Transient electromagnetic scanning detection system and method

Abstract

A transient electromagnetic scanning detection system and method. An all-terrain vehicle carrying an instrument tows a transmitting coil and a receiving coil having an offset distance of 6-8 meters therebetween, wherein the transmitting coil arbitrarily rotates within an XZ plane, the receiving coil moves in an X direction and a Z direction, and GPSs are carried on the coils. During working, the vehicle stays for a short time every few meters, the transmitting coil performs rotary scanning until 180 degrees, and the receiving coil completes scanning detection of one point; and the process is repeated when the vehicle travels to the next detection point. The inversion is implemented by means of a deep learning model, a large number of resistivity models are randomly generated, and on the basis of an observation system, numerical simulation is performed on an induced electromotive force corresponding to each resistivity model. On the basis of induced electromotive force data and the resistivity models, a deep learning framework-based TEM prediction model is established, wherein the induced electromotive force data is used as an input, and a two-dimensional resistivity structure below the coils is used as an output. The vehicle collects TEM data and imports same into a deep learning network to obtain an underground resistivity structure below the coils in real time to perform transient electromagnetic scanning detection.

Description

A transient electromagnetic scanning detection system and method Technical Field

[0001] The present invention belongs to the technical field of non-destructive ground detection, and in particular relates to a transient electromagnetic scanning detection system and method. Background Art

[0002] The Transient Electromagnetics Method (TEM), also known as the Time Domain Electromagnetic Method, is an artificial source geophysical exploration method that developed rapidly in the 1950s. This method emits a pulsed electromagnetic field that diffuses underground, which will stimulate induced currents, also known as eddy currents, inside electrically inhomogeneous bodies; by observing the primary diffusion field and the secondary electromagnetic field generated by the eddy currents inside the anomaly, the underground electrical structure can be inverted. In addition, TEM has the characteristics of convenient operation, high experimental efficiency, and strong adaptability to complex terrain, which makes it have broad application prospects in various infrastructure fields, such as groundwater monitoring, mineral exploration, geological mapping and environmental engineering detection, landslide, collapse and other natural disaster monitoring, etc.

[0003] Existing TEM detection technology primarily uses one-dimensional inversion, then stitching together multiple one-dimensional inversion models to form two- or three-dimensional models to rapidly acquire the underground resistivity structure. However, one-dimensional inversion is based on layered structures for forward modeling, and can only constrain the variation trend of the longitudinal resistivity. Due to data errors, the three-dimensional effects of underground anomalies, and the influence of topographical fluctuations, the inversion results from different observation points may vary significantly, resulting in poor lateral continuity of the stitched resistivity model. When detecting certain structurally complex strata, such as near-vertical faults, obvious false anomalies may be generated, greatly complicating the comprehensive geological and geophysical interpretation. One-dimensional inversion suffers from a strong non-uniqueness problem, primarily due to the inadequate exploitation of effective electromagnetic field signals. Currently, transient electromagnetic advance prediction primarily uses a central loop, collecting only the induced electromotive force in the vertical direction while ignoring electromagnetic field data in other directions. However, the spatial distribution of the electromagnetic field caused by anomalies is three-dimensional. Depending on the different receiving and transmitting locations and the underground resistivity structure, the sensitivity of electromagnetic field signals to anomalies in different directions varies greatly. Therefore, if only the vertical component is used to detect underground anomalies, important effective information will inevitably be lost. Limited electromagnetic field signals and limited observation data are insufficient to constrain the underground resistivity distribution structure, which is a clear mathematical underdetermined problem. On the other hand, to improve the convenience of the center loop device, people usually increase the number of turns of the transmitting coil while reducing the coil area. This leads to a strong mutual inductance effect between the transmitting coil and the measuring coil. The measurement data before the mutual inductance effect disappears is invalid data, which makes it difficult for TEM to measure early valid data, forming a blind spot for close-range measurement.

[0004] Furthermore, to enhance the accuracy of data interpretation, researchers have conducted research on two- and three-dimensional (2D) inversion using ATEM (Airborne Time Domain Electromagnetic) methods, optimizing ATEM inversion algorithms from multiple perspectives, such as improving inversion efficiency and reducing memory usage. While 3D inversion can provide more detailed information on the resistivity distribution of subsurface media, practical applications remain challenging, primarily due to high computational complexity and a tendency to fall into local minima.

