Intelligent imaging method, system, device and storage medium for water conveyance tunnel diseases

Abstract

The present application discloses a method, system, device and storage medium for intelligent imaging of water tunnel diseases, wherein the method comprises the following steps: obtaining ground penetrating radar image data, attention area imaging map and dielectric constant distribution map of the water tunnel; inputting the ground penetrating radar image data and attention area imaging map into a deep learning network to obtain a fused image; determining the target loss function of the deep learning network according to the dielectric constant distribution map and the fused image; adjusting the weight parameters of the deep learning network and returning to the step of "inputting the ground penetrating radar image data and attention area imaging map into the deep learning network to obtain a fused image" until the target loss function is the minimum value, and obtaining a trained intelligent imaging model for water tunnel diseases; inputting the ground penetrating radar image data collected in real time into the trained intelligent imaging model for water tunnel diseases to obtain intelligent imaging of water tunnel diseases. The present application can be widely used in the field of artificial intelligence technology.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, system, device and storage medium for intelligent imaging of water tunnel defects. Background Art

[0002] A water tunnel is an underground tunnel structure that can be used to transport and guide water. During long-term use, water tunnels are eroded and corroded by water flow, and defects such as cavities or hollowing will appear inside their linings. Defects in water tunnels will cause the performance of water tunnels to decline, and may even threaten the structural safety of water tunnels. Therefore, it is necessary to conduct regular defect inspections on water tunnels to ensure timely repair of defects. At present, ground penetrating radar is commonly used to detect defects inside concrete, but the interpretation of ground penetrating radar detection data (B-scan images) of this method relies on professional technicians, resulting in low efficiency. Moreover, the currently commonly used target detection algorithm can only give the radar reflection wave position of the defect, and cannot detect the location, size, and range of the defect. Therefore, there are still technical problems that need to be solved in the relevant technology. Summary of the invention

[0003] The purpose of this application is to solve one of the technical problems existing in the prior art to at least a certain extent.

[0004] To this end, an object of an embodiment of the present application is to provide a method, system, device and storage medium for intelligent imaging of water supply tunnel defects, which can improve imaging efficiency and improve the accuracy and efficiency of defect detection.

[0005] In order to achieve the above technical objectives, the technical solutions adopted by the embodiments of the present application include: a method for intelligent imaging of water conveyance tunnel defects, including: obtaining a simulation data set, the simulation data set including ground penetrating radar image data of the water conveyance tunnel, an imaging map of the attention area, and a dielectric constant distribution map;

[0006] Inputting the ground penetrating radar image data and the attention area imaging map into a deep learning network to obtain a fused image;

[0007] Determining a target loss function of the deep learning network according to the dielectric constant distribution map and the fused image;

[0008] Adjust the weight parameters of the deep learning network and return to the step of "inputting the ground penetrating radar image data and the attention area imaging map into the deep learning network to obtain a fused image" until the target loss function is a minimum value, thereby obtaining a trained water tunnel disease intelligent imaging model;

[0009] The ground penetrating radar image data collected in real time is input into the trained intelligent imaging model of water conveyance tunnel diseases to obtain intelligent imaging of water conveyance tunnel diseases.

[0010] In addition, the intelligent imaging method for water conveyance tunnel defects according to the above embodiment of the present invention may also have the following additional technical features:

[0011] Furthermore, in an embodiment of the present application, the attention area imaging diagram is obtained by the following steps:

[0012] Determining a local entropy and a local standard deviation of the ground penetrating radar image data;

[0013] An attention area imaging map is determined according to the local entropy and the local standard deviation.

[0014] Furthermore, in the embodiment of the present application, the step of determining the local entropy of the ground penetrating radar image data specifically includes:

[0015] Dividing the ground penetrating radar image data into a plurality of windows, and calculating a histogram of pixel values ​​of each window;

[0016] normalizing the histogram to a probability density;

[0017] Determining an entropy value for each window according to the probability density;

[0018] The entropy value of each window is merged into the same matrix as the ground penetrating radar image data to obtain the local entropy of the ground penetrating radar image data.

[0019] Furthermore, in the embodiment of the present application, determining the local standard deviation of the ground penetrating radar image data specifically includes:

[0020] Segmenting the ground penetrating radar image data into a plurality of windows and calculating a histogram of pixel values ​​of each window;

[0021] normalizing the histogram to a probability density;

[0022] Determining a standard deviation of each window based on the probability density;

[0023] The standard deviation of each window is merged into the same matrix as the ground penetrating radar image data to obtain the local standard deviation of the ground penetrating radar image data.

[0024] Further, in the embodiment of the present application, the step of determining the attention area imaging map according to the local entropy and the local standard deviation specifically includes:

[0025] The local entropy and the local standard deviation are input into the attention calculation formula to obtain the attention area imaging; wherein the attention calculation formula is:

[0026] S=ω e ·H+ω s σ;

[0027] Among them, ω e and ω s are the weights of local entropy and local standard deviation respectively, H is the local entropy, and σ is the local standard deviation.

