Earth and rockfill dam termite nest detection method based on MSCCEAUNet and physical constraint combined driving

Abstract

An earth and rockfill dam termite nest detection method based on MSCCEAUNet and physical constraint combined driving comprises the following steps that a miscellaneous geologic model of an earth and rockfill dam structure with a termite nest is simulated, and corresponding b-scan data and underground dielectric constant data are generated through a finite difference method FDTD to serve as a simulation data set; 2, constructing an MSCCEAUNet network with multi-scale cascade convolution (MSC) and an efficient channel attention mechanism (ECA), and inputting the complex geologic model data generated by simulation in the step 1 into the MSCCEAUNet network for training; according to the method, the MSCCEAUNet is trained again by adding Gaussian noise and random medium disturbance data with different signal-to-noise ratios, the structure of the MSCCEAUNet is optimized according to the training condition of the model, and the anti-jamming capability and robustness of the model are improved; a physical residual equation meeting physical information constraints of Maxwell equations of the ground penetrating radar is used as a physical loss function to be fused into an MSCCEAUNet framework, and data and physics combined drive inversion is achieved.

Description

Technical Field

[0001] The present invention relates to the field of geophysical technology, in particular to an intelligent detection technology for termite nests in earth-rock dams, and specifically to an intelligent detection method for termite nests in earth-rock dams based on MSCCEAU-Net and physical information constraint-jointly driven GPR intelligent inversion. Background Art

[0002] Ground Penetrating Radar (GPR) is a non-destructive detection technology widely used in the detection and imaging of underground targets. It obtains physical information inside the medium through the propagation and reflection of electromagnetic waves in the underground medium. However, due to the complexity of the underground environment, the traditional GPR inversion method based on physical models may face limitations in accuracy and resolution. In addition, the GPR inversion process often involves problems such as multi-solution and noise, making it difficult for the inversion results to accurately reflect the real underground structure.

[0003] These challenges are even more prominent for the detection of termite nests in earth-rock dams. Termite nests are often small in size, complex in shape, and unevenly distributed. Their electromagnetic characteristics are weakly different from those of the surrounding medium, making them difficult to reliably identify using traditional methods. At the same time, the spread of termite nests inside the dam body will weaken the stability of the dam structure and pose a hidden danger to the long-term safety of the dam. Therefore, high-precision detection methods for termite nests are crucial for engineering safety monitoring.

[0004] Deep learning methods have shown significant advantages in dealing with complex nonlinear problems, automatic feature extraction, and efficient computing. Applying deep learning to GPR inversion problems can not only improve inversion accuracy, but also reduce reliance on manual intervention and improve the automation and intelligence of data processing. Research in this direction will help promote the intelligent upgrade of GPR technology and provide new solutions for more accurate, fast, and efficient underground target detection and physical property evaluation.

[0005] In the deep learning ground penetrating radar (GPR) inversion research, adding physical constraints not only improves the accuracy and reliability of the model, but also gives the research a deeper scientific significance. Specifically, the combination of deep learning and physical laws provides an efficient and accurate solution for the application of GPR in the detection of termite nests in earth-rock dams. This research is not only of great value in engineering applications, but also lays a scientific foundation for the development of intelligent inversion technology.

[0006] The patent document with application publication number CN117371330A discloses a two-dimensional magnetotelluric inversion method using deep learning inversion, which constructs a magnetotelluric data set using a traditional inversion method, and then uses a deep learning model to perform two-dimensional magnetotelluric inversion. Although the above method takes into account that deep learning inversion can achieve automatic feature extraction and perform efficient calculations, it does not take into account the physical meaning of electromagnetic wave propagation in the medium, resulting in the lack of actual physical meaning of the inversion result in the absence of physical information constraints. Summary of the invention

[0007] The purpose of the present invention is to solve the technical problems that the existing detection technology for termite nests in earth-rock dams has high computational complexity, is sensitive to real-world noise, and the existing inversion technology lacks physical information constraints, resulting in the inversion results lacking actual physical meaning.

[0008] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0009] The earth-rock dam termite nest detection method based on MSCCEAUNet and physical constraints jointly driven includes the following steps:

[0010] Step 1: Simulate the complex geological model of the earth-rock dam structure with termite nests, and use the finite difference method FDTD to generate the corresponding b-scan data and underground dielectric constant data as the simulation data set;

[0011] Step 2: Build a MSCCEAUNet network with multi-scale cascade convolution (MSC) and efficient channel attention mechanism (ECA), and input the complex geological model data simulated and generated by step 1 into the MSCCEAUNet network for training;

[0012] Step 3: Add Gaussian noise and random medium disturbance data with different signal-to-noise ratios to train MSCCEAUNet again. According to the training status of the model, optimize the MSCCEAUNet structure to improve the anti-interference ability and robustness of the model;

[0013] Step 4: Incorporate the physical residual equations that satisfy the physical information constraints of Maxwell's equations for ground penetrating radar into the MSCCEAUNet framework as a physical loss function to achieve joint data and physics driven inversion;

[0014] Step 5: Add Gaussian noise with different signal-to-noise ratios and random medium perturbation data again for data and physics joint driven training, and determine the optimal weights of the data loss, physical loss and structural similarity loss of MSCCEAUNet.

