CN-121997278-A - Ground penetrating radar data generation method, device, equipment and medium based on road material constraint antagonistic neural network

CN121997278ACN 121997278 ACN121997278 ACN 121997278ACN-121997278-A

Abstract

The application discloses a method, a device, equipment and a medium for generating ground penetrating radar data based on a road material constraint antagonistic neural network, which relate to the technical field of road detection and comprise the steps of acquiring ground penetrating radar B-SCAN original data and road material parameters of the same road detection area, respectively standardizing the data, splicing channels to obtain multi-mode characteristics, and fusing noise and geological constraint construction combined input characteristics; and generating single-channel data step by step through a three-level layered generator, constructing a generated sample and a real sample, performing countermeasure training through a two-dimensional discriminator until convergence to obtain a target generator, and inputting data to be detected to output target ground penetrating radar B-SCAN data. The method realizes multi-mode geological constraint and layering generation, has more real and reasonable data, effectively solves the problems of insufficient samples and distortion generation, provides richer ground penetrating radar B-SCAN data for a road damage area, and is beneficial to improving the intelligent detection level of the road internal damage.

Inventors

YU JING
XIONG HAO
LV SONGTAO
YANG WENAN
LU WEIWEI
WEI HUI
TANG SHU
Li Yuanqu

Assignees

长沙理工大学

Dates

Publication Date: 20260508
Application Date: 20260408

Claims (10)