[0028] Further, in the embodiment of the present application, the step of determining the target loss function of the deep learning network according to the dielectric constant distribution map and the fused image specifically includes:

[0029] Extracting a first output vector of each pixel output by a convolution decoder branch in the fused image, a second output vector of each pixel output by a transformer branch decoder, and a third output vector of each pixel of the dielectric constant distribution map;

[0030] Determine a first loss function according to the first output vector, the second output vector, and the third output vector;

[0031] Determine a second loss function according to the first output vector and the second output vector;

[0032] A target loss function is determined according to the first loss function and the second loss function.

[0033] Further, in the embodiment of the present application, determining the first loss function according to the first output vector, the second output vector, and the third output vector includes:

[0034] Determine a first absolute error loss according to the first output vector and the third output vector;

[0035] Determine a second absolute error loss according to the second output vector and the third output vector;

[0036] A first loss function is determined according to the first absolute error loss and the second absolute error loss.

[0037] On the other hand, the embodiment of the present application also provides a water conveyance tunnel disease intelligent imaging system, including:

[0038] An acquisition unit, used to acquire a simulation data set, wherein the simulation data set includes ground penetrating radar image data of the water conveyance tunnel, an imaging map of the attention area, and a dielectric constant distribution map;

[0039] A first processing unit, configured to input the ground penetrating radar image data and the attention area imaging map into a deep learning network to obtain a fused image;

[0040] A second processing unit, configured to determine a target loss function of the deep learning network according to the dielectric constant distribution map and the fused image;

[0041] The third processing unit is used to adjust the weight parameters of the deep learning network and return to the step of "inputting the ground penetrating radar image data and the attention area imaging map into the deep learning network to obtain a fused image" until the target loss function is a minimum value, thereby obtaining a trained water tunnel disease intelligent imaging model;

[0042] The fourth processing unit is used to input the ground penetrating radar image data collected in real time into the trained water conveyance tunnel disease intelligent imaging model to obtain the water conveyance tunnel disease intelligent imaging.

[0043] On the other hand, the present application also provides a water conveyance tunnel disease intelligent imaging device, comprising:

[0044] at least one processor;

[0045] at least one memory for storing at least one program;

[0046] When the at least one program is executed by the at least one processor, the at least one processor implements a method for intelligent imaging of water transfer tunnel diseases as described in any one of the invention contents.

[0047] In addition, the present application also provides a storage medium, which stores processor-executable instructions. When the processor-executable instructions are executed by the processor, they are used to execute a water tunnel disease intelligent imaging method as described in any of the above items.

[0048] The advantages and benefits of the present application will be partially given in the following description, and partially become apparent from the following description, or be understood through the practice of the present application:

[0049] The present application can input the ground penetrating radar image data and the attention area imaging map into a deep learning network to obtain a fused image;

[0050] Determining a target loss function of the deep learning network according to the dielectric constant distribution map and the fused image;

[0051] Adjust the weight parameters of the deep learning network and return to the step of "inputting the ground penetrating radar image data and the attention area imaging map into the deep learning network to obtain a fused image" until the target loss function is the minimum value, and obtain a trained water tunnel disease intelligent imaging model. Finally, the ground penetrating radar image data can be directly converted into water tunnel disease intelligent imaging through the trained model. The present application can obtain the corresponding water tunnel disease intelligent imaging through the artificial intelligence algorithm, reduce manual operation and improve efficiency. At the same time, the artificial intelligence algorithm can improve the accuracy of tunnel disease detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A schematic diagram of the steps of a method for intelligent imaging of water conveyance tunnel diseases in a specific embodiment of the present invention;

[0053] Figure 2 A schematic diagram of the steps of obtaining an attention area imaging diagram in a specific embodiment of the present invention;

[0054] Figure 3 A schematic diagram of the steps of determining the local entropy of ground penetrating radar image data in a specific embodiment of the present invention;

[0055] Figure 4 A schematic diagram of the steps of determining the local standard deviation of ground penetrating radar image data in a specific embodiment of the present invention;

[0056] Figure 5 A schematic diagram of the steps of determining the target loss function of the deep learning network according to the dielectric constant distribution map and the fused image in a specific embodiment of the present invention;

[0057] Figure 6 A schematic diagram of a step of determining a first loss function according to a first output vector, a second output vector, and a third output vector in a specific embodiment of the present invention;

[0058] Figure 7 A flow chart of a method for intelligent imaging of water supply tunnel defects in another specific embodiment of the present invention;

[0059] Figure 8 A network structure diagram of a neural network in a specific embodiment of the present invention;

[0060] Fig. 9 It is a schematic diagram of an image of network training data in a specific embodiment of the present invention;

[0061] Fig.10 It is a schematic diagram of an image of measured data imaging in a specific embodiment of the present invention;

[0062] Fig.11It is a structural schematic diagram of an intelligent imaging system for water conveyance tunnel diseases in a specific embodiment of the present invention;

[0063] Fig.12 The present invention is a schematic diagram of an intelligent imaging system for water conveyance tunnel diseases in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0064] The embodiments of the present invention are described in detail below with reference to the accompanying drawings to explain the principles and processes of the water conveyance tunnel disease intelligent imaging method, system, device and storage medium in the embodiments of the present invention.