[0005] Through the above analysis, the problems and defects of the existing technology are as follows:

[0006] (1) The existing one-dimensional transient electromagnetic technology has problems such as a single observation method and insufficient observation data coverage.

[0007] (2) When using the central loop device for observation, there is a strong mutual inductance effect between the transmitting coil and the measuring coil, which leads to a blind spot in the TEM close-range measurement.

[0008] (3) Inversion is mainly based on a one-dimensional layered model, which is insufficient to describe the three-dimensional inhomogeneous underground space and is prone to false anomalies. In addition, the inversion is unstable and has poor lateral continuity.

[0009] Summary of the Invention

[0010] In order to solve the shortcomings of the existing technology and achieve the purpose of improving detection accuracy, the present invention adopts the following technical solutions:

[0011] A transient electromagnetic scanning detection method comprises the following steps:

[0012] Step S1: Design of the scanning observation system. The TEM scanning detection system is an off-road vehicle carrying the instrument, towing the transmitter frame and receiving coil. It uses a dipole transmitting device to measure the early and late induced electromotive force data after each transmitting pulse.

[0013] Step S2: Construct a training set and use a random resistivity model generation algorithm to generate a large number of resistivity models that conform to actual situations. Then, perform forward calculations on the resistivity models to obtain TEM response data.

[0014] Step S3: Construct a TEM prediction model. During training, the input of the model is TEM response data, and the output is a randomly generated underground resistivity structure. The output of the model is the TEM response corresponding to each random resistivity model. The TEM response data is obtained based on finite volume numerical simulation. In the numerical simulation, the transmitting coil is set to rotate within a certain angle. At each corresponding rotation angle and off sampling moment, the corresponding radial induced electromotive force and tangential induced electromotive force are obtained. The absolute value and logarithm of the induced electromotive force data are taken, and a standardization operation is performed. The radial induced electromotive force data after taking the absolute value and logarithm, the tangential induced electromotive force data after taking the absolute value and logarithm, and the standardized radial and tangential induced electromotive force data are input into the TEM prediction model for prediction. The standardization operation is to subtract the mean value from the induced electromotive force and then divide it by the variance. Based on the true value of the resistivity model as a label and the prediction result, the TEM prediction model is trained to obtain a trained TEM prediction model.

[0015] Furthermore, the random resistivity model generation method in step S2 includes the following steps:

[0016] Step S2.1: Randomly generate a layered resistivity model R(l), where the resistivity increases gradually with depth, and the minimum resistivity range follows a uniform distribution U(log 10 (1),log 10 (100))Ω·m, the resistivity variation range also obeys the uniform distribution U(log 10 (100),log 10 (10000))Ω·m;

[0017] Step S2.2: randomly generate a two-dimensional model grid m, and the value of each grid is randomly distributed;

[0018] Step S2.3: Smoothing the two-dimensional model. A greater degree of smoothing means a lower model resolution.

[0019] Step S2.4: normalize the data of each layer of the smoothed model;

[0020] Step S2.5: Assuming that the resolution of TEM decays exponentially with depth, the model data of each layer is processed as follows: m(l) = m(l)·R(l)·e (-τl)

[0021] Where l represents the number of layers, τ represents the attenuation exponent, the larger the τ, the smoother the model at depth, e represents the natural logarithm, and the final m represents a two-dimensional resistivity model with uneven distribution, and the resistivity changes more slowly with increasing depth;

[0022] Step S2.6: Perform nearest neighbor interpolation on m through the octree grid to obtain the resistivity model for fast numerical simulation. The goal of the TEM scanning detection system is to conduct fine detection of shallow surface, so only 50×50m is needed to be simulated. 2 The target area is divided into two-dimensional space by octree grid. The grid in the simulation area is relatively dense and gradually becomes sparse outward to improve the calculation efficiency.