1. A method for generating ground penetrating radar data based on road material constraint antagonism neural network, the method comprising: Acquiring the original data of a ground penetrating radar B-SCAN and road material parameters of the same road detection area; Respectively carrying out standardized preprocessing on the ground penetrating radar B-SCAN original data and the road material parameters to obtain standardized radar data and standardized road material parameters; channel splicing is carried out on the standardized radar data and the standardized road material parameters to obtain multi-mode characteristics; constructing a joint input feature based on the multi-modal feature fusion random noise vector and the geological constraint vector; Inputting the combined input features into a three-level layered generator, and generating ground penetrating radar B-SCAN single-channel data step by step; Splicing the ground penetrating radar B-SCAN single-channel data with the road material parameters of the corresponding batch to form a generated sample, and splicing the ground penetrating radar B-SCAN data of the real road detection area with the real road material parameters of the corresponding batch to form a real sample; inputting the generated sample and the real sample into a two-dimensional discriminator for countermeasures discrimination, alternately and iteratively training the three-level layered generator and the two-dimensional discriminator until a model meets a preset convergence condition, and outputting the current three-level layered generator as a target three-level layered generator; And inputting the ground penetrating radar B-SCAN original data and the road material parameters of the road detection area to be detected into the target three-level layered generator to obtain target ground penetrating radar B-SCAN data.
2. The method for generating ground penetrating radar data based on road material constraint antagonism neural network according to claim 1, wherein the step of performing standardized preprocessing on the ground penetrating radar B-SCAN raw data and the road material parameters, respectively, to obtain standardized radar data and standardized road material parameters comprises: Converting discrete parameters in the road material parameters into numerical characteristics by adopting single-heat coding to obtain coded discrete parameters; Mapping continuous parameters in the road material parameters to a preset numerical interval through minimum and maximum normalization to obtain normalized continuous parameters; Fusing the coded discrete parameters and the normalized continuous parameters to obtain standardized road material parameters; performing noise elimination operation on the ground penetrating radar B-SCAN original data to obtain denoised radar data; And carrying out amplitude normalization mapping on the denoised radar data to a preset signal interval and completing phase calibration to obtain standardized radar data.
3. The method for generating ground penetrating radar data based on road material constraint antagonistic neural network according to claim 1, wherein the step of performing channel stitching on the standardized radar data and the standardized road material parameters to obtain multi-modal characteristics comprises: calculating the sensitivity coefficient between the ground penetrating radar signal and each road material parameter through the pearson correlation analysis; Generating weight coefficients corresponding to the road material parameter channels according to the sensitivity coefficients; Weighting and enhancing the standardized road material parameters based on the weight coefficients to obtain enhanced road material parameters; Aligning the characteristics of the channel dimension of the enhanced road material parameters and the standardized radar data to obtain aligned characteristics; performing channel splicing operation on the aligned features to obtain initial multi-mode features; And performing feature dimension verification and correction on the initial multi-modal feature to obtain the multi-modal feature.
4. The method for generating ground penetrating radar data based on road material constraint antagonistic neural network according to claim 1, wherein the step of constructing joint input features based on the multi-modal feature fusion random noise vector and geological constraint vector comprises: Randomly sampling random noise vectors subjected to normal distribution according to the size of a preset batch, and extracting geological constraint vectors of corresponding batches from a road material parameter set; carrying out weighted fusion processing on the random noise vector and the geological constraint vector to obtain a noise-constraint fusion vector; Performing feature dimension fusion on the noise-constraint fusion vector and the multi-mode features to obtain a fusion feature vector; Normalizing the fusion feature vector to obtain normalized fusion features; and mapping the normalized fusion features to the input dimension of the three-level hierarchical generator to obtain the joint input features.
5. The method for generating ground penetrating radar data based on road material constraint antagonistic neural network according to claim 1, wherein the step of inputting the joint input feature into a three-stage layered generator to generate ground penetrating radar B-SCAN single channel data step by step comprises: Inputting the combined input features into a geological framework layer of a three-level layered generator to generate a macroscopic geological framework feature map with preset low resolution; inputting the macroscopic geological framework feature map into a target contour layer of a three-level layered generator to generate a target contour feature map with preset medium resolution; inputting the target contour feature map into a signal detail layer of a three-level layered generator to generate a preset high-resolution signal detail feature map; Performing feature smoothing and dimension compression processing on the signal detail feature map to obtain a single-channel feature matrix; And converting the single-channel feature matrix into a ground penetrating radar B-SCAN data format to obtain the ground penetrating radar B-SCAN single-channel data.
6. The method for generating ground penetrating radar data based on road material constraint antagonistic neural network according to claim 5, wherein the step of inputting the joint input feature into a geological framework layer of a three-level hierarchical generator, generating a macroscopic geological framework feature map with a preset low resolution comprises: Inputting the combined input features into a geological parameter coding module of a geological framework layer, and converting the combined input features into a global constraint feature map; Performing convolution processing on the global constraint feature map by adopting large-kernel convolution with a preset kernel size to obtain a convolution feature map; Up-sampling the convolution feature map through a multi-layer transposition convolution, batch normalization and activation function to obtain an up-sampled feature map; mapping the up-sampled feature map to a preset low resolution to obtain an initial geological framework feature map; Carrying out road material constraint suitability verification on the initial geological frame feature map to obtain a macroscopic geological frame feature map; the step of inputting the macroscopic geological framework feature map into a target contour layer of a three-level hierarchical generator and generating a target contour feature map with preset medium resolution comprises the following steps: performing up-sampling treatment on the macroscopic geologic frame feature map through transposed convolution to obtain a middle resolution geologic feature map with preset middle resolution; Performing a target candidate region extraction operation on the medium resolution geologic feature map based on the standardized road material parameters, and outputting a target candidate mask; performing reinforcement learning processing on the features in the target candidate mask through a boundary constraint convolution module of the target contour layer to obtain a reinforcement feature map; Performing geological suitability verification operation of road material constraint on the reinforced feature map to obtain a verified medium-resolution feature map; Performing feature fusion and smoothing on the verified medium-resolution feature map to generate a target contour feature map with preset medium resolution; The step of inputting the target contour feature map into a signal detail layer of a three-level layered generator and generating a preset high-resolution signal detail feature map comprises the following steps: upsampling the target contour feature map to a preset high resolution through transposed convolution to obtain a high resolution contour feature map; Performing radar echo intensity simulation on the high-resolution profile feature map based on a time domain finite difference method model to generate a signal amplitude feature map; generating a self-adaptive clutter signal according to the noise statistical distribution of the real ground penetrating radar data to obtain adaptive clutter characteristics; the adaptive clutter features are added to the signal amplitude feature map to obtain a clutter-containing signal feature map; and carrying out phase and amplitude correction on the clutter-containing signal characteristic diagram to obtain a preset high-resolution signal detail characteristic diagram.
7. The method for generating ground penetrating radar data based on road material constraint antagonism neural network according to claim 1, wherein the step of inputting the generated sample and the real sample into a two-dimensional discriminator to perform antagonism discrimination, alternately and iteratively training the three-stage layered generator and the two-dimensional discriminator until the model satisfies a preset convergence condition, and outputting the current three-stage layered generator as a target three-stage layered generator comprises: Inputting the generated sample and the real sample into a two-dimensional discriminator, and respectively extracting two-dimensional characteristics of a geological structure and radar signals to obtain a generated sample characteristic set and a real sample characteristic set; Comparing and evaluating the generated sample feature set and the real sample feature set through a two-dimensional discriminator, outputting a real discrimination score and a real sample discrimination score, and respectively calculating the antagonism loss function values of a three-level layered generator and the two-dimensional discriminator; Updating network parameters of the three-level layered generator and the two-dimensional discriminator according to the counter-loss function value through a counter-propagation algorithm to obtain an optimized generator; When the fluctuation amplitude of the counter loss function value in the continuous preset iteration batch is smaller than a preset threshold value and the authenticity discrimination score of the generated sample exceeds the preset proportion of the authenticity discrimination score of the real sample, outputting the current optimization generator as a target three-level layering generator.
8. A ground penetrating radar data generating device based on road material constraint antagonism neural network, the device comprising: The data acquisition module is used for acquiring the original data of the ground penetrating radar B-SCAN and the road material parameters of the same road detection area; the standardized preprocessing module is used for respectively executing standardized preprocessing on the ground penetrating radar B-SCAN original data and the road material parameters to obtain standardized radar data and standardized road material parameters; The multi-mode splicing module is used for carrying out channel splicing on the standardized radar data and the standardized road material parameters to obtain multi-mode characteristics; The joint input construction module is used for constructing joint input features based on the multi-modal feature fusion random noise vector and the geological constraint vector; the layering generation module is used for inputting the combined input characteristics into a three-level layering generator and generating ground penetrating radar B-SCAN single-channel data step by step; The sample construction module is used for splicing the ground penetrating radar B-SCAN single-channel data with the road material parameters of the corresponding batch to form a generated sample, and splicing the ground penetrating radar B-SCAN data of the real road detection area with the real road material parameters of the corresponding batch to form a real sample; The countermeasure training module is used for inputting the generated sample and the real sample into a two-dimensional discriminator to perform countermeasure discrimination, alternately and iteratively training the three-level layered generator and the two-dimensional discriminator until the model meets the preset convergence condition, and outputting the current three-level layered generator as a target three-level layered generator; the data generation module is used for inputting the ground penetrating radar B-SCAN original data of the road detection area to be detected and road material parameters into the target three-level layered generator to obtain target ground penetrating radar B-SCAN data.
9. A road material constraint antagonism neural network based ground penetrating radar data generating device, characterized in that the device comprises a memory, a processor and a road material constraint antagonism neural network based ground penetrating radar data generating program stored on the memory and running on the processor, the road material constraint antagonism neural network based ground penetrating radar data generating program being configured to implement the steps of the road material constraint antagonism neural network based ground penetrating radar data generating method as defined in any one of claims 1-7.
10. A storage medium having stored thereon a road material constraint antagonism neural network based ground penetrating radar data generating program which when executed by a processor implements the steps of the road material constraint antagonism neural network based ground penetrating radar data generating method of any one of claims 1 to 7.

