An artificial cavity recognition method based on frequency-domain electromagnetic response

Abstract

The present invention belongs to the field of underground detection and is a method for identifying artificial voids based on frequency-domain electromagnetic responses. A one-dimensional forward modeling detection simulation is performed on the artificial voids, a layered earth medium model is established, and a numerical solution of the electromagnetic field is obtained through Matlab programming. The relationship between the magnetic induction intensity and the transmitting and receiving distance when an artificial void is present underground is analyzed. A three-dimensional forward modeling simulation of artificial void detection is performed based on the one-dimensional forward modeling detection simulation. Based on COMSOL, detection models for different types of artificial voids are established to obtain a large amount of electromagnetic response data under different circumstances. A training set is established using the obtained electromagnetic response data. A convolutional neural network is used through deep learning to determine whether an artificial void exists underground. The method recognizes and classifies the artificial void response and obtains an identification result of whether an artificial void exists underground. This method reduces the cost of manual participation and can improve the accuracy and reliability of underground artificial void detection.

Description

Technical Field

[0001] The present invention relates to the field of underground detection, and in particular to a method for identifying frequency-domain electromagnetic responses of underground artificial cavities. Background Art

[0002] Airborne geophysical detection technology is a fast, efficient, non-destructive modern high-tech civil and military technology. It is a fast, accurate non-contact, non-destructive detection technology that can quickly detect man-made voids (underground arsenals, underground ammunition depots, underground command posts) in urban offensive warfare.

[0003] Artificial cavities are often used in military applications for underground defense facilities, offering advantages such as effective concealment and strong resistance to attack. However, due to their high resistivity, they are difficult to identify during electromagnetic detection. Detecting the presence and classifying artificial cavities can reduce the blindness of urban offensive warfare and provide guidance for the design and construction of artificial underground cavities in urban defense. In civilian applications, the development of underground space is a key trend in urban spatial expansion, and the detection of artificial cavities is crucial for its development and utilization, integrated management, and disaster prevention and mitigation.

[0004] Currently, there is limited research in China on the detection and identification of underground man-made cavities. Existing international technologies, such as airborne electromagnetics, airborne gravity, and airborne gamma spectroscopy based on a range of 0-200 meters, are also inaccessible due to limitations in sensitivity and commercial monopoly. Therefore, conducting research in this area to achieve accurate and efficient identification and classification of underground man-made cavities is of great significance to both national defense construction and urban management. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an artificial cavity identification method based on frequency domain electromagnetic response, so as to solve the problem that the cavity is difficult to be identified during the above-mentioned electromagnetic detection.

[0006] The present invention is achieved in this way:

[0007] A method for identifying artificial voids based on frequency domain electromagnetic response, comprising:

[0008] S1 conducts one-dimensional forward detection simulation of artificial cavities, establishes a layered earth medium model, and obtains the numerical solution of the electromagnetic field through Matlab programming, thereby analyzing the relationship between the magnetic induction intensity and the transmission and reception distance when there is an artificial cavity underground;

[0009] S2 performs a three-dimensional forward simulation of artificial cavity detection based on the one-dimensional forward detection simulation in step S1, establishes detection models for different types of artificial cavities based on COMSOL, and obtains a large amount of electromagnetic response data under different conditions. The electromagnetic response data is a square area selected at high altitude, and the magnetic induction intensity of the area is obtained for the presence or absence of cavities and cavities of various sizes.

[0010] S3 uses the obtained electromagnetic response data to establish a training set, and uses a convolutional neural network through deep learning to determine whether there are artificial cavities underground, thereby realizing the recognition and classification of artificial cavity responses and obtaining the recognition result of whether there are artificial cavities underground.

[0011] Furthermore, establishing a layered earth medium model includes: laying a long electrical wire source on the ground and outputting alternating current to the earth, which can radiate alternating electromagnetic waves within the earth and space. Assuming that the earth is a one-dimensional layered model, the x-axis is defined along the direction of the emission source, the y-axis is defined perpendicular to the emission source, and the z-axis is defined perpendicular to the surface downward to establish a spatial rectangular coordinate system. The sampling coil height is set to 30m.

[0012] Furthermore, the filter coefficient designed by Kong is used as the filter coefficient of Hankel numerical integration. According to the response magnetic field formula of the long wire source, the response magnetic field formula of the long wire source is integrated by Hankel through Matlab programming to obtain the response electromagnetic field of any point in one-dimensional space. The response magnetic field formula of the long wire source is as follows:

[0013]

[0014]

[0015]

[0016] Where μ0 is the vacuum permeability, I is the conductor current, and r TE and r TM is the surface reflection coefficient, L is the length of the wire, h is the height of the emission source, λ is the wavelength, J0 is the 0th order Bessel function, J1 is the 1st order Bessel function, R=[(x-x') 2 +y 2 ] 1 / 2 .