[0015] In step 1, the gprMax simulation software is used to simulate the earth-rock dam structure model with termite nests.

[0016] In step 4, when jointly driving the model and physical information constraints, the following steps are taken:

[0017] Step 4-1) First, the Maxwell equations for electric field and magnetic field propagation that satisfy the ground penetrating radar electromagnetic wave propagation equation are decoupled;

[0018] The Max equation used is as follows:

[0019]

[0020] In the formula, represents the magnetic field curl, represents the curl of the electric field, ε0 represents the dielectric constant, J represents the current density, and μ0 represents the magnetic permeability;

[0021] Step 4-2) Combining the decoupled Maxwell equations with the predicted electric field and dielectric constant to construct a physical residual formula that satisfies the electromagnetic wave propagation equation;

[0022] The constructed physical residual formula is:

[0023]

[0024] Where E is the electric field, σ is the conductivity, ε is the dielectric constant, μ is the magnetic permeability, S E is the electric field source term;

[0025] Step 4-3) Integrate the constructed physical residual formula into the MSCCEAUNet framework as the physical loss function to achieve the joint-driven inversion of data and physics, and ensure that the total loss of the MSCCEAUNet model is minimized.

[0026] In step 5, the following steps are used to determine the optimal weights of the losses of each part of MSCCEAUNet:

[0027] Step 5-1) Using the hyperparameter optimization method, the optimal weights of the data loss, physical loss and structural similarity loss of MSCCEAUNet are automatically selected to ensure that the loss of each part is minimized;

[0028] The constructed training loss function formula is as follows:

[0029] L total =αL data +βL physical +γL SSIM ;

[0030] In the formula, α refers to the weight ratio of data loss, β refers to the weight ratio of physical loss, and γ refers to the weight ratio of structural similarity loss; L total is the data loss, L physical is the loss of the physical equation, L SSIM is the structural similarity loss;

[0031] Step 5-2) After determining the optimal weight ratio of each part of the loss, set the weight parameters of each part of MSCCEAUNet to the optimal starting data and physical joint drive training.

[0032] In step 2, the constructed MSCECAUNet model is specifically:

[0033] The MSCECAUNet network includes the first Transformer Encoder MSC layer to the fifth Transformer Encoder MSC layer, the first Transformer Encoder layer to the fifth Transformer Encoder layer, the first ECA layer to the fourth ECA layer, the first Transformer Decoder MSC layer to the fifth Transformer Decoder MSC layer, the first Transformer Decoder layer to the fifth Transformer Decoder layer, and the FPC module;

[0034] The GPR B-Scan profile is input as an input feature to the first Transformer Encoder MSC layer, the output of the first Transformer Encoder MSC layer is connected to the input of the first Transformer Encoder, the output of the first Transformer Encoder is connected to the input of the first ECA layer and the input of the second Transformer Encoder MSC layer, the output of the first ECA layer and the output of the second Transformer Decoder are connected to the input of the first Transformer Decoder MSC layer, the output of the first Transformer Decoder MSC layer is connected to the input of the first Transformer Decoder, the output of the first Transformer Decoder is connected to the input of the FPC module, and the output of the FPC module is a dielectric constant distribution map;

[0035] The output of the second Transformer Encoder MSC layer is connected to the input of the second Transformer Encoder layer, the output of the second Transformer Encoder layer is connected to the input of the second ECA layer and the input of the third Transformer Encoder MSC layer, the output of the second ECA layer and the output of the third Transformer Decoder layer are connected to the input of the second Transformer Decoder MSC layer, and the output of the second Transformer Decoder MSC layer is connected to the input of the second Transformer Decoder layer;

[0036] The output of the third Transformer Encoder MSC layer is connected to the input of the third Transformer Encoder layer, the output of the third Transformer Encoder layer is connected to the input of the third ECA layer and the input of the fourth Transformer Encoder MSC layer, the output of the third ECA layer and the output of the fourth Transformer Decoder layer are connected to the input of the third Transformer Decoder MSC layer, and the output of the third Transformer Decoder MSC layer is connected to the input of the third Transformer Decoder layer;

[0037] The output of the fourth Transformer Encoder MSC layer is connected to the input of the fourth Transformer Encoder layer, the output of the fourth Transformer Encoder layer is connected to the input of the fourth ECA layer and the input of the fifth Transformer Encoder MSC layer, the output of the fourth ECA layer and the output of the fifth Transformer Decoder layer are connected to the input of the fourth Transformer Decoder MSC layer, and the output of the fourth Transformer Decoder MSC layer is connected to the input of the fourth Transformer Decoder layer;

[0038] The output of the fifth Transformer Encoder MSC layer is connected to the input of the fifth Transformer Encoder layer, the output of the fifth Transformer Encoder layer is connected to the input of the fifth Transformer Decoder MSC layer, and the output of the fifth Transformer Decoder MSC layer is connected to the input of the fifth Transformer Decoder layer.

