; v is the propagation speed of radar waves in each layer; and α is the empirical amplification factor, which is generally taken as 1.3.

[0083] Dielectric constant: The simulation model media include air, water, asphalt concrete, and cement-stabilized crushed stone. The relative dielectric constant of air is 1.0, that of water is 81.0, that of asphalt concrete is 6.0-9.0, and that of cement-stabilized crushed stone is 9.0-12.0.

[0084] Different moisture content conditions are achieved by adjusting the air-to-water ratio in the aggregate phase pores of the model base layer.

[0085] Based on the established road surface structure layer model, the determined radar antenna frequency, model mesh size, sampling time window, and the dielectric constants corresponding to different structural layers, a three-dimensional multiphase random medium electromagnetic simulation model is constructed.

[0086] S2. The simulated radar reflection wave data is input into the denoising model for processing to suppress material heterogeneous scattering noise and multiple reflection interference, obtaining the denoised A-scan waveforms under different moisture contents. In this step, the denoising model consists of two Swin-Conv Blocks (SCBs) and a self-attention deep autoencoder (ADAE) connected sequentially to improve the denoising performance of ground-penetrating radar signals. The first SCB module performs parallel feature extraction of local convolution and window self-attention, the self-attention deep autoencoder performs multi-scale encoding-decoding and attention enhancement restoration, and the second SCB module realizes detail reconstruction and feature alignment. The denoising model architecture is as follows: Figure 4 As shown. Specifically includes:

[0087] The first SCB module combines Residual Convolution (RConv) and Swing Transformer (SwinT) to extract fine-grained and global features of radar signals. RConv ensures the preservation of feature information through residual connections, while SwingT processes global features through a self-attention mechanism. This module ultimately outputs a feature tensor Z.

[0088] ADAE module: After receiving the feature tensor Z output by the first SCB module, the ADAE module performs signal recovery through a self-attention enhanced encoder-decoder structure; this encoder-decoder structure performs multi-scale convolution operations on the signal and uses a self-attention mechanism to selectively emphasize spatially consistent and semantically relevant signal patterns while suppressing irrelevant noise;

[0089] The second SCB module is the inverse operation of the first SCB module, used for further denoising and dimension matching. The second SCB module ensures that the recovered signal is consistent with the structure of the original signal by reconstructing the details and spatial alignment of the signal, thereby more accurately reflecting the important features in the radar signal.

[0090] Specifically, the input reflected wave signal is first processed by a 1×1 convolutional layer for dimensionality reduction. The tensor output by the convolution is uniformly divided into two groups, X1 and X2, which are then processed by SwinT and RConv respectively to obtain outputs Y1 and Y2. Y1 and Y2 are concatenated and further processed by a 1×1 convolution. Finally, residual addition is performed with the input reflected wave signal to obtain the feature tensor Z. The feature tensor Z is input to ADAE for further denoising and recovery. Finally, the denoised reflected wave signal is output through the second SCB module, as shown below. Figure 5 As shown.

[0091] S3. Using a radar signal algorithm based on Hilbert-Huang transform, the mean, standard deviation, and peak value of the reflected wave are extracted from the denoised A-scan waveform under different moisture contents, and a three-variable regression model of reflected wave intensity, dielectric constant, and moisture content is constructed. Specific operations include:

[0092] The denoised A-scan waveform signal is decomposed using EMD, which breaks down the multi-frequency signal into a finite number of single-frequency signals, yielding four fundamental mode components (IMF1-4) and residual component curves, as shown below. Figure 6 As shown;

[0093] The Hilbert transform is performed on each component of the ground-penetrating radar signal to obtain the corresponding Hilbert spectrum. Finally, the Hilbert spectrum of the original ground-penetrating radar signal is obtained by synthesizing the different components.

[0094] Based on the obtained A-scan waveform Hilbert spectrum, statistical analysis is performed on the Hilbert spectrum amplitude within a preset time window corresponding to the base layer reflection interface. The mean, standard deviation, and peak value of the amplitude within the time window are calculated respectively. These are used to characterize the overall energy level, amplitude dispersion, and distribution morphology of the reflected wave, thereby obtaining the mean, standard deviation, and peak value of the reflected wave under different moisture content conditions.

