A method and system for karst geological detection of power transmission projects based on ground penetrating radar

Abstract

The present invention proposes a method and system for karst geological detection of power transmission projects based on ground penetrating radar, which relates to the field of geophysical ground penetrating radar detection technology, including: based on autocorrelated random media, establishing a target body model of karst geology, forward generating ground penetrating radar data, inverting and generating a dielectric constant map, and constructing a karst disaster data set; constructing an inversion network with ground penetrating radar data as input and a dielectric constant map as output, and using the karst disaster data set to perform ground penetrating radar source-independent deep learning that does not rely on wavelets on the inversion network; inverting the acquired measured ground penetrating radar data through the trained inversion network to predict the dielectric constant map of karst geology; the present invention realizes rapid inversion and attribute imaging of karst underground disasters of power transmission lines through the preparation of ground penetrating radar karst disaster data sets based on autocorrelated random media and ground penetrating radar source-independent deep learning inversion that does not rely on wavelets, thereby improving the accuracy of karst geological detection of power transmission projects.

Description

Technical Field

[0001] The present invention belongs to the technical field of geophysical ground penetrating radar detection, and in particular relates to a method and system for karst geological detection of power transmission engineering based on ground penetrating radar. Background Art

[0002] In the construction of power transmission and transformation projects, karst geology often affects the safety and stability of the construction process; ground penetrating radar (GPR) is a geophysical method that uses antennas to transmit and receive electromagnetic waves to detect underground media. As a disaster, underground karst geology in power transmission and transformation projects can be effectively surveyed using ground penetrating radar methods. In the data interpretation of ground penetrating radar, deep learning inversion is one of the effective means of imaging underground media. It constructs a data set by establishing an underground disaster library and a simulation response library of ground penetrating radar, and then achieves rapid inversion imaging through training.

[0003] However, there are many problems in using deep learning for karst disaster inversion: (1) Karst disasters have their own unique characteristics, including the typical characteristics of irregularity, smoothness, and random distribution. At present, it is impossible to perform effective morphological simulation and modeling for karst disasters to meet the data set requirements in deep learning inversion; (2) In actual ground penetrating radar detection, the shape of the actual emitted source wavelet is unknown. It is affected by the detection instrument, antenna form, and detection environment. In numerical simulation, it is usually impossible to establish a simulation data set based on the actual wavelet shape for inversion. In summary, the two problems of karst model simulation and wavelet morphology simulation seriously affect the accuracy of ground penetrating radar deep learning inversion. Summary of the invention

[0004] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a method and system for karst geological detection of power transmission projects based on ground penetrating radar. Through the preparation of a ground penetrating radar karst disaster data set based on autocorrelated random media and a ground penetrating radar source-independent deep learning inversion that does not rely on wavelets, rapid inversion and attribute imaging of karst underground disasters in transmission lines can be achieved, thereby improving the accuracy of karst geological detection of power transmission projects.

[0005] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0006] A first aspect of the present invention provides a method for karst geological detection of power transmission projects based on ground penetrating radar.

[0007] A karst geological detection method for power transmission engineering based on ground penetrating radar, comprising:

[0008] Based on the autocorrelated random medium, a target model of karst geology is established, the ground penetrating radar data of the target model is generated by forward modeling, the dielectric constant map of the target model is generated by inversion, and the karst disaster data set is constructed based on the generated ground penetrating radar data and the corresponding dielectric constant map;

[0009] An inversion network with GPR data as input and dielectric constant map as output is constructed, and the karst hazard dataset is used to perform wavelet-independent GPR source-independent deep learning on the inversion network;

[0010] The trained inversion network is used to invert the measured ground penetrating radar data and predict the dielectric constant map of karst geology.

[0011] Furthermore, the target body model of karst geology is established as follows:

[0012] Introduce the random number model as the initialization background of the random medium and construct the random medium model;

[0013] Autocorrelation filtering of random medium models;

[0014] According to a preset threshold, the filtered random medium model is divided into background medium and target medium to obtain an initial target body model;

[0015] Perform boundary processing and area cleaning on the initial target model;

[0016] The karst medium attributes are randomly assigned to the remaining areas, and the physical property parameters are randomly assigned to the background medium to obtain the final karst geological target model.

