CN116224265A - Ground penetrating radar data inversion method and device, computer equipment and storage medium - Google Patents

Abstract

The application provides a ground penetrating radar data inversion method, a device, computer equipment and a storage medium, comprising the following steps: receiving ground penetrating radar data; invoking a preset one-stage network model to perform inversion operation on the ground penetrating radar data to obtain a first dielectric constant model; and calling a preset two-stage network model to perform inversion operation on the ground penetrating radar data and the first dielectric constant model to obtain a second dielectric constant model. The method and the device ensure the accuracy of the first dielectric constant model, modify the wrong structural characteristics of the first dielectric constant model, obtain the technical effect of the second dielectric constant model which is more accurate, greatly improve the operation efficiency of inversion of the ground penetrating radar data and reduce the consumption of calculation resources.

Description

Ground penetrating radar data inversion method, device, computer equipment and storage medium

technical field

The present application relates to the technical field of artificial intelligence, in particular to a ground penetrating radar data inversion method, device, computer equipment and storage medium.

Background technique

The GPR data inversion is to solve the ill-conditioned problem corresponding to the GPR data and the dielectric constant model space with at least one relative permittivity. The ill-conditioned problem refers to the problem that the output result is very sensitive to the input data, that is, in the input data Small errors may cause great changes in the output results. Generally, the condition number is used to measure the ill-conditioned index of the problem. The larger the condition number, the more serious the problem is.

However, the current GPR data inversion usually adopts the traditional GPR full waveform inversion (FWI, full waveform inversion) method, which is used for qualitative and quantitative reconstruction of stratigraphic structure images. It directly uses the entire received waveform to match the simulated GPR data. It then reconstructs the structure's dielectric distribution by minimizing the mismatch between the two sets of data.

However, the inventors have found that traditional FWI usually uses an iterative approach to reduce the error between the simulated data and the ground-penetrating radar data. Therefore, it takes a lot of computing resources to complete one FWI work; In ground radar data, the eigenvalues that have a sensitive relationship with the dielectric constant model are extracted, resulting in low inversion accuracy of the FWI method.

Contents of the invention

This application provides a ground-penetrating radar data inversion method, device, computer equipment and storage media to solve the problem that the traditional FWI method needs to consume a large amount of computing resources, and the FWI method is difficult to extract from the ground-penetrating radar data. The eigenvalues with sensitive relationship between the permittivity models lead to the problem of low inversion accuracy of the FWI method.

In a first aspect, the present application provides a ground penetrating radar data inversion method, including:

Receive ground-penetrating radar data; wherein, the ground-penetrating radar data is waveform data of electromagnetic waves for non-destructive detection of the shallow surface of the target area;

Invoking the preset one-stage network model to perform an inversion operation on the ground penetrating radar data to obtain a first permittivity model; wherein, the first permittivity model is to determine the formation of the electromagnetic wave in the target area a factor of propagation velocity in the structure; the first permittivity model has at least one relative permittivity; the relative permittivity is a physical parameter characterizing the dielectric properties of the formation structure;

Invoke the preset two-stage network model to invert the ground-penetrating radar data and the first permittivity model to obtain a second permittivity model; wherein, the second permittivity model includes The ground penetrating radar data is the physical parameter after the relative permittivity in the first permittivity model is corrected.

In the above solution, before calling the preset one-stage network model to perform inversion operation on the ground penetrating radar data, the method further includes:

Obtain the first training sample;

The one-stage network model is obtained by training the first initial network model preset by the first training sample.

In the above solution, before invoking the preset two-stage network model to perform the inversion operation on the ground penetrating radar data and the first dielectric constant model, the method further includes:

Obtain a second training sample;

Using the second training samples and the one-stage network model, train the preset second initial network model to obtain the two-stage network model.

In the above solution, before the acquisition of the first training sample, the method further includes:

Obtaining the stratum structure, embedding irregular blocks in the stratum structure to convert the stratum structure into a training permittivity model, and performing forward modeling according to the training permittivity model to obtain training GPR data;

Aggregate a described training permittivity model and its described training GPR data to form a training data;

Summarizing several training data to obtain a training set.

In the above scheme, the acquisition of the first training sample includes:

Acquiring M training data from the training set, and summarizing the M training data to obtain the first training sample; wherein, M is a positive integer, M≥1;

The acquisition of the second training sample includes:

Acquiring N training data from the training set, and summarizing the N training data to obtain the second training sample; wherein, N is a positive integer, and N≧1.

In the above solution, the one-stage network model is obtained by training the first initial network model preset by the first training sample, including:

Using the training GPR data of the training data in the first training sample as the first input information of the first initial network model, and running the first initial network model to invert the first input information Obtaining the first output information through operation;

Using the training permittivity model of the training data in the first training sample as the first reference information of the first initial network model, according to the first output information and the first initial network model through the preset first loss function generating a first loss value with reference to information; wherein, the first loss value represents the degree of difference between the first output information and the first reference information;

The preset optimization model is used to iterate the first initial network model according to the first loss value, so as to adjust the weight of the hidden layer in the first initial network model, so that the first initial network model generated by the first initial network model An output information, the first loss value between the first reference information and the first reference information is within a preset first threshold interval, and the first initial network model after iteration is set as the one-stage network model.

