PEC signal grounding grid multi-parameter detection method based on improved multi-polarization feature extraction aggregation network

Abstract

The invention relates to a PEC signal grounding grid multi-parameter detection method based on an improved multi-polarization feature extraction aggregation network, and belongs to the field of grounding grid nondestructive detection. The method comprises the following steps: acquiring a pulsed eddy current effect time domain signal; an improved multi-polarization feature extraction aggregation network is established for grounding grid parameter estimation, firstly, a multi-layer convolutional neural network is used for performing local feature extraction on an input signal oscillogram, and nonlinearity is introduced through an activation function; fusing the local features by using a feature aggregation module; carrying out feature splicing on all features extracted for multiple times; a plurality of parallel branches are designed for a plurality of parameters of the grounding grid, and each branch is fused by introducing a mixed attention mechanism module; and finally, introducing a graph neural network to integrate correlation information among target parameters, performing regression calculation through a full connection layer, and outputting a parameter estimation value. According to the method, three grounding grid steel bar parameters can be calculated at the same time under the condition that the adverse effect of parameter correlation is reduced.

Description

Technical Field

[0001] The present invention belongs to the field of non-destructive testing of grounding grids, and relates to a multi-parameter detection method for PEC signal grounding grids based on an improved multi-polarization feature extraction and aggregation network. Background Art

[0002] Pulsed eddy current (PEC) technology has been applied to a certain extent in the detection and evaluation of grounding grid structures. Determining the depth, radius, and direction angle of steel bars in a grounding grid is one of the key factors for construction quality and structural health assessment. However, there is a strong correlation between the PEC signals detected by the pulsed eddy current detection coil and the grounding grid steel bar parameters, which makes it a major challenge to detect multiple grounding grid parameters simultaneously.

[0003] Since multiple grounding grid steel bar related parameters jointly affect the waveform characteristics of the PEC signal, establishing a non-linear relationship between the waveform characteristics of the PEC signal and multiple grounding grid steel bar related parameters is a possible solution. Deep learning algorithms have powerful feature extraction and non-linear learning capabilities and are very suitable for this challenging task.

[0004] The current other grounding grid detections also have the following defects:

[0005] 1. Currently, the parameter measurement of the grounding grid is often carried out by direct excavation. This method has a large cost and a long time period, which is not conducive to the rapid detection of the corrosion depth of the grounding grid.

[0006] 2. The main method for grounding grid parameter detection is the ground penetrating radar method. By emitting high-frequency electromagnetic waves into the ground and relying on the reflection of electromagnetic waves between different materials to construct an image of underground objects. However, the spatial resolution of the ground penetrating radar is limited. Especially in complex structures or environments with dense steel bars, it may be difficult to accurately distinguish the specific position and radius of the steel bars.

[0007] 3. Some scholars suggest using electromagnetic detection methods, especially using transient electromagnetic methods to detect and quantify the parameters of the grounding grid. It generates and measures the change of the secondary magnetic field by applying a pulsed magnetic field, sets parameters using a forward model and simulates signals, and extracts characteristic parameters to estimate the position, depth, radius, and azimuth deviation of the grounding grid steel bars. However, the transient electromagnetic technology is affected by soil conductivity and environmental interference, and the accuracy and reliability may be reduced.

[0008] 4. Other detection techniques such as infrared thermal imaging detection and ultrasonic detection can detect the grounding grid parameters to a certain extent, but each has its own limitations: infrared thermal imaging detection is mainly used for grounding grid corrosion imaging, and the detection accuracy is relatively low; the penetration performance of ultrasonic detection in metal conductors is limited, and it is difficult to meet the requirements of high-precision detection for grounding grids with complex structures. Summary of the Invention

[0009] In view of this, the purpose of the present invention is to provide a multi-parameter detection method for grounding grids based on PEC signals using an improved multi-polarimetric feature extraction and aggregation network. The improved multi-polarimetric aggregation network, called IMPA-Net (Improved Multi-Polarimetric Aggregation Network), can simultaneously estimate multiple grounding grid reinforcement-related parameters from the PEC detection signal waveform diagram, including the reinforcement burial depth (d), radius (r), and direction angle (φ). IMPA-Net takes the PEC received signal waveform diagram as input, extracts the information features of the grounding grid reinforcement reflection signal, reduces the adverse effects of parameter correlation, and calculates three grounding grid reinforcement parameters simultaneously.

