High-frequency electromagnetic excitation cable detection image reconstruction method and system

Abstract

The invention provides a high-frequency electromagnetic excitation cable detection image reconstruction method and system, and belongs to the technical field of underground facility multi-field coupling detection and imaging, and the method comprises the steps: applying high-frequency electromagnetic excitation to an underground medium; receiving an electromagnetic signal transmitted through an underground medium, synchronously collecting a sound signal generated by electromagnetic excitation, and preprocessing the electromagnetic signal and the sound signal; performing structure imaging on the electromagnetic signal by adopting a frequency wave number migration imaging algorithm to obtain medium structure distribution; performing electrical parameter imaging on the electromagnetic signal by adopting a Born approximate inverse scattering imaging algorithm to obtain medium electromagnetic characteristic parameter distribution; sound source distribution reconstruction is carried out on the sound signals through a time reversal algorithm, and medium sound source response characteristic distribution is obtained; and inputting imaging results of the three algorithms into a pre-constructed deep reinforcement learning model for dynamic fusion to generate an underground medium comprehensive parameter reconstruction image.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of multi-modal detection and imaging of underground facilities, and particularly relates to a high-frequency electromagnetic excitation cable detection image reconstruction method and system. BACKGROUND

[0002] The large-scale application of urban underground cables puts forward higher requirements for detection technology. The existing methods have significant limitations under complex working conditions: ground penetrating radar method relies on a single physical field signal, and it is difficult to stably reconstruct the cable structure distribution when the soil conditions change; electromagnetic induction method is easily disturbed by adjacent lines, and it is difficult to distinguish between running and non-running states; and the resistive impedance imaging method has insufficient reconstruction efficiency due to the ill-posedness of the inverse problem. These methods generally lack the ability to simultaneously detect multiple parameters.

[0003] The above limitations make it difficult to comprehensively evaluate the cable state in actual operation and maintenance. Defects related to mechanical properties lack effective detection means, and the separation of structure imaging and electrical parameter analysis further increases the risk of misjudgment. SUMMARY

[0004] To solve the problem that the original underground cable detection or imaging method cannot simultaneously use images reflecting the structure and electromagnetic characteristic parameters of the underground cable to quickly and intuitively construct the distribution state of the underground cable, the present application provides a high-frequency electromagnetic excitation cable detection image reconstruction method, which uses the method to carry out research on fast imaging technology of underground cable distribution, and realizes image reconstruction of comprehensive parameters such as underground medium including underground cable, pipeline, soil structure and electrical characteristic parameter distribution.

[0005] To achieve the above technical effects, the technical solution adopted by the present application is as follows:

[0006] The present application provides a high-frequency electromagnetic excitation cable detection image reconstruction method, comprising the following steps:

[0007] Applying high-frequency electromagnetic excitation to the underground medium, the underground medium including cables, pipelines and soil;

[0008] Receiving electromagnetic signals propagated through the underground medium, synchronously collecting acoustic signals generated by electromagnetic excitation, and pre-processing the electromagnetic signals and acoustic signals;

[0009] Performing structure imaging on the electromagnetic signals using a frequency-wavenumber migration imaging algorithm to obtain medium structure distribution;

[0010] Performing electrical parameter imaging on the electromagnetic signals using a Born approximation inverse scattering imaging algorithm to obtain medium electromagnetic characteristic parameter distribution;

[0011] Performing sound source distribution reconstruction on the acoustic signals using a time reversal algorithm to obtain medium sound source response characteristic distribution;

[0012] The medium structure distribution, the medium electromagnetic characteristic parameter distribution, and the medium sound source response characteristic distribution are input into a pre-constructed deep reinforcement learning model for dynamic fusion to generate a comprehensive parameter reconstruction image of the underground medium.

[0013] Optionally, the structure imaging by using the frequency-wavenumber migration imaging algorithm comprises:

[0014] Performing two-dimensional Fourier transform on the preprocessed electromagnetic signal to generate a wavenumber-frequency domain data set;

[0015] Mapping the wavenumber-frequency domain data to a depth-wavenumber domain data set by using Stolt interpolation;

[0016] Performing inverse Fourier transform on the depth-wavenumber domain data set to output a medium structure distribution image.

[0017] Optionally, the electrical parameter imaging by using the Born approximation inverse scattering imaging algorithm comprises:

[0018] Establishing a relationship equation of the total electromagnetic field and the scattering field component to separate the scattering field component from the received total electromagnetic field;

[0019] Linearizing the relationship equation into an integral relationship equation of a target body contrast function and an incident field based on the Born approximation;

[0020] Calculating the incident field distribution generated by the electromagnetic excitation in the underground medium based on the spectral characteristics of the high-frequency electromagnetic excitation and the dyadic Green's function;

[0021] Numerically analyzing the dyadic Green's function by the stationary phase method to construct a wavenumber domain Green's function expression containing medium propagation characteristics;

[0022] Substituting the separated scattering field component, the incident field distribution, and the wavenumber domain Green's function expression into the integral relationship equation to solve a contrast function reflecting the electromagnetic characteristic difference between the target body and the background medium, and obtaining a medium electromagnetic characteristic parameter distribution image.

[0023] Optionally, the numerical analysis of the dyadic Green's function by the stationary phase method comprises:

[0024] For the phase function of the dyadic Green's function, solving a stationary phase point equation of the derivative of the phase function being zero to determine a stationary phase point value of a dominant wavenumber component;

[0025] Calculating the reciprocal of the sum of the vertical wavenumber components of the free space and the background medium in the dyadic Green's function as an amplitude factor;

[0026] Combining the stationary phase point value and the amplitude factor to construct a wavenumber domain Green's function expression.

[0027] Optionally, the sound signal is reconstructed by using a time reversal algorithm to reconstruct the sound source distribution, comprising:

[0028] A transient acoustic field propagation equation is constructed for electromagnetic excitation, which describes the relationship between the time and space distribution of the sound pressure field and the sound source intensity distribution, wherein the background medium sound speed is a fixed value;

[0029] The sound pressure signals collected by the ultrasonic transducer array arranged on the closed surface are preprocessed by band-pass filtering and time domain alignment;

[0030] Based on the transient acoustic field propagation equation, a time reversal operation is performed, comprising:

[0031] The preprocessed closed surface sound pressure signal is taken as an equivalent secondary sound source; the second order time derivative of the equivalent secondary sound source signal is calculated; and the second order time derivative of the equivalent secondary sound source signal is spatially weighted and integrated according to the normal direction of the closed surface and the geometric relationship between the sound source point and the transducer coordinates;

[0032] The sound source intensity distribution in the target region is inverted based on the result of the spatially weighted integration; and the sound source intensity distribution is normalized and output as a medium sound source response characteristic distribution image.

[0033] Optionally, the construction of the deep reinforcement learning model comprises:

[0034] The medium structure distribution image, the electromagnetic characteristic parameter image and the sound source response characteristic distribution image are combined into a spatial three-dimensional input feature tensor, wherein each spatial position contains three channel feature values as the state input of the deep reinforcement learning model;

[0035] A spatially adaptive three-channel fusion weight matrix is defined as the action space output, and the sum of the weights of each channel is constrained to be 1;

[0036] A reward function is established by combining a structure similarity index, an information entropy index and a weight smoothness constraint;

[0037] The policy network parameters of the deep reinforcement learning model are iteratively optimized by a policy gradient algorithm until the reward function converges.