[0065] First, the defects in the related art are explained:

[0066] A water tunnel is an underground tunnel structure used to transport and guide water. During long-term use, water tunnels are eroded and corroded by water flow, and defects such as cavities and hollows will appear inside their linings. Defects in water tunnels will cause the performance of water tunnels to decline, and may even threaten the structural safety of water tunnels. Therefore, it is necessary to conduct regular defect inspections on water tunnels to ensure timely repair of defects. In related technologies, ground penetrating radar is often used to detect defects inside concrete, but the interpretation of ground penetrating radar detection data (B-scan images) relies on professional technicians, which is inefficient. The deep learning-based target detection algorithm commonly used in related technologies can only give the radar reflection wave position of the defect, but cannot give the location, size, and range of the defect. The existing related technology uses the ground penetrating radar inversion method based on convolutional neural network, which can give the location, size and range of the disease. However, due to its single feature extraction path, it only relies on convolution to extract features from the image, and cannot capture various details and complex patterns in the image. In addition, it has poor robustness to noise, and thus cannot accurately give information such as the location, size, and range of the disease.

[0067] In view of the defects of the above-mentioned related technologies, refer to Figure 1 The present application provides a method for intelligent imaging of water supply tunnel defects. The method may include steps S101 to S105.

[0068] S101, obtaining a simulation data set, where the simulation data set includes ground penetrating radar image data of a water diversion tunnel, an imaging map of an attention area, and a dielectric constant distribution map.

[0069] S102: input the ground penetrating radar image data and the attention area imaging map into a deep learning network to obtain a fused image.

[0070] S103. Determine the target loss function of the deep learning network according to the dielectric constant distribution map and the fused image.

[0071] S104, adjusting the weight parameters of the deep learning network and returning to the step of "inputting the ground penetrating radar image data and the attention area imaging map into the deep learning network to obtain a fused image" until the target loss function is minimized, thereby obtaining a trained intelligent imaging model for water diversion tunnel diseases.

[0072] S105, inputting the ground penetrating radar image data collected in real time into the trained intelligent imaging model of water conveyance tunnel diseases to obtain intelligent imaging of water conveyance tunnel diseases.

[0073] It is understood that the simulation data set may include ground penetrating radar image data of the water tunnel, an imaging map of the attention area, and a dielectric constant distribution map. Moreover, the simulation data set may be a collection of historical data used to train an artificial intelligence model.

[0074] In some feasible embodiments of the present application, the processor can obtain a simulation data set from a database. The simulation data set can include ground penetrating radar image data, attention area imaging map and dielectric constant distribution map of the water conveyance tunnel. Then the ground penetrating radar image data and the attention area imaging map are input into the deep learning network to obtain a fused image. The fused image can be transmitted in the model in the form of a vector. Then the processor can determine the target loss function of the deep learning network according to the dielectric constant distribution map and the fused image. The weight parameters of the deep learning network are adjusted according to the target and return to step "S102 inputs the ground penetrating radar image data and the attention area imaging map into the deep learning network to obtain a fused image, and step S103, according to the dielectric constant distribution map and the fused image, determines the target loss function of the deep learning network" to perform a new round of training until the target loss function is the minimum value. After the loss function is obtained to be the minimum value, the model training is stopped. The processor can obtain a trained intelligent imaging model of water conveyance tunnel diseases. Finally, the ground penetrating radar image data collected in real time is input into the trained intelligent imaging model of water conveyance tunnel diseases to obtain intelligent imaging of water conveyance tunnel diseases.

[0075] Further, refer to Figure 2 , the attention area imaging map is obtained through steps S201-S202.

[0076] S201, determining the local entropy and local standard deviation of the ground penetrating radar image data.

[0077] S202: Determine an attention area imaging map according to the local entropy and the local standard deviation.

[0078] In some feasible embodiments of the present application, the processor may determine the local entropy and local standard deviation of the ground penetrating radar image data according to the ground penetrating radar image data, and then the processor may further determine the attention area imaging map according to the local entropy and local standard deviation.

[0079] Further, refer to Figure 3 The step of determining the local entropy of the ground penetrating radar image data specifically includes steps S301 to S304.

[0080] S301, dividing the ground penetrating radar image data into a plurality of windows, and calculating a histogram of pixel values ​​of each window.

