A method, system, equipment, and medium for processing ground-penetrating radar data in water conveyance tunnels.

Abstract

This application discloses a method, system, device, and medium for processing ground-penetrating radar (GPR) data in a water conveyance tunnel. The method includes: constructing a simulated scenario for water conveyance tunnel detection and establishing an interference suppression network; simulating structural defects and interference factors in the simulated scenario to obtain a training dataset; training the interference suppression network based on the training dataset to obtain a trained target network; acquiring initial GPR data, inputting the initial GPR data into the target network for interference suppression processing, and outputting the target GPR data. The embodiments of this application can eliminate interference in the data and improve data quality, and can be widely applied in the field of signal processing technology.

Description

Technical Field

[0001] This application relates to the field of signal processing technology, and in particular to a method, system, equipment and medium for processing ground-penetrating radar data in a water conveyance tunnel. Background Technology

[0002] In related technologies, as water conveyance tunnels age, internal structural defects such as voids, cracks, and cavitation appear. These defects can affect the tunnel's usability and even threaten its structural safety. Therefore, ground-penetrating radar (GPR) is used to inspect the structure of water conveyance tunnels, enabling timely detection and repair of these defects. However, various interferences frequently occur when using GPR to inspect water conveyance tunnels, affecting the effectiveness of the GPR data and interfering with the echoes from defects within the tunnel, making it difficult to accurately locate these defects. In summary, the technical problems existing in these technologies need to be improved. Summary of the Invention

[0003] The main objective of this application is to propose a method, system, device, and medium for processing ground-penetrating radar data in water conveyance tunnels, which can improve data quality.

[0004] To achieve the above objectives, one aspect of this application proposes a method for processing ground-penetrating radar data in a water conveyance tunnel, the method comprising:

[0005] A simulation scenario for water conveyance tunnel detection was constructed, and an interference suppression network was established;

[0006] The simulation of the water conveyance tunnel detection scenario is processed to simulate structural defects and interference factors, and a training dataset is obtained.

[0007] The interference suppression network is trained using the training dataset to obtain the trained target network.

[0008] Initial ground-penetrating radar (GPR) data is acquired, and the initial GPR data is input into the target network for interference suppression processing. The target GPR data is then output.

[0009] In some embodiments, establishing the interference suppression network includes the following steps:

[0010] A shallow feature extraction structure is constructed based on the first convolutional module;

[0011] The encoder-decoder structure is constructed based on the preprocessing module, the sparse window transformer module, the partition fusion downsampling module, the double-layer upsampling module, and the skip connection module.

[0012] The reconstructed structure is obtained by constructing the second convolutional module;

[0013] The shallow feature extraction structure, the encoder-decoder structure, and the reconstruction structure are sequentially connected to obtain the interference suppression network.

[0014] In some embodiments, the step of simulating structural defects and interference factors in the simulated scenario of the water conveyance tunnel detection to obtain a training dataset includes the following steps:

[0015] The cavity simulation process was performed on the simulated water conveyance tunnel exploration scenario to obtain reference data;

[0016] The concrete structure of the simulated water conveyance tunnel detection scenario is subjected to heterogeneous treatment, and random water addition simulation treatment is performed on the surface of the concrete structure to obtain an interference model.

[0017] The interference model is subjected to finite-difference time-domain simulation to obtain ground-penetrating radar simulation data;

[0018] The ground-penetrating radar simulation data is processed to remove channel loss and high-frequency noise, resulting in interference data;

[0019] The reference data and the interference data are determined as the training dataset.

[0020] In some embodiments, the training dataset includes reference data and interference data, and the step of training the interference suppression network based on the training dataset to obtain the trained target network includes the following steps:

[0021] The interference data is input into the interference suppression network, and network recovery data is output.

[0022] The network loss function is constructed based on the mean absolute error, peak signal-to-noise ratio, and structural similarity.

[0023] The difference between the reference data and the network recovery data is calculated based on the network loss function to obtain the loss value.

[0024] The parameters of the interference suppression network are updated based on the loss value to obtain the trained target network.

[0025] In some embodiments, the step of inputting the initial ground-penetrating radar data into the target network for interference suppression processing and outputting target ground-penetrating radar data includes the following steps:

[0026] The initial ground-penetrating radar data is input into the target network, which includes a shallow feature extraction structure, an encoder-decoder structure, and a reconstruction structure.

[0027] The initial ground-penetrating radar data is preliminarily extracted and processed using the shallow feature extraction structure to obtain initial image features;

[0028] The encoder-decoder structure performs feature extraction processing on the initial image features to obtain the target image features. The encoder-decoder structure includes a preprocessing module, a sparse window transformer module, a partition fusion downsampling module, a two-layer upsampling module, and a skip connection module.

[0029] The target image features are reconstructed using the reconstructed structure to obtain target ground-penetrating radar data.

[0030] In some embodiments, the step of performing feature extraction processing on the initial image features through the encoder-decoder structure to obtain target image features includes the following steps:

[0031] The preprocessing module performs partitioning and linear transformation on the initial image features to obtain the first feature.

[0032] The first feature is processed by the sparse window transformer module to extract image features, thereby obtaining the second feature.