[0023] Furthermore, in step S3, the TEM response data is obtained based on finite volume numerical simulation. In the numerical simulation, the transmitting coil is set to rotate within a certain angle. At each corresponding rotation angle and shutdown sampling time, the radial induced electromotive force and tangential induced electromotive force are obtained. The radial direction refers to the direction perpendicular to the coil plane, and the tangential direction refers to the direction of the intersection of the coil plane and the XZ plane. The horizontal axis represents the rotation angle and the vertical axis represents the shutdown time. The radial and tangential induced electromotive force data feature maps are constructed after taking the absolute value and logarithm, respectively, as well as the normalized radial and tangential induced electromotive force data feature maps. The predicted induced electromotive force data feature map is compared with the feature map corresponding to the original electromotive force to train the TEM prediction model.

[0024] Furthermore, the TEM prediction model in step S3 adopts UNet, including a convolution layer, a downsampling layer, an upsampling layer, and a normalization layer. The input induced electromotive force is subjected to step-by-step feature extraction through the convolution layer and downsampling, and then upsampling and feature extraction are performed step by step. Finally, the predicted induced electromotive force is output through the normalization layer, and a direct connection is made between the upsampling and downsampling layers corresponding to each level to improve the back propagation efficiency and prediction accuracy of the model; the output of the convolution layer is activated and normalized through the BN layer to further prevent the gradient from disappearing or exploding, and at the same time increase the regularization effect.

[0025] Furthermore, in step S3, the mean square error between the label and the prediction result is used as an indicator to judge the performance of UNet: mse = ‖predi-label‖ 2 +λ‖predi‖ 2

[0026] in,‖·‖ 2 represents L2 regularization processing, which not only fits the label but also makes it as smooth as possible, which is consistent with the resistivity model designed in the present invention. λ represents the weight between data fitting and data smoothing.

[0027] Furthermore, the method also includes step S3: model prediction, collecting TEM response data, inputting the trained TEM prediction model, and obtaining a predicted resistivity model.

[0028] Furthermore, in step S3, the collected TEM response data is regionally decomposed, and the obtained small-scale regions consistent with the training set are predicted, and then the prediction results are spliced ​​according to the original decomposition order to reconstruct a large-scale inversion model to adapt to the situation where the detection area is too large.

[0029] A transient electromagnetic scanning detection system includes an observation system module, a training set construction module, and a TEM prediction module;

[0030] The observation system module includes an off-road vehicle, a transmitting device, a receiving device, a GPS positioning device and a GIS navigation system;

[0031] The training set construction module uses a random resistivity model generation algorithm to generate a large number of resistivity models that conform to actual situations, and then performs forward calculations through the resistivity models to obtain TEM response data;

[0032] The TEM prediction module obtains TEM response data based on the angle of the rotating emission source corresponding to each sampling point and the shut-off sampling time, obtains the corresponding radial induced electromotive force and tangential induced electromotive force, takes the absolute value and logarithm of the induced electromotive force data, and performs a standardization operation. The radial induced electromotive force data after taking the absolute value and logarithm, the tangential induced electromotive force data after taking the absolute value and logarithm, and the standardized radial and tangential induced electromotive force data are input into the TEM prediction module for prediction. The standardization operation is to subtract the average value from the induced electromotive force and then divide it by the variance. The TEM prediction module is trained based on the true value of the resistivity model as a label and the prediction result to obtain a trained TEM prediction module. The purpose of taking the logarithm is to avoid a large difference in order of magnitude.

[0033] Furthermore, the system includes a transient electromagnetic instrument, a transmitting device, a receiving device and a processing module. The transient electromagnetic instrument collects the pulsed electromagnetic field emitted by the transmitting device through the receiving device to obtain an induced electromotive force;

[0034] The transmitting device is a vertically rotating coil, using a dipole transmitting coil, with a vertical rotation angle range of θ∈[-90°,90°];

[0035] The receiving device includes two sets of coils, vertical and horizontal, which are arranged in conjunction with the transmitting device;

[0036] The processing module includes the trained TEM prediction module, which obtains the induced electromotive force obtained by the receiving device and outputs a predicted resistivity model.

[0037] Furthermore, the mobile device of the system is a vehicle, the positioning device is a GPS positioning device, and the navigation device is a GIS navigation device. The transient electromagnetic instrument, the transmitting device, and the receiving device are arranged on the vehicle and maintained at a certain distance. The GPS positioning device is used to obtain the location information of the transmitting device and the receiving device, and the GIS navigation device is used to navigate the vehicle.