[0010] To achieve the above purpose, the present invention provides the following technical solutions:

[0011] A multi-parameter detection method for grounding grids based on PEC signals using an improved multi-polarimetric feature extraction and aggregation network, the method comprising:

[0012] S1. Obtain the time-domain signal of the pulsed eddy current effect, wherein the characteristic quantities in the time-domain signal of the pulsed eddy current effect include pulse width, pulse peak value, pulse valley value, peak time, and secondary zero-crossing time;

[0013] S2. Establish an improved multi-polarimetric feature extraction and aggregation network for grounding grid parameter estimation, wherein the improved multi-polarimetric feature extraction and aggregation network includes a multi-polarimetric feature extraction and aggregation module and a parameter estimation module, specifically including the following steps:

[0014] S21. First, in the multi-polarimetric feature extraction and aggregation module, use a multi-layer convolutional neural network to perform local feature extraction on the input signal waveform diagram, and introduce non-linearity through an activation function;

[0015] S22. Fuse the local features using the feature aggregation module FA;

[0016] S23. After processing by the activation function and the feature aggregation module with several different features, splice all the extracted features;

[0017] S24. In the parameter estimation module, a number of parallel branches are designed for several parameters of the grounding grid respectively, and each branch is fused by introducing a hybrid attention mechanism module HA;

[0018] S25. Finally, a graph neural network is introduced to integrate the correlation information between target parameters, and regression calculation is performed through a fully connected layer to output the parameter estimation value.

[0019] Further, in step S1, when acquiring the time-domain signal of the pulsed eddy current effect, a pulse generator, an electromagnetic probe, and a data acquisition system are set up in sequence; among them, when setting up the pulse generator, it includes setting the pulse width and frequency; when setting up the electromagnetic probe, it includes installing an excitation coil, a detection coil, and adjusting the lift-off distance; when setting up the data acquisition system, it includes setting its sampling rate;

[0020] After generating a transient electromagnetic field through a trigger pulse excitation, the probe detects the eddy current decay signal in the grounding grid conductor. After pre-amplification and band-pass filtering processing, the time-domain waveform is recorded by a high-speed acquisition card, and finally the characteristic signal is extracted through digital filtering.

[0021] Further, let the distance between the conductor to be measured and the excitation coil be d, and the radius of the excitation coil be r b , and the low-frequency pulsed alternating current applied is I 1 , then a decaying magnetic field B that rapidly decays along the rising edge of the square wave is obtained 1 , and when the decaying magnetic field B 1 contacts the conductor to be measured, a transient eddy current I is generated in the conductor to be measured 2 and an eddy current magnetic field B opposite to the direction of the initial magnetic field generated by the transient eddy current I 2 is generated. Thus, according to Faraday's law of electromagnetic induction, the voltage value at the detection point is obtained: 2

[0022]

[0023] In the formula, V p is the coil voltage value at the detection point, B is the induced magnetic field strength, ||A|| is the induced magnetic field vector norm, S is the cross-sectional area of the detection coil, and c represents the integration path; represents the partial derivative with respect to time t, which is used to describe the instantaneous change rate of the magnetic field or magnetic flux; represents the unit normal vector, that is, the normalized normal vector, and its direction is perpendicular to the plane S.

[0024] Further, in step S21, the convolutional layer uses a weight sharing mechanism to extract the local features of the time-domain signal;

[0025]

[0026] ​The activation function uses the Leaky ReLU activation function, which is expressed as:

[0027] Leaky ReLU(x) = max(αx, x)

[0028] Among them, x i,j is the pixel value of the input PEC signal waveform diagram, is the weight value of the k-th convolution kernel, b k is the bias value of the k-th convolution kernel, f is the activation function Leaky ReLU, x represents the input value of the neurons in this layer, and α is the negative slope coefficient.

[0029] Furthermore, in step S22, the feature aggregation module FA combines the complementary information of the multi-polarization signals and generates a global feature representation using the weighted sum operation of the convolution kernels:

[0030]

[0031] Among them, F k is the input feature of the k-th PEC signal, ω k represents the weight coefficient of the k-th input feature, F agg is the global feature after aggregation of a certain branch.

[0032] Furthermore, in step S23, the feature obtained by concatenating the corresponding features respectively obtained according to the pulse width, pulse peak, pulse valley, peak time, and secondary zero-crossing time is:

[0033]

[0034] Among them, F global represents the feature vector after concatenating the corresponding features, Concatenate(·) is the concatenation function, is the global feature after aggregation of the n-th branch.