[0033] The second feature is downsampled layer by layer by the partition fusion downsampling module to obtain the third feature;

[0034] The third feature is transmitted to the two-layer upsampling module through the skip connection module;

[0035] The third feature is interpolated and convolved by the dual-layer upsampling module to obtain the processing result, and then the processing result is upsampled to obtain the target image feature.

[0036] In some embodiments, the sparse window transformer module includes multiple transformer layers. The step of extracting image features from the first feature using the sparse window transformer module to obtain the second feature includes the following steps:

[0037] The first feature is segmented based on the segmentation window to obtain the window feature;

[0038] The window features are processed by multi-head self-attention calculation using the sparse autocorrelation algorithm to obtain the calculation results.

[0039] The segmentation window is offset, the calculation result is added to the first feature, the segmentation window and the first feature are updated, and the process returns to the step of segmenting the first feature according to the segmentation window to obtain the window feature. This process continues until the number of iterations reaches the number of layers in the transformer layer, and the calculation result is determined as the second feature.

[0040] To achieve the above objectives, another aspect of this application proposes a ground-penetrating radar data processing system for water conveyance tunnels, the system comprising:

[0041] The first module is used to construct a simulation scenario for water conveyance tunnel detection and to establish an interference suppression network;

[0042] The second module is used to simulate structural defects and interference factors in the simulated scenario of the water conveyance tunnel detection to obtain a training dataset.

[0043] The third module is used to perform model training on the interference suppression network based on the training dataset to obtain the trained target network.

[0044] The fourth module is used to acquire initial ground-penetrating radar data, input the initial ground-penetrating radar data into the target network for interference suppression processing, and output the target ground-penetrating radar data.

[0045] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0046] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods described above.

[0047] The embodiments of this application include at least the following beneficial effects: This application provides a method, system, device, and medium for processing ground-penetrating radar data in water conveyance tunnels. This scheme constructs a simulated scenario for water conveyance tunnel detection and establishes an interference suppression network; it simulates structural defects and interference factors in the simulated scenario to obtain a training dataset; it trains the interference suppression network based on the training dataset to obtain a trained target network; it acquires initial ground-penetrating radar data, inputs the initial ground-penetrating radar data into the target network for interference suppression processing, and outputs the target ground-penetrating radar data; the embodiments of this application use neural networks to extract features from the ground-penetrating radar data, thereby eliminating interference in the data and improving data quality. Attached Figure Description

[0048] Figure 1 This is a flowchart of a ground-penetrating radar data processing method for a water conveyance tunnel provided in an embodiment of this application;

[0049] Figure 2 This is a schematic diagram of the structure of a target network provided in an embodiment of this application;

[0050] Figure 3 This is a schematic diagram of a training dataset provided in an embodiment of this application;

[0051] Figure 4 This is a schematic diagram of the structure of a ground-penetrating radar data processing system for a water conveyance tunnel provided in an embodiment of this application. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of systems and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0053] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various concepts, but unless otherwise stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to a determination” as used herein may be interpreted as “when…” or “when…” or “in response to a determination.”

[0054] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0056] Before providing a detailed description of the embodiments of this application, some of the nouns and terms involved in the embodiments of this application will be explained first. The nouns and terms involved in the embodiments of this application are subject to the following interpretations.

[0057] In related technologies, water conveyance tunnels are underground hydraulic structures excavated for the purpose of water conveyance. Over time, water conveyance tunnels develop internal defects such as cavities, cracks, and delamination, which can affect their usability and even threaten their structural safety. Therefore, ground-penetrating radar (GPR), as a non-destructive and efficient detection method, is commonly used for the inspection of water conveyance tunnel structures in order to promptly detect and repair these defects. However, when using ground-penetrating radar (GPR) to inspect water conveyance tunnels, various interferences frequently occur. These interferences mainly originate from the following aspects: During inspection, the surface of the water conveyance tunnel is usually covered with water, which may cause multiple signal reflections; the water conveyance tunnel is made of concrete, a heterogeneous structure, causing multiple reflections and scattering of radar signals within the concrete, resulting in clutter in the received signal; the limited operating space inside the water conveyance tunnel may prevent the ranging wheel from closely adhering to the tunnel surface, causing the wheel to spin idly and resulting in missing radar data; interference from electromagnetic devices in the surrounding environment can also cause high-frequency noise in the GPR data. These interferences during the inspection of water conveyance tunnels affect the effectiveness of the GPR data, interfering with the echoes of defects within the tunnel and making it difficult to accurately locate defects.

[0058] In view of this, this application provides a method, system, device, and medium for processing ground-penetrating radar (GPR) data from water conveyance tunnels. This solution suppresses interference and noise in GPR data from water conveyance tunnels by using deep learning to suppress interference. The method involves constructing a simulated water conveyance tunnel detection scenario and establishing an interference suppression network; simulating structural defects and interference factors in the simulated scenario to obtain a training dataset; training the interference suppression network using the training dataset to obtain a trained target network; acquiring initial GPR data and inputting it into the target network for interference suppression, outputting the target GPR data; continuously extracting features from the GPR data and reconstructing the data to eliminate interference and improve data quality.