[0081] S302: Normalize the histogram to probability density.

[0082] S303: Determine the entropy value of each window according to the probability density.

[0083] S304 , merging the entropy value of each window into the same matrix as the ground penetrating radar image data to obtain the local entropy of the ground penetrating radar image data.

[0084] Further, refer to Figure 4 The step of determining the local standard deviation of the ground penetrating radar image data may specifically include steps S401 to S404.

[0085] S401, dividing the ground penetrating radar image data into a plurality of windows and calculating a histogram of pixel values ​​of each window.

[0086] S402: Normalize the histogram to probability density.

[0087] S403: Determine the standard deviation of each window according to the probability density.

[0088] S404 , merging the standard deviation of each window into the same matrix as the ground penetrating radar image data to obtain the local standard deviation of the ground penetrating radar image data.

[0089] Furthermore, the step of determining the attention area imaging map according to the local entropy and the local standard deviation specifically includes:

[0090] Input the local entropy and local standard deviation into the attention calculation formula to obtain the attention area imaging. The attention calculation formula is:

[0091] S=ω e ·H+ω s ·σ.

[0092] Among them, ω e and ω s are the weights of local entropy and local standard deviation respectively, H is the local entropy, and σ is the local standard deviation.

[0093] Further, refer to Figure 5 , according to the dielectric constant distribution map and the fused image, the step of determining the target loss function of the deep learning network may specifically include S501-S504.

[0094] S501, extracting a first output vector of each pixel output by a convolution decoder branch in a fused image, a second output vector of each pixel output by a transformer branch decoder, and a third output vector of each pixel of a dielectric constant distribution map.

[0095] S502: Determine a first loss function according to the first output vector, the second output vector, and the third output vector.

[0096] S503: Determine a second loss function according to the first output vector and the second output vector.

[0097] S504: Determine a target loss function according to the first loss function and the second loss function.

[0098] Further, refer to Figure 6 The step of determining the first loss function according to the first output vector, the second output vector and the third output vector may include steps S601 to S603.

[0099] S601 . Determine a first absolute error loss according to the first output vector and the third output vector.

[0100] S602: Determine a second absolute error loss according to the second output vector and the third output vector.

[0101] S603: Determine a first loss function according to the first absolute error loss and the second absolute error loss.

[0102] The specific calculation principle of this application is explained below with reference to the accompanying drawings:

[0103] The imaging method of this embodiment can refer to Figure 7 .exist Figure 7 In the method, the method may include the following steps:

[0104] Acquire a simulation data set, where the data set is calculated by a finite difference time domain algorithm, and the simulation data set includes a radar profile and a dielectric constant distribution image;

[0105] The ground penetrating radar data is processed and local attention imaging is performed on the ground penetrating radar data using the local information entropy algorithm and the local coefficient of variation algorithm;

[0106] The attention area image and the original radar image are fused and input into the deep learning network, and the network double-branch output is performed for fusion imaging;

[0107] The output of the network is compared with the actual dielectric constant distribution map to obtain the loss function and update the network;

[0108] Imaging of water tunnel defects based on radar detection data collected in the field

[0109] The deep learning model for dual-branch water tunnel disease imaging is composed of a dual-branch encoder-decoder. Figure 8 As shown in Table 1, the structure has seven parts, namely, convolution input module, convolution branch encoder, transformer branch encoder, encoder feature coupler, convolution branch decoder, transformer branch decoder and output imaging module. Among them, the function of the convolution input module is to input the processed ground penetrating radar image data and the local attention imaging data calculated by the local information entropy algorithm and the local coefficient of variation algorithm into the network and perform feature fusion. The basic composition of the convolution input module is the convolution-batch normalization-activation function (convolution-batchnorm-ReLU, CBR) module, in which the convolution kernel size of each CBR module is 3 and the stride is 1.

[0110] First, the two types of input data are subjected to a CBR operation respectively. After the two data of size (H, W, 3) are subjected to a CBR operation respectively, the output is (H, W, 16). The two outputs are concatenated to obtain a tensor of size (H, W, 32).

[0111] Then the tensor is downsampled by the maximum pooling operation, and the output after downsampling is (H / 2, W / 2, 32). Finally, the feature is subjected to a CBR operation to obtain the fused input feature of the network, with a size of (H / 2, W / 2, 64).

[0112] The fused features obtained by the convolution input module are input into the encoders of the two branches. The role of the convolution branch encoder is to gradually reduce the spatial resolution of the input features while increasing the number of features to extract higher-level features and semantic information. The encoder extracts multi-scale features of the input image through a series of convolution and pooling operations, providing rich feature information for the subsequent upsampling and reconstruction process of the decoder.