[0035] Furthermore, in step S24, the parameter estimation module combines the channel attention mechanism FA and the spatial attention mechanism SA, and the two analyze the importance of the feature map from the channel and spatial dimensions respectively; the fused feature is obtained through the hybrid attention mechanism HA:

[0036] f c = σ(W 2 ·ReLU(W 1 ·F global ))

[0037] f s = σ(Conv 1×1 (F global ))

[0038] FHA = f c ·F global + f s ·F global

[0039] Wherein, W 1 W 2 is the weight matrix, σ represents the Leaky ReLU activation function, f c represents the channel weight, Conv 1×1 represents the 1×1 convolution operation, f s represents the spatial weight, F HA is the fusion feature of the hybrid attention mechanism under double weighted fusion.

[0040] Furthermore, the multi-parameters of the grounding grid include the steel bar burial depth, the steel bar radius, and the steel bar direction angle, and three corresponding fusion branches are designed accordingly.

[0041] Furthermore, the processing processes of the graph neural network GNN and the fully connected layer FC are respectively expressed as:

[0042]

[0043] y = W FC F GNN + b FC

[0044] Wherein, H (l) represents the node feature matrix of the l-th layer, represents the adjacency matrix A of the graph plus the self-loop, that is where I is the identity matrix, represents 's degree matrix, W (l) represents the learnable weight matrix of the l-th layer, σ represents the activation function Leaky ReLU, H (l+1) represents the node feature matrix of the (l + 1)-th layer, the feature vector of the i-th node of the L-th layer, F GNN represents the final output of the graph neural network, N represents the total number of nodes in the graph, W FC is the corresponding weight matrix, b FC is the corresponding bias vector, and y is the final output vector corresponding to the estimated values of the burial depth, radius, and direction angle of the grounding grid steel bars respectively.

[0045] The beneficial effects of the present invention are as follows:

[0046] The present invention extracts the characteristics of the PEC signal waveform diagram under different parameters of the grounding grid through pulsed eddy current testing, and proposes an improved multi-polarization aggregation network for detecting the parameters of the steel bars in the grounding grid. The results show that the mean absolute errors (MAEs) of the steel bar burial depth, radius, and direction are 0.20 cm, 0.13 cm, and 2.54°, respectively, proving that the network can accurately estimate multiple grounding grid steel bar related parameters simultaneously.

[0047] The present invention improves the accuracy of multi-parameter detection of the steel bars in the grounding grid. By improving the MPA-Net model, a spatial attention mechanism is added and combined with the channel attention mechanism to form a hybrid attention mechanism. This dual weighting strategy can more comprehensively capture the relevant features of the target parameters, improve the network's ability to analyze multi-dimensional features, and ultimately enhance the accuracy and stability of parameter estimation. At the same time, a graph neural network model is introduced based on the correlation between the target parameters. The results show that the model improves the ability to analyze complex signal features, making it show stronger robustness and adaptability in the face of highly coupled signal scenarios.

[0048] The present invention can be extended to the detection occasions of other systems. By simply replacing the test piece to be detected, other parameters of the test piece can be detected accurately without excavation and damage.

[0049] Other advantages, objectives, and features of the present invention will be described to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the following specification. Description of the Drawings

[0050] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be described in detail preferably with reference to the accompanying drawings, where:

[0051] Figure 1 is the overall flowchart of the PEC signal grounding grid multi-parameter detection method based on the improved multi-polarization feature extraction aggregation network of the present invention;

[0052] Figure 2 is the schematic diagram of the acquisition principle of the time-domain signal of the pulsed eddy current effect of the present invention;

[0053] Figure 3 is the waveform diagram of the time-domain signal of the pulsed eddy current effect obtained by the present invention;

[0054] Figure 4 is the structural schematic diagram of the improved multi-polarization feature extraction aggregation network of the present invention;

[0055] Figure 5This is a histogram of the estimation error of the relevant parameters of the grounding grid reinforcement bars of the present invention. Specific implementation manners

[0056] The following uses specific specific examples to illustrate the implementation manners of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific implementation manners. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the drawings provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner. Without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0057] Among them, the drawings are only for illustrative purposes, showing only schematic diagrams, not physical diagrams, and cannot be understood as a limitation to the present invention; in order to better illustrate the embodiments of the present invention, some components in the drawings will be omitted, enlarged or reduced, and do not represent the dimensions of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0058] In the drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components; in the description of the present invention, it should be understood that if there are terms such as "upper", "lower", "left", "right", "front", "rear", etc. indicating the orientation or positional relationship, it is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, the terms describing the positional relationship in the drawings are only for illustrative purposes and cannot be understood as a limitation to the present invention. For those of ordinary skill in the art, the specific meanings of the above terms can be understood according to specific circumstances.

