Tunnel concrete ground penetrating radar forward modeling data automatic generation method and system

Abstract

The invention discloses a tunnel concrete ground penetrating radar forward modeling data automatic generation method and system, and relates to the technical field of radar detection. The problems of lack of data sets, low modeling efficiency and insufficient signal generalization ability in the prior art are solved. According to the method, automatic tunnel concrete internal defect model generation and radar signal forward modeling simulation based on a finite difference time domain (FDTD) method are combined, various concrete structure defect models are automatically constructed through a random parameter generator according to an electromagnetic parameter range input by a user, and visual preview and verification functions are provided; and the reasonability of the model is ensured. According to the method, diversified ground penetrating radar simulation B-scan data sets can be efficiently generated, the method has clutter suppression and envelope enhancement functions at the same time, the signal quality is optimized through the filtering and adaptive gain technology, and the interpretability of images is improved. According to the method, reliable data support is provided for subsequent deep learning model training.

Description

Technical Field

[0001] The present invention relates to the field of radar detection technology, and in particular to a method and system for automatically generating tunnel concrete ground penetrating radar forward modeling data. Background Art

[0002] Ground penetrating radar (GPR) is a non-destructive detection tool widely used in underground target object detection, geological structure assessment, and engineering quality inspection. However, the current signal processing and target detection technology based on GPR has the following main problems: Lack of data sets: Due to the high cost and complex environment of real GPR data acquisition, there is often a lack of sufficient high-quality data sets in the process of training efficient target detection algorithms or evaluating algorithm performance. For example, the patent with announcement number CN116563173B discloses a "GPR image denoising method based on multi-noise self-supervised learning". The patent uses a large number of real GPR data sets for supervised learning, but due to the lack of widespread application of GPR, it is very difficult to obtain real data sets, and the image features used as training sets may not be obvious enough, which affects the denoising effect of the algorithm. At the same time, the performance of deep learning models benefits from more diverse and larger data sets, which further highlights the necessity of data generation. Low modeling efficiency: Traditional GPR simulation model construction (such as GPRMAX modeling) relies on manual design, which is inefficient and difficult to meet the needs of large-scale data generation. Insufficient signal generalization ability: Due to the limitations of existing data sets, the algorithms trained by deep learning have poor generalization ability in complex scenarios and are difficult to adapt to diverse actual detection scenarios. Summary of the invention

[0003] In order to solve the problems of lack of data sets, low modeling efficiency and insufficient signal generalization ability in the prior art, the present invention provides a method and system for automatically generating tunnel concrete ground penetrating radar forward modeling data, comprising:

[0004] Step 1: Based on the parameters input by the user, a random parameter generator is used to automatically generate a model parameter combination according to the constructed parameter space; a GPRMAX model interface is called to automatically generate a ground penetrating radar simulation model using the model parameter combination;

[0005] Step 2: Based on the finite difference time domain method, the ground penetrating radar signal is quickly simulated and generated to output the B-scan image;

[0006] Step 3: Generate and optimize the B-scan image using a generative adversarial network to obtain an optimized B-scan image;

[0007] Step 4: Enhance the optimized B-scan image to obtain tunnel concrete ground penetrating radar forward modeling data.

[0008] Further, in step one, the parameter space includes: the minimum and maximum volume of the target object, the shape of the target object, the dielectric constant, the electrical conductivity, the relative magnetic susceptibility, the magnetic loss and the distribution probability of the target object.

[0009] Further, in step 3, the generative adversarial network includes: two neural networks, a generator and a discriminator;

[0010] The output range of the generator is [-1, 1], which is converted to the range of [0, 255] through denormalization;

[0011] The discriminator calculates the classification results of real and generated images through a binary cross entropy loss function.

[0012] Furthermore, the binary cross entropy loss function is:

[0013]

[0014] Obtain, where N is the total number of samples; y i is the true label of sample i; p i is the sample predicted by the model; log is the natural logarithm.

[0015] Further, in step 4, the enhancement processing includes: clutter suppression, envelope enhancement and normalization processing;

[0016] The clutter suppression is achieved by:

[0017] S(t)=Sraw(t)-Sdirect(t)

[0018] Implementation, where S(t) is the suppressed signal, Sraw(t) is the original signal, and Sdirect(t) is the direct wave signal;

[0019] The envelope enhancement is achieved by:

[0020] S enhanced (t) = G(t)*S smooth (t)

[0021] Implementation, where S enhanced (t) is the enhanced signal, G(t) is the adaptive gain coefficient, S smooth (t) is the envelope signal after smoothing.

[0022] A tunnel concrete ground penetrating radar forward modeling data automatic generation system, comprising:

[0023] An automated GPRMAX model generation module is used to automatically generate a model parameter combination based on the parameters input by the user and using a random parameter generator according to the constructed parameter space; calling the GPRMAX model interface and automatically generating a ground penetrating radar simulation model using the model parameter combination;

[0024] FDTD forward simulation module, used to quickly simulate and generate ground penetrating radar signals based on the finite difference time domain method and output B-scan images;

[0025] A GAN image generation module, used to generate and optimize the B-scan image using a generative adversarial network;

[0026] The signal processing and enhancement module is used to enhance the B-scan image to obtain tunnel concrete ground penetrating radar forward modeling data.

