CN-121995337-A - Ground Penetrating Radar (GPR) inversion method based on time-frequency combined physical constraint

CN121995337ACN 121995337 ACN121995337 ACN 121995337ACN-121995337-A

Abstract

The invention discloses a Ground Penetrating Radar (GPR) inversion method based on time-frequency combined physical constraint, and relates to the technical field of ground penetrating radar detection. The method comprises the steps of firstly obtaining GPR common offset B-scan echo data, constructing a data set through anti-noise processing and forward modeling, decomposing the data through two-dimensional discrete wavelet transformation, outputting a fusion feature map through a time-frequency joint feature fusion module, introducing a Maxwell equation set, an electromagnetic wave propagation equation and an electromagnetic field boundary equation to construct multiple constraints, and finally outputting a relative dielectric constant distribution map for detection through visualization and accuracy verification. The method improves noise immunity and small target recognition accuracy through time-frequency depth fusion, guarantees result rationality through physical constraint, relies on DLFUNet network optimization efficiency, is suitable for scenes such as underground pipe network distinction and roadbed disease diagnosis, and solves the problems of insufficient accuracy and poor physical interpretability of the traditional method under low signal-to-noise ratio.

Inventors

SHENG GUANQUN
Hou Zhensong
TAN YUNZHI
TANG JING

Assignees

三峡大学

Dates

Publication Date: 20260508
Application Date: 20260121

Claims (10)

1. The Ground Penetrating Radar (GPR) inversion method based on the time-frequency combined physical constraint is characterized by comprising the following steps of: S1, acquiring B-scan echo data in a GPR common offset detection mode, constructing a forward simulation data set by using simulation software based on a time domain finite difference method, and performing noise resistance and robustness treatment on the data; S2, decomposing the preprocessed time domain B-scan data by adopting two-dimensional discrete wavelet transformation to obtain a high-frequency component and a low-frequency component, and carrying out depth fusion on the time domain space characteristics, the high-frequency component and the low-frequency component by a time-frequency joint characteristic fusion module to output a time-frequency fusion characteristic diagram; S3, introducing a Maxwell equation set as physical information constraint, and constructing a data driving inversion basis; s4, constructing a physical residual error loss function based on an electromagnetic wave propagation equation, and carrying out physical rule constraint on an inversion result; S5, constructing a boundary loss function based on an electromagnetic field boundary equation, and optimizing the boundary definition and the structure fidelity of an inversion result; and S6, outputting a relative dielectric constant distribution diagram corresponding to the spatial size of the input B-scan data, and performing visualization processing and accuracy verification to detect the underground target.
2. The Ground Penetrating Radar (GPR) inversion method based on time-frequency combined physical constraint of claim 1, wherein the anti-noise and robust processing in S1 is specifically that Gaussian white noise with signal-to-noise ratio of-5 dB to 5dB is introduced into a part of B-scan sample, and 1% to 15% of dielectric constant random disturbance is superimposed to simulate the environment electromagnetic interference and medium non-uniformity in real detection.
3. The method of claim 1, wherein the high frequency component in S2 is used to extract edge, texture and small target reflection signals in the image, and the low frequency component is used to extract smooth background and geological hierarchy information in the image.
4. The Ground Penetrating Radar (GPR) inversion method based on time-frequency joint physical constraints is characterized in that the processing flow of the time-frequency joint feature fusion module comprises window segmentation, cross-domain attention operation, definition of Query as time domain space features, key as high-frequency components and Value as low-frequency components, realization of weighted fusion of high-frequency information and low-frequency information guided by the space features, optimization of spatial relationship, addition of relative position bias items in attention score calculation, promotion of feature flow of adjacent windows by using a shift window mechanism, feature restoration and output, reverse splicing of window output results, and output of the time-frequency fusion feature map after convolution layer channel dimension reduction and residual structure fusion.
5. The Ground Penetrating Radar (GPR) inversion method based on time-frequency joint physical constraint according to claim 1, wherein maxwell' S equations in S3 are specifically: ; ; Wherein the method comprises the steps of Is a rotation operator, H is the magnetic field intensity (unit: A/m), The dielectric constant, E is the electric field strength (unit: V/m), t is time, J is current density, Is magnetic permeability.
6. The method of Ground Penetrating Radar (GPR) inversion based on time-frequency joint physical constraints of claim 1, wherein the physical residual loss function expression in S4 is: ; Wherein the method comprises the steps of In order to be of electrical conductivity, And for the Laplace operator, the residual error of the equation is calculated to be used as a physical loss term, so that the inversion result is ensured to accord with the electromagnetic wave propagation rule.
7. The method of Ground Penetrating Radar (GPR) inversion based on time-frequency joint physical constraints of claim 1, wherein the electromagnetic field boundary equation in S5 comprises the following electric field tangential component continuous equation: Magnetic field tangential component equation: And (3) an electric displacement vector normal component equation: equation of continuity of normal component of magnetic induction intensity Wherein , E 1 t、E 2 t is the tangential component of the electric field at the two sides of the interface, H 1 t、H 2 t is the tangential component of the magnetic field at the two sides of the interface, J s is the free surface current density (unit: A/m), D 1n 、D 2n is the normal component of the electric displacement vector at the two sides of the interface, Is the free surface charge density (unit: C/m 2 ）,B 1n 、B 2n is the normal component of magnetic induction intensity at two sides of the interface).
8. The method for Ground Penetrating Radar (GPR) inversion based on time-frequency joint physical constraint according to claim 1, wherein the accuracy verification in S6 adopts Root Mean Square Error (RMSE) and Structural Similarity Index (SSIM) as quantization indexes, and inversion results are subjected to morphological filtering and edge sharpening processing, so that background noise residues are eliminated, and object boundary definition is enhanced.
9. The Ground Penetrating Radar (GPR) inversion method based on time-frequency joint physical constraints according to claim 1, wherein the DLFUNet network comprises an encoding module, a decoding module, a DBottleneck module and an output module, and the DBottleneck module is constructed by adopting convolution layers with expansion rates of 1, 6, 12 and 18 respectively, so that multi-scale feature extraction is achieved.
10. The method for inversion of Ground Penetrating Radar (GPR) based on time-frequency combined physical constraint according to claim 1, wherein the application scene of the method comprises quantitative diagnosis of underground pipe network area and road subgrade faults, wherein the underground pipe network area is used for identifying metal pipelines and nonmetal pipelines, and the quantitative diagnosis of faults is used for early identifying road subgrade void, water-containing subsidence areas and performing volume estimation.