[0059] Please refer to Figures 1 to 5 , which is a multi-parameter detection method for the grounding grid of PEC signals based on an improved multi-polarization feature extraction and aggregation network.

[0060] The present invention proposes a method based on the waveform changes of the pulsed eddy current effect (PEC) time-domain signals caused by grounding grids with different parameters, and analyzes and estimates the parameters through deep learning technology. Research shows that the changes in the PEC time-domain signal waveform diagram are closely related to characteristic quantities such as pulse width, pulse peak value, pulse valley value, peak time, and secondary zero-crossing time. Therefore, this method uses the PEC time-domain signal waveform diagram as the input of the deep learning model, and identifies characteristic quantities such as pulse width, pulse peak value, pulse valley value, peak time, and secondary zero-crossing time in the image through image recognition technology to achieve accurate estimation of steel bar parameters. For this purpose, the present invention introduces a new deep learning model - the improved multi-polarization feature extraction and aggregation network (IMPA-Net). The network architecture of IMPA-Net is divided into two main modules: the multi-polarization feature extraction-aggregation module (MFEM) and the parameter estimation module (PEM). The MFEM module extracts features from each PEC detection signal waveform diagram, making full use of the complementary information of different steel bar parameters carried by different waveform diagrams. These polarization-dependent features are obtained through a multi-polarization aggregation method to obtain a more comprehensive and informative representation, thereby improving the robustness and accuracy of feature extraction. The PEM module designs a three-branch model for three relevant parameters of the grounding grid steel bars (burial depth, radius, and direction angle). Each branch enhances the features related to specific parameters by introducing a hybrid attention mechanism, thereby improving the estimation ability of these parameters. In addition, in order to further improve the estimation accuracy of each parameter, this method also introduces a graph neural network (GNN) mechanism, which reduces the correlation between parameters, enabling the model to more accurately estimate each parameter. As Figure 1 shown, the specific process is as follows:

[0061] S1. Obtain the time-domain signal of the pulsed eddy current effect;

[0062] When obtaining the time-domain signal of the pulsed eddy current effect, it is necessary to sequentially set the pulse generator (set the pulse width of 0.1 - 10 μs and the frequency of 25 Hz - 1 kHz), the electromagnetic probe (install the excitation / detection coil and adjust the lift-off distance of 0.1 - 5 mm), and the data acquisition system (sampling rate of 10 - 100 MS / s). After generating a transient electromagnetic field through pulsed excitation, the probe detects the eddy current decay signal in the grounding grid conductor, which is processed by pre-amplification and band-pass filtering, and the time-domain waveform is recorded by a high-speed acquisition card. Finally, the characteristic signal is extracted through digital filtering.

[0063] Assume that the distance between the conductor under test and the excitation coil is d, and the radius is r b The coil will generate a rapidly decaying magnetic field B along the square wave transition edge when excited by a low-frequency pulsed alternating current I 1 . When the decaying magnetic field contacts the conductor, a transient eddy current I is generated in the conductor 1 . When the decaying magnetic field contacts the conductor, a transient eddy current I is generated in the conductor . When the decaying magnetic field contacts the conductor, a transient eddy current I is generated in the conductor2 An eddy current magnetic field B opposite to the direction of the initial magnetic field will be generated. 2， However, due to the rapid weakening of the eddy current field and its interaction with the excitation magnetic field, the magnetic flux inside the coil will change, thereby changing the initial magnetic field. As the secondary magnetic field decays, an instantaneous voltage signal that changes with time can be induced in the detection coil. The schematic diagram of pulsed eddy current detection is as Figure 2 shown.