[0059] This application provides a method for processing ground-penetrating radar data in a water conveyance tunnel, relating to the field of signal processing technology. This method can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or vehicle-mounted terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the ground-penetrating radar data processing method for a water conveyance tunnel, but is not limited to the above forms.

[0060] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0061] Figure 1 This is an optional flowchart of a ground-penetrating radar data processing method for a water conveyance tunnel provided in an embodiment of this application. Figure 1 The method may include, but is not limited to, steps S101 to S104.

[0062] Step S101: Construct a simulation scenario for water conveyance tunnel detection and establish an interference suppression network;

[0063] Step S102: Perform structural defects and interference factor simulation processing on the simulated scenario of water conveyance tunnel detection to obtain a training dataset;

[0064] Step S103: Perform model training on the interference suppression network based on the training dataset to obtain the trained target network;

[0065] Step S104: Obtain initial ground-penetrating radar data, input the initial ground-penetrating radar data into the target network for interference suppression processing, and output the target ground-penetrating radar data.

[0066] Steps S101 to S104 of this embodiment first involve constructing a simulated scenario for water conveyance tunnel detection and establishing an interference suppression network. A three-dimensional model of the tunnel can be constructed using simulation software to create the simulated scenario, and a multi-layer transformer module is used to build the interference suppression network. Then, based on the constructed simulated scenario, structural defects and interference factors are simulated to generate a training dataset. The interference suppression network is trained, and the optimal network model parameters are determined according to the optimized loss function, resulting in a trained target network. Initial ground-penetrating radar (GPR) data collected in the field is input into the trained target network to obtain target GPR data, which is the GPR image after interference suppression. This embodiment uses a deep learning-based target network to perform data interference suppression processing on GPR data, eliminating interference in the GPR data and thus improving the quality of the GPR data.

[0067] In step S101 of some embodiments, establishing the interference suppression network includes the following steps:

[0068] A shallow feature extraction structure is constructed based on the first convolutional module;

[0069] The encoder-decoder structure is constructed based on the preprocessing module, the sparse window transformer module, the partition fusion downsampling module, the double-layer upsampling module, and the skip connection module.

[0070] The reconstructed structure is obtained by constructing the second convolutional module;

[0071] The shallow feature extraction structure, the encoder-decoder structure, and the reconstruction structure are sequentially connected to obtain the interference suppression network.

[0072] In this embodiment, the interference suppression network is composed of a "U"-shaped network, as shown in the following structure. Figure 2 As shown, the interference suppression network is constructed by sequentially connecting a shallow feature extraction structure, an encoder-decoder structure, and a reconstruction structure. The shallow feature extraction structure initially extracts image features through a first convolutional module, performing image preprocessing. The shallow feature extraction structure extracts shallow features F. shallow After that, F shallowFeatures are input into an encoder-decoder architecture, which includes multiple sparse window transformer modules, a partitioned fusion downsampling module, a two-layer upsampling module, and skip connection modules for extracting deep features. The encoder-decoder architecture also includes a preprocessing module for partitioning and linearly transforming the input features, reducing the shallow features F... shallow The image is transformed into an embedding representation suitable for processing by the sparse window transformer module. The sparse window transformer module is the basic unit for image feature extraction; it processes image features through segmentation windows, self-attention mechanisms, shifted windows, and multilayer perceptrons. The partition fusion downsampling module follows the transformer structure and functions similarly to the pooling layer in a convolutional neural network, used for downsampling the feature layer. Please refer to Table 1, which shows the encoder-decoder structure. Table 1 describes the computational modules, inputs, and outputs corresponding to different levels in the encoder-decoder structure, as shown in the table below:

[0073]

[0074] Table 1 Encoder-Decoder Structure Table

[0075] Then, a dual-layer upsampling module is used to upsample the feature map in the decoder section, thereby restoring the resolution to the original image. In the encoder-decoder structure, a skip connection is made between the encoder and decoder, directly passing the feature layer of the encoder to the feature layer of the decoder. Then, a second convolution module is used to construct the reconstructed structure, and the extracted features are remodeled back into the image through convolution operations. This enables the establishment of an interference suppression network to perform data interference suppression processing on the ground penetrating radar data, eliminating interference in the ground penetrating radar data.

[0076] In some embodiments, the step of simulating structural defects and interference factors in the simulated scenario of the water conveyance tunnel detection to obtain a training dataset includes the following steps:

[0077] The cavity simulation process was performed on the simulated water conveyance tunnel exploration scenario to obtain reference data;

[0078] The concrete structure of the simulated water conveyance tunnel detection scenario is subjected to heterogeneous treatment, and random water addition simulation treatment is performed on the surface of the concrete structure to obtain an interference model.

[0079] The interference model is subjected to finite-difference time-domain simulation to obtain ground-penetrating radar simulation data;

[0080] The ground-penetrating radar simulation data is processed to remove channel loss and high-frequency noise, resulting in interference data;

[0081] The reference data and the interference data are determined as the training dataset.