[0017] Furthermore, the forward modeling generates the ground penetrating radar data of the target body model by using the electromagnetic wave finite difference time domain method to perform numerical simulation based on the established karst geological target body model to generate the ground penetrating radar two-dimensional B-SCAN data;

[0018] The dielectric constant map of the target body model generated by inversion is converted into a two-dimensional dielectric constant map based on the established karst geological target body model and wave velocity calculation.

[0019] Furthermore, the inversion network includes an encoder and a decoder;

[0020] The encoder extracts a feature map from input ground penetrating radar data based on a spatial pyramid structure of multi-scale dilated convolution;

[0021] The decoder reconstructs the characteristic map into a dielectric constant map.

[0022] Furthermore, the method also includes: before the ground penetrating radar data is input into the inversion network, a cross convolution operation is performed.

[0023] Furthermore, when using the karst hazard dataset for deep learning, the direct wave reference track in the measured ground penetrating radar data is cross-convolved with the simulated ground penetrating radar data.

[0024] Furthermore, when inverting the acquired measured GPR data, the standard wavelet in the numerical simulation is cross-convolved with the measured GPR data.

[0025] A second aspect of the present invention provides a karst geological detection system for power transmission projects based on ground penetrating radar.

[0026] A karst geological detection system for power transmission engineering based on ground penetrating radar, including a data construction module, a network construction module and an inversion prediction module:

[0027] The data construction module is configured to: establish a target body model of karst geology based on autocorrelated random media, forward generate ground penetrating radar data of the target body model, inverse generate a dielectric constant map of the target body model, and construct a karst disaster data set based on the generated ground penetrating radar data and the corresponding dielectric constant map;

[0028] The network construction module is configured to: construct an inversion network with GPR data as input and dielectric constant map as output, and use the karst hazard dataset to perform wavelet-independent GPR source-independent deep learning on the inversion network;

[0029] The inversion prediction module is configured to: invert the acquired measured ground penetrating radar data through the trained inversion network, and predict the dielectric constant map of karst geology.

[0030] The third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of a method for karst geological detection of power transmission projects based on ground penetrating radar as described in the first aspect of the present invention.

[0031] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in the karst geological detection method for power transmission projects based on ground penetrating radar as described in the first aspect of the present invention are implemented.

[0032] One or more of the above technical solutions have the following beneficial effects:

[0033] The present invention proposes a method for preparing a ground penetrating radar karst hazard dataset and a ground penetrating radar source-independent deep learning inversion method; the dataset preparation method is based on the autocorrelation random medium modeling technology to establish a large number of extensive karst hazard datasets; the ground penetrating radar source-independent deep learning inversion method is based on cross convolution, and performs rapid inversion without relying on source wavelets, solving the problem that "in numerical simulation, it is usually impossible to establish a simulation dataset for inversion based on the real wavelet shape"; through the above method, rapid inversion and attribute imaging of karst underground disasters in transmission lines can be achieved.

[0034] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0036] Figure 1 This is a flow chart of the method of the first embodiment.

[0037] Figure 2 This is a flow chart of the autocorrelated random medium modeling method of the first embodiment.

[0038] Figure 3 Schematic diagram of the source-independent deep learning inversion of ground penetrating radar in the first embodiment. DETAILED DESCRIPTION

[0039] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0040] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0041] Embodiment 1

[0042] In one embodiment of the present disclosure, a method for karst geological detection of power transmission projects based on ground penetrating radar is provided. Figure 1 As shown, the following steps are included:

[0043] Step S1: Based on the autocorrelated random medium, a target model of karst geology is established, ground penetrating radar data of the target model is generated by forward modeling, a dielectric constant map of the target model is generated by inversion, and a karst disaster data set is constructed based on the generated ground penetrating radar data and the corresponding dielectric constant map.

[0044] Furthermore, the target body model of karst geology is established as follows:

[0045] Introduce the random number model as the initialization background of the random medium and construct the random medium model;

[0046] Autocorrelation filtering of random medium models;

[0047] According to a preset threshold, the filtered random medium model is divided into background medium and target medium to obtain an initial target body model;

[0048] Perform boundary processing and area cleaning on the initial target model;

[0049] The karst medium attributes are randomly assigned to the remaining areas, and the physical property parameters are randomly assigned to the background medium to obtain the final karst geological target model.

[0050] Furthermore, the forward modeling generates the ground penetrating radar data of the target body model by using the electromagnetic wave finite difference time domain method to perform numerical simulation based on the established karst geological target body model to generate the ground penetrating radar two-dimensional B-SCAN data;

[0051] The dielectric constant map of the target body model generated by inversion is converted into a two-dimensional dielectric constant map based on the established karst geological target body model and wave velocity calculation.