[0064] According to Faraday's law of electromagnetic induction, the voltage value at the detection point is obtained:

[0065]

[0066] where V p is the coil voltage value at the detection point, B is the induced magnetic field strength, ||A|| is the induced magnetic field vector norm, S is the cross-sectional area of the detection coil, represents the partial derivative with respect to time t and is used to describe the instantaneous change rate of the magnetic field or magnetic flux, t represents time, represents the unit normal vector (normalized normal vector), the direction is perpendicular to the plane S, and c represents the integration path. If the grounding grid parameters change, it will cause the induced magnetic field strength B to change, and ultimately cause the transient induced voltage on the detection coil to change. Since the pulse contains a very wide spectrum and the pulsed eddy current decays more slowly than a single-frequency sinusoidal eddy current, the transient induced voltage signal contains important information about the grounding grid parameters.

[0067] By collecting the voltage signal obtained from the detection coil and performing denoising processing, a PEC signal waveform diagram is obtained, as Figure 3 shown. The PEC signal waveform diagram can reflect characteristic quantities such as the pulse width, pulse peak, pulse valley, peak time, and secondary zero-crossing time of the signal.

[0068] S2. Establish an improved multi-polarization feature extraction and aggregation network for grounding grid parameter estimation. Among them, IMPA-Net includes a multi-polarization feature extraction and aggregation module (MFEM) and a parameter estimation module (PEM). Specifically, as Figure 4 shown, it includes the following specific steps:

[0069] S21. First, in the multi-polarization feature extraction and aggregation module, a multi-layer convolutional neural network CNN is used to perform local feature extraction on the input PEC signal waveform diagram, and the extracted features are introduced with non-linearity through an activation function. Specifically, the convolutional layer uses a weight sharing mechanism to extract local features of the time-domain signal;

[0070]

[0071] The activation function uses the Leaky ReLU activation function, which is expressed as:

[0072] Leaky ReLU(x) = max(αx, x)

[0073] where x i,j is the pixel value of the input PEC signal waveform diagram, is the weight value of the k-th convolutional kernel, b k is the bias value of the k-th convolutional kernel, f is the activation function Leaky ReLU, x represents the input value of the neurons in this layer, α is the negative slope coefficient, usually set to 0.01 to prevent negative-valued features from being completely ignored during propagation.

[0074] S22. Then, the local features are fused using the Feature Aggregation Module FA, which combines the complementary information of the multi-polarization signals and generates a global feature representation using the weighted sum operation of the convolutional kernels:

[0075]

[0076] where F k is the input feature of the k-th PEC signal, ω k represents the weight coefficient of the k-th input feature, and F agg is the global feature aggregated by a certain branch.

[0077] S23. After being processed by the activation functions and the feature aggregation module for several different features, all the extracted features are concatenated. Among them, the extracted features include the corresponding features obtained according to the pulse width, pulse peak, pulse valley, peak time, and secondary zero-crossing time respectively. The finally concatenated feature is:

[0078]

[0079] where F global represents the feature vector after concatenating the corresponding features.

[0080] S24. In the parameter estimation module, several parallel branches are designed for several parameters of the grounding grid respectively. Each branch is fused by introducing a Hybrid Attention mechanism (HA) module, which combines the Channel Attention mechanism (FA) and the Spatial Attention mechanism (SA). The two analyze the importance of the feature map from the channel and spatial dimensions respectively to improve the parsing quality of the target parameters; the fused feature is obtained through the Hybrid Attention mechanism HA:

[0081] f c = σ(W 2 ·ReLU(W 1 ·F global ))

[0082] f s = σ(Conv 1×1 (F global ))

[0083] F HA = f c ·F global + f s ·F global

[0084] where W 1 W 2 is the weight matrix, σ represents the Leaky ReLU activation function, f c represents the channel weight, Conv 1×1 represents the 1×1 convolution operation, f s represents the spatial weight, and F HA is the fused feature of the hybrid attention mechanism under double weighted fusion.

[0085] The channel attention mechanism compresses the spatial information of the input feature map, extracts the overall characteristics of each channel, performs non-linear transformation and weighted processing to generate channel importance scores, thereby amplifying relevant channel features and suppressing irrelevant channel features. The spatial attention mechanism aggregates the feature map in the channel dimension, extracts the global context information of each spatial position, generates a spatial weight map, enhances the features of key regions, and weakens the features of background or irrelevant regions. This double weighted strategy can capture the highly correlated features of target parameters more comprehensively, improve the multi-dimensional feature analysis ability of the network, and ultimately enhance the accuracy and stability of parameter estimation. Since there is usually a certain correlation between target parameters, capturing these correlations helps the network to perform more reasonable feature allocation and information sharing during the parameter estimation process.