[0082] In this embodiment, it is necessary to perform void simulation processing on the simulated scenario of water conveyance tunnel detection to obtain reference data. This reference data is used to compare the model output data in subsequent model training. This embodiment simulates potential structural defects in the simulated scenario of water conveyance tunnel detection. Common defects inside water conveyance tunnels include internal voids, looseness, and delamination. These defects manifest as voids or loose areas in the material. In this embodiment, during modeling, the distribution area of ​​defects is determined by generating two-dimensional Gaussian distribution sample data, and voids are generated at the sample data points to simulate voids or loose areas in the water conveyance tunnel. Specifically, the mean vector μ of the generated two-dimensional Gaussian distribution sample data is... x ,μ y ] and covariance matrix Where, μ x and μ y These are the means in the x and y directions, respectively. and This represents the variance in the corresponding direction. The center point location of each small cavity in a loose region, meeting specified parameter requirements, is generated using a multivariate normal distribution formula, as shown below:

[0083]

[0084] In the formula, x is an N-dimensional random vector, N is a positive integer, the mean vector determines the average position (center position) of all small cavities, and the covariance matrix determines the variance of the distance between the center of each small cavity and the average position (center position) of all small cavities. Then, using these generated sample data as the centers of the small cavities, the shape and size of the cavities are determined by random variables to simulate non-dense regions of different shapes and sizes in reality. And by specifying... and The size can generate different types of defects in water conveyance tunnels. When and When all values ​​are relatively large, each small cavity is far from the average position (center position) of all cavities, and the cavity distribution is relatively scattered. In this case, the resulting disease is a loosely compacted area; when... and When all values ​​are relatively small, each small cavity is close to the average position (center position) of all cavities, and the cavity distribution is relatively diverse. In this case, the resulting defect is an internal cavity. relatively When the voids are large, each small void is relatively concentrated in the y-direction and relatively dispersed in the x-direction, resulting in a defect called voiding. This application embodiment also simulates interference data using interference factors from a water conveyance tunnel. The surface of the water conveyance tunnel contains water, which can be simulated by adding a layer of water of random thickness to the surface of the structural concrete. The randomly generated water thickness is 1 to 5 cm, and the dielectric constant and conductivity of the water are set to ε = 81 and σ = 0.001 S / m, respectively. The concrete is a heterogeneous structure composed of aggregate and cement. During modeling, cement is considered a homogeneous medium, and aggregate is placed into the cement background, where the aggregate is a polygon with random side lengths. The simulation of missing channels is performed after forward modeling. When imaging ground-penetrating radar data, a random error rate is first generated, and then the number of channels to be disordered is calculated based on the total number of channels and the error rate. Then, the size b of the disordered block is randomly selected, and the starting position s and the position p to be exchanged of the disordered block are randomly selected based on b.

[0085] s = rand(0, n-b+1);

[0086] p = rand(0, n-b+1);

[0087] In the formula, rand represents randomly selecting an integer within the interval, and n represents the total number of radar channels. Then, the corresponding columns in the matrix are swapped, as shown in the following formula:

[0088] matrix[:,s:s+b],matrix[:,p:p+b]=matrix[:,p:p+b],matrix[:,s:s+b];

[0089] In the formula, matrix represents the ground-penetrating radar data matrix. Through these steps, the function simulates the random distortion of the signal channels, making the generated signal data closer to the complex situations that may occur in reality. Gaussian noise is superimposed on the original ground-penetrating radar data to simulate high-frequency noise in the data. This is achieved by generating a noise matrix N, where each element is a noise source with a mean of 0 and a variance of σ. 2 For random variables drawn from a normal distribution, the noise matrix is ​​expressed as: N = [n ij ],n ij ~N(0,σ 2 ). Reference Figure 3 , Figure 3 This is a schematic diagram of a training dataset provided in an embodiment of this application. Figure 3In this application, 301 represents interference-free reference data, 302 represents simulated data with surface water, 303 represents simulated data with a heterogeneous structure, 304 represents simulated data with missing channels, and 305 represents simulated data with high-frequency noise. This application embodiment can construct different simulated data for different interference factors, or it can simulate interference data with all interference factors. Specifically, the concrete structure in the water conveyance tunnel detection simulation scenario can be heterogeneously processed, and random water addition simulation processing can be performed on the surface of the concrete structure to obtain an interference model. This application embodiment uses the Finite-Difference Time-Domain (FDTD) method to simulate ground-penetrating radar waves, where the time-domain form of Maxwell's equations is shown below:

[0090]

[0091] Where E is the electric field strength, H is the magnetic field strength, μ is the permeability of the medium, ∈ is the permittivity of the medium, and σ is the conductivity of the medium. The space and events are discretized, with spatial discretization steps of Δx, Δy, and Δz, and event discretization steps of Δt. The electric and magnetic fields are represented at the discrete grid points as E(i,j,k,n) and H(i,j,k,n). Maxwell's equations are discretized using a difference scheme to obtain the difference equations for updating the electric and magnetic fields. Taking the update of the magnetic and electric fields in the x-direction as an example:

[0092]

[0093]

[0094] During the simulation, the Ricker waveform is used as the excitation source, and an electric field is applied at specific time steps and spatial locations. As the time step increases, the electric and magnetic fields are updated at each time step. By recording the electric and magnetic fields at each time step, the propagation of electromagnetic waves in the medium is obtained. Finally, the ground-penetrating radar simulation data is processed for missing traces and high-frequency noise to obtain interference data. In this embodiment, reference data and interference data are determined as the training dataset. The training dataset includes two sets of data: one set is ground-penetrating radar data containing various interference factors (surface water, structural heterogeneity, missing traces, high-frequency noise interference), and the other set is data considering only the impact of defects in water conveyance tunnels. The two sets of data appear in pairs as training data for the network to update the network parameters.