[0038] Optionally, the dynamic fusion of the medium structure distribution, the medium electromagnetic characteristic parameter distribution and the medium sound source response characteristic distribution into the pre-constructed deep reinforcement learning model comprises:

[0039] The medium structure distribution image, the medium electromagnetic characteristic parameter distribution image and the medium sound source response characteristic distribution image are input into the pre-constructed deep reinforcement learning model, which performs the following operations:

[0040] The input three images are spatially aligned and histogram normalized;

[0041] The spatial detail features of the three processed images are extracted by a local convolutional network, and the global semantic features of the three processed images are extracted by combining a global attention mechanism;

[0042] The spatial detail features and the global semantic features are spliced and input into a full connection network to generate three groups of original fusion weight matrices;

[0043] Based on the conductivity noise level of the medium electromagnetic characteristic parameter distribution image, the three groups of original fusion weight matrices are exponentially decayed and corrected;

[0044] Based on the corrected fusion weight matrix, the pixel-level weighted fusion is performed on the normalized medium structure distribution image, the electromagnetic characteristic parameter distribution image and the sound source response characteristic distribution image to generate a preliminary reconstruction image;

[0045] An anisotropic diffusion filter is used to denoise and smooth the preliminary reconstruction image, and an underground medium comprehensive parameter reconstruction image is output.

[0046] The second aspect of the application provides a high-frequency electromagnetic excitation cable detection image reconstruction system based on the high-frequency electromagnetic excitation cable detection image reconstruction method of the first aspect of the application, and the system comprises:

[0047] A high-frequency electromagnetic excitation module is used to apply a high-frequency electromagnetic excitation signal to an underground medium, and the underground medium comprises a cable, a pipeline and soil;

[0048] A signal receiving module is used to receive an electromagnetic signal propagated through the underground medium and synchronously collect an acoustic signal generated by electromagnetic excitation;

[0049] A preprocessing module is used to preprocess the electromagnetic signal and the acoustic signal;

[0050] A structure imaging module is used to perform structure imaging on the electromagnetic signal by using a frequency-wavenumber migration imaging algorithm to obtain a medium structure distribution;

[0051] An electromagnetic parameter imaging module is used to perform electric parameter imaging on the electromagnetic signal by using a Born approximation inverse scattering imaging algorithm to obtain a medium electromagnetic characteristic parameter distribution;

[0052] An acoustic characteristic imaging module is used to perform sound source distribution reconstruction on the acoustic signal by using a time reversal algorithm to obtain a medium sound source response characteristic distribution;

[0053] A dynamic fusion module comprises a deep reinforcement learning model and is used to input the medium structure distribution, the medium electromagnetic characteristic parameter distribution and the medium sound source response characteristic distribution into a pre-constructed deep reinforcement learning model to perform dynamic fusion and generate an underground medium comprehensive parameter reconstruction image.

[0054] Optionally, the structure imaging module comprises:

[0055] a two-dimensional Fourier transform unit configured to perform a two-dimensional Fourier transform on the pre-processed electromagnetic signals to generate a wave-number-frequency domain dataset;

[0056] a Stolt interpolation unit configured to map the wave-number-frequency domain dataset to a depth-wave-number domain dataset using Stolt interpolation;

[0057] an inverse Fourier transform unit configured to perform an inverse Fourier transform on the depth-wave-number domain dataset to output a medium structure distribution image.

[0058] Optionally, the electromagnetic parameter imaging module comprises:

[0059] a scattered field separation unit configured to establish a relationship equation between a total electromagnetic field and a scattered field component, and separate the scattered field component from the received total electromagnetic field;

[0060] a linearization processing unit configured to linearize the relationship equation to an integral relationship equation between a target body contrast function and an incident field based on Born approximation;

[0061] an incident field calculation unit configured to calculate an incident field distribution generated by the high-frequency electromagnetic excitation in the underground medium based on spectral characteristics of the high-frequency electromagnetic excitation and a dyadic Green's function;

[0062] a dyadic Green's function analysis unit configured to numerically analyze the dyadic Green's function by a stationary phase method to construct a wave-number domain Green's function expression containing medium propagation characteristics;

[0063] a contrast function solving unit configured to substitute the separated scattered field component, the incident field distribution and the wave-number domain Green's function expression into the integral relationship equation to solve a contrast function reflecting electromagnetic characteristic differences between the target body and the background medium, and obtain a medium electromagnetic characteristic parameter distribution image.

[0064] Optionally, the dyadic Green's function analysis unit comprises:

[0065] a phase function processing sub-unit configured to solve a stationary phase point equation with a derivative of a phase function of the dyadic Green's function being zero for the phase function, and determine a stationary phase point value of a dominant wave-number component;

[0066] an amplitude factor calculation sub-unit configured to calculate a reciprocal of a sum of vertical wave-number components of free space and the background medium in the dyadic Green's function as an amplitude factor;

[0067] an expression construction sub-unit configured to combine the stationary phase point value and the amplitude factor to construct the wave-number domain Green's function expression.

[0068] Optionally, the acoustic characteristic imaging module comprises:

[0069] a sound field modeling unit configured to construct a transient sound field propagation equation generated by electromagnetic excitation, the equation describing a relationship between a sound pressure field spatio-temporal distribution and a sound source intensity distribution, wherein a background medium sound speed is a fixed value;

[0070] a signal preprocessing unit configured to perform band-pass filtering and time-domain alignment preprocessing on sound pressure signals collected by an ultrasonic transducer array arranged on a closed curved surface;

[0071] a time reversal operation unit configured to perform a time reversal operation based on the transient sound field propagation equation, including:

[0072] taking the preprocessed closed curved surface sound pressure signal as an equivalent secondary sound source; calculating a second-order time derivative of the equivalent secondary sound source signal; and performing spatial weighted integration on the second-order time derivative of the equivalent secondary sound source signal according to a normal direction of the closed curved surface and a geometric relationship between a sound source point and a transducer coordinate;

[0073] a sound source reconstruction unit configured to inverse the sound source intensity distribution in a target region based on a result of the spatial weighted integration; and outputting a medium sound source response characteristic distribution image after normalizing the sound source intensity distribution.

[0074] Optionally, the deep reinforcement learning model in the dynamic fusion module includes:

[0075] a state input unit configured to combine a medium structure distribution image, an electromagnetic characteristic parameter image, and a sound source response characteristic distribution image into a spatial three-dimensional input feature tensor, wherein each spatial position contains three channel feature values as state input of the deep reinforcement learning model;

[0076] an action output unit defining a spatially adaptive three-channel fusion weight matrix as action space output, with a constraint that the sum of the weights of each channel is 1;

[0077] a reward mechanism unit establishing a reward function jointly with a structure similarity index, an information entropy index, and a weight smoothness constraint;

[0078] a parameter optimization unit iteratively optimizing policy network parameters of the deep reinforcement learning model through a policy gradient algorithm until the reward function converges.

[0079] Optionally, the dynamic fusion module includes:

[0080] an input preprocessing unit configured to input a medium structure distribution image, a medium electromagnetic characteristic parameter distribution image, and a medium sound source response characteristic distribution image into a pre-constructed deep reinforcement learning model, the model performing the following operations:

[0081] spatial alignment and histogram normalization processing on the input three images;

[0082] The spatial detail features of the three processed images are extracted by a local convolution network, and the global semantic features of the three processed images are extracted by combining a global attention mechanism;

[0083] The weight generation unit is configured to input the spatial detail features and the global semantic features after splicing into a full connection network to generate three groups of original fusion weight matrices;

[0084] The weight optimization unit is configured to perform exponential decay correction on the three groups of original fusion weight matrices based on the conductivity noise level of the medium electromagnetic characteristic parameter distribution image;

[0085] The image fusion unit is configured to perform pixel-level weighted fusion on the normalized medium structure distribution image, the electromagnetic characteristic parameter distribution image and the sound source response characteristic distribution image based on the corrected fusion weight matrix to generate a preliminary reconstruction image.

[0086] The post-processing unit performs denoising and smoothing processing on the preliminary reconstruction image by using an anisotropic diffusion filter to output an underground medium comprehensive parameter reconstruction image.

[0087] The third aspect of the present application provides an electronic device, including a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the computer program, when loaded into the processor, implements the image reconstruction method of the high-frequency electromagnetic excitation cable detection according to the first aspect of the present application.

[0088] The fourth aspect of the present application provides a computer readable storage medium, the computer readable storage medium stores a computer program, characterized in that the computer program, when executed by a processor, implements the image reconstruction method of the high-frequency electromagnetic excitation cable detection according to the first aspect of the present application.