[0113] The convolutional branch encoder consists of a series of convolutional modules and downsampling modules. The convolutional module contains two consecutive CBR modules. The function of the convolutional layer is to extract features and enable the network to express more complex features. Each convolutional module is followed by a downsampling module to reduce the spatial resolution of the feature map while retaining important features.

[0114] In the dual-branch deep learning structure, the convolutional branch encoder is responsible for extracting local features, which include image edges, textures and other information.

[0115] The fused features obtained by the convolution input module are input into the transformer branch encoder. The function of the transformer branch encoder is also to gradually reduce the spatial resolution of the input features while increasing the number of features to extract higher-level features and semantic information.

[0116] The basic unit of the Transformer branch encoder is the transformer module. In the transformer module, the input is first subjected to layer normalization, which normalizes the feature dimension of each sample. Specifically, it normalizes the feature dimension of each sample so that the mean of each sample is 0 and the variance is 1. The calculation process is:

[0117]

[0118] In the above formula, γ and β are learnable parameters corresponding to each feature dimension of the input; ε is a small constant used to prevent division by zero; μ and σ are the mean and variance of the input x, respectively.

[0119] After layer normalization, the features are input into the multi-head self-attention module, where each attention head first transforms the input into query Q, key K and value V vectors through linear transformation:

[0120] Q=XW Q ;

[0121] K=XW K ;

[0122] V=XW V ;

[0123] Where X is the matrix representation of the input vector. Then the dot product is used to calculate the similarity between the query and the key to get the attention score

[0124]

[0125] in is a scaling factor used to stabilize the gradient. Multiple attention heads perform multiple self-attention operations in parallel, and then concatenate the outputs of these heads and obtain the final output through linear transformation.

[0126] Feature x is self-attentioned to obtain the attention output y and attention weight, and then the attention output y is regularized and added to x to form the first layer of residual connection, and then the result of the first layer of residual connection is layer normalized and passed into a multilayer perception (MLP) to obtain the MLP output, and then the MLP output is regularized and added to the output of the previous layer to form the second layer of residual connection. Finally, the output x and attention weight are returned.

[0127] The transformer branch encoder consists of a series of transformer modules and downsampling modules. The role of the transformer module is to extract features so that the network can express more complex features. Each transformer module is followed by a downsampling module to reduce the spatial resolution of the feature map while retaining important features.

[0128] In the dual-branch deep learning structure, the transformer branch can extract global information, including overall shape, relationship between objects, global texture, etc.

[0129] In the dual-branch deep learning structure, the convolution branch calculates the features of the local area by sliding the convolution kernel on the input features, while the self-attention mechanism in the transformer module can calculate the dependency between each position and all other positions, so that the features of each position include not only its own information, but also the relationship information with other positions, thereby obtaining global features. An encoder feature coupler is set between the two branches of the dual-branch structure. The function of the coupler is to fuse the features of the convolution branch encoder and the transformer branch encoder, so that the model can use both local and global information.

[0130] The encoder feature coupler consists of a convolution module, an up / down sampling module, a normalization module and an activation function module, and is interconnected between the convolution branch encoder and the transformer branch encoder: the features extracted by the convolution branch encoder are connected to the transformer module of the next layer in the transformer branch through the convolution module, the down sampling module, the normalization module and the activation function module; the features extracted by the transformer branch encoder are connected to the convolution module of the previous layer in the convolution branch through the convolution module, the up sampling module, the normalization module and the activation function module.

[0131] In the dual-branch deep learning model, the role of the decoder is to gradually restore the features compressed and extracted by the encoder, and finally generate an output image of the same size as the input. In the convolutional branch encoder, after each layer extracts features through a CBS module, it performs an upsampling to gradually restore the low-resolution features to the resolution of the original input. And in each layer of the CBS module, it will receive the feature map from the corresponding layer of the encoder for splicing. This splicing allows the decoder to not only rely on the results of its own layer-by-layer upsampling, but also combine the feature information of different levels in the encoder stage, so as to better restore the image details.

[0132] In the transformer branch decoder, the extracted features are processed and restored through the transformer module, and high-resolution output is generated by continuous upsampling, ultimately generating an output image of the same size as the input image.

[0133] The convolution output module uses a two-dimensional convolution layer to perform two-dimensional imaging on the output results of the transformer branch decoder and the convolution branch decoder respectively, and dynamically combines the results obtained by the two branch decoders.

[0134] Out=α×y CNN +β×y trans ,α,β∈(0,1)

[0135] In the above formula, α, β are the binding coefficients, y CNN ,y trans They are the outputs of the convolutional branch decoder and the transformer branch decoder respectively.