[0017] Furthermore, it also includes: creating a graphical user interface based on the layered earth medium model, inputting the resistivity of each layer of underground medium, the thickness of each layer of medium and the transmission frequency parameters in the interface, calling the one-dimensional forward function, and calculating the reflection coefficient of the air layer and the bottom wave impedance Then use the following formula to iterate layer by layer starting from the bottom layer to calculate the surface reflection coefficient:

[0018]

[0019]

[0020]

[0021]

[0022] In the above formula, for the bottommost n layers, there is: Starting from the bottommost layer and iterating layer by layer, at the m-th layer (0 < m ≤ n), there is: σ m = 1 / ρ m , μ m = μ0 = 4π×10 -7 , ε m = ε0 = 8.85×10 -12 F / m, h m is the thickness of the m-th layer, ρ m is the resistivity of the m-th layer;

[0023] According to the surface reflection coefficient, using the magnetic field response formula, calculate the magnetic induction intensity at each observation point, compare the relationship curves of the magnetic induction intensity and the transceiver distance with and without underground artificial cavities, and obtain the values of the magnetic induction intensity at any point on the y-axis in the two cases of with and without cavities.

[0024] Furthermore, in step S2, establish an artificial cavity detection model based on COMSOL, perform mesh partitioning on the area of the artificial cavity detection model, set the material property values for each part of the model, where the setting of the material property values includes conductivity, relative magnetic permeability, and relative permittivity, set the current and frequency of the transmitting antenna, and obtain the magnetic induction intensity distribution result of the target area through simulation calculation, where the target area refers to the three-dimensional space where the artificial cavity detection model conducts simulation.

[0025] Furthermore, the mesh partitioning adopts unstructured tetrahedral mesh partitioning, and the core area is encrypted.

[0026] Furthermore, in step S3, obtain a classifier by training a convolutional neural network, and perform identification and classification of underground artificial cavities through the trained classifier, including:

[0027] Obtain a sample set through multiple frequency-domain electromagnetic three-dimensional forward simulation, construct an original convolutional neural network, and obtain a classifier for detecting underground artificial cavities through training with the sample set;

[0028] Obtain the frequency-domain electromagnetic three-dimensional forward data image;

[0029] The data image is input into the trained convolutional neural network to obtain the classification and recognition results of underground artificial cavities.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] Due to the high resistivity of air, the response of artificial voids is classified as high-resistance anomaly according to electrical characteristics, and is not easy to be identified during detection. The method of the present invention reduces the cost of manual participation and can improve the accuracy and reliability of underground artificial void detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a one-dimensional layered earth medium model diagram in an embodiment of the present invention.

[0033] Figure 2 3 is a comparison curve diagram of the relationship between magnetic induction intensity and transmitting and receiving distance when there is a cavity and when there is no cavity in the embodiment of the present invention.

[0034] Figure 3 This is a three-dimensional model diagram of artificial cavity detection in an embodiment of the present invention.

[0035] Figure 4 It is a grid division diagram of the three-dimensional model of artificial cavity detection in an embodiment of the present invention.

[0036] Figure 5 1 is a magnetic induction intensity distribution diagram of the XZ section where the center of the artificial cavity is located in an embodiment of the present invention, (a) is the whole, and (b) is the XZ section.

[0037] Figure 6 This is a distribution diagram of magnetic induction intensity on a cross-section line in the x direction at a height of 30 m directly above the artificial cavity in an embodiment of the present invention.

[0038] Figure 7 This is a flow chart of the classification and identification of underground artificial cavities according to the present invention. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0040] See also Figure 7 As shown, a method for identifying artificial voids based on frequency domain electromagnetic response includes:

[0041] S1 conducts one-dimensional forward detection simulation of artificial cavities, establishes a layered earth medium model, and obtains the numerical solution of the electromagnetic field through Matlab programming, thereby analyzing the relationship between the magnetic induction intensity and the transmission and reception distance when there is an artificial cavity underground;

[0042] S2 performs a three-dimensional forward simulation of artificial cavity detection based on the one-dimensional forward detection simulation in step S1, establishes detection models for different types of artificial cavities based on COMSOL, and obtains a large amount of electromagnetic response data under different conditions. The electromagnetic response data is a square area selected at high altitude, and the magnetic induction intensity of the area is obtained for the presence or absence of cavities and cavities of various sizes.