In the above scheme, the second initial network model is trained to obtain the second-stage network model through the second training samples and the one-stage network model, including:

Using the training ground penetrating radar data of the training data in the second training sample as the second input information, running the one-stage network model to perform an inversion operation on the second input information to obtain a one-stage output information, and running the The second initial network model performs an inversion operation on the first-stage output information to obtain second output information;

Using the training permittivity model of the training data in the second training sample as the second reference information of the second initial network model, according to the second output information and the first Two reference information generates a second loss value; wherein, the second loss value represents the degree of difference between the second output information and the second reference information;

The preset optimization model is used to iterate the second initial network model according to the second loss value, so as to adjust the weight of the hidden layer in the second initial network model, so that the second initial network model generated by the second initial network model The second loss value between the second output information and the second reference information is within a preset second threshold interval, and the second initial network model after iteration is set as the two-stage network model.

In a second aspect, the present application provides a ground penetrating radar data inversion device, including:

The input module is used to receive ground-penetrating radar data; wherein, the ground-penetrating radar data is waveform data of electromagnetic waves for non-destructive detection of the shallow surface of the target area;

The first inversion module is used to invoke the preset one-stage network model to invert the ground penetrating radar data to obtain the first permittivity model; wherein, the first permittivity model is to determine the A factor of electromagnetic wave propagation speed in the formation structure of the target area; the first permittivity model has at least one relative permittivity; the relative permittivity is a physical parameter characterizing the dielectric properties of the formation structure ;

The second inversion module is used to invoke the preset two-stage network model to invert the ground penetrating radar data and the first permittivity model to obtain a second permittivity model; wherein, the first The two-permittivity model includes physical parameters after the relative permittivity in the first permittivity model has been corrected according to the ground-penetrating radar data.

In a third aspect, the present application provides a computer device, including: a processor and a memory communicatively connected to the processor;

the memory stores computer-executable instructions;

The processor executes the computer-executable instructions stored in the memory, so as to realize the ground penetrating radar data inversion method as claimed in the above claims.

In a fourth aspect, the present application provides a computer-readable storage medium, where computer-executable instructions are stored in the computer-readable storage medium, and when the computer-executable instructions are executed by a processor, they are used to realize the above-mentioned GPR data inversion method.

In a fifth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, the above ground penetrating radar data inversion method is realized.

A ground penetrating radar data inversion method, device, computer equipment, and storage medium provided by the present application perform inversion operations on the ground penetrating radar data by calling a preset one-stage network model, and extract ground penetrating radar data An eigenvalue having a sensitive relationship with the dielectric constant model, and a first dielectric constant model is obtained according to the eigenvalue, so as to ensure the accuracy of the first dielectric constant model.

By invoking the preset two-stage network model, an inversion operation is performed on the ground penetrating radar data and the first permittivity model to modify the wrong structural features in the first permittivity model and obtain more accurate The technical effect of the second permittivity model to obtain the second permittivity model based on the first permittivity model.

By using the network model to perform inversion calculation on the ground penetrating radar data, the calculation efficiency of the ground penetrating radar data inversion is greatly improved, and the consumption of computing resources is reduced.

Description of drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and together with the description serve to explain the principles of the application.

FIG. 1 is a schematic diagram of an application scenario provided by an embodiment of the present application;

Fig. 2 is the flow chart of Embodiment 1 of a ground penetrating radar data inversion method provided by the embodiment of the present application;

Fig. 3 is the image of the second dielectric constant model in a kind of GPR data inversion method provided by the embodiment of the present application;

Fig. 4 is the flowchart of Embodiment 2 of a ground penetrating radar data inversion method provided by the embodiment of the present application;

Fig. 5 is the image of the training dielectric constant model in Embodiment 2 of a ground penetrating radar data inversion method provided by the embodiment of the present application;

FIG. 6 is an image of the training ground-penetrating radar data in Embodiment 2 of a ground-penetrating radar data inversion method provided in an embodiment of the present application;

7 is a schematic structural diagram of the first initial network model in Embodiment 2 of a ground penetrating radar data inversion method provided in the embodiment of the present application;

Fig. 8 is a graph of the relationship between the first loss value and the number of iterations during the training process of the first initial network model;

FIG. 9 is a schematic structural diagram of the second initial network model in Embodiment 2 of a ground penetrating radar data inversion method provided in the embodiment of the present application;

Fig. 10 is a graph of the relationship between the second loss value and the number of iterations during the training process of the second initial network model;

Fig. 11 is a schematic diagram of program modules of a ground penetrating radar data inversion device provided by the present invention;

Fig. 12 is a schematic diagram of the hardware structure of the computer equipment in the computer equipment of the present invention.

By means of the above drawings, specific embodiments of the present application have been shown, which will be described in more detail hereinafter. These drawings and text descriptions are not intended to limit the scope of the concept of the application in any way, but to illustrate the concept of the application for those skilled in the art by referring to specific embodiments.

Detailed ways

Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numerals in different drawings refer to the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with aspects of the present application as recited in the appended claims.

The specific application scenarios of this application are:

Ground Penetrating Radar (GPR) is a shallow surface geophysical prospecting technology that uses electromagnetic waves for non-destructive detection. It has been widely used in many fields, including glaciology, archaeology and geotechnical engineering. GPR can convert the electromagnetic information in the underground medium into information related to the characteristics of the geological medium (such as position, shape and dielectric properties), which is very important for geological exploration and analysis of the shallow surface.

The existing GPR inversion method focuses on roughly inferring the position, size and size of the detection object based on the observed GPR data. Full waveform inversion (FWI) of traditional GPR is considered as a solution to qualitatively and quantitatively reconstruct images of stratigraphic structures. It directly uses the entire received waveform to match the simulated GPR data. It then reconstructs the structure's dielectric distribution by minimizing the mismatch between the two sets of data. FWI originated in the field of seismic exploration and has since been rapidly adapted to process radar data. However, because the actual stratigraphic structure always has irregular geometric features and complex distribution patterns, the received GPR data are generally interlaced and messy, accompanied by discontinuous and distorted echoes. In addition, some multiple reflections caused by abnormal bodies in the underground medium will also cover up the characteristics of the radar signal, and there are often some signal features with messy shapes in the image. On the other hand, traditional FWI usually uses iterative methods to reduce the error between simulated data and ground penetrating radar data. Therefore, it takes a lot of computing resources to complete one FWI job. Therefore, traditional FWI not only consumes a large amount of computing resources, but also has a lot of room for improvement in the accuracy of its inversion model.