[0038] The advantages and beneficial effects of the present invention are:

[0039] A transient electromagnetic scanning detection system and method of the present invention adopts a 180° ground scanning detection mode, which improves the coverage of observation data and reduces the non-uniqueness of inversion. In addition, the traditional TEM detection method can only detect the underground resistivity structure directly below, and approximates the inversion model as a one-dimensional layered medium, which is insufficient to describe the lateral heterogeneity of the underground medium. The present invention scans the medium below the coil on a 2D profile, which can better detect the underground lateral heterogeneous structure and avoid erroneous model interpretation. In the present invention, the transmitting coil and the receiving coil are separated by 8 to 10 meters, which avoids electromagnetic coupling between the two coils, greatly reduces the detection blind area, and improves the signal-to-noise ratio of the shallow part. Finally, TEM scanning detection uses a deep learning model to realize the prediction of underground two-dimensional structure. Once the training is completed, the target can be predicted in seconds, which greatly improves the computing efficiency and lays the foundation for the realization of real-time inversion imaging. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] FIG1 is a schematic diagram of the structure of a system in an embodiment of the present invention.

[0041] FIG2 is a flow chart of a method according to an embodiment of the present invention.

[0042] FIG3 is a UNet framework structure diagram of the TEM scanning detection model in an embodiment of the present invention.

[0043] FIG4a is a comparison result diagram of part of the real model of the test set and the model predicted by UNet in an embodiment of the present invention.

[0044] FIG4 b is a second diagram showing the comparison results of some real models of the test set and the models predicted by UNet in an embodiment of the present invention.

[0045] FIG4c is a third diagram showing the comparison results of some real models of the test set and the models predicted by UNet in an embodiment of the present invention.

[0046] FIG4 d is a fourth diagram showing the comparison results of some real models of the test set and the models predicted by UNet in an embodiment of the present invention.

[0047] FIG5 is a true resistivity model of the synthetic model 1 and its corresponding inversion model in an embodiment of the present invention.

[0048] FIG6 is a true resistivity model of the synthetic model 2 and its corresponding inversion model according to an embodiment of the present invention.

[0049] FIG7 is a schematic structural diagram of a device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The following describes the specific embodiments of the present invention in detail with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention and are not intended to limit the present invention.

[0051] A transient electromagnetic scanning survey system was designed to provide a fast, high-precision shallow-surface survey system with high lateral and vertical resolution (approximately 10 meters). It can resolve resistivity structures from the top 2-3 meters to at least 40 meters, which is the limit of electromagnetic field resolution based on diffusion equations. The system boasts high signal acquisition efficiency and efficient inversion imaging computation speed, enabling scanning surveys of areas ranging from a few hectares to approximately 10 square kilometers within a reasonable time and cost. As shown in Figure 1, the instrument is carried by an all-terrain vehicle (ATV), which tows the transmitter frame (Tx coil) and the receiver coil (Rx coil). The two coils are mounted on the vehicle with a 10-meter offset. The ATV driver uses a tablet computer for navigation and data collection. The navigation software utilizes a geographic information system (GIS) and includes survey path and line numbering, status parameters, terrain relief, and integration with real-time GPS input. The geolocation of the coils is recorded using a two-satellite augmented GPS system.

[0052] The transmitter and receiver instruments of the TEM scanning system employ a design similar to that of the tTEM system to achieve the design objectives. The scanning detection system uses a dipole transmitter to measure the induced electromotive force data at the early and late stages after each transmit pulse. The receiving coils are two multi-turn coils with a radius of 0.6 meters, used to measure the induced electromotive force of the vertical Z component and the horizontal X component, respectively. The transmitting coil Tx uses a rotating transmission mode, sampling 19 times within the range θ∈[-90°, 90°] (Figure 1) at intervals of 10°.

[0053] The current emitted by a TEM system diffuses rapidly into the subsurface, making it crucial to achieve very close-to-surface resolution using a TEM system a fast shutdown and immediate, unbiased signal recording. The shutdown time depends primarily on the Tx, the Tx coil's capacitance and self-inductance, and the coil's damping. With a resistance of 200 Ω·m, the system shuts down the approximately 3A transmit current in approximately 2.5 μs, and an unbiased time gate is achieved just 1-2 μs after the shutdown is complete.