[0052] Step S2: construct an inversion network with GPR data as input and dielectric constant map as output, and use the karst disaster dataset to perform wavelet-independent GPR source-independent deep learning on the inversion network.

[0053] Furthermore, the inversion network includes an encoder and a decoder;

[0054] The encoder extracts a feature map from input ground penetrating radar data based on a spatial pyramid structure of multi-scale dilated convolution;

[0055] The decoder reconstructs the characteristic map into a dielectric constant map.

[0056] Step S3: Invert the acquired measured ground penetrating radar data through the trained inversion network to predict the dielectric constant map of karst geology.

[0057] Furthermore, the method also includes: before the ground penetrating radar data is input into the inversion network, a cross convolution operation is performed.

[0058] Furthermore, when using the karst hazard dataset for deep learning, the direct wave reference track in the measured ground penetrating radar data is cross-convolved with the simulated ground penetrating radar data.

[0059] Furthermore, when inverting the acquired measured GPR data, the standard wavelet in the numerical simulation is cross-convolved with the measured GPR data.

[0060] The following is a detailed description of the implementation process of a method for karst geological detection of power transmission projects based on ground penetrating radar in this embodiment.

[0061] This embodiment proposes a method for preparing a ground penetrating radar karst disaster dataset based on autocorrelated random media and a ground penetrating radar source-independent deep learning inversion method that does not rely on wavelets. Aiming at the ground penetrating radar inversion problem of karst disasters, firstly, a typical karst disaster database, i.e., a karst disaster dataset, is established by adopting an autocorrelated random medium modeling method. Then, the direct wave reference channel in the measured data is selected for cross-convolution with the simulation data, and the inversion network is trained using the cross-convolved data. Finally, the standard wavelet selected for numerical simulation is cross-convolved with the measured data, and the convolved data is input into the trained inversion network for prediction, thereby finally realizing the inversion imaging of underground karst disasters.

[0062] A more specific example is provided, which is divided into two parts: karst modeling and source-independent inversion. Specifically:

[0063] 1. Preparation method of ground penetrating radar karst hazard dataset based on autocorrelated random medium

[0064] The autocorrelated random medium modeling method is used to establish a karst hazard dataset for the training of the inversion network, such as Figure 2 As shown, the specific steps are:

[0065] Step 1: Create model dimensions.

[0066] The karst target body in the uniform rock background has the typical characteristics of irregularity, smoothness and random distribution. This embodiment models the tunnel karst geology based on the autocorrelation method, that is, the karst geological model: first, the size of the model is planned according to the actual karst geological environment (that is, the environment to be detected) and physical properties. The model size should meet the computing resource requirements of the server, and the grid size should meet the calculation stability conditions of the time domain finite difference method.

[0067] Step 2: Random noise initialization.

[0068] Introducing extensive random noise in the modeling to simulate the randomly distributed heterogeneous environment requires introducing a random number model of the same size as the model size as the initialization background of the random medium. The range of random numbers is -1~1, thus obtaining a random medium model.

[0069] Step 3: Autocorrelation coefficient control.

[0070] The autocorrelation method controls the roughness by changing the autocorrelation coefficient, thereby controlling the distribution form of the generated heterogeneous bodies. For the random medium model, the filtering autocorrelation coefficient is 30.

[0071] Step 4: Autocorrelation Gaussian filtering.

[0072] Only the Gaussian autocorrelation is applied to the initialized background for filtering and smoothing; due to the extensiveness of the data and the normal distribution of the initialized random numbers, the distribution range of the filtered random numbers should be within the range of 0.49~0.51.

[0073] Step 5: Threshold segmentation modeling.

[0074] In order to establish the target body model of karst geology, it is necessary to segment the filtered random medium model to distinguish the target medium from the background medium. In this embodiment, 0.503 is used as the threshold to act on the filtered random number to segment the model into a uniform background medium and a closed target medium.

[0075] Step 6: Boundary processing.

[0076] According to the actual detection environment, there should be no karst medium near the ground, so the medium near the ground is smoothed to remove the karst medium near the ground.

[0077] Step 7: Connected body marking.