[0086] In this embodiment, the multi-parameters of the grounding grid include the steel bar burial depth, steel bar radius, and steel bar direction angle, and three corresponding fusion branches are designed accordingly.

[0087] S25. Finally, a graph neural network (GNN) is introduced to effectively integrate the correlation information between target parameters, and regression calculation is performed through a fully connected layer (FC) to output the parameter estimation value, thereby realizing the collaborative estimation of each parameter.

[0088]

[0089] y = W FC F GNN + b FC ,

[0090] where H (l) represents the node feature matrix of the l-th layer. Denote the adjacency matrix A of the graph plus self-loops (i.e., where I is the identity matrix), Denote as the degree matrix of (l) , \(W^{ (l)}\) denote the learnable weight matrix of the \(l\)-th layer, \(\sigma\) denote the activation function Leaky ReLU, \(H^{ (l+1)}\) (l+1) denote the node feature matrix of the \((l + 1)\)-th layer, the feature vector of the \(i\)-th node in the \(L\)-th (last) layer, \(F^{ GNN}\) GNN denote the final output of the graph neural network, \(N\) denote the total number of nodes in the graph, \(W^{ FC}\) FC is the corresponding weight matrix, \(b^{ FC}\) FC is the corresponding bias vector, and \(y\) is the final output vector corresponding to the estimated values of the burial depth, radius, and orientation angle of the grounding grid steel bars respectively.

[0091] In this embodiment, the IMPA-Net model after training was used to estimate the grounding grid steel bar parameters of 500 sets of test data. The histogram of the estimation errors of the 3 grounding grid steel bar related parameters is as Figure 5 shown. The mean absolute errors (MAEs) of the steel bar burial depth, radius, and orientation angle in 500 sets of test data are 0.20 cm, 0.13 cm, and 2.54°, respectively. The relatively low MAEs in the test data prove that this network can accurately estimate multiple grounding grid steel bar related parameters simultaneously.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the purpose and scope of the present technical solution, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A multi-parameter detection method for PEC signal grounding grid based on an improved multi-polarization feature extraction aggregation network, characterized in that: The method comprises: S1. Acquire a pulse eddy current effect time domain signal, wherein the characteristic quantities in the pulse eddy current effect time domain signal include pulse width, pulse peak value, pulse valley value, peak time and secondary zero crossing time; S2. Establishing an improved multi-polarization feature extraction and aggregation network to estimate grounding grid parameters, wherein the improved multi-polarization feature extraction and aggregation network includes a multi-polarization feature extraction and aggregation module and a parameter estimation module, specifically including the following steps: S21. First, in the multi-polarization feature extraction and aggregation module, a multi-layer convolutional neural network is used to extract local features of the input signal waveform, and nonlinearity is introduced through an activation function; S22, fusing the local features using a feature aggregation module FA; S23, after being processed by activation functions of different features and feature aggregation modules for several times, all the extracted features are spliced; S24, in the parameter estimation module, a plurality of parallel branches are designed for a plurality of parameters of the grounding grid, and each branch is fused by introducing a hybrid attention mechanism module HA; S25. Finally, the graph neural network is introduced to integrate the correlation information between the target parameters, and regression calculation is performed through the fully connected layer to output the parameter estimation value.

2. According to claim 1, a PEC signal grounding grid multi-parameter detection method based on an improved multi-polarization feature extraction aggregation network is characterized in that: In step S1, when obtaining the pulse eddy current effect time domain signal, the pulse generator, the electromagnetic probe and the data acquisition system are set in sequence; wherein, when setting the pulse generator, it includes setting the pulse width and frequency; when setting the electromagnetic probe, it includes installing the excitation coil, the detection coil, and adjusting the lifting distance; when setting the data acquisition system, it includes setting its sampling rate; After the transient electromagnetic field is generated by trigger pulse excitation, the probe detects the eddy current attenuation signal in the grounding grid conductor. After pre-amplification and bandpass filtering, the time domain waveform is recorded by a high-speed acquisition card, and finally the characteristic signal is extracted through digital filtering.