[0136]

[0137]

[0138]

[0139] Table 1. Network layers and input and output characteristics

[0140] In this embodiment, the ground penetrating radar wave can be simulated by the finite difference time domain method. The time domain form of Maxwell's equations is as follows

[0141]

[0142] Where E is the electric field intensity, H is the magnetic field intensity, μ is the magnetic permeability of the medium, ∈ is the dielectric constant of the medium, and σ is the conductivity of the medium. Discretize space and events, with the spatial discretization step size being Δx, Δy, Δz, and the event discretization step size being Δt. The electric and magnetic fields are represented as E(i,j,k,n) and H(i,j,k,n) at discrete grid points. Discretize the Maxwell equations using the difference format to obtain the difference equations for updating the electric and magnetic fields. Take the update of the magnetic field and electric field in the x direction as an example:

[0143]

[0144]

[0145] In this embodiment, when simulating ground penetrating radar waves, the Ricker waveform is used as the excitation source, and an electric field is applied at a specific time step and spatial position. As the time step increases, the electric field and magnetic field are updated at each time step, and the propagation of electromagnetic waves in the medium is obtained by recording the electric field and magnetic field at each time step.

[0146] When simulating the GPR wave, randomly generate cavity defects of random size, shape and position in the model, and use the finite difference time domain method to perform forward simulation on each model to generate the corresponding radar profile. At the same time, the radar image is subjected to attention region imaging calculation. Attention region imaging highlights the important feature areas in the image by calculating the local entropy and local standard deviation. Areas with high local entropy mean that the area has a large amount of information (variable), while areas with high local standard deviations indicate that the area has high contrast or edge information. These features can help identify the key parts of the image, and by calculating the attention score, some small noise in the GPR image can be smoothed out while retaining larger changes.

[0147] Specifically, the steps of local attention area imaging calculation are:

[0148] First, the original image is cut into windows of size window_size×window_size. Then, for each small window W ij , calculate the histogram of its pixel values ​​and normalize it to probability density. Assume that there are N pixel values ​​in the window, and its histogram is expressed by the following formula:

[0149]

[0150] Among them, p k is the probability density of the k-th pixel value, count(x k ) is the pixel value x k The number of occurrences in the window. The entropy in the window is calculated using the Shannon entropy formula. For a window Wij , whose entropy H ij The calculation formula is:

[0151]

[0152] Finally, the entropy values ​​of all windows are merged into a matrix of the same size as the original image to obtain the local entropy image.

[0153] At the same time, when calculating the local entropy value, each small window W is also calculated ij The standard deviation is calculated as:

[0154]

[0155] In the above formula, μ i,j is the mean of the pixel values ​​in the window, σ i,j is the standard deviation. Finally, the standard deviation values ​​of all windows are merged into a matrix of the same size as the original image to obtain a local standard deviation image.

[0156] After obtaining the local entropy and local standard deviation of the radar data, attention area imaging is performed. The attention calculation formula is:

[0157] S=ω e ·H+ω s ·σ

[0158] In the above formula, ω e and ω s are the weights of local entropy and local standard deviation, H is the local entropy, and σ is the local standard deviation. Fig. 9 Shown is an attention area imaging image, where the highlighted area is the imaging area that requires more attention.

[0159] During the calculation process of the neural network, the output of the neural network is compared with the actual dielectric constant distribution map to obtain the loss function. The neural network updates the network parameters according to the loss function. The updated parameters can be weights or network structures. The neural network has a dual-branch decoder, so there will be a convolution decoder branch output and a transformer branch decoder output. The dual-branch outputs are compared with the actual dielectric constant distribution map and the interpolation is added and averaged to obtain the first loss function:

[0160]

[0161] In the above formula, L1(y CNN ,gt) and L1(y trans ,gt) are L1 losses, that is, absolute error losses, and their calculation formula is:

[0162]

[0163] At the same time, based on the difference in receptive fields between the convolutional branch and the transformer branch, pseudo labels are generated by dynamically mixing the output prediction results of the two branches, and then the two predictions are supervised separately. The pseudo labels are obtained by mixing the output results of the two branches in random proportions:

[0164] Y=α×y CNN +β×y trans ,α,β∈(0,1)

[0165]

[0166] The final loss function of the network is:

[0167] L total =λ1×L(gt,y CNN ,y trans )+λ2×L(Y,y CNN ,y trans )

[0168] In the above formula, λ1 and λ2 are coefficients, which can be used to adjust the proportion of the two types of loss functions.

[0169] Finally, the measured images are collected in the water diversion tunnel and input into the trained deep learning network to obtain the following Fig.10 Imaging results are shown.

[0170] In summary, this application provides a deep learning model with a dual-branch structure. This solution fully learns the information in the radar detection data through the dual-branch encoding-decoding structure, and realizes automatic imaging of the defects in the tunnel. This method has high imaging accuracy and strong anti-interference ability, as well as fast processing speed.