[0054] TEM instruments are typically very sensitive and can easily introduce noise / bias due to coupling from conductive objects placed near the instrument or near the Tx and / or Rx coils. Therefore, the Tx frame and Tx and Rx sleds are constructed from non-conductive components (composite materials, wood, and plastic) to avoid any interference from external signals. However, the instrument, cables, GPS receiver, ATV, and other nearby conductive objects can all be potential sources of noise / coupling. The ATV is positioned a certain distance from the Tx / Rx coils, as shown in Figure 1, to minimize the mutual inductance caused by electromagnetic coupling between the instruments and ensure that no significant bias signals are introduced into the data.

[0055] As shown in Figure 2, a transient electromagnetic scanning detection method is proposed in this paper. This method proposes a TEM prediction model based on a deep learning framework. The model input is the scanned TEM data, including the induced electromotive force before and after normalization, and the output is the two-dimensional resistivity structure below the coil. The deep learning model is trained based on a large amount of simulation data. Therefore, it is necessary to first generate a large number of random resistivity models. Then, numerical simulations are performed based on these resistivity models to obtain the electromagnetic field data under TEM scanning observation. The network is then trained and optimized. Specifically, the steps include:

[0056] Step S1: Constructing a training set. The construction of the training set includes two parts, namely obtaining a large number of underground resistivity models and performing forward calculations on the resistivity models. Since the real underground resistivity distribution data is very scarce, the present invention synthesizes a theoretical resistivity model based on the real underground geological situation. Since the performance of the neural network depends to a large extent on the diversity and rationality of the training samples, in order to achieve generalized detection, a sufficient number of sufficiently diverse training data sets must be provided to take into account all possible underground spatial structures. Therefore, the complexity of the synthetic resistivity model is the first issue to be considered. If the model is too complex and its resolution completely exceeds the detection accuracy of TEM, the neural network will not be able to learn its characteristics and the prediction effect will be poor; if the model is too simple, the inversion operator will have a low resolution or will not be universal.

[0057] Based on the above principles, the present invention designs a random model generation method with the following steps:

[0058] Step S1.1: Randomly generate a layered resistivity model R(l), where the resistivity increases gradually with depth, and the minimum resistivity range follows a uniform distribution U(log 10 (1),log 10 (100))Ω·m, the resistivity variation range also obeys the uniform distribution U(log 10 (100),log 10 (10000))Ω·m;

[0059] Step S1.2: randomly generate a two-dimensional model grid m, and the value of each grid is randomly distributed;

[0060] Step S1.3: Smoothing the two-dimensional model. The greater the degree of smoothing, the lower the model resolution.

[0061] Step S1.4: Normalize the data of each layer of the smoothed model;

[0062] Step S1.5: Assuming that the resolution of TEM decays exponentially with depth, the model data of each layer is processed as follows: m(l) = m(l)·R(l)·e (-τl)

[0063] Where l represents the number of layers, τ represents the attenuation exponent, the larger τ is, the smoother the deep model is, e is the natural logarithm, and finally m is the two-dimensional non-uniformly distributed resistivity model, and the resistivity changes more slowly with increasing depth.

[0064] Step S1.6: Perform nearest neighbor interpolation on m using the octree grid to obtain a model that can be used for fast numerical simulation.

[0065] After generating a sufficient number of resistivity models in the above way, the finite volume method is used to perform forward calculations on the models to obtain their TEM responses. The goal of the TEM scanning detection system is to conduct fine detection of the shallow surface, so it is only necessary to simulate a 50×50m 2 The target area is divided into two-dimensional space by octree grid. The grid in the simulation area is relatively dense and gradually becomes sparse outward to improve the calculation efficiency.