[0078] Due to the randomness of the initialization noise, there are a certain number of tiny karst targets in the modeling results. In numerical simulation, tiny karst bodies will form obvious diffraction waves and multiple reflections, affecting the authenticity of the simulation data. At the same time, in actual engineering construction, tiny karst bodies have little impact on the project and can be ignored; therefore, it is necessary to remove these tiny karst target bodies. Therefore, the regional marking algorithm is used to mark all closed connected bodies in the model.

[0079] Step 8: Calculate the area of ​​connected bodies.

[0080] According to the pixels, the area of ​​each closed connected body is calculated, counted and marked.

[0081] Step 9: Clean the connecting body.

[0082] Connected bodies whose mesh area is smaller than the specified area are removed, and the specific form is to directly assign coverage processing to connected bodies whose area is smaller than the specified area.

[0083] Step 10: Generate random attributes.

[0084] According to the statistical connected body area and serial number coordinates, the karst medium attributes are randomly assigned, including two types: cavity caves and filled caves. The medium attributes of each cave are randomly distributed and arranged. The attributes of the cavity cave are the same as those of the air, and the electromagnetic wave speed of the filled cave is between 0.03m / ns and 0.06m / ns.

[0085] Step 11: Assign values ​​to connected body attributes.

[0086] Physical property parameters are assigned to each karst area connected body, and the physical property parameters are random values ​​generated in step ten.

[0087] Step 12: Assign background medium parameters.

[0088] The physical parameters are randomly assigned to the background medium, and the electromagnetic wave speed of the background medium is between 0.08m / ns and 0.12m / ns.

[0089] At this point, the target model of karst geology has been completed.

[0090] Step 13: Forward simulation.

[0091] According to the established karst geological target model, the electromagnetic wave time-domain finite difference method is used for numerical simulation to generate two-dimensional B-SCAN data of ground penetrating radar.

[0092] Step 14: Generate label data.

[0093] According to the established target body model of karst geology, the wave velocity is calculated and converted into a two-dimensional dielectric constant map as a label for the ground penetrating radar data.

[0094] Step 15: Generation of karst hazard dataset.

[0095] The forward simulated ground penetrating radar data and the corresponding dielectric constant map are combined into a karst hazard dataset, and the karst hazard dataset is preprocessed by normalization and normalized to the range of 0-1 by proportional scaling.

[0096] 2. Source-independent deep learning inversion method for ground penetrating radar without relying on wavelets

[0097] like Figure 3As shown in the figure, firstly, based on the generated karst disaster data set, the direct wave reference channel in the measured data and the simulated data in the karst disaster data set are selected for cross convolution, and the inversion network is trained with the cross-convolved data; then, the standard wavelet selected for numerical simulation is cross-convolved with the measured data, and the convolved data is input into the trained inversion network for prediction, and finally the inversion imaging of underground karst disasters is realized. The specific steps are as follows:

[0098] Step 1: Inversion network construction.

[0099] The constructed inversion network consists of an encoder and a decoder. The encoder is based on the spatial pyramid structure of multi-scale dilated convolution. Its function is to extract feature maps based on the input cross-convolution data. The function of the decoder is to reconstruct the feature maps into dielectric constant maps.

[0100] The encoder, for the input data, first, uses three dilated convolutions to convert it into three feature maps of different scales; then, a fused residual convolution block processes the feature map of the smallest scale and fuses it with the feature maps of other scales. The fused feature map is further processed by the residual convolution block; finally, the extracted feature map is generated. Its core feature is to apply the dilated convolution and spatial pyramid pooling modules to the feature map before the decoder to capture features of multiple scales.

[0101] Compared with traditional convolution, this network can increase or decrease the receptive field by adjusting the convolution expansion rate, capturing the neighborhood and global features of the feature map; through the combination of convolutions with multiple expansion rates, it can capture neighborhood information in multi-scale space, which is helpful for learning deep features.

[0102] Step 2: Parameter adjustment.

[0103] Adjust deep learning parameters such as learning rate, batchsize, number of training rounds, and training set allocation.

[0104] Step 3: Collect measured data.

[0105] This method has certain requirements for the data acquisition process. Based on the basic assumption of common-offset ground penetrating radar data, the spacing between the transmitting and receiving antennas should be assumed to be zero. Therefore, during data acquisition, the transmitting and receiving antennas should be self-excited and self-receiving antennas, or a ground penetrating radar antenna layout mode with a smaller transmitting and receiving distance should be adopted. At the same time, during the acquisition process, the moving step distance of the antenna should be kept the same.