[0039] The Transformer Encoder MSC layer of the MSCECAUNet model includes a single-core convolutional neural network layer SK-ConvNet, a three-dimensional convolutional neural network layer 3D-ConvNet, a two-level cascaded convolutional neural network layer C2-ConvNet, a three-level cascaded convolutional neural network layer C3-ConvNet, and an FFM module.

[0040] The input features are respectively input to the SK-ConvNet, 3D-ConvNet, C2-ConvNet and C3-ConvNet layers, and then the outputs of the above convolutional neural network layers are connected to the input of the FFM module, the output of the FFM module is connected to the input of the 3D-ConvNet, and the output of the 3D-ConvNet is connected to the input of the Kth Transformer Encoder layer of the MSCECAUNet model; where K represents the number of layers of the MSCECAUNet model.

[0041] The ECA layer of the MSCECAUNet model includes the input feature module IFM, the global average pooling module GAPM, the channel number adaptive K value module CKM, the K×K one-dimensional convolution module K×K 1D Conv, the Sigmoid activation function module SAFM, and the output feature module OFM;

[0042] The output of the Transformer Encoder layer is connected to the input of the ECA layer and input into the input feature module IFM. The output of the input feature module IFM is connected to the input of the global average pooling module GAPM. The output of GAPM is connected to the input of the channel number adaptive K value module CKM. The output of CKM is connected to the input of the K×K one-dimensional convolution module K×K 1D Conv. The output of K×K1D Conv is connected to the input of the Sigmoid activation function module SAFM. The output of SAFM is connected to the input of the output feature module OFM. The output of OFM is connected to the input of the Kth Transformer Decoder MSC layer.

[0043] Compared with the prior art, the present invention has the following technical effects:

[0044] 1) The network of the present invention adopts the traditional U-Net network as the framework, and the transformer encoder-decoder as the basic structure, integrating the multi-scale cascade convolution module and the efficient channel attention mechanism module. The network can effectively extract the key features of the termite nest in the earth-rock dam and realize the precise positioning of the termite nest.

[0045] 2) The present invention introduces physical information constraints: the Maxwell equations that conform to the propagation of electromagnetic waves underground are added as physical constraints in the deep learning model, thereby ensuring the accuracy and physical interpretability of the inversion results of termite nests in earth-rock dams.

[0046] 3) The present invention is based on automatic annotation of finite difference simulation data: by using gprMax software for geological modeling, annotation data is automatically generated. This method sets the simulation size, spatial step size, and location, size, and dielectric constant of the underground anomaly of the model. With these known parameters, the underground structure map can be automatically annotated, thereby greatly reducing the time and workload of manual annotation. This innovative technology provides accurate and efficient annotation data for termite nest detection in earth-rock dams, significantly improves the training effect and prediction accuracy of the inversion model, and promotes the application and development of intelligent detection technology.

[0047] 4) The intelligent system for earth-rock dam health monitoring of the present invention: The model can be integrated into the earth-rock dam health monitoring system to achieve real-time monitoring and early warning of termite nest expansion, thereby ensuring the long-term safety of the dam structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0049] Figure 1 It is the overall flow chart of the present invention;

[0050] Figure 2 It is a structural diagram of the MSCCEAUNet model in the present invention;

[0051] Figure 3 Schematic diagram of the structure of the efficient channel attention mechanism module (ECA) in the present invention;

[0052] Figure 4 Schematic diagram of the structure of the multi-scale cascade convolution module (MSC) in the present invention;

[0053] Figure 5 is the actual geological model diagram in the embodiment of the present invention;

[0054] Figure 6 It is a B-Scan waveform diagram corresponding to the geological model in the embodiment of the present invention;

[0055] Figure 7 Schematic diagram of underground dielectric constant inversion structure in an embodiment of the present invention;

[0056] Figure 8 Schematic diagram of deep learning model training loss with and without physical loss in the present invention. DETAILED DESCRIPTION

[0057] The earth-rock dam termite nest detection method based on MSCCEAUNet and physical constraints jointly driven includes the following steps:

[0058] Step 1: Simulate the complex geological model of the earth-rock dam structure with termite nests, and use the finite difference method FDTD to generate the corresponding b-scan data and underground dielectric constant data as the simulation data set;

