A method for detecting embankment leakage hazards based on transient electromagnetic three-dimensional detection using unmanned aerial vehicles (UAVs)

Abstract

The present invention discloses a method for detecting levee leakage hazards using transient electromagnetic three-dimensional (TEEM) technology, carried out by an unmanned aerial vehicle (UAV). The method comprises the following steps: S1: scanning the detection area to obtain remote sensing images and point cloud image data of the levee terrain, and setting the UAV flight trajectory based on a preset detection area; S2: performing a mission flight, transmitting transient electromagnetic pulse signals of varying frequencies into the levee's underground area via a transmitting coil according to a transmission current frequency range; S3: synchronously collecting secondary electromagnetic response signals induced by the underground medium using a receiving coil, and discretizing and sampling the collected signals at sampling intervals to obtain a sampled data sequence; S4: constructing a three-dimensional resistivity model of the levee's underground medium using a three-dimensional inversion algorithm; and S5: extracting multi-dimensional features of the TEEM using the constructed three-dimensional resistivity model to identify areas of levee leakage hazards. The present invention rapidly identifies areas of potential leakage hazards, improving detection efficiency and accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of embankment leakage hazard detection technology, and in particular to a method for embankment leakage hazard detection based on transient electromagnetic three-dimensional detection by unmanned aerial vehicle (UAV). Background Art

[0002] The embankment has been subjected to long-term erosion and infiltration by water, and leakage risks are very likely to occur inside.

[0003] Traditional methods for detecting leakage risks in embankments mainly include geological drilling and electrical sounding. The geological drilling method is a method of directly obtaining rock and soil samples from the inside of the embankment, and determining whether there are leakage risks by analyzing the physical properties of the samples. The advantage of this method is that it can intuitively understand the geological conditions inside the embankment, and the sample data obtained is true and reliable. However, it also has many disadvantages. On the one hand, geological drilling is a destructive detection method that will cause a certain degree of damage to the embankment structure, especially in the case of frequent drilling, which affects the overall stability of the embankment. On the other hand, the drilling process is costly and requires a lot of manpower, material resources and time. In addition, the selection of drilling points is limited, and it cannot fully reflect the situation of the entire embankment, and it is easy to miss some key leakage risk areas.

[0004] The electrical depth sounding method uses the differences in electrical conductivity between different strata to detect underground structures. Its advantages are that it is relatively simple to operate and has a lower cost than geological drilling. It can, to a certain extent, reflect changes in geological structures within a certain depth range underground. However, this method also has obvious drawbacks. The electrical depth sounding method can only obtain approximate resistivity distribution information below the detection point. It does not accurately reflect the lateral changes in the detection area, making it difficult to accurately locate leakage hazards. Moreover, its detection results are easily affected by terrain undulations and surrounding electromagnetic interference factors, resulting in low detection accuracy. In complex geological conditions, misjudgments and missed detections often occur.

[0005] Currently, with the development of science and technology, new detection technologies are constantly emerging, but there are still shortcomings in the detection of embankment leakage risks. For example, although ground-penetrating radar technology can quickly obtain image information at a certain depth underground, its ability to identify leakage risks in low-resistivity media is weak, and its detection depth is limited. In addition, traditional manual inspection methods rely on the experience and sense of responsibility of inspectors, are inefficient and easily influenced by subjective factors, making it difficult to detect some hidden leakage risks. When faced with large-scale embankment detection tasks, existing detection technologies often cannot balance detection efficiency and accuracy, and cannot meet the needs of actual engineering projects for the rapid and accurate detection of embankment leakage risks. Summary of the Invention

[0006] Based on the above, this application discloses a method for detecting embankment leakage hazards using transient electromagnetic three-dimensional detection using a drone, including:

[0007] S1. Scan the detection area with detection equipment to obtain remote sensing images and point cloud image data of the embankment terrain, and set the flight trajectory of the drone according to the preset detection area;

[0008] S2. Perform mission flight according to the flight trajectory of the UAV. During the flight, transient electromagnetic pulse signals of different frequencies are emitted to the underground of the embankment through the transmitting coil according to the frequency range of the transmitting current;

[0009] S3, synchronously collecting secondary electromagnetic response signals induced by the underground medium using a receiving coil, discretizing the collected signals according to a sampling time interval to obtain a sampling data sequence;

[0010] S4. Preprocessing the sampled data sequence to remove noise interference and enhance effective signal characteristics to obtain preprocessed data, and constructing a three-dimensional resistivity model of the underground medium of the embankment through a three-dimensional inversion algorithm;

[0011] S5. Through the constructed three-dimensional resistivity model, the multi-dimensional characteristics of leakage hazards are extracted and the areas with leakage hazards in the embankment are identified.