[0095] S4. Acquire measured highway radar echo signal data using 3D ground-penetrating radar. Based on line tracking theory, invert the measured dielectric constant by measuring the two-way travel time of the reflected wave and the layer thickness. Substitute the measured dielectric constant into the three-variable regression model of reflected wave intensity-dielectric constant-water content to obtain the measured water content data at the corresponding measuring point. In this step, the 3D ground-penetrating radar is a 24-channel 3D array ground-penetrating radar system. The radar antenna of this 3D array ground-penetrating radar system emits electromagnetic waves along the direction of lane travel, obtaining high-density 3D electromagnetic wave data in 3DRA format composed of multiple parallel B-scans.

[0096] In this step, based on line tracing theory, the dielectric constant is inverted by measuring the two-way travel time of the reflected wave and the layer thickness. The calculation formula is as follows:

[0097]

[0098]

[0099] in, Let be the thickness of the i-th layer; This is the two-way travel time for this layer; For the speed of transmission; The speed of light; Let be the dielectric constant of the i-th layer.

[0100] S5. Using the covariance matrix decomposition method, a completely random field model is constructed to characterize the spatial variability of water content. In the completely random field model, each point in space is regarded as a random variable following a specific probability distribution to describe the spatial variability of the water content parameter. For example... Figure 7 As shown, the specific model building process is as follows:

[0101] A spatial discrete model of a test section of cement-stabilized crushed stone base course for highways was constructed using numerical modeling software. The test section was divided into several spatial computational units for random field modeling, and the spatial coordinates of the center point of each spatial computational unit were output.

[0102] The mean of the moisture content parameter follows a log-normal distribution. and standard deviation Convert to the mean of the corresponding normal distribution and standard deviation The calculation formula is as follows:

[0103]

[0104]

[0105] The coordinates of the center points of each model unit are substituted into the exponential autocorrelation function to calculate the covariance matrix. The covariance matrix Perform Cholesky decomposition This yields the lower triangular matrix L.

[0106] The exponential autocorrelation function is:

[0107] ;

[0108] In the formula, l is the autocorrelation distance, and h is the relative distance between two sample points;

[0109] If a column vector S is formed by randomly generating n independent random samples that follow a standard normal distribution, then the standard normal random field P can be expressed as: A mathematical transformation is performed on the standard normal random field to obtain a field that follows the mean value of ; The sum and variance are Normally distributed random fields: Finally, taking the logarithm of the normally distributed random field yields the log-normally distributed completely random field of the water content parameter.

[0110] By using the Kriging interpolation method, the measured moisture content is used as a sample point to constrain and calibrate the random field model, generating a visualized distribution map of the moisture content of the entire highway base layer.

[0111] In this step, Kriging interpolation is calculated using the following formula:

[0112]

[0113]

[0114] in, This represents the value of the variation function between positions i and j. The weighting coefficients are assigned to the moisture content sample data at position i; m is the number of known moisture content sample points. The value of the variation function between position i and the predicted point. For Lagrange daily numbers; for The actual measured moisture content at the location; For prediction points The estimated value.

[0115] Example 2:

[0116] Based on Embodiment 1, this invention proposes a system for detecting the overall moisture content of highway base courses using three-dimensional ground-penetrating radar. This method includes a data acquisition module, a noise preprocessing module, a moisture content detection module, and a moisture content visualization module.

[0117] The data acquisition module is used for acquiring and storing multi-channel data from 3D ground-penetrating radar.

[0118] The noise preprocessing module is used to remove material heterogeneity scattering noise and multiple reflection interference from the raw 3D ground-penetrating radar data.

[0119] The moisture content detection module is used to automatically detect the moisture content parameters of the road base layer from the input three-dimensional ground penetrating radar data;

[0120] The moisture content visualization module is used to automatically summarize the detected moisture content parameters and generate a visualized distribution map of the moisture content of the entire base layer of the tested road section.