[0066] Step S2: Construct a UNet neural network to predict the nonlinear mapping between the ATEM signal and the model space. Compared with traditional neural networks, UNet introduces skip connections to solve problems such as gradient vanishing and gradient exploding, allowing the network to learn at a deeper level. As shown in Figure 3, the input is the TEM response obtained by forward calculation of the resistivity model. During the numerical simulation, the rotating emitter is sampled at 24 points within θ∈[-90°,90°], and the sampling time is off at [10 -5 ,10 -2] range, a radial induced EMF and a tangential induced EMF are obtained for each corresponding θ and shutdown sampling instant. Therefore, the data received within a channel has a dimension of 24*24*2. The data is then taken in absolute value and logarithm, or normalized (mean value subtracted and divided by variance), resulting in a total data dimension of 24*24*4. Forward modeling, or numerical simulation of electromagnetic fields, can employ the finite volume method. In the figure, the first and second 24*24 plots (from top to bottom) show the radial induced EMF data and the tangential induced EMF data after taking the absolute value and logarithm. The radial direction refers to data perpendicular to the coil plane, while the tangential direction refers to the intersection of the coil plane and the XZ plane. Logarithms are taken to avoid large differences in order of magnitude. The horizontal and vertical axes represent the rotation angle and shutdown instant, respectively. The third and fourth 24*24 plots show the radial and tangential induced EMF data after normalization, which preserves the directional information of the electromagnetic field. The length and width of the data are 24, which are the number of time channels and the number of sampling points of the rotation angle respectively.

[0067] The network has 4 layers on each side, including 14 convolutional layers, 3 pooling layers, and 3 transposed convolutional layers. The RELU activation function is applied to the output of the convolutional layers. Batch_normilization is used after the convolution operation to normalize the data to further prevent gradient vanishing or gradient exploding, while also increasing the regularization effect. To reduce data overfitting, L2 regularization is applied to the weights of the output layer neurons. Finally, the mean squared error between the label label and the predicted result predi is used as the metric to judge the performance of the UNet: mse = ‖predi-label‖ 2 +λ‖predi‖ 2

[0068] The second term in the above equation represents L2 regularization of the predicted result, predi, ensuring that it not only fits the label but also is as smooth as possible. This is consistent with the resistivity model designed in this paper. λ represents the weight between data fitting and data smoothing, and the optimal parameter here is 0.01, as determined by experimentation. Furthermore, the step size and filter kernel size can be optimized through multiple tests to select optimal hyperparameters.

[0069] Step S3: Model prediction: Based on the transient electromagnetic scanning detection system, the TEM response is collected and input into the trained UNet neural network to obtain the predicted resistivity model;

[0070] In order to verify the performance of the ATEM two-dimensional inversion algorithm based on deep learning proposed in the present invention, as shown in Figures 4a and 4b, the comparison results of some real models in the test set and the models predicted by UNet are shown. It can be seen that the model predicted by UNet is very consistent with the real model, which shows that the UNet model can learn the features in the training set very well and has strong generalization ability.

[0071] The area of ​​the neural network prediction model is fixed to 50×50m 2 However, in actual exploration, the detected area may exceed this area. For this reason, we proposed a partitioned inversion method. During inversion, the entire area is decomposed into multiple small-scale areas for inversion. The area size and measurement point distribution should be consistent with the training set. The output (or label) of the neural network model refers to the partitioned small-scale area. The inversion results are reassembled in the original order. Different small-scale areas can overlap, and the inversion results of the overlapping parts are processed as a weighted average. Finally, the large-scale inversion model can be reconstructed.

[0072] To this end, the present invention designed two synthetic models as shown in Figures 5 and 6 to verify the algorithm. During observation, the initial transmitting coil position was set to -50m, the receiving coil position was set to -40m, the two coils moved synchronously in the positive X direction, and data was recorded every 10m, resulting in a total of 9 sets of data. Therefore, the final inversion model can be spliced ​​together from 9 inversion models. It can be seen from the inversion results that the inverted model can better outline the spatial position of the anomaly, and the numerical values ​​are in good agreement. This further illustrates the feasibility and performance superiority of the ATEM two-dimensional inversion algorithm based on deep learning in a transient electromagnetic scanning detection method proposed by the present invention.

[0073] Corresponding to the aforementioned embodiment of a transient electromagnetic scanning detection method, the present invention also provides an embodiment of a transient electromagnetic scanning detection device.

[0074] Referring to Figure 7, an embodiment of the present invention provides a transient electromagnetic scanning detection device, which includes a memory and one or more processors. The memory stores executable code, and when the one or more processors execute the executable code, they are used to implement a transient electromagnetic scanning detection method in the above embodiment.