[0059] Step 2: Build a MSCCEAUNet network with multi-scale cascade convolution (MSC) and efficient channel attention mechanism (ECA), and input the complex geological model data simulated and generated by step 1 into the MSCCEAUNet network for training;

[0060] Step 3: Add Gaussian noise and random medium disturbance data with different signal-to-noise ratios to train MSCCEAUNet again. According to the training status of the model, optimize the MSCCEAUNet structure to improve the anti-interference ability and robustness of the model;

[0061] Step 4: Incorporate the physical residual equations that satisfy the physical information constraints of Maxwell's equations for ground penetrating radar into the MSCCEAUNet framework as a physical loss function to achieve joint data and physics driven inversion;

[0062] Step 5: Add Gaussian noise with different signal-to-noise ratios and random medium perturbation data again for data and physics joint driven training, and determine the optimal weights of the data loss, physical loss and structural similarity loss of MSCCEAUNet.

[0063] In step 1, the gprMax simulation software is used to simulate the earth-rock dam structure model with termite nests.

[0064] In step 4, when jointly driving the model and physical information constraints, the following steps are taken:

[0065] Step 4-1) First, the Maxwell equations for electric field and magnetic field propagation that satisfy the ground penetrating radar electromagnetic wave propagation equation are decoupled;

[0066] The Max equation used is as follows:

[0067]

[0068] In the formula, represents the magnetic field curl, represents the curl of the electric field, ε0 represents the dielectric constant, J represents the current density, and μ0 represents the magnetic permeability;

[0069] Step 4-2) Combining the decoupled Maxwell equations with the predicted electric field and dielectric constant to construct a physical residual formula that satisfies the electromagnetic wave propagation equation;

[0070] The constructed physical residual formula is:

[0071]

[0072] Where E is the electric field, σ is the conductivity, ε is the dielectric constant, μ is the magnetic permeability, S E is the electric field source term;

[0073] Step 4-3) Integrate the constructed physical residual formula into the MSCCEAUNet framework as the physical loss function to achieve the joint-driven inversion of data and physics, and ensure that the total loss of the MSCCEAUNet model is minimized.

[0074] In step 5, the following steps are used to determine the optimal weights of the losses of each part of MSCCEAUNet:

[0075] Step 5-1) Using the hyperparameter optimization method, the optimal weights of the data loss, physical loss and structural similarity loss of MSCCEAUNet are automatically selected to ensure that the loss of each part is minimized;

[0076] The constructed training loss function formula is as follows:

[0077] L total =αL data +βL physical +γL SSIM ;

[0078] In the formula, α refers to the weight ratio of data loss, β refers to the weight ratio of physical loss, and γ refers to the weight ratio of structural similarity loss; L total is the data loss, L physical is the loss of the physical equation, L SSIM is the structural similarity loss;

[0079] Step 5-2) After determining the optimal weight ratio of each part of the loss, set the weight parameters of each part of MSCCEAUNet to the optimal starting data and physical joint drive training.

[0080] In step 2, the constructed MSCECAUNet model is specifically:

[0081] The MSCECAUNet network includes the first Transformer Encoder MSC layer to the fifth Transformer Encoder MSC layer, the first Transformer Encoder layer to the fifth Transformer Encoder layer, the first ECA layer to the fourth ECA layer, the first Transformer Decoder MSC layer to the fifth Transformer Decoder MSC layer, the first Transformer Decoder layer to the fifth Transformer Decoder layer, and the FPC module;

[0082] The GPR B-Scan profile is input as an input feature to the first Transformer Encoder MSC layer, the output of the first Transformer Encoder MSC layer is connected to the input of the first Transformer Encoder, the output of the first Transformer Encoder is connected to the input of the first ECA layer and the input of the second Transformer Encoder MSC layer, the output of the first ECA layer and the output of the second Transformer Decoder are connected to the input of the first Transformer Decoder MSC layer, the output of the first Transformer Decoder MSC layer is connected to the input of the first Transformer Decoder, the output of the first Transformer Decoder is connected to the input of the FPC module, and the output of the FPC module is a dielectric constant distribution map;

[0083] The output of the second Transformer Encoder MSC layer is connected to the input of the second Transformer Encoder layer, the output of the second Transformer Encoder layer is connected to the input of the second ECA layer and the input of the third Transformer Encoder MSC layer, the output of the second ECA layer and the output of the third Transformer Decoder layer are connected to the input of the second Transformer Decoder MSC layer, and the output of the second Transformer Decoder MSC layer is connected to the input of the second Transformer Decoder layer;