[0027] The beneficial effects of the present invention are as follows:

[0028] (1) Significantly reduce manual modeling time through automated GPR model generation technology;

[0029] (2) Generate a large number of high-quality B-scan images of concrete internal defects based on FDTD forward simulation, covering a variety of scenarios and concrete defect characteristics;

[0030] (3) Generative adversarial networks are used to further expand the dataset and make up for the problem of insufficient data;

[0031] (4) Integrate direct wave suppression and envelope enhancement functions to improve the effectiveness of simulation data;

[0032] (5) The generated dataset is suitable for fields such as tunnel detection and helps to improve the performance of deep learning in the field of radar signal recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A flow chart of a method for automatically generating forward modeling data of tunnel concrete ground penetrating radar;

[0034] Figure 2 Generate images for generative adversarial networks;

[0035] Figure 3 To generate a schematic diagram of the concrete defect model;

[0036] Figure 4 is a schematic diagram of a processed radar simulation image;

[0037] Figure 5 A structural diagram of a system for automatically generating forward modeling data of a tunnel concrete ground penetrating radar.

[0038] In the figure, 1. water cave, 2. bedrock layer, 3. air layer, 4. concrete layer, 5. ice cave, 6. cavity, 7. defect echo, 8. defect echo, 9. time depth, 10. number of radar channels. DETAILED DESCRIPTION

[0039] Example 1: Reference Figure 1 This embodiment describes a method for automatically generating tunnel concrete ground penetrating radar forward modeling data, comprising:

[0040] Step 1: Based on the parameters input by the user, a random parameter generator is used to automatically generate a model parameter combination according to the constructed parameter space; a GPRMAX model interface is called to automatically generate a ground penetrating radar simulation model using the model parameter combination;

[0041] Step 2: Based on the finite difference time domain method, the ground penetrating radar signal is quickly simulated and generated to output the B-scan image;

[0042] Step 3: Generate and optimize the B-scan image using a generative adversarial network to obtain an optimized B-scan image;

[0043] Step 4: Enhance the optimized B-scan image to obtain tunnel concrete ground penetrating radar forward modeling data.

[0044] In step 1, the parameter space includes: the minimum and maximum volume of the target object, the shape of the target object, the dielectric constant, the electrical conductivity, the relative magnetic susceptibility, the magnetic loss and the distribution probability of the target object.

[0045] Specifically, the generation rule of the random parameter generator is: randomly extract the size and position distribution of the target object from the parameter space to construct the concrete internal simulation model according to the set probability distribution.

[0046] The function of the generative adversarial network is to make the initially generated B-scan image closer to the real ground penetrating radar data;

[0047] The finite-difference time-domain method is implemented by:

[0048]

[0049] Implementation, where E z is the electric field strength; H x , H y is the magnetic field intensity; μ, ε are the magnetic permeability and dielectric constant of the medium respectively, and t represents the time step.

[0050] The optimized B-scan image is as follows Figure 2 As shown; the concrete defect image generated by the ground penetrating radar simulation model is as follows Figure 3 As shown;

[0051] The enhanced B-scan image is as follows Figure 4 As shown. ;

[0052] In step 3, the generative adversarial network includes two neural networks: a generator and a discriminator;

[0053] The output range of the generator is [-1, 1], which is converted to the range of [0, 255] through denormalization;

[0054] The discriminator calculates the classification results of real and generated images through a binary cross entropy loss function.

[0055] Specifically, in the generative adversarial network, the generator receives a random noise vector and outputs an image with a distribution similar to that of a real B-scan image. The discriminator is trained to distinguish between real images and generated images. The loss function is optimized for the discriminator and the generator respectively, and the two are updated alternately. The loss function is printed every 500 iterations to observe whether the model gradually converges. The final B-scan image is generated with random noise.

[0056] The binary cross entropy loss function is:

[0057]

[0058] Obtained, where N is the total number of samples; y i is the true label of sample i; p i is the sample predicted by the model; log is the natural logarithm.

[0059] In step 4, the enhancement processing includes: clutter suppression, envelope enhancement and normalization processing;

[0060] The clutter suppression is achieved by:

[0061] S(t)=Sraw(t)-Sdirect(t)

[0062] Implementation, where S(t) is the suppressed signal, Sraw(t) is the original signal, and Sdirect(t) is the direct wave signal;

[0063] The envelope enhancement is achieved by:

[0064] S enhanced (t) = G(t)*S smppth (t)

[0065] Implementation, where S enhanced (t) is the enhanced signal, G(t) is the adaptive gain coefficient, S smooth (t) is the envelope signal after smoothing.