3. A multi-parameter detection method for a PEC signal grounding grid based on an improved multi-polarization feature extraction aggregation network according to claim 2, characterized in that: Assume that the distance between the conductor under test and the excitation coil is d, and the radius of the excitation coil is r. b , the applied low-frequency pulse alternating current is I1, then a decaying magnetic field B1 that decays rapidly along the transition edge of the square wave is obtained, and when the decaying magnetic field B1 contacts the conductor under test, a transient eddy current I2 and an eddy current magnetic field B2 generated by the transient eddy current I2 in the opposite direction to the initial magnetic field are generated in the conductor under test, and thus the voltage value of the detection point is obtained according to Faraday's law of electromagnetic induction: Where V p is the coil voltage value at the detection point, B is the induced magnetic field intensity, ||A|| is the induced magnetic field vector norm, S is the cross-sectional area of ​​the detection coil, and c represents the integration path; represents the partial derivative with respect to time t, which is used to describe the instantaneous rate of change of the magnetic field or magnetic flux; Represents the unit normal vector, that is, the normalized normal vector, whose direction is perpendicular to the surface S.

4. The multi-parameter detection method of a PEC signal grounding grid based on an improved multi-polarization feature extraction aggregation network according to claim 1, characterized in that: In step S21, the convolution layer uses a weight sharing mechanism to extract local features of the time domain signal; The activation function uses the Leaky ReLU activation function, which is expressed as: Leaky ReLU(x)=max(αx,x) Among them, x i,j is the pixel value of the input PEC signal waveform, is the weight value of the kth convolution kernel, b k is the bias value of the kth convolution kernel, f is the activation function Leaky ReLU, x represents the input value of the neuron in this layer, and α is the negative slope coefficient.

5. The multi-parameter detection method of a PEC signal grounding grid based on an improved multi-polarization feature extraction aggregation network according to claim 4 is characterized in that: In step S22, the feature aggregation module FA combines the complementary information of the multi-polarization signals and generates a global feature representation using the weighted sum operation of the convolution kernel: Among them, F k is the kth PEC signal input feature, ω k represents the weight coefficient of the kth input feature, F agg It is the global feature after a branch is aggregated.

6. A multi-parameter detection method for PEC signal grounding grid based on improved multi-polarization feature extraction aggregation network according to claim 5, characterized in that: In step S23, the corresponding features obtained according to the pulse width, pulse peak value, pulse valley value, peak time and secondary zero crossing time are spliced ​​to obtain the following features: Among them, F global represents the feature vector after the corresponding features are concatenated, Concatenate(·) is the concatenation function, is the global feature after the aggregation of the nth branch.

7. The multi-parameter detection method of a PEC signal grounding grid based on an improved multi-polarization feature extraction aggregation network according to claim 6 is characterized in that: In step S24, the parameter estimation module combines the channel attention mechanism FA and the spatial attention mechanism SA, which analyze the importance of the feature map from the channel and spatial dimensions respectively; and obtains the fused feature through the hybrid attention mechanism HA: f c =σ(W2·ReLU(W1·F global )) f s =σ(Conv 1×1 (F global )) F HA =f c ·F global +f s ·F global Among them, W1W2 is the weight matrix, σ represents the Leaky ReLU activation function, and f c Represents channel weight, Conv 1×1 represents a 1×1 convolution operation, f s represents the spatial weight, F HA Fusion features of hybrid attention mechanism under double weighted fusion.

8. The multi-parameter detection method of a PEC signal grounding grid based on an improved multi-polarization feature extraction aggregation network according to claim 7, characterized in that: The multiple parameters of the grounding grid include the buried depth of the steel bars, the radius of the steel bars, and the direction angle of the steel bars, and three corresponding fused branches are designed based on these parameters.

9. The method for multi-parameter detection of a PEC signal grounding grid based on an improved multi-polarization feature extraction aggregation network according to claim 7, characterized in that: The processing of the graph neural network GNN and the fully connected layer FC are respectively expressed as: Among them, H (l) represents the node feature matrix of the lth layer, Represents the adjacency matrix A of the graph plus self-loops, that is, where I is the identity matrix, express The degree matrix, W (l) represents the learnable weight matrix of the lth layer, σ represents the activation function LeakyReLU, H (l+1) represents the feature matrix of the l+1th layer node, The feature vector of the i-th node in the L-th layer, F GNN represents the final output of the graph neural network, N represents the total number of nodes in the graph, and W FC is the corresponding weight matrix, b FC is the corresponding bias vector, and y is the final output vector corresponding to the estimated values ​​of the buried depth, radius and direction angle of the grounding grid steel bars.

Images