In recent years, deep neural network (DNN) has been developed rapidly in seismic denoising, signal processing, and geophysical inversion. DNNs automatically learn high-level features from training data, and are then able to estimate nonlinear mappings between input image data and various data domains. With the rapid development of deep learning technology, the breadth and depth of intelligent inversion work in the field of geosciences are also expanding. Using Convolutional Neural Networks (CNNs) to Predict High-Resolution Impedance. Li et al. used multi-task learning to achieve super-resolution velocity image prediction. The two-dimensional inversion imaging of GPR and earthquake is realized by end-to-end learning method.

So far, although the GPR inversion work based on DNN has made some progress, there is still a lot of room for development. To reconstruct the complex structure of the electrical properties of subsurface media, the difficulty lies in extracting effective features from complex GPR data and preserving the spatial alignment between the input and output.

This application proposes to use incremental learning to predict the subsurface permittivity model. The one-stage network extracts the initial permittivity model from the GPR data in an end-to-end manner. Then, a dual-channel two-stage network model is built, and the initial permittivity model is regarded as prior information, and combined with GPR data as input for inversion prediction. The one-stage network can extract the signal features of GPR from adjacent trajectories, and the two-stage network can effectively modify the wrong structural features in the permittivity model. This method combines the idea of incremental learning and proposes a new algorithm for the learning mechanism in the GPR inversion task.

Specifically, referring to Fig. 1, the application proposes a server 2 that operates a GPR data inversion method, and receives GPR data from a GPR device 3; The waveform data of electromagnetic waves for non-destructive detection of shallow ground; call the preset one-stage network model to invert the GPR data to obtain the first permittivity model; where the first permittivity model is to determine the electromagnetic wave in the A factor of propagation velocity in the formation structure of the target area; the first permittivity model has at least one relative permittivity; the relative permittivity is a physical parameter characterizing the dielectric properties of the formation structure;

And call the preset two-stage network model to invert the ground-penetrating radar data and the first permittivity model to obtain the second permittivity model; wherein, the second permittivity model includes ground-penetrating radar data, The physical parameter after the relative permittivity in the first permittivity model is corrected.

Therefore, this application implements GPR inversion based on incremental learning. GPR inversion is an ill-conditioned problem of solving the correspondence between ground penetrating radar data and the dielectric constant model space. This application realizes the GPR inversion task by learning new knowledge from GPR data many times. The inversion method used in this method can not only effectively improve the accuracy of GPR inversion, but also has the ability to suppress false anomalies and restore deep formation structures.

This application uses the one-stage network model and the two-stage network model to accurately analyze the ground-penetrating radar data and obtain the second permittivity model with a relative permittivity that can accurately characterize the dielectric properties of the stratum structure. Electrical properties refer to the geological information that shows the storage and loss of electrostatic energy by formation structure under the action of electric field.

Therefore, according to the second dielectric constant model, the eigenvalues of the dielectric properties of the ground-penetrating radar data are analyzed and extracted to obtain geological information, and the precise analysis and extraction of the eigenvalues of the ground-penetrating radar data are realized.

The technical solution of the present application and how the technical solution of the present application solves the problems in the prior art will be described in detail below with specific embodiments. The following specific embodiments may be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. Embodiments of the present application will be described below in conjunction with the accompanying drawings.

Example 1:

Please refer to Figure 2, this application provides a ground penetrating radar data inversion method, including:

S101: Receive ground-penetrating radar data; wherein, the ground-penetrating radar data is waveform data of electromagnetic waves for non-destructively detecting shallow ground surfaces in a target area.

In this step, Ground Penetrating Radar (GPR) is a geophysical method that uses antennas to transmit and receive high-frequency electromagnetic waves to detect the properties and distribution of materials inside the medium. GPR data refers to GPR data. GPR data is a shallow surface geophysical prospecting technology that uses electromagnetic waves for non-destructive detection. It has been widely used in many fields, including glaciology, archaeology, and geotechnical engineering. GPR data can convert the electromagnetic information in the underground medium into information related to the characteristics of the geological medium (such as position, shape and dielectric properties), which is very important for geological exploration and analysis of the shallow surface.

S102: Call the preset one-stage network model to invert the GPR data to obtain the first permittivity model; where the first permittivity model is a factor that determines the propagation speed of electromagnetic waves in the formation structure of the target area ; the first permittivity model has at least one relative permittivity; the relative permittivity is a physical parameter characterizing the dielectric properties of the formation structure.

In this example, by calling the one-stage network model to invert the ground-penetrating radar data to obtain the first dielectric constant model, the stratum structure of the target area can be predicted based on the ground-penetrating radar data, and the first permittivity model can be obtained to characterize the stratum structure. A dielectric constant model. Among them, the one-stage network model extracts the first permittivity model from GPR data in an end-to-end manner.

Relative permittivity is a physical parameter that characterizes the dielectric properties of dielectric materials. Dielectric performance refers to the property of showing the storage and loss of electrostatic energy under the action of an electric field, usually expressed by dielectric constant and dielectric loss.

Optionally, the one-stage network model can extract signal features of the GPR from adjacent trajectories, and then use the signal features to obtain the first permittivity model.