[0043] S3 uses the obtained electromagnetic response data to establish a training set, and uses a convolutional neural network through deep learning to determine whether there are artificial cavities underground, thereby realizing the recognition and classification of artificial cavity responses and obtaining the recognition result of whether there are artificial cavities underground.

[0044] First, a one-dimensional forward detection simulation of artificial cavities was performed, and a layered earth medium model was established. The numerical solution of the electromagnetic field was obtained through Matlab calculations, and the relationship between the magnetic induction intensity and the receiving and transmitting distance was analyzed when there were artificial cavities in the ground. Based on COMSOL, a detection model for different types of artificial cavities was established, and a three-dimensional forward simulation of artificial cavity detection was performed. A large amount of electromagnetic response data under different situations was obtained. The above electromagnetic response data was used as the training set for the classifier, and a convolutional neural network was trained to determine whether there were artificial cavities in the ground, thereby realizing the recognition and classification of artificial cavity responses.

[0045] Specifically, the present invention first establishes a layered earth medium model. By laying out a long electrical wire source on the ground and outputting an alternating current to the earth, alternating electromagnetic waves can be radiated within the earth and space. During the propagation of electromagnetic waves, when there are abnormal changes in the electrical structure underground, the propagation path and the amplitude and phase of the electromagnetic field will also change accordingly. Assuming that the earth is a one-dimensional layered model, the x-axis is defined along the direction of the emission source, the y-axis is defined perpendicular to the emission source, and the z-axis is defined perpendicular to the surface to establish a spatial rectangular coordinate system. Taking the three-layer earth model as an example, the spatial rectangular coordinate system and the parameters of each layer of the earth are as follows Figure 1 shown.

[0046] The solution to the finite source in the layered half-space can be solved by Maxwell's frequency domain equations combined with boundary conditions. Mathematically, the general solution to the boundary value problem is the sum of the complementary solution of the homogeneous Helmholtz equation in the source-free region and the special solution of the inhomogeneous Helmholtz equation in the source-containing region. To simplify the solution of Maxwell's equations, the Sekunoff vector potential is introduced: Where F is the electric vector potential and a is the magnetic vector potential. The electromagnetic field in a uniform passive region can always be decomposed into two parts: one part is the electric field component perpendicular to a certain axis (TE mode), and the other part is the magnetic field component perpendicular to the same axis (TM mode). The axis direction can be defined arbitrarily. If the axis direction is defined along the z-axis, then the Sekunoff vector potential is also along the z-axis, that is: A = Au z (TE mode), F = Fu z(TM mode). Therefore, the problem of solving Maxwell's equations can be simplified to the problem of solving the Sekunoff potential function. First, the homogeneous Helmholtz equation in the passive region is established based on the Sekunoff vector potential, and the complementary solution in the passive region is solved. Then, the special solution generated by the horizontal electric dipole at any point in the active space is solved. The general solution of the response potential function generated by the horizontal electric dipole in space can be obtained by adding the special solution and the complementary solution. Referring to the general solution generated by the electric dipole in the entire space, the reflection coefficient of the interface can be used to determine the unknown coefficients in the complementary solution, and finally the response electromagnetic field of the horizontal electric dipole at any point above the one-dimensional layered half space is determined, that is, the response magnetic field formula of the long wire source is obtained:

[0047]

[0048]

[0049]

[0050] Where μ0 is the vacuum permeability, I is the conductor current, and r TE and r TM is the surface reflection coefficient, L is the length of the wire, h is the height of the emission source, λ is the wavelength, J0 is the 0th order Bessel function, J1 is the 1st order Bessel function, R=[(x-x') 2 +y 2 ] 1 / 2 .

[0051] The filter coefficient designed by Kong is used as the filter coefficient of Hankel numerical integration. According to the response magnetic field formula of the long wire source, the response magnetic field formula of the long wire source is integrated through Hankel integration through Matlab programming to obtain the response electromagnetic field of any point in one-dimensional space.

[0052] In response to the surface reflection coefficient in the magnetic field formula, a graphical user interface is created based on the layered earth medium model. The resistivity of each layer of underground medium, the thickness of each layer of medium, and the transmission frequency parameters are input into the interface, and the one-dimensional forward function is called to calculate the reflection coefficient of the air layer. and the bottom wave impedance Then use the following formula to iterate layer by layer starting from the bottom layer to calculate the surface reflection coefficient:

[0053]

[0054]

[0055]

[0056]

[0057] In the above formula, for the bottom n layers: Starting from the bottom layer and iterating layer by layer, at the m-th layer (0 < m ≤ n), there are: σ m = 1 / ρ m and μ m = μ0 = 4π×10 -7 and ε m = ε0 = 8.85×10 -12 F / m, h m is the thickness of the m-th layer, and ρ m is the resistivity of the m-th layer;

[0058] According to the surface reflection coefficient, using the magnetic field response formula, calculate the magnetic induction intensity at each observation point, and compare the relationship curves of the magnetic induction intensity with the transceiver distance in the case of underground artificial cavities or not, to obtain the numerical values of the magnetic induction intensity at any point on the y-axis in the two cases of with or without cavities.