[0095] In some embodiments, the training dataset includes reference data and interference data, and the step of training the interference suppression network based on the training dataset to obtain the trained target network includes the following steps:

[0096] The interference data is input into the interference suppression network, and network recovery data is output.

[0097] The network loss function is constructed based on the mean absolute error, peak signal-to-noise ratio, and structural similarity.

[0098] The difference between the reference data and the network recovery data is calculated based on the network loss function to obtain the loss value.

[0099] The parameters of the interference suppression network are updated based on the loss value to obtain the trained target network.

[0100] In this embodiment, the interfering input data is defined as "input," the network reference data is defined as the interference-free data as "target," the network-restored data is defined as "restored," and the network is defined as F. After the data is input into the network, the network's output expression is as follows:

[0101] restored = F(input);

[0102] The network's output data is then compared with the target data, and the differences between the two are compared in three aspects: L1 loss, peak signal-to-noise ratio (PSNR), and structural similarity (SSIM), forming the network's loss function. The L1 loss is... Peak signal-to-noise ratio loss is Structural similarity loss is

[0103] Among them, y i It is the true value of the i-th sample point. L is the network prediction value for the i-th sample point. 2 It is the maximum value in the reference data, μ x and μ y These are the average values ​​of the output image x and the reference image y, respectively, σ x and σ y These are the standard deviations of the output image x and the reference image y, respectively, σ xy C1 and C2 are small constants used to avoid a distribution of 0 between the output image x and the reference image y.

[0104] This application's embodiments consider the differences between the restored image and the target reference image from three dimensions: PSNR is the ratio between the maximum signal value and noise, focusing only on pixel-to-pixel differences; SSIM considers the structural characteristics of the image; L1 loss also considers pixel-to-pixel differences but is insensitive to outliers. These three loss functions together constitute the network training loss function, the expression of which is shown below:

[0105] Loss=λ1PSNR+λ2SSIM+λ3L1;

[0106] Wherein, λ1, λ2, and λ3 are the weights of the three types of losses in the total loss, respectively.

[0107] The difference between the reference data and the network recovery data is calculated using the network loss function to obtain the loss value. The parameters of the interference suppression network are updated based on the loss value to obtain the trained target network. The gradient is calculated layer by layer from the output layer to the input layer using the backpropagation algorithm. The calculated gradient is used to update the network parameters, thereby adjusting the parameters to optimize the network performance and obtaining the trained target network.

[0108] In some embodiments, the step of inputting the initial ground-penetrating radar data into the target network for interference suppression processing and outputting target ground-penetrating radar data includes the following steps:

[0109] The initial ground-penetrating radar data is input into the target network, which includes a shallow feature extraction structure, an encoder-decoder structure, and a reconstruction structure.

[0110] The initial ground-penetrating radar data is preliminarily extracted and processed using the shallow feature extraction structure to obtain initial image features;

[0111] The encoder-decoder structure performs feature extraction processing on the initial image features to obtain the target image features. The encoder-decoder structure includes a preprocessing module, a sparse window transformer module, a partition fusion downsampling module, a two-layer upsampling module, and a skip connection module.

[0112] The target image features are reconstructed using the reconstructed structure to obtain target ground-penetrating radar data.

[0113] In this embodiment, initial ground-penetrating radar data is input into a trained target network. The target network includes a shallow feature extraction structure, an encoder-decoder structure, and a reconstruction structure. The shallow feature extraction structure uses a convolutional module to initially extract image features and performs image preprocessing to obtain initial image features. Radar images F with various types of interference... input The size is (H, W, 3), where H and W are the resolutions of the image in two directions, respectively. The shallow feature extraction structure uses a single 3*3 convolutional layer to obtain low-frequency information such as color or texture of the image, and its initial image features F shallow The calculation expression is as follows:

[0114] F shallow =conv(F input );

[0115] Wherein, the output F after convolution calculation shallow The size is (H, W, 96).

[0116] Then the initial image features F extracted by the shallow feature extraction structure are... shallow The data is input to an encoder-decoder structure, which includes a preprocessing module, a sparse window transformer module, a partition fusion downsampling module, a two-layer upsampling module, and a skip connection module, used to extract deep-level features to obtain target image features. Finally, the target image features are reconstructed using a reconstruction structure to obtain the target ground-penetrating radar data.

[0117] In some embodiments, the step of performing feature extraction processing on the initial image features through the encoder-decoder structure to obtain target image features includes the following steps:

[0118] The preprocessing module performs partitioning and linear transformation on the initial image features to obtain the first feature.

[0119] The first feature is processed by the sparse window transformer module to extract image features, thereby obtaining the second feature.