[0089] The beneficial effects of the present application are as follows:

[0090] 1. The present application solves the problems of low detection resolution and inability to consider different medium and structure characteristics in the prior art by using high-frequency electromagnetic wave excitation technology, combining frequency-wave number migration imaging algorithm and wave number domain Green function analysis optimized by standing phase method, significantly improves the resolution and non-contact detection capability of image reconstruction, and is especially suitable for rapid imaging of soil and multi-material underground medium, and intuitively constructs cable and pipeline distribution network.

[0091] 2. The present application solves the problems of being susceptible to electromagnetic interference and single dependence on electromagnetic parameters in the prior art by using high-frequency electromagnetic excitation and synchronous acquisition of acoustic signals, combining time reversal tomography algorithm to reconstruct sound velocity distribution, improves the detection accuracy and anti-interference ability of underground cable mechanical characteristic parameters, and provides reliable basis for cable mechanical state evaluation.

[0092] 3. This invention uses a deep reinforcement learning model to dynamically fuse the three-modal imaging results of structural distribution, electromagnetic characteristic parameters and mechanical characteristic parameters, which solves the limitations of existing technologies that only reflect structural or electrical parameters. It achieves efficient fusion and visualization of comprehensive parameters of underground media, provides multi-dimensional information support for cable operation and maintenance management, and comprehensively improves the timeliness of hidden danger identification and operation and maintenance level. Attached Figure Description

[0093] Figure 1 A schematic diagram of an image reconstruction method for high-frequency electromagnetic excitation cable detection;

[0094] Figure 2 The steps of image reconstruction method for high-frequency electromagnetic excitation cable detection;

[0095] Figure 3 This is a schematic diagram of a high-frequency electromagnetic excitation cable testing system. Detailed Implementation

[0096] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0097] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0098] This invention proposes a high-frequency electromagnetic excitation cable detection image reconstruction method and system. The method is used to conduct research on rapid imaging technology for underground cable distribution, and realizes image reconstruction of comprehensive parameters of underground media, including underground cables, pipelines, soil structure and electrical characteristic parameter distribution.

[0099] like Figure 1 The diagram illustrates a method for image reconstruction in high-frequency electromagnetically excited cable inspection. Using high-frequency pulsed electromagnetic excitation, a narrow-pulse high-frequency electromagnetic wave is transmitted underground via a directional antenna. After propagating through the underground medium (cable, pipeline, and soil), the electromagnetic signal is received by a broadband receiving antenna, and the acoustic signal generated by the electromagnetically excited underground cable or pipeline is received by an acoustic probe. The electromagnetic and acoustic signals are then used to perform comprehensive parametric image reconstruction of the electromagnetic characteristic parameters and structural distribution of the underground medium. In the diagram, ε... b σ b and μ b These represent the dielectric constant, electrical conductivity, and relative magnetic permeability of the underground medium, respectively.

[0100] In Example 1, the present invention provides a method for image reconstruction of high-frequency electromagnetic excitation cable detection.Figure 2 The high-frequency electromagnetic excitation cable detection image reconstruction method steps are shown, including:

[0101] Step 1, applying a high-frequency electromagnetic excitation to the underground medium.

[0102] Preferably, in the step 1, the underground medium includes a cable, a pipeline and soil.

[0103] Preferably, in the step 1, the applied high-frequency electromagnetic excitation can be implemented by using a plane wave antenna, a dipole antenna, etc., and the electromagnetic excitation mode can be a single sine wave with a fixed main frequency, a continuous sine wave, a Ricker wavelet and a Gaussian pulse, etc.

[0104] Exemplarily, the waveform of the high-frequency electromagnetic excitation is a Gaussian pulse, the spectral center frequency range is 100 MHz to 1 GHz, and the pulse width is 1 ns to 10 ns.

[0105] Step 2, receiving an electromagnetic signal propagated through the underground medium, synchronously collecting an acoustic signal generated by the electromagnetic excitation, and pre-processing the electromagnetic signal and the acoustic signal.

[0106] Preferably, in the step 2, the pre-processing of the signal includes:

[0107] de-noising the electromagnetic signal and time-frequency analysis of the acoustic signal;

[0108] Specifically, the de-noising of the electromagnetic signal uses a wavelet threshold de-noising method, and the time-frequency analysis of the acoustic signal extracts spectral features through a short-time Fourier transform.

[0109] Step 3, using a frequency-wavenumber migration imaging algorithm to perform structural imaging on the electromagnetic signal to obtain a medium structure distribution.

[0110] Preferably, in the step 3, the structural imaging by the frequency-wavenumber migration imaging algorithm includes:

[0111] performing two-dimensional Fourier transform on the pre-processed electromagnetic signal to generate a wavenumber-frequency domain data set;

[0112]

[0113] wherein ω is a frequency, k = ω / v is an amplitude of a wavenumber vector, k x and k z respectively represent components of the wavenumber vector in horizontal and vertical directions;

[0114] mapping the wavenumber-frequency domain data to a depth-wavenumber domain data set by using Stolt interpolation; specifically, performing interpolation calculation to map S(k x , z = 0, ω) from (k x , ω) space to B(kx ,k z ), B(k x ,k z Perform a two-dimensional inverse fast Fourier transform to obtain s(x,z,t=0);

[0115]

[0116] The above formula represents the value of each pixel after the offset, k z This represents the vertical component of the wavenumber vector;

[0117]

[0118] Where v is the propagation speed of electromagnetic waves in the medium;

[0119] Perform an inverse Fourier transform on the depth-wavenumber domain dataset to output a medium structure distribution image:

[0120]

[0121] That is, obtain the medium structure distribution image I1(x,z)=F -1 [B(k x ,k z )).

[0122] It is worth noting that, in response to the problems of low structural imaging resolution and reliance on complex iterative algorithms in existing technologies, this invention adopts a frequency wavenumber shift imaging algorithm (step 3) to perform two-dimensional Fourier transform and Stolt interpolation mapping on the preprocessed electromagnetic signal, and directly generates structural images through inverse Fourier transform. This avoids the iterative solution process in traditional inverse scattering methods, which can improve the reconstruction efficiency of underground medium structure distribution, and is especially suitable for the rapid positioning of cables and pipelines in complex soil environments.

[0123] Step 4: Use the Born approximate inverse scattering imaging algorithm to perform electrical parameter imaging on the electromagnetic signal and obtain the distribution of electromagnetic property parameters of the medium.

[0124] Preferably, in step 4, the electrical parameter imaging using the Born approximate inverse scattering imaging algorithm includes:

[0125] Step 4.1: Based on electromagnetic field theory, establish the relationship equation between the total electromagnetic field and the scattered field components, and separate the scattered field components from the received total electromagnetic field.

[0126]

[0127] Where, r s Let r be the antenna position, r be the imaging area, ω be the angular frequency, and k be the frequency. b E represents the background wavenumber. s(r s ,ω) is the scattering field, Ge(r s , r, ω) represents the dyadic Green's function, E tot (r, r s , ω) represents the total electromagnetic field, χ(r) is the contrast function between the target and the background medium, and is defined as:

[0128]

[0129] where ε eq (r) is the equivalent permittivity of the target at position r, ε eqb is the equivalent permittivity of the background medium, and is used to reflect the difference in electromagnetic properties between the target and the background medium;

[0130] Step 4.2, linearizing the relationship equation into an integral relationship equation of the target contrast function and the incident field based on the Born approximation;

[0131] Further preferably, the step 4.2 comprises:

[0132] The electromagnetic inverse scattering nonlinear integral equation is converted into a linear integral equation using the Born approximation, which is based on the premise that E tot ≈ E in , i.e., the value of the scattering field is small compared to the incident field and can be ignored, at which time the scattering model can be represented as:

[0133]

[0134] where E in (r, r s , ω) is the incident field;

[0135] Step 4.3, calculating the incident field distribution generated by the electromagnetic excitation in the underground medium based on the spectral characteristics of the high-frequency electromagnetic excitation and the dyadic Green's function;

[0136] Specifically, when the excitation source is known, the incident field can be obtained as:

[0137] E in (r, r s , ω) = iωμ0I(ω)Ge(r, r s , ω)

[0138] where I(ω) represents the spectrum of the high-frequency electromagnetic excitation applied in step 1, i represents the imaginary unit, μ0 represents the vacuum magnetic permeability, Ge(r, r s , ω) represents the dyadic Green's function, and after substitution, we obtain:

[0139]

[0140] Step 4.4, constructing the wave number domain Green's function expression containing the medium propagation characteristics by numerically analyzing the dyadic Green's function through the stationary phase method;

[0141] Further preferably, the step 4.4 comprises:

[0142] Step 4.4.1, solving the dominant wave number component through the stationary phase equation:

[0143] For the phase function of the dyadic Green's function, solving the stationary phase equation of the phase function derivative being zero to determine the stationary phase point value of the dominant wave number component, comprising:

[0144] Solving the stationary phase equation of the phase function Φ(k x ) = k x (x s -x) + k bz (z s -z)

[0145] Solving the stationary phase point k x of the phase function Φ(k x0 ), and the corresponding dominant wave number component is:

[0146]

[0147] Wherein, z s represents the position of the antenna in the z-axis direction, z represents the longitudinal coordinate of the point in the imaging area, k0 and k b respectively represent the wave numbers of free space and uniform underground medium, k bz represents the component of the wave number of the uniform underground medium in the z direction:

[0148]

[0149] Step 4.4.2, calculating the amplitude factor:

[0150] Calculating the reciprocal of the sum of the vertical wave number components of free space and background medium in the dyadic Green's function as the amplitude factor, comprising:

[0151] The amplitude factor is determined by the reciprocal of the sum of the vertical wave number components of free space and background medium:

[0152]

[0153] Wherein, k 0z represents the component of the wave number of free space in the z direction;

[0154] Step 4.4.3, constructing the wave number domain Green's function expression:

[0155] Combining the stationary phase point value and the amplitude factor, a wave number domain Green function expression is constructed, including:

[0156] Through stationary phase approximation, a wave number domain expression of dyadic Green function is obtained:

[0157]

[0158] Wherein, is the value of the phase function derivative at the stationary phase point; F(k x0 ) is the amplitude factor corresponding to the stationary phase point;

[0159] Step 4.5, substituting the separated scattering field component, incident field distribution and wave number domain Green function expression into the integral relationship equation, solving the contrast function reflecting the electromagnetic characteristic difference of the target body and the background medium, and obtaining the medium electromagnetic characteristic parameter distribution image.

[0160] Specifically, since the scattering field E s (r s , ω) is measurable, the incident field E in (r, r s , ω) can be solved according to the excitation characteristics and parameters, then χ(r) containing the underground medium electric characteristic parameter information can be solved:

[0161]

[0162] Through least square or regularization method to invert χ(r), the medium electromagnetic characteristic parameter distribution image is generated as:

[0163] I2(x, z) = χ(x, z).

[0164] In view of the problems of complex calculation and low solving efficiency of dyadic Green function in the prior art, the stationary phase method is used to numerically analyze and solve the dyadic Green function, the scattering field integral equation is linearized by combining the Born approximation, the wave number domain Green function expression is constructed, the calculation complexity of the electric parameter imaging is reduced, and the high-frequency electromagnetic signal inversion efficiency is improved.

[0165] Step 5, the time reversal algorithm is used to reconstruct the sound source distribution of the acoustic signal, and the medium sound source response characteristic distribution is obtained.

[0166] Preferably, in the step 5, the sound source distribution reconstruction through the time reversal algorithm comprises:

[0167] (1) Acoustic wave propagation model construction

[0168] A transient acoustic field propagation equation generated by electromagnetic excitation is constructed, which describes the correlation between the acoustic pressure field space-time distribution and the sound source intensity distribution, wherein the background medium sound velocity is a fixed value, including:

[0169] The sound field propagation equation generated by transient electromagnetic excitation is established:

[0170]

[0171] Where p(r, t) is the time-domain distribution of the sound pressure field at spatial position r, c s is the fixed sound speed of the background medium (such as the average sound speed of soil), S(r) is the sound source intensity distribution in the target body region, reflecting the acoustic emission source generated by electromagnetic excitation, and δ(t) is the Dirac impulse function, representing the transient excitation characteristics of the sound source.

[0172] (2) Signal preprocessing and time delay alignment

[0173] The sound pressure signals collected by the ultrasonic transducer array arranged on the closed surface are preprocessed by band-pass filtering and time-domain alignment, including:

[0174] The sound pressure signals p(r i , t) received by the ultrasonic transducer array on the plane S are collected, where r i is the coordinate of the i-th transducer;

[0175] The signals are band-pass filtered to filter out non-target frequency band noise, retain cable discharge characteristic frequency band, and suppress environmental noise and low-frequency interference.

[0176] Based on the transient sound field propagation equation, time reversal operation is performed, including:

[0177] The preprocessed closed surface sound pressure signals are used as equivalent secondary sound sources; the second-order time derivative of the equivalent secondary sound source signals is calculated; and the second-order time derivative of the equivalent secondary sound source signals is spatially weighted integrated according to the normal direction of the closed surface and the geometric relationship between the sound source point and the transducer coordinates. Specifically, it includes:

[0178] (3) Calculate the second-order time derivative of sound pressure

[0179] For each ultrasonic transducer received signal, the second-order derivative at the time delay is calculated:

[0180]

[0181] Where the time delay term represents the time for the sound wave to propagate from the sound source point r' to the transducer r i ; the second-order derivative p" extracts the transient characteristics of the sound source, which matches the δ(t) excitation characteristics in the sound wave propagation model.

[0182] (4) Sound source distribution reconstruction and discretization implementation

[0183] Based on the focusing characteristics of time reversal wave field, the sound source distribution is inverted by the sound pressure integration on the closed surface:

[0184]

[0185] Wherein, Σ is the plane where the ultrasonic transducer is located, and n is the normal vector of the plane Σ at the sound source point r';

[0186] Discretization implementation:

[0187]

[0188] Wherein, ΔS i is the plane microelement area corresponding to the i-th transducer.

[0189] Specifically, the sound source of any point in the sound field can be reconstructed by the sound pressure signal and its derivative information collected on a closed surface outside the sound source. When reconstructing the sound source, any closed surface is selected outside the sound source to surround the sound source to be reconstructed, the sound pressure signal is collected on the closed surface, which is used as a secondary sound source, the initial sound source is removed, the internal sound pressure of the closed surface is calculated, and then the original sound source distribution information is obtained. The inversion field in the closed surface can be obtained by using the sound pressure on the closed surface.

[0190] Exemplarily, the closed surface is a plane or a geometric surface (such as a spherical surface, a cylindrical surface) composed of an ultrasonic transducer array, and the geometric shape is selected according to the distribution range of the target sound source and the detection environment. For the underground cable detection scene, the closed surface is preferably a plane composed of a linear array of ultrasonic transducers arranged on the ground, and the array spacing is set to 0.1-0.5 meters according to the detection resolution requirement. The closed surface needs to completely surround the sound source area to be reconstructed to ensure the completeness of the sound field information;

[0191] The secondary sound source is defined as the equivalent sound source distribution obtained by time domain alignment and filtering of the sound pressure signal collected on the closed surface, and its physical meaning is to reconstruct the internal sound source distribution by using the sound pressure signal on the closed surface after removing the initial sound source (transient sound field generated by electromagnetic excitation).

[0192] (5) Generating medium sound source response characteristic image

[0193] Based on the result of spatial weighted integration, the sound source intensity distribution in the target area is inverted; after normalization processing, the medium sound source response characteristic distribution image is output, including:

[0194] According to the sound source distribution S(x, z) obtained by inversion, the normalized output is the medium sound source response characteristic distribution image:

[0195]

[0196] I3(x,z) reflects the spatial distribution of acoustic emission sources generated by electromagnetic excitation in the underground medium, and the high-value area corresponds to the abnormal acoustic source position of cable joint discharge, metal fatigue, etc. The image resolution is determined by the sensor array density and the geometric accuracy of the closed integration surface Σ.