Description

Ground Penetrating Radar (GPR) inversion method based on time-frequency combined physical constraint Technical Field The invention relates to the technical field of ground penetrating radar detection, in particular to a Ground Penetrating Radar (GPR) inversion method based on time-frequency combined physical constraint. Background Ground Penetrating Radar (GPR) inversion is a core nondestructive detection technology in the field of modern engineering geology and municipal maintenance, and by transmitting and receiving electromagnetic waves, the three-dimensional structure of the underground space can be reconstructed, and physical parameters such as dielectric constants and the like can be accurately obtained, so that the Ground Penetrating Radar (GPR) inversion plays a key role in avoiding construction risks, checking underground hidden danger, combing municipal pipe network and other scenes. However, the existing GPR inversion technology still faces the problems of insufficient inversion precision and poor physical interpretability under the condition of low signal-to-noise ratio in practical application: The inversion accuracy is insufficient at low signal-to-noise ratios. The inhomogeneity of the underground medium causes severe scattering and attenuation in the electromagnetic wave propagation process, and the effective signals are very weak and are easily covered by environmental noise and ground clutter. The traditional inversion method or the time-frequency combined method relying on Fast Fourier Transform (FFT) has the limitation that the resolution of the time domain and the frequency domain are difficult to be compatible, and the target signal in the noise cannot be effectively extracted, so that the inversion accuracy of a small-size target is low, the noise immunity is weak, and the missed judgment or the false judgment is easy to occur. The physical interpretability is poor. The existing GPR inversion mostly adopts a pure data driving mode, and lacks effective constraint on the electromagnetic wave propagation physical rule. Although the method can utilize deep learning to process mass data, the inversion result is often deviated from a real electromagnetic propagation rule, so that the physical rationality of the inversion result is difficult to explain, and the application of the technology in engineering scenes with higher requirements on the reliability of the detection result is limited. Disclosure of Invention Aiming at the problems of insufficient precision and poor physical interpretability of the existing GPR inversion technology under low signal-to-noise ratio, the invention provides the GPR inversion method based on time-frequency combined physical constraint, the noise immunity and the precision are improved through time-frequency depth fusion, and the physical rationality of a result is ensured by combining the physical constraint and the boundary constraint, so that the efficient and accurate underground target detection is realized. In order to solve the problems, the technical scheme of the invention is as follows: A Ground Penetrating Radar (GPR) inversion method based on time-frequency joint physical constraint comprises the following steps: S1, acquiring B-scan echo data in a GPR common offset detection mode, constructing a forward simulation data set by using simulation software based on a time domain finite difference method, and performing noise resistance and robustness treatment on the data; S2, decomposing the preprocessed time domain B-scan data by adopting two-dimensional discrete wavelet transformation to obtain a high-frequency component and a low-frequency component, and carrying out depth fusion on the time domain space characteristics, the high-frequency component and the low-frequency component by a time-frequency joint characteristic fusion module to output a time-frequency fusion characteristic diagram; S3, introducing a Maxwell equation set as physical information constraint, and constructing a data driving inversion basis; s4, constructing a physical residual error loss function based on an electromagnetic wave propagation equation, and carrying out physical rule constraint on an inversion result; S5, constructing a boundary loss function based on an electromagnetic field boundary equation, and optimizing the boundary definition and the structure fidelity of an inversion result; and S6, outputting a relative dielectric constant distribution diagram corresponding to the spatial size of the input B-scan data, and performing visualization processing and accuracy verification to detect the underground target. Further, the noise immunity and robustness processing in S1 specifically comprises the steps of introducing Gaussian white noise with a signal-to-noise ratio of-5 dB to 5dB into a part of B-scan sample, and superposing 1% to 15% of dielectric constant random disturbance to simulate the environment electromagnetic interference and medium non-uniformity in r