Claims

1. A method for detecting the overall moisture content of highway base courses based on three-dimensional ground-penetrating radar, characterized in that: Includes the following steps: By considering aggregate gradation and pore structure, a three-dimensional multiphase random medium electromagnetic simulation model is constructed; the three-dimensional multiphase random medium electromagnetic simulation model includes an air layer, an asphalt concrete layer and a base layer from top to bottom. Based on the constructed three-dimensional multiphase random medium electromagnetic simulation model, the propagation process of radar waves in the base layer under different moisture contents was simulated, and simulated radar reflection wave data under different moisture contents were obtained. Simulated radar reflection wave data under different moisture contents are converted into A-scan waveforms under different moisture contents. The mean, standard deviation, and peak value of reflected waves are extracted from A-scan waveforms under different moisture contents. These mean, standard deviation, and peak value are used as characterization indicators of reflected wave intensity. Using the dielectric constant and reflected wave intensity characteristic parameters corresponding to different moisture contents as sample data, a mapping relationship between reflected wave intensity, dielectric constant, and moisture content is established through multiple regression analysis, thereby constructing a three-variable regression model of reflected wave intensity-dielectric constant-moisture content. Measured radar reflection wave data are obtained. Based on line tracking theory, the dielectric constant of the measured data is inverted by the measured two-way travel time of the reflected wave and the layer thickness. The dielectric constant of the measured data is substituted into the three-variable regression model of reflected wave intensity-dielectric constant-water content to obtain the measured water content data at the corresponding measuring point. A fully random field model is established based on the covariance matrix decomposition method. In the fully random field model, the water content parameter at any point in space is regarded as a random variable that follows a preset probability distribution. The correlation between different spatial locations is described by the autocorrelation function and the autocorrelation distance, thereby characterizing the spatial variability of the water content parameter. Using the measured moisture content data as known constraints, the Kriging interpolation method is used to constrain and calibrate the distribution of the completely random field model, so that the completely random field model matches the measured moisture content data while maintaining consistent statistical laws, thereby obtaining a visual distribution map of moisture content across the entire grassroots area.

2. The method for detecting the overall moisture content of highway base course based on three-dimensional ground-penetrating radar according to claim 1, characterized in that: The aforementioned construction of a three-dimensional multiphase random medium electromagnetic simulation model by considering aggregate gradation and pore structure specifically includes: A pavement structure layer model is established from top to bottom, including an air layer, an asphalt concrete layer, and a base course. The air layer is used to simulate the positional relationship between the radar antenna and the asphalt concrete layer. The asphalt concrete layer is characterized using a random medium model. The base course is constructed by reconstructing the three-dimensional geometry of the aggregate particles and randomly placing the reconstructed aggregate particles according to the aggregate gradation range to form an aggregate phase. A vertical force is applied to compact the aggregate phase, causing the aggregate particles to form irregularly distributed unfilled spaces during the contact and rearrangement process. These unfilled spaces serve as a porous phase and together with the aggregate phase, constitute the multiphase random porous structure of the base course. Determine the radar antenna frequency; The model mesh size is determined based on the minimum wavelength of radar wave propagation, the dimensions of the road structure layer model, and the pixels of the base layer profile. The sampling window is determined based on the propagation time of radar waves within the road surface structure layer model; Determine the dielectric constants of different structural layers in the pavement structural layer model; Based on the established road surface structure layer model, the determined radar antenna frequency, model mesh size, sampling time window, and the dielectric constants corresponding to different structural layers, a three-dimensional multiphase random medium electromagnetic simulation model is constructed.

3. The method for detecting the overall moisture content of highway base course based on three-dimensional ground-penetrating radar according to claim 2, characterized in that: Different moisture content conditions are achieved by adjusting the air-to-water ratio in the aggregate phase pores of the base layer.

4. The method for detecting the overall moisture content of highway base course based on three-dimensional ground-penetrating radar according to claim 1, characterized in that: The specific operation of converting simulated radar reflection wave data under different moisture contents into A-scan waveforms under different moisture contents includes: Simulated radar reflection wave data under different moisture content conditions are input into the denoising model to obtain denoised simulated radar reflection wave data. The denoised simulated radar reflection wave data is converted into A-scan waveforms; The denoising model consists of a first SCB module, a self-attention deep autoencoder, and a second SCB module connected in sequence. In the first SCB module, the input simulated radar reflection wave data is first processed by a 1×1 convolutional layer for dimensionality reduction. The tensor output by the convolution is uniformly divided into two groups, X1 and X2, and then processed by SwinTransformer and residual convolution to obtain outputs Y1 and Y2 respectively. The outputs Y1 and Y2 are concatenated together and further processed by 1×1 convolution. Finally, residual addition is performed with the input simulated radar reflection wave data to obtain the feature tensor Z. In the self-attention deep autoencoder, the feature tensor output by the first SCB module is denoised and restored through a self-attention enhanced encoder-decoder structure. The second SCB module is the inverse operation of the first SCB module, and the second SCB module outputs the denoised simulated radar reflection wave data.