[0012] Preferably, the flight trajectory in S1 is preset by an ant colony-genetic hybrid optimization algorithm, and the levee length, shape complexity, and terrain undulation are obtained through a three-dimensional spatial information model to determine the effective detection range of the detection equipment. As a basis, set the path planning fitness function , the formula is: ,in is the complexity of the embankment shape, is the length of the embankment, is the terrain undulation; through the path planning fitness function , dynamically adjust the search direction of the ant colony and the crossover mutation probability of the genetic algorithm in each iteration; dynamically determine the distance between adjacent track lines according to path planning , the formula is: :in It is an adjustment factor that forms a flight trajectory with adaptive variable density characteristics to ensure that the UAV flight trajectory fully covers the embankment area to be inspected.

[0013] Preferably, the three-dimensional spatial information model is an image feature matrix extracted by satellite remote sensing image features. And the point cloud feature matrix after processing the lidar point cloud image data , constructed by spatial registration data fusion, the formula is: ,in is the spatial transformation matrix of satellite remote sensing images, is the spatial transformation matrix of the lidar point cloud data, ⊙ is the multiplication of the corresponding elements of the matrix, is the matrix norm, by fusion matrix Construct a three-dimensional spatial information model.

[0014] Preferably, the frequency range of the emission current in S2 is determined according to the geological characteristics of the dike, the detection depth requirements and the performance parameters of the detection equipment, specifically:

[0015] Obtain the conductivity distribution of different depth layers of the embankment ,in The effective detection depth of the dike is combined with the detection equipment and minimum resolvable depth Determine the minimum and maximum frequencies of the emission current using the formula: , ,in and are the maximum and minimum resistivity within the dike, is the vacuum permeability, and the emission current frequency range is obtained The transmitting coil selects currents of different frequencies within the frequency range and transmits transient electromagnetic pulse signals to the underground of the embankment.

[0016] Preferably, in S3, the receiving coil collects the secondary electromagnetic response signal through a phase-locked loop and multi-channel synchronous sampling technology, and phase-locks each channel of the receiving coil array with the trigger signal of the transmitting coil. The moment when the transmitting coil emits the transient electromagnetic pulse signal is used as a reference, so that the phase difference between the sampling clock and the trigger signal of each channel is controlled within ±100ps. The adaptive gain adjustment algorithm is used to adjust the gain of each channel in real time according to the preset detection depth-signal strength model. The formula is: ,in To detect depth, is the initial gain coefficient, The attenuation coefficient is set, and the signal acquisition window is dynamically divided according to the emission current frequency. In each frequency emission cycle, the optimal sampling time interval is automatically matched to collect the secondary electromagnetic response signal induced by the underground medium.

[0017] Preferably, the sampling time interval in S3 is dynamically adjusted according to the signal change intensity, and the formula is: ,in is the minimum sampling period of the detection equipment, is the signal strength change rate, is the adjustment coefficient.

[0018] Preferably, the signal collected in S3 is discretized and sampled by a dynamic adaptive sampling strategy, the frequency component and intensity change rate of the secondary electromagnetic response signal are analyzed in real time, and the sampling is discretized by a multi-resolution sampling mechanism, and the high-frequency part of the signal is oversampled. The formula is: ,in is the highest frequency component of the signal, is the conventional Nyquist sampling frequency, and the undersampling mechanism is used for the low-frequency stable part. The formula is: , The lowest frequency component of the signal is sampled in segments. The number of sampling points in each time window is dynamically determined according to the complexity of the signal to obtain a sampling data sequence containing different frequency and intensity characteristics.