[0197] It should be noted that, in order to solve the problem that the prior art relies on a single electromagnetic signal and is susceptible to interference and cannot accurately locate the active acoustic emission source (such as the electromagnetic induction method only reflects electrical characteristics, and passive acoustic detection relies on the uniformity assumption of the medium), the present application directly reconstructs the spatial distribution of the acoustic source by the time reversal method, and correlates the electromagnetic excitation and the acoustic field propagation characteristics based on the closed integration formula, solving the problem of blurred acoustic source positioning and insufficient anti-complex medium interference capability of traditional methods, and realizing non-invasive accurate positioning of active acoustic emission events such as cable partial discharge and metal fracture. Combined with electromagnetic-acoustic source multi-modal data fusion, the reliability and spatial resolution of underground cable defect diagnosis are effectively improved, and cross-physical field collaborative analysis capability is provided for cable condition assessment in complex environments.

[0198] Step 6: input the imaging results of the above three algorithms into a pre-constructed deep reinforcement learning model for dynamic fusion to generate a comprehensive parameter reconstruction image of the underground medium, including the cable, the pipeline and the soil, and the comprehensive parameters include the structural distribution, the electromagnetic characteristic parameter distribution and the acoustic source response characteristic parameter distribution.

[0199] Preferably, in step 6, the construction of the deep reinforcement learning model includes:

[0200] (1) State space construction

[0201] The medium structure distribution image, the electromagnetic characteristic parameter image and the acoustic source response characteristic distribution image are combined into a spatial three-dimensional input feature tensor, wherein each spatial position contains three channel feature values, which are used as the state input of the deep reinforcement learning model, including:

[0202] Define a three-dimensional fusion feature tensor:

[0203] S(x,z) = [I1(x,z), I2(x,z), I3(x,z)] ∈ R H×W×3

[0204] Wherein, I1(x,z), I2(x,z), I3(x,z) are respectively the medium structure distribution image, the medium electromagnetic characteristic parameter distribution image and the medium acoustic source response characteristic distribution image; HxW is the spatial resolution of the imaging area.

[0205] (2) Action space design

[0206] Define a spatially adaptive three-channel fusion weight matrix as the action space output, and constrain the sum of the weights of each channel to be 1, including:

[0207] Generating pixel-level fusion weights:

[0208] A(x,z)=[w1(x,z),w2(x,z),w3(x,z)]∈[0,1] H×W×3

[0209] The constraint condition is:

[0210]

[0211] (3) Reward function design

[0212] The reward function is established by combining the joint structure similarity index, information entropy index and weight smoothness constraint, including:

[0213] Multi-objective optimization function:

[0214]

[0215] Where, I ref is a two-dimensional real image for simulation or experimental verification; SSIM is a structural similarity index (fidelity constraint), EN is an information entropy (information constraint), is the weight gradient L2 norm (smoothness constraint);

[0216] (4) Parameter optimization

[0217] The policy network parameters of the deep reinforcement learning model are updated using the policy gradient algorithm, and the optimization goal is to maximize the reward function value. When the change of the reward value R is less than the preset convergence threshold (for example, the default setting is 0.1%) for three consecutive iterations, it is determined that the model training converges, and the optimization process ends.

[0218] Specifically, the policy network of the deep reinforcement learning model includes:

[0219] Spatial feature extraction network: composed of multiple 3x3 convolution layers, used to extract local spatial detail features of the input image;

[0220] Global attention network: based on the SE (Squeeze-and-Excitation, channel attention) mechanism, used to extract global semantic features of the input image;

[0221] Weight generation network: composed of fully connected layers, used to fuse and process the aforementioned extracted local spatial features and global semantic features, and finally output a three-channel fusion weight matrix with the same spatial resolution as the input image.

[0222] The above network components work together to realize dynamic fusion of the imaging results of the three algorithms.

[0223] Further preferably, the using the policy gradient algorithm to update the policy network parameters of the deep reinforcement learning model comprises:

[0224] The learnable parameters of the above model network components are iteratively optimized using the policy gradient algorithm, including the weights and biases of each convolutional layer, the parameters of the compression and excitation fully connected layers in the attention mechanism, and the weights and biases of the fully connected layer. The dynamic fusion comprises:

[0225] Dynamic fusion weighting:

[0226]

[0227] where N(·) represents histogram equalization preprocessing.

[0228] Preferably, the dynamic fusion and reconstruction of the imaging results of the above three algorithms by the deep reinforcement learning model comprises:

[0229] Step 6.1, spatial alignment and histogram normalization processing is performed on the medium structure distribution image, the medium electromagnetic property parameter distribution image, and the medium mechanical property distribution image;

[0230] Further preferably, in step 6.1, the preprocessing includes spatial alignment and normalization;

[0231] The step 6.1 further comprises generating spatial position encoding, comprising:

[0232] P(x,z)=[sin(2πx / H),cos(2πz / W)]

[0233] where W and H are the width and height of the image, respectively;

[0234] Step 6.2, spatial detail features of the three processed images are extracted by a local convolutional network, and global semantic features of the three processed images are extracted by combining a global attention mechanism, comprising:

[0235] Local feature extraction and global semantic feature extraction are performed on the three preprocessed multi-modal images, and fusion features are generated by fusing the two;

[0236] Further preferably, the step 6.2 comprises:

[0237] A three-dimensional fusion feature tensor is constructed:

[0238] S(x,z)=[I1(x,z),I2(x,z),I3(x,z)]∈R H×W×3

[0239] A double-channel convolutional network is used to perform local feature extraction and global feature extraction, respectively, and the local feature extraction comprises:

[0240]

[0241] where S is a three-dimensional fusion feature tensor, P is a spatial position encoding, is a tensor concatenation operation;

[0242] The global feature extraction includes:

[0243] F global = SEBlock(Conv 1×1 (S))

[0244] Fusion of local features and global features:

[0245] F combined = Concat(F local ,F global )∈R H×W×256

[0246] Step 6.3, input the spatial detail features and the global semantic features after splicing into a full connection network to generate three groups of original fusion weight matrices, based on the conductivity noise level of the medium electromagnetic characteristic parameter distribution image, the three groups of original fusion weight matrices are exponentially decayed and corrected, including:

[0247] Further preferably, the step 6.3 includes:

[0248] (1) input the fusion features after flattening into a double-layer full connection network:

[0249] f flat = Flatten(F combined )∈R HW×256

[0250] w raw = Softmax(W1·ReLU(W2·f flat ))∈R HW×3

[0251] where f flat is the flattened feature matrix, W1 and W2 are the first and second layer weights of the double-layer full connection network respectively, Softmax is applied along the channel dimension, w raw is the original fusion weight matrix;

[0252] (2) correct the weight based on prior knowledge:

[0253] According to the conductivity information of I2(x,z), the weight of the high noise area is suppressed:

[0254]

[0255] where, The original fusion weight of the kth imaging result, The modified fusion weight of the kth imaging result, λ is the attenuation coefficient (default 0.1), and σ(x,z) is the conductivity amplitude.

[0256] Step 6.4, based on the modified fusion weight matrix, pixel-level weighted fusion is performed on the normalized medium structure distribution image, electromagnetic characteristic parameter distribution image and sound source response characteristic distribution image to generate a preliminary reconstruction image, anisotropic diffusion filtering is used to perform denoising and smoothing processing on the preliminary reconstruction image, and an underground medium comprehensive parameter reconstruction image is output.

[0257] Further preferably, the step 6.4 comprises:

[0258] Histogram equalization enhances contrast:

[0259]

[0260] Pixel-level weighted fusion:

[0261]

[0262] Post-processing optimization, anisotropic diffusion filtering smooths noise:

[0263]

[0264] Wherein, κ=0.1 is the gradient threshold, and the iteration is 5 times.