5. The method for detecting the overall moisture content of highway base course based on three-dimensional ground-penetrating radar according to claim 1, characterized in that: The specific operations for extracting the mean, standard deviation, and peak value of reflected waves from A-scan waveforms under different moisture content conditions include: The A-scan waveforms under different moisture contents were decomposed by EMD to obtain the components, including: basic mode components and residual components. Perform Hilbert transform on each component to obtain the corresponding Hilbert spectrum; Hilbert spectra of different components were synthesized to obtain Hilbert spectra of A-scan waveforms under different moisture contents; Based on the Hilbert spectrum of the obtained A-scan waveform, statistical analysis is performed on the amplitude of the Hilbert spectrum within a preset time window corresponding to the base layer reflection interface. The mean, standard deviation, and peak value of the amplitude within the time window are calculated respectively. These are used to characterize the overall energy level, amplitude dispersion, and distribution morphology of the reflected wave, thereby obtaining the mean, standard deviation, and peak value of the reflected wave under different moisture content conditions.

6. The method for detecting the overall moisture content of highway base course based on three-dimensional ground-penetrating radar according to claim 1, characterized in that: The acquisition of measured radar reflected wave data, based on line tracking theory, involves inverting the dielectric constant through the measured two-way travel time of the reflected wave and the layer thickness. Specific operations include: According to the following formula, based on line tracing theory, the dielectric constant is inverted by measuring the two-way travel time of the reflected wave and the layer thickness: in, Let be the thickness of the i-th layer. For the two-way travel time of this layer, For the speed of transmission, At the speed of light, Let be the dielectric constant of the i-th layer.

7. The method for detecting the overall moisture content of highway base course based on three-dimensional ground-penetrating radar according to claim 1, characterized in that: The method for establishing a completely random field model based on covariance matrix factorization specifically includes: A spatial discrete model of the grassroots test section is constructed, the grassroots test section is divided into several spatial discrete units for random field modeling, and the spatial coordinates of the center point of each spatial discrete unit are output. The mean of the moisture content parameter follows a log-normal distribution. and standard deviation Convert to the mean of the corresponding normal distribution and standard deviation The calculation formula is as follows: The coordinates of the center points of each model unit are substituted into the exponential autocorrelation function to calculate the covariance matrix. The covariance matrix Perform Cholesky decomposition This yields the lower triangular matrix L; The exponential autocorrelation function is: ； In the formula, l is the autocorrelation distance, and h is the relative distance between two sample points; If a column vector S is formed by randomly generating n independent random samples that follow a standard normal distribution, then the standard normal random field P can be expressed as: ; A mathematical transformation is performed on the standard normal random field P to obtain a field with a mean of . The sum and variance are Normally distributed random fields: ; Taking the logarithm of a normally distributed random field yields a completely random field model.

8. The method for detecting the overall moisture content of highway base course based on three-dimensional ground-penetrating radar according to claim 1, characterized in that: The Kriging interpolation is calculated using the following formula: in, This represents the value of the variation function between positions i and j. The weighting coefficients are assigned to the moisture content sample data at position i; m is the number of known moisture content sample points. The value of the variation function between position i and the predicted point. For Lagrange daily numbers; for The actual measured moisture content at the location; For prediction points The estimated value.

9. A method for detecting the overall moisture content of highway base course based on three-dimensional ground-penetrating radar according to claim 1, characterized in that: The base course is a cement-stabilized crushed stone base course; the cement-stabilized crushed stone base course is a semi-rigid road base course structure formed by mixing, spreading and compacting graded crushed stone as the main aggregate and cement as the binding material.

Images