[0120] The second feature is downsampled layer by layer by the partition fusion downsampling module to obtain the third feature;

[0121] The third feature is transmitted to the two-layer upsampling module through the skip connection module;

[0122] The third feature is interpolated and convolved by the dual-layer upsampling module to obtain the processing result, and then the processing result is upsampled to obtain the target image feature.

[0123] In this embodiment, the initial image features are partitioned and linearly transformed by a preprocessing module to obtain the first feature. The partitioning operation divides the input initial image features into fixed-size partitions. Assuming the size of the input feature layer is (H, W, C), each 4x4 adjacent feature area constitutes a partition, and each partition is a small image region containing local image information. Then, each block is flattened along the channel dimension. After the partitioning operation, the size of the feature layer becomes (H / 4, W / 4, C*16). Then, a linear transformation maps the 16 channels of each pixel to the embedding dimension; this step is achieved through matrix multiplication.

[0124] Embedding=W·Flatten(patch)+b;

[0125] Where W is the weight matrix, b is the bias vector, and patch is each small partition. After a linear transformation, the size of the feature layer becomes (H / 4, W / 4, C), resulting in the first feature.

[0126] The first feature, with dimensions (H / 4, W / 4, C), is then input into the sparse window transformer module. The sparse window transformer module is the basic unit for image feature extraction. It processes image features through segmentation windows, self-attention mechanisms, shifted windows, and multilayer perceptrons to obtain the second feature. The second feature is then downsampled layer by layer by the partition fusion downsampling module to obtain the third feature. This partition fusion downsampling module, which follows the transformer structure, functions similarly to the pooling layer in a convolutional neural network, used to downsample the feature layer. Its purpose is to reduce the spatial size of the feature layer, thereby reducing computational complexity for subsequent calculations, and to increase the number of channels through layer-by-layer downsampling, thus extracting high-level image features. This module performs the following operations on the feature layer: first, it divides the feature map into 2*2 partitions, then flattens each partition and stitches them together. Assuming the input feature layer size of this module is (H, W, C), the module flattens each 2*2 partition of the input feature layer into a vector and concatenates these vectors along the channel dimension to form a feature layer of size (H / 2, W / 2, 4*C). Then, a linear transformation is performed on this feature layer, finally outputting a feature layer of size (H / 2, W / 2, 2*C), which is the third feature. The dual-layer upsampling module is used to upsample the feature map in the decoder section, thereby restoring the resolution to the original image. This module achieves finer feature reconstruction through a two-step upsampling process. First, the resolution of the feature map is increased to a portion of the original image using interpolation and convolution methods, and then upsampling is performed again to restore the resolution of the feature map to the original image resolution. Assuming the input feature layer size of this module is (H, W, C), after upsampling by this module, assuming an upsampling rate of 2, the shape of the output feature is (2H, 2W, C / 4); assuming an upsampling rate of 4, the shape of the output feature is (4H, 4W, C / 2), thus obtaining the target image features. In the encoder-decoder architecture, skip connections are used between the encoder and decoder, directly passing the encoder's feature layers to the decoder's feature layers. Skip connections serve several purposes: During encoding, feature maps are downsampled layer by layer, gradually decreasing spatial resolution. While capturing more high-level features, some detailed information is lost. Skip connections directly pass the high-resolution feature maps from the encoder layer to the decoder layer, helping to recover this detailed information; by concatenating the feature maps from the encoder and decoder layers, skip connections can fuse feature representations from different levels, enhancing feature diversity and expressive power; and skip connections provide additional paths, allowing gradients to propagate more effectively back through the network, mitigating the vanishing gradient problem and stabilizing the training process.

[0127] In some embodiments, the sparse window transformer module includes multiple transformer layers. The step of extracting image features from the first feature using the sparse window transformer module to obtain the second feature includes the following steps:

[0128] The first feature is segmented based on the segmentation window to obtain the window feature;

[0129] The window features are processed by multi-head self-attention calculation using the sparse autocorrelation algorithm to obtain the calculation results.

[0130] The segmentation window is offset, the calculation result is added to the first feature, the segmentation window and the first feature are updated, and the process returns to the step of segmenting the first feature according to the segmentation window to obtain the window feature. This process continues until the number of iterations reaches the number of layers in the transformer layer, and the calculation result is determined as the second feature.

[0131] In this embodiment, the self-attention mechanism of the transformer module requires calculating the similarity between the query (Q) and key (K) at each position in the input sequence, which causes the computational complexity to increase quadratically with the length of the input sequence. However, directly processing the feature layer using the transformer would be computationally expensive. Therefore, the sparse window transformer module in this embodiment first divides the input features into multiple fixed-size windows, and then performs a transformer calculation on the features within each window. After one transformer module calculation, the second transformer module calculation moves the window by a certain offset (e.g., half the window size), so that the window positions of the two transformer modules are different. Therefore, this embodiment performs offset processing on the segmented window, adds the calculation result to the first feature, updates the segmented window and the first feature, and returns to perform feature segmentation processing on the first feature based on the segmented window to obtain the window feature, until the number of iterations reaches the transformer layer level, and the calculation result is determined as the second feature. This allows features across windows to be exchanged, capturing global dependency information. Specifically, firstly, the first feature is segmented according to the segmentation window to obtain window features. The segmentation window divides the input feature into multiple fixed-size windows. Then, for the features within each window, its self-attention is calculated. During the self-attention calculation, for the features within each window, a query (Q), key (K), and value (V) matrix is ​​calculated:

[0132] Q = XW Q ;

[0133] K = XW K ;

[0134] V = XW V ;

[0135] Where X is the input feature, W Q W K W V This is a learnable weight matrix. The query (Q) matrix represents the features of the currently processed element. The query vector of each element is used to perform a dot product operation with the key vectors of other elements to calculate the similarity of that element to other elements; the keys (K) represent the features of all elements, used for matching the query vector. The key vector of each element measures the importance of other elements to the current query vector; the values ​​(V) represent the actual information of all elements, used for weighted averaging based on the attention weights to generate the output vector. The value vector is the fundamental information used to calculate the final output. When calculating the attention weights, a dot product operation is performed on each query vector (Q) and all key vectors (K). Simultaneously, to further reduce computational complexity, a sparse autocorrelation strategy is adopted when calculating autocorrelation. Multi-head self-attention processing is performed on the window features according to the sparse autocorrelation algorithm to obtain the calculation results. That is, selecting the few keys (K) most relevant to each query, instead of all keys, is achieved through a pre-screening step. The attention weights (Attention) are obtained through the softmax function, and the expression of the attention function is as follows:

[0136]

[0137] in, is a scaling factor used to stabilize the gradient. M is a sparse mask matrix, obtained by filtering the keys (K) using a set threshold, i.e., only the selected keys are set to 1, and the rest are set to 0. After obtaining the attention weights, the value vector V is weighted and averaged according to the attention weights to obtain the final output. Simultaneously, through a multi-head mechanism, the input features are divided into multiple subspaces, attention is calculated for each subspace, and the results are then concatenated and linearly transformed to obtain the final output, as shown in the following expression:

[0138] MultiHeadAttention(Q,K,V)=concat(head1,head2,...,head n W head ;

[0139] Among them, head i It is the result of individual self-attention for each head, W headIt is a linear transformation matrix. The multi-head attention mechanism within the window can capture rich local features (such as edges, textures, etc.) within the window, as well as the relationships between pixels within the window, which helps in modeling the contextual information of local regions.

[0140] Please see Figure 4 This application also provides a ground-penetrating radar data processing system for water conveyance tunnels, which can implement the above-mentioned ground-penetrating radar data processing method for water conveyance tunnels. The system includes:

[0141] The first module 401 is used to construct a simulation scenario for water conveyance tunnel detection and to establish an interference suppression network;

[0142] The second module 402 is used to simulate structural defects and interference factors in the water conveyance tunnel detection simulation scenario to obtain a training dataset.

[0143] The third module 403 is used to perform model training on the interference suppression network based on the training dataset to obtain the trained target network.

[0144] The fourth module 404 is used to acquire initial ground-penetrating radar data, input the initial ground-penetrating radar data into the target network for interference suppression processing, and output the target ground-penetrating radar data.

[0145] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0146] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method for processing ground-penetrating radar data in a water conveyance tunnel. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0147] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0148] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for processing ground-penetrating radar data in a water conveyance tunnel.

[0149] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0150] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0151] This application provides a method, system, device, and medium for processing ground-penetrating radar (GPR) data in a water conveyance tunnel. The scheme involves constructing a simulated water conveyance tunnel detection scenario and establishing an interference suppression network; simulating structural defects and interference factors in the simulated scenario to obtain a training dataset; training the interference suppression network using the training dataset to obtain a trained target network; acquiring initial GPR data; inputting the initial GPR data into the target network for interference suppression processing; and outputting the target GPR data. This application embodiment uses a neural network to extract features from the GPR data, thereby eliminating interference and improving data quality.

[0152] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0153] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0154] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0155] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0156] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0157] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0158] In the embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0159] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0160] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0161] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0162] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for processing ground-penetrating radar data in a water conveyance tunnel, characterized in that, The method includes the following steps: A simulation scenario for water conveyance tunnel detection was constructed, and an interference suppression network was established; The simulation of the water conveyance tunnel detection scenario is processed to simulate structural defects and interference factors, and a training dataset is obtained. The interference suppression network is trained using the training dataset to obtain the trained target network. Acquire initial ground-penetrating radar data, input the initial ground-penetrating radar data into the target network for interference suppression processing, and output the target ground-penetrating radar data. The process of simulating structural defects and interference factors in the simulated scenario of the water conveyance tunnel detection to obtain a training dataset includes the following steps: The simulation of the water conveyance tunnel detection scenario is processed by void simulation. By generating two-dimensional Gaussian distribution sample data during modeling, the distribution area of ​​the defects is determined, and voids are generated at the sample data points to simulate voids or loose areas in the water conveyance tunnel, so as to obtain reference data. The concrete structure of the simulated water conveyance tunnel detection scenario is subjected to heterogeneous treatment, and random water addition simulation is performed on the surface of the concrete structure to obtain an interference model; The interference model is subjected to finite-difference time-domain simulation to obtain ground-penetrating radar simulation data; The ground-penetrating radar simulation data is processed to remove channel loss and high-frequency noise. A random error rate is generated, and the number of channels to be disordered is calculated based on the total number of channels and the error rate. The size of the disordered block is randomly selected, and the starting position and the position to be swapped of the disordered block are randomly selected based on the size of the disordered block to simulate random disorder of the signal channels and obtain interference data. The reference data and the interference data are determined as the training dataset.