In this embodiment, the one-stage network model obtains at least one relative permittivity by extracting the eigenvalues representing the dielectric properties of the stratum structure from the ground penetrating radar data, and performing an inversion operation based on the eigenvalues, and according to at least one relative permittivity Permittivity A first permittivity model is generated to ensure the accuracy of the first node constant model.

S103: Invoke the preset two-stage network model to perform an inversion operation on the ground-penetrating radar data and the first permittivity model to obtain a second permittivity model; wherein, the second permittivity model includes ground-penetrating radar data, The physical parameter after the relative permittivity in the first permittivity model is corrected.

In this example, by building a dual-channel two-stage network model, the first permittivity model is regarded as prior information, and the first permittivity model and GPR data are input as the two-stage network model, so that the second-stage The network model can adjust the first dielectric constant model according to the ground penetrating radar data (that is, the original waveform of the shallow surface of the target area represented by the GPR data), so as to realize the inversion prediction. Among them, the second-stage network model can be effectively modified erroneous structural features in the first permittivity model, and the technical effect of obtaining a more accurate second permittivity model. This method combines the idea of incremental learning, and proposes a new algorithm for the learning mechanism in the GPR data inversion task. The second dielectric constant model is shown in Figure 3, wherein the vertical axis on the left in Figure 3 is the depth of the stratum structure, the horizontal axis is the position of the target area, and the vertical axis on the right is the relative value of the second dielectric constant model. dielectric constant.

Specifically, the subsurface permittivity model is predicted by using incremental learning. The one-stage network model extracts the first permittivity model from the GPR data in an end-to-end manner. Then, a dual-channel two-stage network model is built, and the first dielectric constant model is regarded as prior information, and combined with GPR data as input for inversion prediction. The one-stage network model can extract the signal features of GPR from adjacent trajectories, and the two-stage network model can effectively modify the wrong structural features in the permittivity model. This method combines the idea of incremental learning, and proposes a new algorithm for the learning mechanism in the GPR data inversion task.

This application uses an incremental learning-based approach to realize ground-penetrating radar data inversion. GPR data inversion is an ill-conditioned problem of solving the correspondence between GPR data and the dielectric constant model space. This application realizes the GPR data inversion task by learning new knowledge from GPR data many times. The inversion method used in this method can not only effectively improve the inversion accuracy of ground penetrating radar data, but also has the ability to suppress false anomalies and restore deep formation structures.

Therefore, the constraints of the inversion process are realized by using the prediction results of the one-stage network model as the prior information of the two-stage network model. Compared with the traditional FWI algorithm, the advanced nature of this application lies in effectively improving the inversion accuracy and ensuring proper calculation efficiency. Compared with the current GPR data inversion algorithm based on deep learning, the advanced nature of this application is that the prediction results are in better agreement with the model parameters, and can well reflect the structural distribution of deep layers.

In this embodiment, the one-stage network model extracts the eigenvalues representing the dielectric properties of the stratum structure from the ground penetrating radar data, and creates the first dielectric constant model based on the eigenvalues to ensure the accuracy of the first node constant model Spend.

The two-stage no-network model further modifies the relative permittivity in the first permittivity model by comparing the first permittivity model and detecting all the eigenvalues in the radar data to obtain the second permittivity model , so that the relative permittivity in the second permittivity model matches the GPR data more closely, thereby ensuring the accuracy of each relative permittivity in the second permittivity model.

At the same time, since this embodiment uses the first-stage network model and the second-stage network model to directly calculate the ground-penetrating radar data, it does not need to use multiple iterations to reduce the size of the simulated data as in the FWI method in the prior art. The error between the ground penetrating radar data makes it necessary to consume a large amount of computing resources to complete one FWI job. Therefore, compared with the prior art, this application greatly reduces the consumption of computing resources.

Example 2:

Please refer to Figure 4, this application provides a ground penetrating radar data inversion method, including:

S201: Receive ground-penetrating radar data; wherein, the ground-penetrating radar data is waveform data of electromagnetic waves for non-destructive detection of shallow ground surfaces in a target area.

This step is the same as S101 in Embodiment 1, so it will not be repeated here.

S202: Obtain the stratum structure, embed irregular blocks in the stratum structure to convert the stratum structure into a training permittivity model, and perform forward modeling according to the training permittivity model to obtain training GPR data;

Aggregate a training permittivity model and its training GPR data to form a training data set;

Summarize several training data to obtain a training set.

In this example, by generating data similar to the actual ground penetrating radar data scene characteristics as training data, it is used to effectively improve the ability of the network to solve practical problems. As shown in Figure 5, the training dielectric constant model is based on exploration conditions and mathematical methods Randomly generated, the training dielectric constant model contains randomly simulated undulating formation structures, and the dielectric parameters gradually increase from top to bottom. The stochastic equivalent medium technology is used to describe the inhomogeneity of the formation medium, and at the same time, it is combined with the abrasive particle technology to simulate different Regular blocks are embedded in the formation, and the range of medium parameters in the model is randomly generated within a reasonable interval.

Consider local random feature anomaly features. This application adopts the heterogeneous mixed function random medium modeling method to simulate the layer structure, and the function is expressed by the following formula:

Among them, r represents the fuzzy factor, a, b, and c represent the autocorrelation lengths in the x, y, and z directions. By selecting the local disturbance radius (a, b, c) and local disturbance strength (r), each Different forms of training permittivity models can be used to realize the diversity of training permittivity models.

After obtaining the training permittivity model, the FDTD algorithm is used to carry out the forward modeling simulation of the training ground penetrating radar data (GPR), as shown in Figure 6. The size of the dielectric constant model is 4.5m×4m, the cell grid size is 0.025m×0.025m, the sampling time interval is 0.0201ns, and the main frequency of the transmitting antenna is 650MHz. The sample set used in this application contains a total of 1000 pairs of GPR data and dielectric constant models.