[0171] In addition, refer to Fig.11 ,and Figure 1Corresponding to the method, an embodiment of the present application also provides a water tunnel disease intelligent imaging system. The system may include an acquisition unit 1001, a first processing unit 1002, a second processing unit 1003, a third processing unit 1004 and a fourth processing unit 1005. Among them, the acquisition unit 1001 can be used to acquire a simulation data set, which includes ground penetrating radar image data, attention area imaging map and dielectric constant distribution map of the water conveyance tunnel; the first processing unit 1002 can be used to input the ground penetrating radar image data and the attention area imaging map into the deep learning network to obtain a fused image; the second processing unit 1003 can be used to determine the target loss function of the deep learning network according to the dielectric constant distribution map and the fused image; the third processing unit 1004 can be used to adjust the weight parameters of the deep learning network and return to the step of "inputting the ground penetrating radar image data and the attention area imaging map into the deep learning network to obtain a fused image" until the target loss function is the minimum value, and a trained water conveyance tunnel disease intelligent imaging model is obtained; the fourth processing unit 1005 can be used to input the ground penetrating radar image data collected in real time into the trained water conveyance tunnel disease intelligent imaging model to obtain water conveyance tunnel disease intelligent imaging.

[0172] It should be noted that the contents of the above-mentioned embodiments of the intelligent imaging method for water conveyance tunnel defects are applicable to the embodiments of the present intelligent imaging system for water conveyance tunnel defects. The functions specifically implemented by the embodiments of the present intelligent imaging system for water conveyance tunnel defects are the same as those of the above-mentioned embodiments of the intelligent imaging method for water conveyance tunnel defects, and the beneficial effects achieved are also the same as those achieved by the above-mentioned embodiments of the intelligent imaging method for water conveyance tunnel defects.

[0173] and Figure 1 Corresponding to the method, the embodiment of the present application also provides a water conveyance tunnel disease intelligent imaging device, the specific structure of which can be referred to Fig.12 ,include:

[0174] at least one processor 1011;

[0175] At least one memory 1012, used to store at least one program;

[0176] When the at least one program is executed by the at least one processor, the at least one processor implements the intelligent imaging method for water transfer tunnel diseases.

[0177] It should be noted that the contents of the above method embodiments are all applicable to the present device embodiments, the functions specifically implemented by the present device embodiments 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.

[0178] and Figure 1 Corresponding to the method, an embodiment of the present application further provides a storage medium, in which processor-executable instructions are stored, and the processor-executable instructions are used to execute the water conveyance tunnel disease intelligent imaging method when executed by the processor.

[0179] It should be noted that the contents of the above-mentioned embodiments of the intelligent imaging method for water conveyance tunnel defects are all applicable to the embodiments of this storage medium. The functions specifically implemented by this storage medium embodiment are the same as those in the above-mentioned embodiments of the intelligent imaging method for water conveyance tunnel defects, and the beneficial effects achieved are also the same as those achieved by the above-mentioned embodiments of the intelligent imaging method for water conveyance tunnel defects.

[0180] In some optional embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the application is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Optional embodiments are expected, wherein the order of various operations is changed and the sub-operation described as a part of a larger operation is performed independently.

[0181] In addition, although the present application is described in the context of functional modules, it should be understood that, unless otherwise specified, one or more of the functions and / or features can be integrated into a single physical device and / or software module, or one or more functions and / or features can be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present application. More specifically, in view of the properties, functions, and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional techniques of the engineer. Therefore, those skilled in the art can implement the present application set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present application, which is determined by the full scope of the attached claims and their equivalents.

[0182] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several programs to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0183] The logic and / or steps represented in the flowchart or otherwise described herein, for example, may be considered as an ordered list of executable programs for implementing the logical functions, and may be embodied in any computer-readable medium for use by a program execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch and execute a program from a program execution system, device or apparatus), or in conjunction with such program execution systems, devices or apparatuses. For purposes of this specification, a "computer-readable medium" may be any device that can contain, store, communicate, propagate or transmit a program for use by a program execution system, device or apparatus, or in conjunction with such program execution systems, devices or apparatuses.

[0184] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0185] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable program execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0186] In the above description of this specification, the description with reference to the terms "one embodiment / example", "another embodiment / example" or "certain embodiments / examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0187] Although the embodiments of the present application have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present application, and that the scope of the present application is defined by the claims and their equivalents.

[0188] The above is a specific description of the preferred implementation of the present application, but the present application is not limited to the described embodiments. Technical personnel familiar with the field may make various equivalent modifications or substitutions without violating the spirit of the present application. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present application.