[0265] The prior art discloses a non-contact pulse electromagnetic underground cable detection imaging method in the technical field of electromagnetic detection and imaging, which fuses and images the electrical property parameters and structural distribution of the underground cable by combining high-frequency pulse electromagnetic excitation with received signals, but it depends on an iterative algorithm to solve the electrical property parameters of the underground medium and needs to directly process a complex equation containing dyadic Green's function, and there are problems of low calculation efficiency, slow image reconstruction speed and difficulty in synchronously obtaining mechanical characteristic parameters. In addition, the existing method depends on electromagnetic parameters alone, is susceptible to electromagnetic interference, and cannot comprehensively reflect the structure, electromagnetic characteristics and mechanical characteristic parameters of the underground medium, thereby limiting the detection accuracy and operation and maintenance decision support capability. In view of the problems of subjective deviation caused by the multi-modal data fusion depending on artificial experience, low multi-modal image fusion efficiency and uneven weight distribution in the prior art, the present application constructs a deep reinforcement learning model, takes structural similarity, information entropy and weight smoothness as a reward function, and dynamically generates a spatial self-adaptive weight matrix to dynamically optimize the fusion weight matrix of the three modal images. Meanwhile, the fusion weight is corrected in combination with electromagnetic noise distribution, not only solving the image artifact problem caused by the fixed weight of the traditional fusion method, but also realizing adaptive weighted fusion of the structure, electromagnetic characteristics and acoustic source response characteristic parameters, effectively improving the noise resistance of the comprehensive parameter reconstruction image of the underground medium, the definition of the comprehensive parameter reconstruction image of the underground cable, and the information integrity and visualization effect of the two, thereby providing a high credibility visual basis for operation and maintenance decision.

[0266] The present application provides a high-frequency electromagnetic excitation cable detection image reconstruction system in embodiment 2, based on the high-frequency electromagnetic excitation cable detection image reconstruction method described in embodiment 1, the system comprises:

[0267] A high-frequency electromagnetic excitation module is used to apply a high-frequency electromagnetic excitation signal to the underground medium, and the underground medium includes a cable, a pipeline and soil.

[0268] A signal receiving module is used to receive electromagnetic signals propagated through the underground medium and synchronously collect acoustic signals generated by electromagnetic excitation.

[0269] A preprocessing module is used to preprocess the electromagnetic signals and acoustic signals.

[0270] A structure imaging module is used to perform structure imaging on the electromagnetic signals by using a frequency-wavenumber migration imaging algorithm to obtain the structural distribution of the medium.

[0271] An electromagnetic parameter imaging module is used to perform electrical parameter imaging on the electromagnetic signals by using a Born approximation inverse scattering imaging algorithm to obtain the electromagnetic characteristic parameter distribution of the medium.

[0272] An acoustic characteristic imaging module is used to perform acoustic source distribution reconstruction on the acoustic signals by using a time reversal algorithm to obtain the acoustic source response characteristic distribution of the medium.

[0273] The dynamic fusion module comprises a deep reinforcement learning model, and is configured to input a medium structure distribution, a medium electromagnetic characteristic parameter distribution and a medium sound source response characteristic distribution into a pre-constructed deep reinforcement learning model to perform dynamic fusion, and generate a comprehensive parameter reconstruction image of the underground medium.

[0274] Preferably, the structure imaging module comprises:

[0275] A two-dimensional Fourier transform unit is configured to perform two-dimensional Fourier transform on the pre-processed electromagnetic signal to generate a wave number-frequency domain data set;

[0276] A Stolt interpolation unit is configured to map the wave number-frequency domain data to a depth-wave number domain data set by Stolt interpolation;

[0277] An inverse Fourier transform unit is configured to perform inverse Fourier transform on the depth-wave number domain data set to output a medium structure distribution image.

[0278] Preferably, the electromagnetic parameter imaging module comprises:

[0279] A scattered field separation unit is configured to establish a relationship equation between a total electromagnetic field and a scattered field component, and separate the scattered field component from the received total electromagnetic field;

[0280] A linearization processing unit is configured to linearize the relationship equation into an integral relationship equation between a target body contrast function and an incident field based on the Born approximation;

[0281] An incident field calculation unit is configured to calculate an incident field distribution generated by the high-frequency electromagnetic excitation in the underground medium based on spectral characteristics of the high-frequency electromagnetic excitation and a dyadic Green's function;

[0282] A dyadic Green's function analysis unit is configured to numerically analyze the dyadic Green's function by the stationary phase method to construct a wave number domain Green's function expression containing medium propagation characteristics;

[0283] A contrast function solving unit is configured to substitute the separated scattered field component, the incident field distribution and the wave number domain Green's function expression into the integral relationship equation to solve a contrast function reflecting electromagnetic characteristic differences between the target body and the background medium, and obtain a medium electromagnetic characteristic parameter distribution image.

[0284] Preferably, the dyadic Green's function analysis unit comprises:

[0285] A phase function processing sub-unit is configured to solve a stationary phase point equation with a derivative of a phase function of the dyadic Green's function being zero for the phase function, and determine a stationary phase point value of a dominant wave number component;

[0286] a magnitude factor calculation subunit configured to calculate the reciprocal of the sum of the vertical wave number components of the free space and the background medium in the dyadic Green's function as a magnitude factor;

[0287] an expression construction subunit configured to combine the stationary phase point value and the magnitude factor to construct an expression of the wave number domain Green's function.

[0288] Preferably, the acoustic property imaging module comprises:

[0289] a sound field modeling unit configured to construct a transient sound field propagation equation of electromagnetic excitation, which describes the correlation between the sound pressure field spatio-temporal distribution and the sound source intensity distribution, wherein the sound velocity of the background medium is a fixed value;

[0290] a signal preprocessing unit configured to perform band-pass filtering and time domain alignment preprocessing on the sound pressure signals collected by the ultrasonic transducer array arranged on the closed curved surface;

[0291] a time reversal operation unit configured to perform time reversal operation based on the transient sound field propagation equation, comprising:

[0292] taking the preprocessed closed curved surface sound pressure signal as an equivalent secondary sound source; calculating the second-order time derivative of the equivalent secondary sound source signal; and performing spatial weighted integration on the second-order time derivative of the equivalent secondary sound source signal according to the normal direction of the closed curved surface and the geometric relationship between the sound source point and the transducer coordinates;

[0293] a sound source reconstruction unit configured to inverse the sound source intensity distribution in the target region based on the result of the spatial weighted integration; and outputting the medium sound source response characteristic distribution image after normalization processing of the sound source intensity distribution.

[0294] Preferably, the deep reinforcement learning model in the dynamic fusion module comprises:

[0295] a state input unit configured to combine the medium structure distribution image, the electromagnetic characteristic parameter image and the sound source response characteristic distribution image into a spatial three-dimensional input feature tensor, wherein each spatial position contains three channel feature values as the state input of the deep reinforcement learning model;

[0296] an action output unit defining a spatially adaptive three-channel fusion weight matrix as the action space output, which constrains the sum of the weights of each channel to be 1;

[0297] a reward mechanism unit establishing a reward function jointly with the structure similarity index, the information entropy index and the weight smoothness constraint;

[0298] a parameter optimization unit iteratively optimizing the policy network parameters of the deep reinforcement learning model through a policy gradient algorithm until the reward function converges.

[0299] Preferably, the dynamic fusion module comprises:

[0300] An input preprocessing unit is configured to input the medium structure distribution image, the medium electromagnetic characteristic parameter distribution image and the medium sound source response characteristic distribution image into a pre-constructed deep reinforcement learning model, which performs the following operations:

[0301] The three input images are subjected to spatial alignment and histogram normalization processing;

[0302] The spatial detail features of the three processed images are extracted by a local convolutional network, and the global semantic features of the three processed images are extracted by combining a global attention mechanism;

[0303] A weight generation unit is configured to input the spatial detail features and the global semantic features after splicing into a fully connected network to generate three groups of original fusion weight matrices;

[0304] A weight optimization unit is configured to perform exponential decay correction on the three groups of original fusion weight matrices based on the conductivity noise level of the medium electromagnetic characteristic parameter distribution image;

[0305] An image fusion unit is configured to perform pixel-level weighted fusion on the normalized medium structure distribution image, the electromagnetic characteristic parameter distribution image and the sound source response characteristic distribution image based on the corrected fusion weight matrix to generate a preliminary reconstruction image;

[0306] A post-processing unit is configured to perform denoising and smoothing processing on the preliminary reconstruction image by an anisotropic diffusion filter to output an underground medium comprehensive parameter reconstruction image.