[0075] An embodiment of a transient electromagnetic scanning detection device of the present invention can be applied to any device with data processing capabilities, and the any device with data processing capabilities can be a device or apparatus such as a computer. The device embodiment can be implemented through software, or through hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, it is formed by the processor of any device with data processing capabilities in which it is located to read the corresponding computer program instructions in the non-volatile memory into the memory and run them. From the hardware level, as shown in Figure 7, it is a hardware structure diagram of any device with data processing capabilities in which a transient electromagnetic scanning detection device of the present invention is located. In addition to the processor, memory, network interface, and non-volatile memory shown in Figure 7, the any device with data processing capabilities in which the device in the embodiment is located can also include other hardware according to the actual function of the device with data processing capabilities, which will not be described in detail.

[0076] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0077] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present invention. A person of ordinary skill in the art can understand and implement the present invention without inventive work.

[0078] An embodiment of the present invention further provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, a transient electromagnetic scanning detection method in the above embodiment is implemented.

[0079] The computer-readable storage medium may be an internal storage unit of any device with data processing capabilities described in any of the aforementioned embodiments, such as a hard disk or memory. The computer-readable storage medium may also be an external storage device of any device with data processing capabilities, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capabilities. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and may also be used to temporarily store data that has been output or is to be output.

[0080] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A transient electromagnetic scanning detection method, characterized in that The steps include: Step S1: Design of scanning observation system: The TEM scanning detection system carries a mobile device of the instrument, pulling a transmitting device and a receiving device, and uses a dipole transmitting device to measure the early and late induced electromotive force data after each transmitting pulse; Step S2: construct a training set, use a random resistivity model generation algorithm to generate a large number of resistivity models that meet the actual situation, perform numerical simulation calculations on the resistivity model based on the observation system, and obtain transient electromagnetic TEM response data; Step S3: construct a TEM prediction model, wherein the TEM response data is based on the rotating emission source, and at the turn-off sampling moment of each sampling point at different angles, the corresponding radial induced electromotive force and tangential induced electromotive force are obtained, and the absolute value and logarithm of the induced electromotive force data are taken, as well as the standardization operation. The radial induced electromotive force data after taking the absolute value and logarithm, the tangential induced electromotive force data after taking the absolute value and logarithm, and the standardized radial and tangential induced electromotive force data are input into the TEM prediction model for prediction. The standardization operation is to subtract the mean value from the induced electromotive force and then divide it by the variance. The TEM prediction model is trained based on the true value of the resistivity model as a label and the prediction result to obtain a trained TEM prediction model.

2. A transient electromagnetic scanning detection method according to claim 1, characterized in that In the step S1, the transmitting device uses a transmitting coil, and the receiving device uses a receiving coil installed on the mobile device, with an offset of 6-10 meters between them. The mobile device navigates through the navigation device and collects data at the same time. The receiving coil is used to measure the induced electromotive force of the vertical Z component and the horizontal X component. The transmitting coil adopts a rotating transmission mode and samples in the range of θ∈[-90°,90°]; The random resistivity model generation method in step S2 includes the following steps: Step S2.1: Randomly generate a layered resistivity model R(l), the resistivity increases gradually with depth, the minimum resistivity range obeys the uniform distribution U(log10(1),log10(100))Ω·m, and the resistivity variation range also obeys the uniform distribution U(log10(100),log10(10000))Ω·m; Step S2.2: randomly generate a two-dimensional model grid m, and the value of each grid is randomly distributed; Step S2.3: smoothing the two-dimensional model; Step S2.4: normalize each layer of data of the smoothed model; Step S2.5: Process each layer of model data as follows: m(l)=m(l)·R(l)·e (-τl) Where l represents the number of layers, τ represents the decay exponent, the larger the τ, the smoother the depth model, and e represents the self The logarithm of m finally obtained represents a two-dimensional non-uniformly distributed resistivity model, and the resistivity changes more slowly with increasing depth; Step S2.6: Perform nearest neighbor interpolation on m through the octree grid to obtain a resistivity model for fast numerical simulation.