[0106] Under the above settings, the antenna of the ground penetrating radar is used to transmit and receive electromagnetic waves to detect the underground medium and obtain the measured ground penetrating radar data.

[0107] Step 4: Measured data processing.

[0108] The measured ground penetrating radar data is subjected to preliminary data processing, including distance correction, depth correction, waste road elimination, bandpass filtering, etc.

[0109] Step 5: Non-negative matrix factorization.

[0110] The non-negative matrix decomposition method is used to separate the direct wave information and underground echo reflection information in the measured ground penetrating radar data. Here, the direct wave is regarded as a low-rank sparse matrix component, and other underground echo information is regarded as a high-rank non-sparse component. This method separates the stable direct wave by separating the low-rank sparse matrix components.

[0111] Step 6: Construct the extraction window function.

[0112] In order to effectively extract the effective reference trace signal from the measured data, a time window function W is introduced to extract the direct wave signal in the reference trace. The extraction window function needs to be constructed as follows:

[0113]

[0114] in, , are the extracted direct wave signal and time value respectively. and are the length and order of the time windows, respectively.

[0115] In practical applications, parameters are selected according to actual conditions to suppress noise and keep only direct waves. The length of the time window will affect the accuracy of the algorithm.

[0116] Step 7: Create an extraction window.

[0117] According to the constructed extraction window function, an extraction window is constructed, and the direct wave signal is separated through the extraction window.

[0118] Step 8: Reference track selection.

[0119] From the numerous direct wave signals separated from the detection data, select 5-10 reference channel direct wave signals with smooth waveforms and stable amplitudes.

[0120] Step 9: Reference trace extraction.

[0121] The direct waves are extracted from the selected reference channels, and the signal values ​​are averaged, and the average reference channel is used as the final reference channel.

[0122] Step 10: Reference track calibration.

[0123] The first arrival position of the reference track is corrected so that its amplitude starting point is as close as possible to the amplitude starting point of the simulation data; at the same time, the overall starting point of the field measured data is adjusted through the zero-point correction method. Like the reference track, the overall starting point of the field measured data is adjusted to be roughly the same as the amplitude starting point of the detection data.

[0124] Step 11: Cross convolution of reference channels.

[0125] The corrected reference track is cross-convolved with each simulated ground penetrating radar data in the karst hazard dataset to obtain the convolved data.

[0126] Step 12: Generate cross convolution training set.

[0127] The ground penetrating radar data after cross convolution of the reference channel and the original labels are combined to form a new dataset, which is divided into training set, test set and validation set according to the ratio of 8:1:1.

[0128] Step 13: Network training.

[0129] The constructed training set is input into the inversion network constructed in step one, and the network training is performed using the training parameters determined in step two.

[0130] Step 14: Network verification and training parameter saving.

[0131] For the trained network, reference trace extraction, cross convolution and prediction verification are performed on the karst disaster dataset to ensure the accuracy of the trained network and then save the parameters.

[0132] Step 15: Numerical simulation wavelet extraction.

[0133] The antenna receiving and transmitting distance is adjusted to a certain distance, and numerical simulation is performed under the background condition of air medium (conductivity = 0) to obtain the standard sub-wave in the numerical simulation.

[0134] Step 16: Cross convolution of real data.

[0135] The numerical simulation standard wavelet extracted in the previous step is cross-convolved with the measured ground penetrating radar data to obtain the convolved measured data.

[0136] Step 17: Inverse prediction.

[0137] The measured data after cross convolution in the previous step is input into the trained inversion network, and deep learning inversion prediction is performed without relying on source wavelets to obtain the dielectric constant image of the karst geological body.

[0138] Embodiment 2

[0139] In one embodiment of the present disclosure, a karst geological detection system for power transmission engineering based on ground penetrating radar is provided, including a data construction module, a network construction module and an inversion prediction module:

[0140] The data construction module is configured to: establish a target body model of karst geology based on autocorrelated random media, forward generate ground penetrating radar data of the target body model, inverse generate a dielectric constant map of the target body model, and construct a karst disaster data set based on the generated ground penetrating radar data and the corresponding dielectric constant map;

[0141] The network construction module is configured to: construct an inversion network with GPR data as input and dielectric constant map as output, and use the karst hazard dataset to perform wavelet-independent GPR source-independent deep learning on the inversion network;

[0142] The inversion prediction module is configured to: invert the acquired measured ground penetrating radar data through the trained inversion network, and predict the dielectric constant map of karst geology.