[0307] Preferably, the system further comprises a display module configured to output the underground cable reconstruction image.

[0308] Embodiment 3 of the present application provides an electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the computer program, when loaded into the processor, implements the high-frequency electromagnetic excitation cable detection image reconstruction method according to Embodiment 1.

[0309] Embodiment 4 of the present application provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program, when executed by a processor, implements the high-frequency electromagnetic excitation cable detection image reconstruction method according to Embodiment 1.

[0310] Embodiment 5 of the present application provides an application example of a high-frequency electromagnetic excitation cable detection system, which is based on the high-frequency electromagnetic excitation cable detection image reconstruction system according to Embodiment 2, and the system comprises Figure 3As shown, it comprises a high-frequency electromagnetic excitation source, a transmitting antenna, a receiving antenna, an electromagnetic receiving system, an acoustic sensor, an acoustic signal receiving module, an imaging algorithm 1 module, an imaging algorithm 2 module, an imaging algorithm 3 module, a structure and electromagnetic characteristic parameter fusion unit and an imaging and display unit. The output end of the high-frequency electromagnetic excitation source is connected to the transmitting antenna, and the high-frequency electromagnetic field is sent to the target body through the transmitting antenna. The receiving antenna receives the electromagnetic signal propagated through the underground medium. The output end of the receiving antenna is connected to the input end of the electromagnetic receiving system. The output end of the electromagnetic receiving system is connected to the input end of the imaging algorithm 1 module and the imaging algorithm 2 module. The acoustic sensor receives the acoustic signal. The output end of the acoustic sensor is connected to the acoustic signal receiving module. The output end of the imaging algorithm 1 module, the imaging algorithm 2 module and the imaging algorithm 3 module is connected to the input end of the structure and electromagnetic characteristic parameter fusion unit. The output end of the structure and electromagnetic characteristic parameter fusion unit is connected to the imaging and display unit.

[0311] In a further preferred but non-limiting embodiment, the high-frequency electromagnetic excitation cable detection image reconstruction method utilizes the imaging algorithm 1, the imaging algorithm 2 and the imaging algorithm 3 to jointly realize the structure and electromagnetic characteristic parameter imaging of the underground medium.

[0312] Finally, it should be noted that the above examples are only used to illustrate the technical solutions of the present application and not to limit it. Although the present application has been described in detail with reference to the above examples, those skilled in the art should understand that the specific embodiments of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application. Any modification or equivalent replacement without departing from the spirit and scope of the present application should be covered within the protection scope of the claims of the present application.

Claims

1. A method of high frequency electromagnetic excitation cable detection image reconstruction, characterized in that, The method comprises the following steps: applying high-frequency electromagnetic excitation to the underground medium, the underground medium comprising a cable, a pipeline and soil; receiving electromagnetic signals propagated through the underground medium, synchronously collecting acoustic signals generated by the electromagnetic excitation, and pre-processing the electromagnetic signals and the acoustic signals; performing structural imaging on the electromagnetic signals by using a frequency-wavenumber migration imaging algorithm to obtain a medium structure distribution; performing electrical parameter imaging on the electromagnetic signals by using a Born approximation inverse scattering imaging algorithm to obtain a medium electromagnetic characteristic parameter distribution; performing acoustic source distribution reconstruction on the acoustic signals by using a time reversal algorithm to obtain a medium acoustic source response characteristic distribution; inputting the medium structure distribution, the medium electromagnetic characteristic parameter distribution and the medium acoustic source response characteristic distribution into a pre-constructed deep reinforcement learning model for dynamic fusion to generate a comprehensive parameter reconstruction image of the underground medium.

2. The high-frequency electromagnetic excitation cable detection image reconstruction method according to claim 1, wherein: the structural imaging by using the frequency-wavenumber migration imaging algorithm comprises: performing two-dimensional Fourier transform on the pre-processed electromagnetic signals to generate a wavenumber-frequency domain data set; mapping the wavenumber-frequency domain data into a depth-wavenumber domain data set by using Stolt interpolation; performing inverse Fourier transform on the depth-wavenumber domain data set to output a medium structure distribution image.

3. The high-frequency electromagnetic excitation cable detection image reconstruction method according to claim 1, wherein: the electrical parameter imaging by using the Born approximation inverse scattering imaging algorithm comprises: establishing a relationship equation of total electromagnetic field and scattered field components to separate the scattered field components from the received total electromagnetic field; linearizing the relationship equation into an integral relationship equation of target body contrast function and incident field based on Born approximation; calculating the incident field distribution generated by the high-frequency electromagnetic excitation in the underground medium based on the spectral characteristics of the high-frequency electromagnetic excitation and dyadic Green's function; numerically analyzing the dyadic Green's function by using the stationary phase method to construct a wavenumber domain Green's function expression containing medium propagation characteristics; substituting the separated scattered field components, the incident field distribution and the wavenumber domain Green's function expression into the integral relationship equation to solve the contrast function reflecting the electromagnetic characteristic difference between the target body and the background medium, and obtaining a medium electromagnetic characteristic parameter distribution image.

4. The high-frequency electromagnetic excitation cable detection image reconstruction method according to claim 3, wherein: the numerical analysis of the dyadic Green's function by using the stationary phase method comprises: solving a stationary phase point equation with the derivative of the phase function of the dyadic Green's function being zero to determine the stationary phase point value of the dominant wavenumber component; calculating the reciprocal of the sum of the vertical wavenumber components of free space and background medium in the dyadic Green's function as an amplitude factor; combining the stationary phase point value and the amplitude factor to construct the wavenumber domain Green's function expression.

5. The high-frequency electromagnetic excitation cable detection image reconstruction method according to claim 1, wherein: the acoustic source distribution reconstruction on the acoustic signals by using the time reversal algorithm comprises: A transient acoustic field propagation equation is constructed to describe the relationship between the time-space distribution of the acoustic pressure field and the intensity distribution of the acoustic source, wherein the acoustic velocity of the background medium is a fixed value; The acoustic pressure signals collected by the ultrasonic transducer array arranged on the closed surface are band-pass filtered and time-domain aligned for preprocessing; Based on the transient acoustic field propagation equation, a time reversal operation is performed, including: The preprocessed acoustic pressure signals on the closed surface are taken as equivalent secondary sound sources, and the second-order time derivative of the equivalent secondary sound source signals is calculated; the second-order time derivative of the equivalent secondary sound source signals is spatially weighted and integrated according to the normal direction of the closed surface and the geometric relationship between the sound source point and the transducer coordinates; The intensity distribution of the acoustic source in the target region is inverted based on the result of the spatially weighted integration; and the intensity distribution of the acoustic source is normalized and output as a medium source response characteristic distribution image.

6. The high-frequency electromagnetic excitation cable detection image reconstruction method according to claim 1, wherein: The construction of the deep reinforcement learning model includes: The medium structure distribution image, the electromagnetic characteristic parameter image, and the sound source response characteristic distribution image are combined into a spatial three-dimensional input feature tensor, wherein each spatial position contains three channel feature values as the state input of the deep reinforcement learning model; A spatially adaptive three-channel fusion weight matrix is defined as the action space output, and the sum of the weights of each channel is constrained to be 1; A reward function is established by combining a structural similarity index, an information entropy index, and a weight smoothness constraint; The policy network parameters of the deep reinforcement learning model are iteratively optimized through a policy gradient algorithm until the reward function converges.