It should be noted that the stochastic equivalent medium technology simulates the deposit structure and embeds the deposit structure into the random undulating stratum structure to obtain a geological model including the deposit structure, and converts the geological model into a resistivity model based on the existing geological data , and obtain a sample data set according to the resistivity model. The sample data set includes randomly simulated undulating stratum structure data, ore deposit structure data and corresponding resistivity data, and then train the initial resistivity model reconstruction network based on the sample data set, and obtain The resistivity model reconstruction network is used to perform deep learning on the first electromagnetic inversion data to obtain the second electromagnetic inversion data.

Abrasive particle technology is a computer algorithm that generates a three-dimensional array based on matlab or python, and generates irregular shapes based on the three-dimensional array.

FDTD algorithm is a finite-difference time-domain method (Finite-Difference Time-Domain, FDTD) is a common method in the field of electromagnetic field calculation. The model basis of the finite-difference time-domain method is the most basic Maxwell's equation in electrodynamics (Maxwell'sequation). After the FDTD method was proposed, with the development of computing technology, especially electronic computer technology, the FDTD method has been greatly developed and has been widely used in electromagnetics, electronics, optics and other fields.

S203: Obtain a first training sample, and perform training with a first initial network model preset in the first training sample to obtain a one-stage network model.

In a preferred embodiment, obtaining the first training sample includes:

Acquiring M training data from the training set, and summarizing the M training data to obtain the first training sample; wherein, M is a positive integer, and M≥1.

In this example, M pieces of training data are acquired from the training set and aggregated to obtain the first training sample through the preset number M, which ensures the controllability of the amount of training data in the sample.

In a preferred embodiment, the first initial network model preset by the first training sample is used for training to obtain a one-stage network model, including:

Using the training ground penetrating radar data of the training data in the first training sample as the first input information of the first initial network model, and running the first initial network model to invert the first input information to obtain the first output information;

The training permittivity model of the training data in the first training sample is used as the first reference information of the first initial network model, and the first loss value is generated according to the first output information and the first reference information through the preset first loss function ; Wherein, the first loss value represents the degree of difference between the first output information and the first reference information;

The first initial network model is iterated according to the first loss value through the preset optimization model to adjust the weight of the hidden layer in the first initial network model, so that the first output information generated by the first initial network model is consistent with the first reference The first loss value among the information is within the preset first threshold interval, and the first initial network model after iteration is set as a one-stage network model.

In this example, a deep neural network is used as the first initial network model. The first initial network model is shown in FIG. 7 , and the first initial network model uses a down-sampling codec structure. The network performs four downsampling operations, implemented by convolutional layers with a "stride" parameter of 2. There are 4 sets of feature maps of different scales in the network, and the size ratios are 8:4:2:1. The decoding layer is similar to the encoding layer. Deconvolution also has 4 repeated structures to form each repeated structure. Deconvolution is used before each repeated structure. After each deconvolution, the number of feature channels is halved, and the size of the feature map is doubled. After deconvolution, the result of deconvolution is stitched together with the feature map of the corresponding step of the encoding part. The convolution kernel of the last layer is a 1x1 convolution kernel, which converts the 64-channel feature map into the result of a specific number of categories. The input of the first-stage network model is single channel (GPR data), the input of the second-stage network model is two-channel (output of the first-stage network model and GPR data), and the output of both networks is the dielectric constant model .

The first initial network model uses the mean square error MSE as the loss function of the deep neural network, uses the Adam optimizer as the optimization model to train the first initial network model, and uses a decaying learning rate during the training process. After putting in the training data, record the number of training times, loss function value, and learning rate, and store the network parameters to a specific file. After the one-stage network model prediction model is integrated with the GPR data, it is put into the training of the two-stage network model for prediction.

The expression of the first loss function is:

Wherein, loss_1 is the first loss value, y is the first output information, r is the first reference information, and 1/n represents the mean value. Put in the training data, use the Adam optimizer for training, and use the decaying learning rate during the training process. As shown in Figure 8, the vertical axis is the first loss value loss_1, and the horizontal and vertical axis is the iteration number Epoch of the first initial network model, wherein Figure 8 includes the first loss value of the training set in the first training sample during the iteration process The training set curve formed, and the verification set curve formed by the first loss value of the verification set in the iterative process in the first training sample.

The learning rate is characterized by three parameters: initial learning rate η ₀ , decay period T, and decay rate α. The real-time learning rate expression during training is

η _i =α _i η ₀

Where i is the current learning rate decay times. Record training times, loss function values, and learning rates in text files, and store network parameters to specific files.

It should be noted that the deep neural network is a technology in the field of machine learning (ML, Machine Learning). The benefit of multiple layers is that complex functions can be expressed with fewer parameters. In supervised learning, the problem of the previous multi-layer neural network is that it is easy to fall into local extremum points. If the training samples are sufficient to adequately cover future samples, then the learned multi-layer weights can be well used to predict new test samples.

S204: Obtain a second training sample; use the second training sample and the first-stage network model to train the preset second initial network model to obtain a second-stage network model.

In a preferred embodiment, obtaining a second training sample includes:

Acquiring N training data from the training set, and summarizing the N training data to obtain a second training sample; wherein, N is a positive integer, and N≥1.

In this example, through the preset number N, N training data are obtained from the training set and aggregated to obtain the first training sample, which ensures the controllability of the number of training data in the sample.