2. The method according to claim 1, characterized in that, The establishment of the interference suppression network includes the following steps: A shallow feature extraction structure is constructed based on the first convolutional module; The encoder-decoder structure is constructed based on the preprocessing module, the sparse window transformer module, the partition fusion downsampling module, the double-layer upsampling module, and the skip connection module. The reconstructed structure is obtained by constructing the second convolutional module; The shallow feature extraction structure, the encoder-decoder structure, and the reconstruction structure are sequentially connected to obtain the interference suppression network.

3. The method according to claim 1, characterized in that, The training dataset includes reference data and interference data. The step of training the interference suppression network using the training dataset to obtain the trained target network includes the following steps: The interference data is input into the interference suppression network, and network recovery data is output. The network loss function is constructed based on mean absolute error, peak signal-to-noise ratio, and structural similarity. The difference between the reference data and the network recovery data is calculated based on the network loss function to obtain the loss value. The parameters of the interference suppression network are updated based on the loss value to obtain the trained target network.

4. The method according to claim 1, characterized in that, The process of inputting the initial ground-penetrating radar data into the target network for interference suppression processing and outputting the target ground-penetrating radar data includes the following steps: The initial ground-penetrating radar data is input into the target network, which includes a shallow feature extraction structure, an encoder-decoder structure, and a reconstruction structure. The initial ground-penetrating radar data is preliminarily extracted and processed using the shallow feature extraction structure to obtain initial image features; The encoder-decoder structure performs feature extraction processing on the initial image features to obtain the target image features. The encoder-decoder structure includes a preprocessing module, a sparse window transformer module, a partition fusion downsampling module, a two-layer upsampling module, and a skip connection module. The target image features are reconstructed using the reconstructed structure to obtain target ground-penetrating radar data.

5. The method according to claim 4, characterized in that, The step of extracting features from the initial image features using the encoder-decoder structure to obtain target image features includes the following steps: The preprocessing module performs partitioning and linear transformation on the initial image features to obtain the first feature. The first feature is processed by the sparse window transformer module to extract image features, resulting in the second feature; The second feature is downsampled layer by layer by the partition fusion downsampling module to obtain the third feature; The third feature is transmitted to the two-layer upsampling module through the skip connection module; The third feature is interpolated and convolved by the dual-layer upsampling module to obtain the processing result, and then the processing result is upsampled to obtain the target image feature.

6. The method according to claim 5, characterized in that, The sparse window transformer module includes multiple transformer layers. The process of extracting image features from the first feature using the sparse window transformer module to obtain the second feature includes the following steps: The first feature is segmented based on the segmentation window to obtain the window feature; The window features are processed by multi-head self-attention calculation using the sparse autocorrelation algorithm to obtain the calculation results. The segmentation window is offset, the calculation result is added to the first feature, the segmentation window and the first feature are updated, and the process returns to the step of segmenting the first feature according to the segmentation window to obtain the window feature. This process continues until the number of iterations reaches the number of layers in the transformer layer, and the calculation result is determined as the second feature.

7. A ground-penetrating radar data processing system for a water conveyance tunnel, characterized in that, The system includes: The first module is used to construct a simulation scenario for water conveyance tunnel detection and to establish an interference suppression network; The second module is used to simulate structural defects and interference factors in the simulated scenario of the water conveyance tunnel detection to obtain a training dataset. The third module is used to perform model training on the interference suppression network based on the training dataset to obtain the trained target network. The fourth module is used to acquire initial ground-penetrating radar data, input the initial ground-penetrating radar data into the target network for interference suppression processing, and output the target ground-penetrating radar data. The second module is used to simulate structural defects and interference factors in the simulated scenario of the water conveyance tunnel detection, and obtain a training dataset, including: The simulation of the water conveyance tunnel detection scenario is processed by void simulation. By generating two-dimensional Gaussian distribution sample data during modeling, the distribution area of ​​the defects is determined, and voids are generated at the sample data points to simulate voids or loose areas in the water conveyance tunnel, so as to obtain reference data. The concrete structure of the simulated water conveyance tunnel detection scenario is subjected to heterogeneous treatment, and random water addition simulation is performed on the surface of the concrete structure to obtain an interference model; The interference model is subjected to finite-difference time-domain simulation to obtain ground-penetrating radar simulation data; The ground-penetrating radar simulation data is processed to remove channel loss and high-frequency noise. A random error rate is generated, and the number of channels to be disordered is calculated based on the total number of channels and the error rate. The size of the disordered block is randomly selected, and the starting position and the position to be swapped of the disordered block are randomly selected based on the size of the disordered block to simulate random disorder of the signal channels and obtain interference data. The reference data and the interference data are determined as the training dataset.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

Images