Claims

1. A method for intelligent imaging of water supply tunnel defects, characterized in that: include: A simulation data set is obtained, wherein the simulation data set includes ground penetrating radar image data of a water conveyance tunnel, an imaging map of an attention area, and a dielectric constant distribution map; the imaging map of the attention area is obtained by the following steps: Determining a local entropy and a local standard deviation of the ground penetrating radar image data; Determine an attention area imaging map according to the local entropy and the local standard deviation; Inputting the ground penetrating radar image data and the attention area imaging map into a deep learning network to obtain a fused image; Determining the target loss function of the deep learning network according to the dielectric constant distribution map and the fused image; determining the target loss function of the deep learning network according to the dielectric constant distribution map and the fused image includes: Extracting a first output vector of each pixel output by a convolution decoder branch in the fused image, a second output vector of each pixel output by a transformer branch decoder, and a third output vector of each pixel of the dielectric constant distribution map; Determine a first loss function according to the first output vector, the second output vector, and the third output vector; Determine a second loss function according to the first output vector and the second output vector; Determine a target loss function according to the first loss function and the second loss function; Adjust the weight parameters of the deep learning network and return to the step of "inputting the ground penetrating radar image data and the attention area imaging map into the deep learning network to obtain a fused image" until the target loss function is the minimum value, thereby obtaining a trained intelligent imaging model for water tunnel diseases; The ground penetrating radar image data collected in real time is input into the trained intelligent imaging model of water conveyance tunnel diseases to obtain intelligent imaging of water conveyance tunnel diseases.

2. According to claim 1, a method for intelligent imaging of water supply tunnel diseases is characterized in that: Determining the local entropy of the ground penetrating radar image data includes: Dividing the ground penetrating radar image data into a plurality of windows, and calculating a histogram of pixel values ​​of each window; normalizing the histogram to a probability density; Determining an entropy value for each window according to the probability density; The entropy value of each window is merged into the same matrix as the ground penetrating radar image data to obtain the local entropy of the ground penetrating radar image data.

3. According to claim 1, a method for intelligent imaging of water supply tunnel diseases is characterized in that: Determining a local standard deviation of the ground penetrating radar image data includes: Segmenting the ground penetrating radar image data into a plurality of windows and calculating a histogram of pixel values ​​of each window; normalizing the histogram to a probability density; Determining a standard deviation of each window based on the probability density; The standard deviation of each window is merged into the same matrix as the ground penetrating radar image data to obtain the local standard deviation of the ground penetrating radar image data.

4. According to claim 1, a method for intelligent imaging of water supply tunnel diseases is characterized in that: Determining the attention area imaging map according to the local entropy and the local standard deviation includes: The local entropy and the local standard deviation are input into the attention calculation formula to obtain the attention area imaging; wherein the attention calculation formula is: ； in, and are the weights of local entropy and local standard deviation, respectively. is the local entropy, is the local standard deviation.

5. According to claim 1, a method for intelligent imaging of water supply tunnel diseases is characterized in that: The determining a first loss function according to the first output vector, the second output vector, and the third output vector comprises: Determine a first absolute error loss according to the first output vector and the third output vector; Determine a second absolute error loss according to the second output vector and the third output vector; A first loss function is determined according to the first absolute error loss and the second absolute error loss.

6. An intelligent imaging system for water supply tunnel diseases, characterized in that: include: The acquisition unit is used to acquire a simulation data set, wherein the simulation data set includes ground penetrating radar image data of the water conveyance tunnel, an imaging map of the attention area, and a dielectric constant distribution map; the imaging map of the attention area is obtained by the following steps: Determining a local entropy and a local standard deviation of the ground penetrating radar image data; Determine an attention area imaging map according to the local entropy and the local standard deviation; A first processing unit, configured to input the ground penetrating radar image data and the attention area imaging map into a deep learning network to obtain a fused image; The second processing unit is configured to determine a target loss function of the deep learning network according to the dielectric constant distribution map and the fused image; wherein the determining the target loss function of the deep learning network according to the dielectric constant distribution map and the fused image comprises: Extracting a first output vector of each pixel output by a convolution decoder branch in the fused image, a second output vector of each pixel output by a transformer branch decoder, and a third output vector of each pixel of the dielectric constant distribution map; Determine a first loss function according to the first output vector, the second output vector, and the third output vector; Determine a second loss function according to the first output vector and the second output vector; Determine a target loss function according to the first loss function and the second loss function; The third processing unit is used to adjust the weight parameters of the deep learning network and return to the step of "inputting the ground penetrating radar image data and the attention area imaging map into the deep learning network to obtain a fused image" until the target loss function is a minimum value, thereby obtaining a trained water tunnel disease intelligent imaging model; The fourth processing unit is used to input the ground penetrating radar image data collected in real time into the trained water conveyance tunnel disease intelligent imaging model to obtain the water conveyance tunnel disease intelligent imaging.

7. An intelligent imaging device for water tunnel diseases, characterized in that include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the intelligent imaging method for water conveyance tunnel diseases as described in any one of claims 1 to 5.

8. A storage medium storing instructions executable by a processor, characterized in that: The processor-executable instructions are used to execute the intelligent imaging method for water conveyance tunnel defects as described in any one of claims 1 to 5 when executed by the processor.

Images