[0059] According to the one-dimensional forward detection simulation, conduct a three-dimensional forward simulation for artificial cavity detection. Based on COMSOL, establish detection models for different types of artificial cavities, and obtain a large amount of electromagnetic response data. The electromagnetic response data is the magnetic induction intensity of a square area at high altitude, obtained respectively in the cases of with or without cavities and cavities of various sizes. See Figure 3 as shown.

[0060] The present invention establishes the following example layered earth model according to the above method: In the depth range of 50 - 200 meters, with a resolution of 15%, establish a simulation model of a homogeneous medium and a three-layered medium. The resistivity of the homogeneous medium is selected as 1Ω.m (clay, medium 1), 100Ω.m (ordinary rock, medium 2), 10000Ω.m (granite, medium 3), and the three-layered medium is combined with the above three. The artificial cavity is simulated as completely air (resistivity 3*1013Ω.m), and 15% of the depth is used as the depth.

[0061] In the above situation, use MATLAB to create a GUI graphical user interface, as Figure 2 shown. Enter the resistivity of each layer medium, the thickness of each layer medium, the emission frequency, and the flight altitude in the interface. After the numerical calculation of the program, the interface will display the corresponding ground magnetic induction intensity curve, the magnetic induction intensity curve at 10m high altitude, the magnetic induction intensity curve at 30m high altitude, and the comparison chart of the results of the three cases.

[0062] Based on a one-dimensional forward modeling, the present invention establishes a three-dimensional artificial cavity detection simulation model consisting of air, earth, and anomalies. A long surface wire source is arranged along the x-direction, with its midpoint located at the origin of the coordinate system. A COMSOL-based artificial cavity detection model is established, the artificial cavity detection model area is meshed, and material property values are set for each part of the model. The material property values include conductivity, relative magnetic permeability, and relative dielectric constant. The transmitting antenna current and frequency are set, and the magnetic induction intensity distribution results of the target area are obtained through simulation calculation. In this embodiment, the cavity is buried at two depths of 50 and 200 meters. The large earth layer size is set to 2000m*2000m*2000m, the atmospheric layer size is set to 2000m*2000m*500m, and a spherical cavity is buried underground at y = 1500m, with a radius d = 7.5m, an outer metal wall thickness of 1.5m, a transmitting pole distance of 500m, a transmitting current of 50A, and a transmitting and receiving distance of 1500m. The air resistivity is set to 10000Ω*m and the metal resistivity is set to 1Ω*m. The mesh is divided as follows Figure 4 The figure shows that the unstructured tetrahedral mesh is used and the core area is encrypted to improve the calculation accuracy. The suitable transmission frequency under different earth resistivity conditions can be obtained by Deduced. Figure 5 The magnetic induction intensity distribution of the XZ section where the cavity center is located in the above example, 5(a) is the whole, 5(b) is the XZ section, Figure 6 It is the value of magnetic induction intensity on the x-direction cross-section line at a height of 30m directly above the cavity in the above example (the lower line is the case without cavity, and the upper line is the case with cavity).

[0063] The underground artificial cavity identification method provided by the present invention obtains a sample set through multiple frequency-domain electromagnetic three-dimensional forward simulations as a training set for the neural network, trains the weights and bias parameters of each layer of the neural network, and adjusts parameters such as the learning rate and the number of learning iterations to improve the accuracy of the CNN neural network in determining the presence or absence of underground artificial cavities. This method can achieve the identification and classification of the presence or absence of underground artificial cavities. Compared with traditional manual identification, the use of convolutional neural networks for classification greatly improves the efficiency of underground artificial cavity identification and classification and has higher accuracy. A classifier is obtained by training the convolutional neural network, and the trained classifier is used to identify and classify underground artificial cavities, specifically including:

[0064] A sample set was obtained through multiple frequency-domain electromagnetic three-dimensional forward simulations, and an original convolutional neural network was constructed. This sample set was then used for training to obtain a classifier for detecting underground artificial cavities.

[0065] Obtain frequency domain electromagnetic three-dimensional forward modeling data images;

[0066] The data image is input into the trained convolutional neural network to obtain the classification and recognition results of underground artificial cavities.