[0084] The output of the third Transformer Encoder MSC layer is connected to the input of the third Transformer Encoder layer, the output of the third Transformer Encoder layer is connected to the input of the third ECA layer and the input of the fourth Transformer Encoder MSC layer, the output of the third ECA layer and the output of the fourth Transformer Decoder layer are connected to the input of the third Transformer Decoder MSC layer, and the output of the third Transformer Decoder MSC layer is connected to the input of the third Transformer Decoder layer;

[0085] The output of the fourth Transformer Encoder MSC layer is connected to the input of the fourth Transformer Encoder layer, the output of the fourth Transformer Encoder layer is connected to the input of the fourth ECA layer and the input of the fifth Transformer Encoder MSC layer, the output of the fourth ECA layer and the output of the fifth Transformer Decoder layer are connected to the input of the fourth Transformer Decoder MSC layer, and the output of the fourth Transformer Decoder MSC layer is connected to the input of the fourth Transformer Decoder layer;

[0086] The output of the fifth Transformer Encoder MSC layer is connected to the input of the fifth Transformer Encoder layer, the output of the fifth Transformer Encoder layer is connected to the input of the fifth Transformer Decoder MSC layer, and the output of the fifth Transformer Decoder MSC layer is connected to the input of the fifth Transformer Decoder layer.

[0087] The Transformer Encoder MSC layer of the MSCECAUNet model includes a single-core convolutional neural network layer SK-ConvNet, a three-dimensional convolutional neural network layer 3D-ConvNet, a two-level cascaded convolutional neural network layer C2-ConvNet, a three-level cascaded convolutional neural network layer C3-ConvNet, and an FFM module.

[0088] The input features are respectively input to the SK-ConvNet, 3D-ConvNet, C2-ConvNet and C3-ConvNet layers, and then the outputs of the above convolutional neural network layers are connected to the input of the FFM module, the output of the FFM module is connected to the input of the 3D-ConvNet, and the output of the 3D-ConvNet is connected to the input of the Kth Transformer Encoder layer of the MSCECAUNet model; where K represents the number of layers of the MSCECAUNet model.

[0089] The ECA layer of the MSCECAUNet model includes the input feature module IFM, the global average pooling module GAPM, the channel number adaptive K value module CKM, the K×K one-dimensional convolution module K×K 1D Conv, the Sigmoid activation function module SAFM, and the output feature module OFM;

[0090] The output of the Transformer Encoder layer is connected to the input of the ECA layer and input into the input feature module IFM. The output of the input feature module IFM is connected to the input of the global average pooling module GAPM. The output of GAPM is connected to the input of the channel number adaptive K value module CKM. The output of CKM is connected to the input of the K×K one-dimensional convolution module K×K 1D Conv. The output of K×K1D Conv is connected to the input of the Sigmoid activation function module SAFM. The output of SAFM is connected to the input of the output feature module OFM. The output of OFM is connected to the input of the Kth Transformer Decoder MSC layer.

[0091] Example:

[0092] This embodiment includes the following steps:

[0093] Step 1: Use gprMax simulation software to simulate the earth-rock dam structure model with termite nests, and generate corresponding b-scan data and underground dielectric constant data as training data for the model. Below are the variable names and sizes of various parameters given by the established complex geological model.

[0094] Table 1: Geological model parameters

[0095]

[0096]

[0097] Step 2: Input the complex geological model data of the simulated earth-rock dam termite nest structure into the MSCECAUNet model for training.

[0098] The MSCECAUNet model includes: an encoder-decoder network with a U-Net architecture, a multi-scale cascade convolution module (MSC) and an efficient channel attention mechanism module (ECA), which can effectively extract the key features of termite nests in earth-rock dams.

[0099] (I) The encoder-decoder network of the U-Net architecture contains a multi-scale cascade convolution module (MSC), up-convolution operation and deconvolution operation, and the calculation formula is:

[0100] 1. Input layer: GPR B-scan signal is used as the input X0 of the model;

[0101] 2. Feature fusion layer: X0 passes through a multi-scale cascade convolution module to perform a multi-resolution feature fusion operation;

[0102] X1 = MSC (X0);

[0103] 3. Pooling layer: perform maximum pooling operation on the feature map;

[0104] X2=MaxPool(X1);

[0105] 4. Downsampling layer: downsample each layer and apply ReLU activation function

[0106] Xn+1=Convn(Xn)+ReLU

[0107] After each layer of downsampling and convolution operations, the number of channels changes as follows: 64->128->256->512->1024 channels.

[0108] 5. Bottleneck layer: In the deepest layer, convolution operations are applied to feature maps with 1024 channels

[0109] Xbottom=Conv(X1024)+ReLU

[0110] 6. Upsampling convolution (decoder path)

[0111] In the upsampling path, the operation of each layer is as follows:

[0112] 1) Use upconvolution (deconvolution) operations to increase the spatial dimension.

[0113] Xup=UpConv(Xbottom)

[0114] 2) Obtain feature maps from the downsampling path and perform feature fusion (skip connection).