Preferably, the preset second initial network model is trained to obtain a two-stage network model through the second training sample and the one-stage network model, including:

Using the training ground penetrating radar data of the training data in the second training sample as the second input information, run the first-stage network model to invert the second input information to obtain the first-stage output information, and run the second initial network model for the first-stage performing an inversion operation on the output information to obtain second output information;

The training permittivity model of the training data in the second training sample is used as the second reference information of the second initial network model, and the second loss value is generated according to the second output information and the second reference information through the preset second loss function ; Wherein, the second loss value represents the degree of difference between the second output information and the second reference information;

The second initial network model is iterated according to the second loss value through the preset optimization model to adjust the weight of the hidden layer in the second initial network model, so that the second output information generated by the second initial network model is consistent with the second reference The second loss value between the information is within the preset second threshold interval, and the second initial network model after iteration is set as a two-stage network model.

In this example, a deep neural network is used as the second initial network model. The second initial network model is shown in FIG. 9 , and the second initial network model uses a downsampled codec structure. The network performs four downsampling operations, implemented by convolutional layers with a "stride" parameter of 2. There are 4 sets of feature maps of different scales in the network, and the size ratios are 8:4:2:1. The decoding layer is similar to the encoding layer. Deconvolution also has 4 repeated structures to form each repeated structure. Deconvolution is used before each repeated structure. After each deconvolution, the number of feature channels is halved, and the size of the feature map is doubled. After deconvolution, the result of deconvolution is stitched together with the feature map of the corresponding step of the encoding part. The convolution kernel of the last layer is a 1x1 convolution kernel, which converts the 64-channel feature map into the result of a specific number of categories. The input of the first-stage network model is single channel (GPR data), the input of the second-stage network model is two-channel (output of the first-stage network model and GPR data), and the output of both networks is the dielectric constant model .

The second initial network model uses the mean square error MSE as the loss function of the deep neural network, uses the Adam optimizer as the optimization model to train the second initial network model, and uses a decaying learning rate during the training process. After putting in the training data, record the number of training times, loss function value, and learning rate, and store the network parameters to a specific file. After the one-stage network model prediction model is integrated with the GPR data, it is put into the training of the two-stage network model for prediction.

The expression of the second loss function is:

Wherein, loss_2 is the second loss value, x is the second output information, r is the second reference information, and 1/n represents the mean value. Put in the training data, use the Adam optimizer for training, and use the decaying learning rate during the training process. As shown in FIG. 10 , the vertical axis is the second loss value loss_2, and the vertical axis is the iteration number Epoch of the second initial network model.

The learning rate is characterized by three parameters: initial learning rate η ₀ , decay period T, and decay rate α. The real-time learning rate expression during training is

η _i =α ⁱ η ₀

Where i is the current learning rate decay times. Record training times, loss function values, and learning rates in text files, and store network parameters to specific files.

It should be noted that the deep neural network is a technology in the field of machine learning (ML, Machine Learning). The benefit of multiple layers is that complex functions can be expressed with fewer parameters. In supervised learning, the problem of the previous multi-layer neural network is that it is easy to fall into local extremum points. If the training samples are sufficient to adequately cover future samples, then the learned multi-layer weights can be well used to predict new test samples.

S205: Call the preset one-stage network model to invert the GPR data to obtain the first permittivity model; where the first permittivity model is a factor that determines the propagation speed of electromagnetic waves in the formation structure of the target area ; the first permittivity model has at least one relative permittivity; the relative permittivity is a physical parameter characterizing the dielectric properties of the formation structure.

This step is the same as S102 in Embodiment 1, so it will not be repeated here.

S206: Invoking the preset two-stage network model to perform an inversion operation on the ground-penetrating radar data and the first permittivity model to obtain a second permittivity model; wherein, the second permittivity model includes ground-penetrating radar data, The physical parameter after the relative permittivity in the first permittivity model is corrected.

This step is the same as S103 in Embodiment 1, so it will not be repeated here.

An experimental result based on the above technical solution includes:

In evaluating the inversion algorithm used in the present invention, the prediction results of the one-stage network model and the two-stage network model are compared. Compared with the prediction results of the first-stage network, the second-stage network has higher accuracy for the detection of deep layers and abnormal targets. The GPR inversion method used in the present invention draws on the idea of incremental learning, requires the dielectric constant predicted by the first-stage network as a constraint, and improves the prediction accuracy of the second-stage network for the medium model. From the indicators of MSE, PSNR and SSIM, the second-stage network prediction model has significantly improved compared with the first-stage network.

The quantitative comparison of the test results is shown in the table below:

algorithm MSE↓ PSNR↑ SSIM↑ one-stage network 0.1719 55.7793 0.9957 two-stage network 0.0859 58.7929 0.9979

Among them, MSE: mean-square error (mean-square error, MSE) is a measure that reflects the degree of difference between the estimator and the estimated quantity. Let t be an estimator of the population parameter θ determined according to the sample, and the mathematical expectation of (θ-t)2 is called the mean square error of the estimator t.

PSNR: Peak signal-to-noise ratio (English: Peak signal-to-noise ratio, often abbreviated as PSNR) is an engineering term that represents the ratio of the maximum possible power of a signal to the destructive noise power that affects its representation accuracy.

SSIM (Structural Similarity), structural similarity, is an index to measure the similarity of two images. This metric was first proposed by the Laboratory for Image and Video Engineering at UT Austin. Of the two images used by SSIM, one is an uncompressed undistorted image and the other is a distorted image.