[0067] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for identifying artificial voids based on frequency domain electromagnetic response, characterized in that: include: S1 conducts one-dimensional forward detection simulation of artificial cavities, establishes a layered earth medium model, and obtains the numerical solution of the electromagnetic field through Matlab programming, thereby analyzing the relationship between the magnetic induction intensity and the transmission and reception distance when there is an artificial cavity underground; S2 performs a three-dimensional forward simulation of artificial cavity detection based on the one-dimensional forward detection simulation in step S1, establishes detection models for different types of artificial cavities based on COMSOL, and obtains a large amount of electromagnetic response data under different conditions. The electromagnetic response data is a square area selected at high altitude, and the magnetic induction intensity of the area is obtained for the presence or absence of cavities and cavities of various sizes. S3 uses the obtained electromagnetic response data to establish a training set and uses a convolutional neural network through deep learning to determine whether there is an artificial cavity underground, thereby realizing the recognition and classification of the artificial cavity response and obtaining the recognition result of whether there is an artificial cavity underground; The establishment of a layered earth medium model involves laying out a long electrical wire source on the ground and outputting alternating current to the earth, radiating alternating electromagnetic waves within the earth and space. Assuming the earth is a one-dimensional layered model, the x-axis is defined along the emission source, the y-axis is defined perpendicular to the emission source, and the z-axis is defined perpendicular to the surface downward to establish a spatial rectangular coordinate system. The sampling coil height is set to 30m. The filter coefficient designed by Kong is used as the filter coefficient of Hankel numerical integration. According to the response magnetic field formula of the long wire source, the response magnetic field formula of the long wire source is integrated through Hankel integration through Matlab programming to obtain the response electromagnetic field of any point in one-dimensional space.

2. The artificial cavity identification method based on frequency domain electromagnetic response according to claim 1 is characterized in that: The response magnetic field formula of a long wire source is as follows: Where μ0 is the vacuum permeability, I is the conductor current, and r TE and r TM is the surface reflection coefficient, L is the length of the wire, h is the height of the emission source, λ is the wavelength, J0 is the 0th order Bessel function, J1 is the 1st order Bessel function, R=[(x-x') 2 +y 2 ] 1 / 2 .

3. The artificial cavity identification method based on frequency domain electromagnetic response according to claim 2 is characterized in that: Also includes: Create a graphical user interface based on the layered earth medium model, input the resistivity of each layer of underground medium, the thickness of each layer of medium and the transmission frequency parameters in the interface, call the one-dimensional forward function, and calculate the reflection coefficient of the air layer and the bottom wave impedance Then use the following formula to iterate layer by layer starting from the bottom layer to calculate the surface reflection coefficient: In the above formula, for the bottom n layers: Starting from the bottom layer, at layer m we have: σ m =1 / ρ m , μ m =μ0=4π×10 -7 , ε m =ε0=8.85×10 -12 F / m,h m is the thickness of the m layer, ρ m is the resistivity of layer m, where 0 <m≤n; According to the surface reflection coefficient, the magnetic field response formula is used to calculate the magnetic induction intensity of each observation point. The relationship curves between the magnetic induction intensity and the transmitting and receiving distance with and without underground artificial cavities are compared to obtain the value of the magnetic induction intensity at any point on the y-axis with and without cavities.

4. The artificial cavity identification method based on frequency domain electromagnetic response according to claim 1, characterized in that: In step S2, an artificial cavity detection model based on COMSOL is established, the artificial cavity detection model area is meshed, and the material property values of each part of the model are set. The material property values include electrical conductivity, relative magnetic permeability and relative dielectric constant. The transmitting antenna current and frequency are set, and the magnetic induction intensity distribution result of the target area is obtained through simulation calculation. The target area refers to the three-dimensional space where the artificial cavity detection model is simulated.

5. The artificial cavity identification method based on frequency domain electromagnetic response according to claim 4 is characterized in that: The grid generation adopts unstructured tetrahedral grid generation, and the core area is encrypted.

6. The artificial cavity identification method based on frequency domain electromagnetic response according to claim 1 is characterized in that: Step S3 obtains a classifier by training a convolutional neural network, and uses the trained classifier to identify and classify underground artificial cavities, including: A sample set was obtained through multiple frequency-domain electromagnetic three-dimensional forward simulations, and an original convolutional neural network was constructed. This sample set was then used for training to obtain a classifier for detecting underground artificial cavities. Obtain frequency domain electromagnetic three-dimensional forward modeling data images; The data image is input into the trained convolutional neural network to obtain the classification and recognition results of underground artificial cavities.

Images