[0115] Xconcat=Concat(Xeca,Xdown)

[0116] 3) Apply convolution operation and ReLU activation function to reduce the number of channels.

[0117] Xconv=Conv(Xconcat)+ReLU

[0118] During the upsampling process, the number of channels is gradually halved: 1024->512->256->128->64.

[0119] 7. Output layer: The output of this layer is used to predict the dielectric constant distribution.

[0120] Xfinal=Conv(X64)

[0121] (II) Multi-scale Cascade Convolution Module (MSC)

[0122] The multi-scale cascade convolution module mainly performs a multi-resolution feature fusion operation of termite nests in earth-rock dams, which mainly includes 4 types of convolutions with acceptance field sizes of 1*1, 3*3, 5*5, and 7*7. Considering the impact of computational complexity, the 5*5 and 7*7 convolutions are replaced by 2 cascades of 3*3 convolutions and 3 cascades of 3*3 convolutions. Finally, all feature channel maps are fused and a total feature map is extracted through a 3*3 convolution module.

[0123] (1) Input layer: Input B-scan signal data X0

[0124] (2) Branch operation:

[0125] X1=Conv3×3(X0)

[0126] X2=Conv3×3(X0)

[0127] X3=(Conv3×3)2(X0)

[0128] X4=(Conv3×3)3(X0)

[0129] Among them, X1, X2, X3, and X4 represent the features after 4 branch convolutions respectively.

[0130] (3) Splicing layer:

[0131] Xconcat=Concat(X1,X2,X3,X4)

[0132] Where contact represents the splicing operation,

[0133] (4) Final output layer:

[0134] Y = Conv3 × 3 (Xconcat)

[0135] (III) The efficient channel attention mechanism module (ECA) includes the following steps:

[0136] (1) Perform local cross-channel interaction between the channel features and the channel features of its neighbors. The calculation formula is:

[0137]

[0138] where |t odd represents the odd number closest to t; c represents the total number of channels; γ and b are fixed values, which are 2 and 1 respectively.

[0139] (2) Perform weighted summation on all channels, and the calculation formula is:

[0140]

[0141] Where: Represents and y i A set of k adjacent channels; For i The adjacent j-th channel output; α j Represents the weight parameter shared by all channels; δ is the sigmoid function.

[0142] Step 3: Add Gaussian noise with different signal-to-noise ratios and random medium disturbance earth-rock dam termite nest data to train MSCCEAUNet again. According to the training situation of the model, optimize the MSCCEAUNet structure to improve the anti-interference ability and robustness of the model;

[0143] Step 4: Incorporate the physical residual equation that satisfies the physical information constraints of Maxwell's equations of ground penetrating radar into the MSCCEAUNet framework as a physical loss function to achieve the joint-driven inversion of data and physics and ensure the inversion accuracy of termite nests in earth-rock dams;

[0144] Ground penetrating radar uses electromagnetic waves for detection. However, the propagation of electromagnetic waves underground follows Maxwell's equations, which are used to describe the interaction between electric and magnetic fields. These equations are the basis for describing electromagnetic phenomena, so they can be used to guide the data processing and output of deep learning models in GPR inversion. The propagation of electromagnetic waves is described by Maxwell's equations. In GPR, the changes in electric and magnetic fields are expressed by the following equations:

[0145]

[0146]

[0147] In the formula represents the magnetic field curl, represents the curl of the electric field, ε0 represents the dielectric constant, J represents the current density, and μ0 represents the magnetic permeability.

[0148] The constraints of the physical equations used in this method are based on the prediction results of the electric field and dielectric constant, and the physical residuals of the electromagnetic wave propagation equation are introduced to ensure that the prediction results meet the physical laws. The residual equation form is as follows:

[0149]

[0150] Where E is the electric field, σ is the conductivity, ε is the dielectric constant, μ is the magnetic permeability, S E is the electric field source term.

[0151] Step 5: Add termite nest data with different signal-to-noise ratios of Gaussian noise and random medium disturbance again for data and physics joint driven training, and determine the optimal weights of data loss, physical loss and structural similarity loss of MSCCEAUNet. The total loss function formula is as follows:

[0152] L total =αL data +βL physical +γL SSIM

[0153] Among them, L total is the data loss, L physical is the loss of the physical equation, L SSIM is the structural similarity loss. α, β and γ are the weights of the three parts of loss respectively.