[0143] Embodiment 3

[0144] The purpose of this embodiment is to provide a computer-readable storage medium.

[0145] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in a method for karst geological detection of power transmission projects based on ground-penetrating radar as described in the first embodiment of the present disclosure.

[0146] Embodiment 4

[0147] The purpose of this embodiment is to provide an electronic device.

[0148] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in a method for karst geological detection of power transmission projects based on ground penetrating radar as described in the first embodiment of the present disclosure are implemented.

[0149] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for karst geological detection of power transmission engineering based on ground penetrating radar, characterized in that: include: Step S1: Based on the autocorrelated random medium, a target model of karst geology is established, ground penetrating radar data of the target model is generated by forward modeling, a dielectric constant map of the target model is generated by inversion, and a karst disaster data set is constructed based on the generated ground penetrating radar data and the corresponding dielectric constant map; Step S2: construct an inversion network with GPR data as input and dielectric constant map as output, and use the karst hazard dataset to perform wavelet-independent GPR source-independent deep learning on the inversion network; Step S3: inverting the acquired measured ground penetrating radar data through the trained inversion network to predict the dielectric constant map of karst geology; Before the ground penetrating radar data is input into the inversion network, a cross convolution operation is performed; in the training stage, the cross convolution operation is to cross-convolute the calibrated reference track with each simulated ground penetrating radar data in the karst disaster data set, and after the training is completed, the extracted numerical simulation standard wavelet is cross-convolved with the measured ground penetrating radar data; The specific extraction method of the calibrated reference track is as follows: (1) Perform non-negative matrix decomposition on the measured ground penetrating radar data to separate the direct wave information and underground echo reflection information in the measured ground penetrating radar data; (2) Construct an extraction window function W, construct an extraction window, and extract the direct wave signal in the reference trace. The extraction window function is as follows: in, , are the extracted direct wave signal and time value respectively. and are the length and order of the time windows, respectively; (3) Reference channel selection: select a number of reference channel direct wave signals from the numerous direct wave signals separated from the detection data; (4) Reference channel extraction: extract the direct waves of the selected reference channels and calculate the average of the signal values. The calculated average reference channel is used as the final reference channel. (5) Reference track correction: the reference track’s initial arrival position is corrected so that its amplitude starting point is as close as possible to the amplitude starting point of the simulation data. At the same time, the overall starting point of the field measured data is adjusted through the zero point correction method. Like the reference track, the overall starting point of the field measured data is adjusted to be roughly the same as the amplitude starting point of the detection data. The specific extraction method of the numerical simulation standard wavelet is as follows: The antenna receiving and transmitting distance is adjusted to a certain distance, and numerical simulation is performed under air medium background conditions to obtain the standard sub-wave in the numerical simulation.

2. The method for karst geological detection of power transmission engineering based on ground penetrating radar according to claim 1, characterized in that: The inversion network includes an encoder and a decoder; The encoder extracts a feature map from input ground penetrating radar data based on a spatial pyramid structure of multi-scale dilated convolution; The decoder reconstructs the characteristic map into a dielectric constant map.

3. A karst geological detection system for power transmission engineering based on ground penetrating radar, using the karst geological detection method for power transmission engineering based on ground penetrating radar as described in any one of claims 1-2, characterized in that: Including data building module, network building module and inversion prediction module: The data construction module is configured to: establish a target body model of karst geology based on autocorrelated random media, forward generate ground penetrating radar data of the target body model, inverse generate a dielectric constant map of the target body model, and construct a karst disaster data set based on the generated ground penetrating radar data and the corresponding dielectric constant map; The network construction module is configured to: construct an inversion network with GPR data as input and dielectric constant map as output, and use the karst hazard dataset to perform wavelet-independent GPR source-independent deep learning on the inversion network; The inversion prediction module is configured to: invert the acquired measured ground penetrating radar data through the trained inversion network, and predict the dielectric constant map of karst geology.

4. An electronic device, comprising: a memory for non-transitory storage of computer readable instructions; a processor for executing the computer-readable instructions; Wherein, when the computer-readable instructions are executed by the processor, the method according to any one of claims 1-2 is executed.

5. A storage medium, characterized in that: The computer-readable instructions are non-transitorily stored, wherein when the computer-readable instructions are executed by a computer, the method according to any one of claims 1 to 2 is performed.

Images