Example 3:

Please refer to FIG. 11, the present application provides a ground penetrating radar data inversion device 1, including:

The input module 11 is used to receive ground-penetrating radar data; wherein, the ground-penetrating radar data is waveform data of electromagnetic waves for non-destructive detection of the shallow surface of the target area;

The first inversion module 15 is used to call the preset one-stage network model to invert the ground penetrating radar data to obtain the first permittivity model; wherein, the first permittivity model is to determine the electromagnetic wave in the target area A factor of propagation velocity in the formation structure; the first permittivity model has at least one relative permittivity; the relative permittivity is a physical parameter characterizing the dielectric properties of the formation structure;

The second inversion module 16 is used to call the preset two-stage network model to invert the GPR data and the first permittivity model to obtain the second permittivity model; wherein, the second permittivity model It includes the physical parameters after the relative permittivity in the first permittivity model is corrected according to the ground penetrating radar data.

Optionally, the GPR data retrieval device 1 also includes:

The sample construction module 12 is used to obtain the stratum structure, embed irregular blocks in the stratum structure so that the stratum structure is converted into a training permittivity model, and perform forward modeling according to the training permittivity model to obtain training ground penetrating radar data; summary A training permittivity model and its training ground penetrating radar data form a training data; a number of training data are aggregated to obtain a training set.

The first training module 13 is configured to obtain a first training sample, and perform training on a first initial network model preset by the first training sample to obtain a one-stage network model.

The second training module 14 is used to obtain a second training sample; through the second training sample and the first-stage network model, train the preset second initial network model to obtain a second-stage network model.

Example 4:

To achieve the above purpose, the present application also provides a computer device 4, including: a processor 42 and a memory 41 communicatively connected to the processor 42; the memory stores computer-executed instructions;

The processor executes the computer-executed instructions stored in the memory 41 to realize the above-mentioned ground penetrating radar data inversion method, wherein the components of the ground penetrating radar data inversion device can be dispersed in different computer equipment, and the computer equipment 4 can be executed Smartphones, tablet computers, laptops, desktop computers, rack servers, blade servers, tower servers, or cabinet servers (including independent servers, or server clusters composed of multiple application servers), etc. The computer device in this embodiment at least includes but is not limited to: a memory 41 and a processor 42 that can be communicatively connected to each other through a system bus, as shown in FIG. 12 . It should be noted that FIG. 12 only shows a computer device with components - but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead. In this embodiment, the memory 41 (that is, a readable storage medium) includes a flash memory, a hard disk, a multimedia card, a card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), Programmable Read Only Memory (PROM), Magnetic Memory, Magnetic Disk, Optical Disk, etc. In some embodiments, the memory 41 may be an internal storage unit of a computer device, such as a hard disk or internal memory of the computer device. In some other embodiments, the memory 41 can also be an external storage device of the computer equipment, such as a plug-in hard disk equipped on the computer equipment, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, flash memory card (Flash Card), etc. Of course, the memory 41 may also include both the internal storage unit of the computer device and its external storage device. In this embodiment, the memory 41 is usually used to store the operating system and various application software installed in the computer equipment, such as the program code of the ground penetrating radar data inversion device in the third embodiment. In addition, the memory 41 can also be used to temporarily store various types of data that have been output or will be output. The processor 42 may be a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, a microprocessor, or other data processing chips in some embodiments. The processor 42 is generally used to control the overall operation of the computer device. In this embodiment, the processor 42 is used to run the program codes stored in the memory 41 or process data, for example, run a GPR data inversion device, so as to realize the GPR data inversion method in the above embodiment.

The above-mentioned integrated modules implemented in the form of software function modules can be stored in a computer-readable storage medium. The above-mentioned software function modules are stored in a storage medium, and include several instructions to make a computer device (which may be a personal computer, server, or network device, etc.) or a processor execute some steps of the methods in various embodiments of the present application. It should be understood that the above-mentioned processor may be a central processing unit (Central Processing Unit, referred to as CPU), and may also be other general-purpose processors, a digital signal processor (Digital Signal Processor, referred to as DSP), an application specific integrated circuit (Application Specific Integrated Circuit, referred to as ASIC) and so on. A general-purpose processor may be a microprocessor, or the processor may be any conventional processor, or the like. The steps of the method disclosed in conjunction with the application can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor. The storage may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk storage, and may also be a U disk, a mobile hard disk, a read-only memory, a magnetic disk, or an optical disk.

To achieve the above object, the present application also provides a computer-readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory ( SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, server, App application store, etc., on which storage The computer executes instructions, and the program realizes corresponding functions when executed by the processor 42 . The computer-readable storage medium of this embodiment is used to store computer-executable instructions for implementing the GPR data inversion method, and when executed by the processor 42, implements the GPR data inversion method of the above-mentioned embodiment.

The above-mentioned storage medium can be realized by any type of volatile or non-volatile storage device or their combination, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable In addition to programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. A storage media may be any available media that can be accessed by a general purpose or special purpose computer.

An exemplary storage medium is coupled to the processor such the processor can read information from, and write information to, the storage medium. Of course, the storage medium may also be a component of the processor. The processor and the storage medium may be located in application specific integrated circuits (Application Specific Integrated Circuits, ASIC for short). Of course, the processor and the storage medium can also exist in the electronic device or the main control device as discrete components.

The present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, the above ground penetrating radar data inversion method is realized.

It should be noted that, in this document, the term "comprising", "comprising" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article or apparatus comprising a set of elements includes not only those elements, It also includes other elements not expressly listed, or elements inherent in the process, method, article, or device. Without further limitations, an element defined by the phrase "comprising a ..." does not preclude the presence of additional identical elements in the process, method, article, or apparatus comprising that element.

Other embodiments of the present application will be readily apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any modification, use or adaptation of the application, these modifications, uses or adaptations follow the general principles of the application and include common knowledge or conventional technical means in the technical field not disclosed in the application . The specification and examples are to be considered exemplary only, with a true scope and spirit of the application indicated by the following claims.