[0066] Specifically, the envelope enhancement adopts adaptive gain control and envelope detection technology, and dynamically compensates for target reflection signals at different depths by adjusting the signal gain in real time to compensate for the characteristics of strong shallow signals and weak deep signals. The window function is used to calculate the mean and variance of the local signal amplitude, and the gain coefficient is adjusted in real time according to the change of the signal. The signal envelope is extracted using the Hilbert transform to eliminate high-frequency oscillation components and highlight the outline of the reflecting target. The signal is smoothed in combination with envelope detection to suppress random noise and improve the smoothness and continuity of the envelope curve.

[0067] In the envelope enhancement method, the envelope is obtained by calculating the Hilbert transform of the signal:

[0068]

[0069] Wherein, H(S(t)) represents the Hilbert transform of S(t), and Senv(t) represents the envelope.

[0070] The gain is achieved by the following adaptive gain control formula:

[0071]

[0072] Where G(t) is the gain coefficient, α is the gain adjustment factor, and μ(t) is the average value of the local signal.

[0073] The envelope is enhanced after smoothing by the following formula:

[0074]

[0075] Among them, S smooth (t) is the smoothed envelope signal, N is the size of the smoothing window, and i is the index within the window.

[0076] Embodiment 2, combined Figure 5 This embodiment describes a system for automatically generating tunnel concrete ground penetrating radar forward modeling data, comprising:

[0077] An automated GPRMAX model generation module is used to automatically generate a model parameter combination based on the parameters input by the user and using a random parameter generator according to the constructed parameter space; calling the GPRMAX model interface and automatically generating a ground penetrating radar simulation model using the model parameter combination;

[0078] FDTD forward simulation module, used to quickly simulate and generate ground penetrating radar signals based on the finite difference time domain method and output B-scan images;

[0079] A GAN image generation module, used to generate and optimize the B-scan image using a generative adversarial network;

[0080] The signal processing and enhancement module is used to enhance the B-scan image to obtain tunnel concrete ground penetrating radar forward modeling data.

Claims

1. A method for automatically generating tunnel concrete ground penetrating radar forward modeling data, characterized in that: include: Step 1: Based on the parameters input by the user, a random parameter generator is used to automatically generate a model parameter combination according to the constructed parameter space; a GPRMAX model interface is called to automatically generate a ground penetrating radar simulation model using the model parameter combination; Step 2: Based on the finite difference time domain method, the ground penetrating radar signal is quickly simulated and generated to output the B-scan image; Step 3: Generate and optimize the B-scan image using a generative adversarial network to obtain an optimized B-scan image; Step 4: Enhance the optimized B-scan image to obtain tunnel concrete ground penetrating radar forward modeling data.

2. The method for automatically generating tunnel concrete ground penetrating radar forward modeling data according to claim 1 is characterized in that: In step 1, the parameter space includes: the minimum and maximum volume of the target object, the shape of the target object, the dielectric constant, the electrical conductivity, the relative magnetic susceptibility, the magnetic loss and the distribution probability of the target object.

3. The method for automatically generating tunnel concrete ground penetrating radar forward modeling data according to claim 1 is characterized in that: In step 3, the generative adversarial network includes two neural networks: a generator and a discriminator; The output range of the generator is [-1, 1], which is converted to the range of [0, 255] through denormalization; The discriminator calculates the classification results of real and generated images through a binary cross entropy loss function.

4. The method for automatically generating tunnel concrete ground penetrating radar forward modeling data according to claim 1, characterized in that: The binary cross entropy loss function is: Obtained, where N is the total number of samples; y i is the true label of sample i; p i is the sample predicted by the model; log is the natural logarithm.

5. The method for automatically generating tunnel concrete ground penetrating radar forward modeling data according to claim 1, characterized in that: In step 4, the enhancement processing includes: clutter suppression, envelope enhancement and normalization processing; The clutter suppression is achieved by: S(t)=Sraw(t)-Sdirect(t) Implementation, where S(t) is the suppressed signal, Sraw(t) is the original signal, and Sdirect(t) is the direct wave signal; The envelope enhancement is achieved by: S enhanced (t)=G(t)*S smooth (t) Implementation, where S enhanced (t) is the enhanced signal, G(t) is the adaptive gain coefficient, S smooth (t) is the envelope signal after smoothing.

6. A tunnel concrete ground penetrating radar forward modeling data automatic generation system, characterized in that: include: An automated GPRMAX model generation module is used to automatically generate a model parameter combination based on the parameters input by the user and using a random parameter generator according to the constructed parameter space; calling the GPRMAX model interface and automatically generating a ground penetrating radar simulation model using the model parameter combination; FDTD forward simulation module, used to quickly simulate and generate ground penetrating radar signals based on the finite difference time domain method and output B-scan images; A GAN image generation module, used to generate and optimize the B-scan image using a generative adversarial network; The signal processing and enhancement module is used to enhance the B-scan image to obtain tunnel concrete ground penetrating radar forward modeling data.

Images