Description

Ground penetrating radar data generation method, device, equipment and medium based on road material constraint antagonistic neural network Technical Field The invention relates to the technical field of road detection, in particular to a method, a device, equipment and a medium for generating ground penetrating radar data based on a road material constraint countermeasure neural network. Background At present, the generation and enhancement of the ground penetrating radar B-SCAN data are carried out by adopting a traditional generation countermeasure neural network model in the industry, and the method mainly comprises the steps of firstly carrying out basic preprocessing such as simple amplitude normalization on the collected ground penetrating radar original data, then sending the preprocessed radar data into a generator with a single hierarchical structure as a single input, generating simulated ground penetrating radar data by the generator through random noise and simple feature mapping, then inputting the generated data and real data into a discriminator for true and false discrimination, and enabling the generator to gradually approximate to real data distribution through alternate iteration training of the generator and the discriminator, thus finally realizing the generation of the ground penetrating radar data. Some of the improvements would simply stitch a small number of geologic parameters in the data input, but do not build a targeted constraint mechanism for the road material parameters. In the conventional generation of the anti-neural network model, the hierarchical characteristic characteristics of ground penetrating radar data are not considered in the generation process of the ground penetrating radar B-SCAN data, the adopted single-hierarchy generator structure cannot realize accurate modeling of different hierarchy characteristics of the data, the problems of global geological structure distortion, underground target outline blurring and signal detail deficiency are prone to occur, meanwhile, an effective constraint mechanism is not constructed for road material parameters in the conventional method, the discrete type and continuous type parameters of the road material are not standardized, the constraint effect of the road material on radar signals is not fused in the generation process, the adaptability of the generated data and the actual road geological background is poor, radar echo signal response differences corresponding to different road materials cannot be reduced, in addition, the conventional discriminator only discriminates the whole false of the generated data in a single dimension, does not evaluate the geological logic rationality and the signal physical reality of the data at the same time, the generator is difficult to form effective double supervision, and the stability of model training and the reliability of the generated data are low. Therefore, how to construct a targeted constraint mechanism by combining road material parameters, to realize hierarchical accurate generation of the ground penetrating radar B-SCAN data, and to improve road detection precision and applicability becomes a problem to be solved urgently. Disclosure of Invention The application mainly aims to provide a method, a device, equipment and a medium for generating ground penetrating radar data based on road material constraint antagonism neural network, which aim to solve the technical problem of how to improve the accuracy of the generated ground penetrating radar data. In order to achieve the above object, the present application provides a method for generating ground penetrating radar data based on road material constraint antagonism neural network, comprising: Acquiring the original data of a ground penetrating radar B-SCAN and road material parameters of the same road detection area; Respectively carrying out standardized preprocessing on the ground penetrating radar B-SCAN original data and the road material parameters to obtain standardized radar data and standardized road material parameters; channel splicing is carried out on the standardized radar data and the standardized road material parameters to obtain multi-mode characteristics; constructing a joint input feature based on the multi-modal feature fusion random noise vector and the geological constraint vector; Inputting the combined input features into a three-level layered generator, and generating ground penetrating radar B-SCAN single-channel data step by step; Splicing the ground penetrating radar B-SCAN single-channel data with the road material parameters of the corresponding batch to form a generated sample, and splicing the ground penetrating radar B-SCAN data of the real road detection area with the real road material parameters of the corresponding batch to form a real sample; inputting the generated sample and the real sample into a two-dimensional discriminator for countermeasures discrimination, alternately and it