[0154] according to Figure 5 , Figure 6 , Figure 7 The actual underground model structure diagram of the termite nest in the earth-rock dam, the corresponding B-Scan waveform diagram, and the underground dielectric constant structure diagram inverted by the deep learning model with physical constraints show that the position, size, shape, etc. of the inverted termite nest are very similar to the actual structure diagram. Figure 8 The training loss graph shows that adding physical constraints not only improves the effect and accuracy of deep learning model inversion, but also the underground dielectric constant value of termite nests in earth-rock dams is very similar to the actual value, which satisfies the physical interpretability of underground electromagnetic wave propagation. In summary, the ground penetrating radar inversion method driven by the joint model and physical information constraints shows excellent results. It can not only adapt to complex geological structures, but also effectively deal with underground noise interference. This method provides an innovative technical path for the detection of termite nests in earth-rock dams, further improves the accuracy and reliability of dam monitoring, and provides a solid guarantee for the long-term safety and stable operation of the dam.

[0155] On the one hand, the present invention aims at the complexity of underground models and possible noise problems. By establishing a complex geological model of termite nests in earth-rock dams and using data with added noise and random medium disturbance, a deep learning inversion model with strong noise resistance and strong applicability can be constructed, which can effectively extract the waveform characteristics of termite nests in earth-rock dams and ensure the accurate detection and positioning of termite nests in earth-rock dams. On the other hand, by integrating physical information constraints into the model in an appropriate way, the physical consistency of the inversion results is effectively improved, ensuring the inversion accuracy and physical interpretability in the detection of termite nests in earth-rock dams.

[0156] The present invention adopts the U-Net framework in the constructed deep learning model, and introduces the multi-scale cascade convolution module (MSC) and the efficient channel attention mechanism module (ECA), which not only improves the inversion effect of the model, but also improves the anti-interference ability of the model. The physical information constraint introduction method of the present invention adopts the prediction results based on the electric field and dielectric constant, introduces the physical residual of the electromagnetic wave propagation equation, and ensures that the prediction results meet the physical laws. The present invention provides an innovative solution for the application of GPR in the field of ant nest detection in earth-rock dams. This method not only improves the accuracy of ant nest detection, but also enables intelligent monitoring and risk warning of dam structures, providing technical guarantees for the long-term safety and stable operation of dams.

Claims

1. A termite nest detection method for earth-rock dam based on MSCCEAUNet and physical constraints, characterized in that: The following steps are involved: Step 1: Simulate the complex geological model of the earth-rock dam structure with termite nests, and use the finite difference method FDTD to generate the corresponding b-scan data and underground dielectric constant data as the simulation data set; Step 2: Construct the MSCCEAUNet network with multi-scale cascade convolution MSC and efficient channel attention mechanism ECA, and input the complex geological model data simulated and generated by step 1 into the MSCCEAUNet network for training; Step 3: Add Gaussian noise and random medium disturbance data with different signal-to-noise ratios to train MSCCEAUNet again. According to the training status of the model, optimize the MSCCEAUNet structure to improve the anti-interference ability and robustness of the model; Step 4: Incorporate the physical residual equations that satisfy the physical information constraints of Maxwell's equations for ground penetrating radar into the MSCCEAUNet framework as a physical loss function to achieve joint data and physics driven inversion; Step 5: Add Gaussian noise with different signal-to-noise ratios and random medium perturbation data again for data and physics joint driven training, and determine the optimal weights of the data loss, physical loss and structural similarity loss of MSCCEAUNet.

2. The method according to claim 1, characterized in that In step 1, the gprMax simulation software is used to simulate the earth-rock dam structure model with termite nests.

3. The method according to claim 1, characterized in that In step 4, when jointly driving the model and physical information constraints, the following steps are taken: Step 4-1) First, the Maxwell equations for electric field and magnetic field propagation satisfying the ground penetrating radar electromagnetic wave propagation equation are decoupled; The Max equation used is as follows: In the formula, represents the magnetic field curl, represents the curl of the electric field, ε0 represents the dielectric constant, J represents the current density, and μ0 represents the magnetic permeability; Step 4-2) Combining the decoupled Maxwell equations with the predicted electric field and dielectric constant to construct a physical residual formula that satisfies the electromagnetic wave propagation equation; The constructed physical residual formula is: Where E is the electric field, σ is the conductivity, ε is the dielectric constant, μ is the magnetic permeability, S E is the electric field source term; Step 4-3) Integrate the constructed physical residual formula into the MSCCEAUNet framework as the physical loss function to achieve the joint-driven inversion of data and physics, and ensure that the total loss of the MSCCEAUNet model is minimized.

4. The method according to claim 1, characterized in that In step 5, the following steps are used to determine the optimal weights of the losses of each part of MSCCEAUNet: Step 5-1) Using the hyperparameter optimization method, the optimal weights of the data loss, physical loss and structural similarity loss of MSCCEAUNet are automatically selected to ensure that the loss of each part is minimized; The constructed training loss function formula is as follows: L total =αL data +βL physical +γL SSIM ; In the formula, α refers to the weight ratio of data loss, β refers to the weight ratio of physical loss, and γ refers to the weight ratio of structural similarity loss; L total is the data loss, L physical is the loss of the physical equation, L SSIM is the structural similarity loss; Step 5-2) After determining the optimal weight ratio of each part of the loss, set the weight parameters of each part of MSCCEAUNet to the optimal starting data and physical joint drive training.