7. The high-frequency electromagnetic excitation cable detection image reconstruction method according to claim 1, wherein: The input of the medium structure distribution, the medium electromagnetic characteristic parameter distribution, and the medium sound source response characteristic distribution into the pre-constructed deep reinforcement learning model for dynamic fusion includes: The medium structure distribution image, the medium electromagnetic characteristic parameter distribution image, and the medium sound source response characteristic distribution image are input into the pre-constructed deep reinforcement learning model, which performs the following operations: The input three images are spatially aligned and histogram normalized; The spatial detail features of the processed three images are extracted through a local convolutional network, and the global semantic features of the processed three images are extracted through a global attention mechanism; The spatial detail features and global semantic features are concatenated and input into a fully connected network to generate three groups of original fusion weight matrices; Based on the conductivity noise level of the medium electromagnetic characteristic parameter distribution image, the three groups of original fusion weight matrices are exponentially decayed and corrected; Based on the corrected fusion weight matrices, the normalized medium structure distribution image, the electromagnetic characteristic parameter distribution image, and the sound source response characteristic distribution image are pixel-level weighted and fused to generate a preliminary reconstruction image; An anisotropic diffusion filter is used to denoise and smooth the preliminary reconstruction image, and an underground medium comprehensive parameter reconstruction image is output.

8. A high frequency electromagnetic excitation cable detection image reconstruction system based on the high frequency electromagnetic excitation cable detection image reconstruction method according to any one of claims 1 to 7, characterized in that, The system includes: A high-frequency electromagnetic excitation module for applying a high-frequency electromagnetic excitation signal to an underground medium, the underground medium including a cable, a pipeline, and soil; The signal receiving module receives electromagnetic signals propagated through the underground medium and synchronously collects acoustic signals generated by the electromagnetic excitation; The preprocessing module pre-processes the electromagnetic signals and the acoustic signals; The structural imaging module is configured to perform structural imaging on the electromagnetic signals by using a frequency-wavenumber migration imaging algorithm to obtain a medium structure distribution; The electromagnetic parameter imaging module is configured to perform electromagnetic parameter imaging on the electromagnetic signals by using a Born approximation inverse scattering imaging algorithm to obtain a medium electromagnetic characteristic parameter distribution; The acoustic characteristic imaging module is configured to perform acoustic source distribution reconstruction on the acoustic signals by using a time reversal algorithm to obtain a medium acoustic source response characteristic distribution; The dynamic fusion module includes a deep reinforcement learning model, configured to input the medium structure distribution, the medium electromagnetic characteristic parameter distribution and the medium acoustic source response characteristic distribution into a pre-constructed deep reinforcement learning model to perform dynamic fusion and generate a comprehensive parameter reconstruction image of the underground medium.

9. The high-frequency electromagnetic excitation cable detection image reconstruction system according to claim 8, wherein: The structural imaging module includes: a two-dimensional Fourier transform unit configured to perform two-dimensional Fourier transform on the pre-processed electromagnetic signals to generate a wavenumber-frequency domain data set; a Stolt interpolation unit configured to map the wavenumber-frequency domain data to a depth-wavenumber domain data set by using Stolt interpolation; an inverse Fourier transform unit configured to perform inverse Fourier transform on the depth-wavenumber domain data set to output a medium structure distribution image.

10. The high-frequency electromagnetic excitation cable detection image reconstruction system according to claim 8, wherein: The electromagnetic parameter imaging module includes: a scattered field separation unit configured to establish a relationship equation between a total electromagnetic field and a scattered field component, and separate the scattered field component from the received total electromagnetic field; a linearization processing unit configured to linearize the relationship equation into an integral relationship equation of a target body contrast function and an incident field based on Born approximation; an incident field calculation unit configured to calculate an incident field distribution generated by the high-frequency electromagnetic excitation in the underground medium based on spectral characteristics of the high-frequency electromagnetic excitation and a dyadic Green's function; a dyadic Green's function analysis unit configured to numerically analyze the dyadic Green's function by using a stationary phase method to construct a wavenumber domain Green's function expression containing medium propagation characteristics; a contrast function solving unit configured to substitute the separated scattered field component, the incident field distribution and the wavenumber domain Green's function expression into the integral relationship equation to solve a contrast function reflecting electromagnetic characteristic differences between the target body and the background medium, and obtain a medium electromagnetic characteristic parameter distribution image.

11. The high-frequency electromagnetic excitation cable detection image reconstruction system according to claim 10, wherein: The dyadic Green's function analysis unit includes: a phase function processing subunit configured to solve a stationary phase point equation with a derivative of a phase function of the dyadic Green's function being zero for the phase function, and determine a stationary phase point value of a dominant wavenumber component; an amplitude factor calculation subunit configured to calculate a reciprocal of a sum of vertical wavenumber components of free space and the background medium in the dyadic Green's function as an amplitude factor. An expression building subunit for combining the stationary point values and the amplitude factors to build a wave-number domain Green function expression.

12. The system of claim 8, wherein the dynamic fusion module comprises: The acoustic property imaging module comprises: A sound field modeling unit for building a transient sound field propagation equation generated by electromagnetic excitation, which describes the correlation between the sound pressure field space-time distribution and the sound source intensity distribution, wherein the background medium sound speed is a fixed value; A signal preprocessing unit for performing band-pass filtering and time-domain alignment preprocessing on the sound pressure signals collected by the ultrasonic transducer array arranged on the closed curved surface; A time reversal operation unit for performing time reversal operation based on the transient sound field propagation equation, including: Taking the preprocessed closed curved surface sound pressure signal as an equivalent secondary sound source; calculating the second-order time derivative of the equivalent secondary sound source signal; and performing spatial weighted integration on the second-order time derivative of the equivalent secondary sound source signal according to the normal direction of the closed curved surface and the geometric relationship between the sound source point and the transducer coordinates; A sound source reconstruction unit for inverting the sound source intensity distribution in the target region based on the result of spatial weighted integration; and outputting a medium sound source response characteristic distribution image after normalization processing of the sound source intensity distribution.

13. The system of claim 8, wherein the dynamic fusion module comprises: The deep reinforcement learning model in the dynamic fusion module comprises: A state input unit for combining the medium structure distribution image, the electromagnetic characteristic parameter image and the sound source response characteristic distribution image into a spatial three-dimensional input feature tensor, wherein each spatial position contains three channel feature values as the state input of the deep reinforcement learning model; An action output unit for defining a spatially adaptive three-channel fusion weight matrix as the action space output, and constraining the sum of the weights of each channel to be 1; A reward mechanism unit for establishing a reward function jointly with the structure similarity index, the information entropy index and the weight smoothness constraint; A parameter optimization unit for iteratively optimizing the policy network parameters of the deep reinforcement learning model through a policy gradient algorithm until the reward function converges.

14. The system of claim 8, wherein the dynamic fusion module comprises: The input preprocessing unit for inputting the medium structure distribution image, the medium electromagnetic characteristic parameter distribution image and the medium sound source response characteristic distribution image into the pre-built deep reinforcement learning model, which performs the following operations: Performing spatial alignment and histogram normalization processing on the input three images; Extracting the spatial detail features of the processed three images through a local convolutional network, and extracting the global semantic features of the processed three images through a global attention mechanism; The weight generation unit for concatenating the spatial detail features and the global semantic features and inputting them into a fully connected network to generate three groups of original fusion weight matrices; The weight optimization unit for performing exponential decay correction on the three groups of original fusion weight matrices based on the conductivity noise level of the medium electromagnetic characteristic parameter distribution image; ​ An image fusion unit is configured to perform pixel-level weighted fusion on the normalized medium structure distribution image, the electromagnetic characteristic parameter distribution image and the sound source response characteristic distribution image based on the corrected fusion weight matrix to generate a preliminary reconstruction image; A post-processing unit is configured to perform denoising and smoothing processing on the preliminary reconstruction image by using an anisotropic diffusion filter to output an underground medium comprehensive parameter reconstruction image.

15. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The computer program, when loaded into the processor, implements the image reconstruction method of the high-frequency electromagnetic excitation cable detection according to any one of claims 1-7.

16. A computer-readable storage medium, the computer-readable storage medium storing a computer program, characterized in that, The computer program, when executed by the processor, implements the image reconstruction method of the high-frequency electromagnetic excitation cable detection according to any one of claims 1-7.

Images