It should be understood that the present application is not limited to the precise constructions which have been described above and shown in the accompanying drawings, and various modifications and changes may be made without departing from the scope thereof. The scope of the application is limited only by the appended claims.

Claims (10)

1. The ground penetrating radar data inversion method is characterized by comprising the following steps of:

receiving ground penetrating radar data; the ground penetrating radar data are waveform data of electromagnetic waves for carrying out nondestructive detection on the shallow earth surface of the target area;

invoking a preset one-stage network model to perform inversion operation on the ground penetrating radar data to obtain a first dielectric constant model; wherein the first permittivity model is a factor determining a propagation speed of the electromagnetic wave in a stratum structure of the target area; the first dielectric constant model has at least one relative dielectric constant; the relative permittivity is a physical parameter that characterizes the dielectric properties of the formation structure;

Invoking a preset two-stage network model to perform inversion operation on the ground penetrating radar data and the first dielectric constant model to obtain a second dielectric constant model; the second dielectric constant model comprises physical parameters after the relative dielectric constant in the first dielectric constant model is corrected according to the ground penetrating radar data.

2. The method of claim 1, wherein before the invoking the preset one-stage network model to perform the inversion operation on the ground penetrating radar data, the method further comprises:

acquiring a first training sample;

and training through a first initial network model preset by the first training sample to obtain the one-stage network model.

3. The method of inversion of ground penetrating radar data according to claim 2, wherein before the invoking the preset two-stage network model to perform inversion operation on the ground penetrating radar data and the first dielectric constant model, the method further comprises:

acquiring a second training sample;

and training a preset second initial network model through the second training sample and the first-stage network model to obtain the second-stage network model.

4. A ground penetrating radar data inversion method according to claim 3 wherein prior to said obtaining a first training sample, the method further comprises:

acquiring a stratum structure, embedding irregular blocks in the stratum structure to convert the stratum structure into a training dielectric constant model, and performing forward modeling according to the training dielectric constant model to obtain training ground penetrating radar data;

summarizing one training dielectric constant model and the training ground penetrating radar data to form training data;

and summarizing a plurality of training data to obtain a training set.

5. A ground penetrating radar data inversion method according to claim 3 wherein said obtaining a first training sample comprises:

m training data are obtained from the training set, and the M training data are summarized to obtain the first training sample; wherein M is a positive integer, M is more than or equal to 1;

the obtaining a second training sample includes:

acquiring N training data from the training set, and summarizing the N training data to obtain the second training sample; wherein N is a positive integer, and N is more than or equal to 1.

6. The ground penetrating radar data inversion method according to claim 2, wherein the training by the first initial network model preset by the first training sample to obtain the one-stage network model includes:

The ground penetrating radar data of training data in the first training sample are used as first input information of the first initial network model, and the first initial network model is operated to carry out inversion operation on the first input information to obtain first output information;

the training dielectric constant model of training data in the first training sample is used as first reference information of the first initial network model, and a first loss value is generated according to the first output information and the first reference information through a preset first loss function; wherein the first loss value characterizes a degree of difference between the first output information and the first reference information;

iterating the first initial network model according to the first loss value through a preset optimizing model to adjust the weight of a hidden layer in the first initial network model, so that the first loss value between the first output information generated by the first initial network model and the first reference information is in a preset first threshold interval, and setting the first initial network model after iteration as the first-stage network model.

7. A ground penetrating radar data inversion method according to claim 3 wherein training a preset second initial network model through said second training sample and said one-stage network model to obtain said two-stage network model comprises:

Taking the training ground penetrating radar data of the training data in the second training sample as second input information, operating the first-stage network model to perform inversion operation on the second input information to obtain first-stage output information, and operating the second initial network model to perform inversion operation on the first-stage output information to obtain second output information;

the training dielectric constant model of training data in the second training sample is used as second reference information of the second initial network model, and a second loss value is generated according to the second output information and the second reference information through a preset second loss function; wherein the second loss value characterizes a degree of difference between the second output information and the second reference information;

and iterating the second initial network model according to the second loss value through a preset optimizing model to adjust the weight of a hidden layer in the second initial network model, so that the second loss value between second output information generated by the second initial network model and the second reference information is in a preset second threshold interval, and setting the iterated second initial network model as the two-stage network model.

8. A ground penetrating radar data inversion apparatus, comprising:

the input module is used for receiving ground penetrating radar data; the ground penetrating radar data are waveform data of electromagnetic waves for carrying out nondestructive detection on the shallow earth surface of the target area;

the first inversion module is used for calling a preset one-stage network model to perform inversion operation on the ground penetrating radar data to obtain a first dielectric constant model; wherein the first permittivity model is a factor determining a propagation speed of the electromagnetic wave in a stratum structure of the target area; the first dielectric constant model has at least one relative dielectric constant; the relative permittivity is a physical parameter that characterizes the dielectric properties of the formation structure;

the second inversion module is used for calling a preset two-stage network model to perform inversion operation on the ground penetrating radar data and the first dielectric constant model to obtain a second dielectric constant model; the second dielectric constant model comprises physical parameters after the relative dielectric constant in the first dielectric constant model is corrected according to the ground penetrating radar data.

9. A computer device, comprising: a processor and a memory communicatively coupled to the processor;

The memory stores computer-executable instructions;

the processor executes computer-executable instructions stored by the memory to implement the ground penetrating radar data inversion method of any one of claims 1 to 7.

10. A computer readable storage medium having stored therein computer executable instructions which when executed by a processor are for implementing the ground penetrating radar data inversion method of any one of claims 1 to 7.

Images