5. The method according to claim 1, characterized in that In step 2, the constructed MSCECAUNet model is specifically: The MSCECAUNet network includes the first Transformer Encoder MSC layer to the fifth Transformer Encoder MSC layer, the first Transformer Encoder layer to the fifth Transformer Encoder layer, the first ECA layer to the fourth ECA layer, the first Transformer Decoder MSC layer to the fifth Transformer Decoder MSC layer, the first Transformer Decoder layer to the fifth Transformer Decoder layer, and the FPC module; The GPR B-Scan profile is input as an input feature to the first Transformer Encoder MSC layer, the output of the first Transformer Encoder MSC layer is connected to the input of the first Transformer Encoder, the output of the first Transformer Encoder is connected to the input of the first ECA layer and the input of the second Transformer Encoder MSC layer, the output of the first ECA layer and the output of the second Transformer Decoder are connected to the input of the first Transformer Decoder MSC layer, the output of the first Transformer Decoder MSC layer is connected to the input of the first Transformer Decoder, the output of the first Transformer Decoder is connected to the input of the FPC module, and the output of the FPC module is a dielectric constant distribution map; The output of the second Transformer Encoder MSC layer is connected to the input of the second Transformer Encoder layer, the output of the second Transformer Encoder layer is connected to the input of the second ECA layer and the input of the third Transformer Encoder MSC layer, the output of the second ECA layer and the output of the third Transformer Decoder layer are connected to the input of the second Transformer Decoder MSC layer, and the output of the second Transformer Decoder MSC layer is connected to the input of the second Transformer Decoder layer; The output of the third Transformer Encoder MSC layer is connected to the input of the third Transformer Encoder layer, the output of the third Transformer Encoder layer is connected to the input of the third ECA layer and the input of the fourth Transformer Encoder MSC layer, the output of the third ECA layer and the output of the fourth Transformer Decoder layer are connected to the input of the third Transformer Decoder MSC layer, and the output of the third Transformer Decoder MSC layer is connected to the input of the third Transformer Decoder layer; The output of the fourth Transformer Encoder MSC layer is connected to the input of the fourth Transformer Encoder layer, the output of the fourth Transformer Encoder layer is connected to the input of the fourth ECA layer and the input of the fifth Transformer Encoder MSC layer, the output of the fourth ECA layer and the output of the fifth Transformer Decoder layer are connected to the input of the fourth Transformer Decoder MSC layer, and the output of the fourth Transformer Decoder MSC layer is connected to the input of the fourth Transformer Decoder layer; The output of the fifth Transformer Encoder MSC layer is connected to the input of the fifth Transformer Encoder layer, the output of the fifth Transformer Encoder layer is connected to the input of the fifth Transformer Decoder MSC layer, and the output of the fifth Transformer Decoder MSC layer is connected to the input of the fifth Transformer Decoder layer.

6. The method according to claim 5, characterized in that The TransformerEncoder MSC layer of the MSCECAUNet model includes a single-core convolutional neural network layer SK-ConvNet, a three-dimensional convolutional neural network layer 3D-ConvNet, a two-level cascaded convolutional neural network layer C2-ConvNet, a three-level cascaded convolutional neural network layer C3-ConvNet, and an FFM module; The input features are respectively input to the SK-ConvNet, 3D-ConvNet, C2-ConvNet and C3-ConvNet layers, and then the outputs of the above convolutional neural network layers are connected to the input of the FFM module, the output of the FFM module is connected to the input of the 3D-ConvNet, and the output of the 3D-ConvNet is connected to the input of the Kth Transformer Encoder layer of the MSCECAUNet model; where K represents the number of layers of the MSCECAUNet model.

7. The method according to claim 5, characterized in that The ECA layer of the MSCECAUNet model includes the input feature module IFM, the global average pooling module GAPM, the channel number adaptive K value module CKM, the K×K one-dimensional convolution module K×K1D Conv, the Sigmoid activation function module SAFM, and the output feature module OFM; The output of the Transformer Encoder layer is connected to the input of the ECA layer and input into the input feature module IFM. The output of the input feature module IFM is connected to the input of the global average pooling module GAPM. The output of GAPM is connected to the input of the channel number adaptive K value module CKM. The output of CKM is connected to the input of the K×K one-dimensional convolution module K×K 1D Conv. The output of K×K 1DConv is connected to the input of the Sigmoid activation function module SAFM. The output of SAFM is connected to the input of the output feature module OFM. The output of OFM is connected to the input of the Kth Transformer Decoder MSC layer.

Images