3. A transient electromagnetic scanning detection method according to claim 1, characterized in that The method comprises the following steps: in step S3, a transmitting device adopts a transmitting coil, a receiving device adopts a receiving coil, TEM response data is obtained by numerical simulation based on an observation system, during simulation, sampling is performed at each corresponding transmitting coil rotation angle, each sampling includes the shut-off moments of multiple channels, and the receiving coil obtains radial induced electromotive force and tangential induced electromotive force, wherein radial refers to a direction perpendicular to the coil plane, and tangential refers to a direction of an intersection line between the coil plane and an XZ plane; the horizontal coordinate represents the rotation angle, and the vertical coordinate represents the shut-off moment, and radial and tangential induced electromotive force data feature graphs after taking absolute values ​​and logarithms, as well as standardized radial and tangential induced electromotive force data feature graphs are respectively constructed, and the predicted induced electromotive force data feature graph is compared with the feature graph corresponding to the original electromotive force, and a TEM prediction model is trained.

4. A transient electromagnetic scanning detection method according to claim 1, characterized in that The method comprises the following steps: the TEM prediction model in step S3 adopts a UNet neural network, including a convolution layer, a downsampling layer, an upsampling layer, and a normalization layer; the input induced electromotive force is subjected to step-by-step feature extraction through the convolution layer and downsampling, and then upsampling and feature extraction are performed step by step; finally, the predicted induced electromotive force is output through the normalization layer, and a direct connection is performed between the upsampling and downsampling layers corresponding to each level; an activation operation is performed on the output of the convolution layer, and normalization is performed through a BN layer.

5. A transient electromagnetic scanning detection training method according to claim 1, characterized in that The following steps are included: In step S3, the mean square error between the label and the prediction result is used as an indicator to judge the performance of UNet: mse = ‖predi-label‖ 2 +λ‖predi‖ 2 in,‖·‖ 2 represents L2 regularization processing, and λ represents the weight between data fitting and data smoothing.

6. A transient electromagnetic scanning detection method according to claim 1, characterized in that The method also includes step S3: model prediction, collecting TEM response data, inputting the trained TEM prediction model, and obtaining a predicted resistivity model.

7. A transient electromagnetic scanning detection method according to claim 6, characterized in that: In step S3, the collected TEM response data is regionally decomposed, the small-scale regions obtained that are consistent with the training set are predicted, and the prediction results are then spliced ​​in the original decomposition order.

8. A transient electromagnetic scanning detection system, comprising an observation system module, a training set construction module, and a TEM prediction module, characterized in that: The observation system module includes a mobile device, a transmitting device, a receiving device, a positioning device and a navigation system; The training set construction module obtains the model resistivity through a random model graph generation algorithm, and performs forward calculation through the resistivity model to obtain TEM response data; The TEM prediction module obtains TEM response data based on the angle of the rotating emission source at each sampling point and the shutdown sampling time, obtains the corresponding radial induced electromotive force and tangential induced electromotive force, takes the absolute value and logarithm of the induced electromotive force data, and performs standardization operations. The radial induced electromotive force data after taking the absolute value and logarithm, the tangential induced electromotive force data after taking the absolute value and logarithm, and the standardized radial and tangential induced electromotive force data are input into the TEM prediction module for prediction. The standardization operation is to subtract the average value from the induced electromotive force and then divide it by the variance. Based on the true value of the resistivity model as a label and the prediction result, the TEM prediction module is trained to obtain a trained TEM prediction module.

9. A transient electromagnetic scanning detection system according to claim 8, comprising a transient electromagnetic instrument, a transmitter, a receiver and a processing module, wherein the transient electromagnetic instrument collects the pulse electromagnetic field emitted by the transmitter through the receiver to obtain an induced electromotive force, and is characterized in that: The transmitting device is a vertically rotating coil; The receiving device includes two sets of coils, vertical and horizontal, which are arranged in coordination with the transmitting device; The processing module includes the trained TEM prediction module, which obtains the induced electromotive force obtained by the receiving device and outputs a predicted resistivity model.

10. A transient electromagnetic scanning detection system according to claim 8, characterized in that: The mobile device of the system is a vehicle, the positioning device is a GPS positioning device, the navigation device is a GIS navigation system, the transient electromagnetic instrument, the transmitting device, and the receiving device are arranged on the vehicle and keep a certain distance, the GPS positioning device is used to obtain the position information of the transmitting device and the receiving device, and the GIS navigation system is used to navigate the vehicle.

Images