[0019] Preferably, in said S4, a three-dimensional resistivity model of the underground medium of the embankment is constructed by a three-dimensional inversion algorithm, specifically:

[0020] The sampled data sequence is decomposed into multiple scales according to the frequency and time dimensions to obtain signal characteristic components of different resolutions. The structural tensor constraint term is introduced to construct the inversion objective function, which is formulated as follows: ,in is the measured data, Predict data for the model, is the data weight, is the total number of data, is the regularization parameter, The model smoothing constraint term of the structure tensor is used. The three-dimensional model space is divided into multiple sub-regions through a distributed parallel computing architecture. The Jacobian matrix of each sub-region is calculated in parallel using acceleration technology, and the model parameters are updated through an iterative optimization algorithm. During the iteration process, the model parameters are updated through the global search capability of the simulated annealing algorithm and the local optimization advantage of the conjugate gradient method. The formula is: ,in is the step length, The Hessian matrix is used, and the three-dimensional resistivity model of the underground medium of the embankment is constructed through iterative calculation until the error between the model response and the actual sampling data meets the preset threshold.

[0021] Preferably, the multi-dimensional characteristics of leakage hazards in S5 include the gradient change rate of resistivity and the spatial distribution variance of resistivity; the formula for the gradient change rate of resistivity is: ,in is the resistivity, is the three-dimensional spatial coordinate; the spatial distribution variance formula of resistivity is: ,in is the number of leakage risk data points, is the average leakage potential resistivity, For the The resistivity of a leakage hazard.

[0022] Preferably, the step of identifying areas of the dike with potential leakage risks in S5 is as follows:

[0023] A deep convolutional neural network model is constructed, taking multi-dimensional features as input. The deep convolutional neural network model contains multiple convolutional layers, pooling layers, and fully connected layers. Convolution kernels of different scales are used in the convolutional layers to capture resistivity feature information in different spatial ranges. The pooling layer is used to reduce the feature dimension and reduce the amount of calculation. The fully connected layer maps the features to the leakage hazard probability space. The deep convolutional neural network model is trained using a large number of three-dimensional resistivity model samples of embankments with known leakage hazards and those without leakage hazards. The network parameters are optimized using the cross-entropy loss function. The formula is: ,in is the number of samples, is the true label, To predict the probability; after the training is completed, the three-dimensional resistivity model to be tested is input into the trained network, and the leakage hidden danger probability is output according to the output , set the probability threshold , when the leakage probability of the area When the leakage risk is detected, the area is judged to have leakage risk and the area with leakage risk in the embankment is accurately identified.

[0024] Compared with the prior art, the technical solution of this application has the following technical effects:

[0025] This invention achieves efficient and comprehensive coverage of levee areas. In the early stages of surveying, remote sensing images and point cloud image data of the levee terrain are acquired through detection equipment, and drone flight paths are predefined using an ant colony-genetic hybrid optimization algorithm. This algorithm leverages a three-dimensional spatial information model, taking into account factors such as levee length, shape complexity, and terrain relief. It sets a path planning fitness function based on the effective detection range of the detection equipment, dynamically adjusts the ant colony search direction and the genetic algorithm crossover probability accordingly, and simultaneously determines the spacing between adjacent track lines along a path with adaptive variable density. This ensures that the drone's flight path fully covers the levee area to be inspected, avoiding detection blind spots.

[0026] This invention ensures accurate acquisition of secondary electromagnetic response signals. By utilizing a phase-locked loop and multi-channel synchronous sampling technology with the receiving coil, each channel is phase-locked to the trigger signal of the transmitting coil. This ensures that the phase difference between the sampling clock and the trigger signal of each channel is controlled within ±100 ps, ensuring high-precision synchronization of the sampling. Furthermore, an adaptive gain adjustment algorithm adjusts the gain of each channel in real time based on a preset detection depth-signal strength model, enabling stable and accurate signal acquisition at various detection depths. For example, when probing deeper areas, the gain is automatically increased to enhance weak signals, while in shallower areas, the gain is appropriately reduced to avoid signal saturation.

[0027] The present invention constructs a three-dimensional resistivity model to extract the multi-dimensional characteristics of leakage hazards, including the gradient change rate of resistivity and the spatial distribution variance of resistivity, and uses a deep convolutional neural network model to identify leakage hazards, effectively avoiding the misjudgment and missed judgment problems caused by subjective judgment or single feature analysis in traditional methods, providing a reliable basis for the timely discovery and treatment of embankment leakage hazards, and ensuring the safe and stable operation of the embankment.

[0028] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application so that it can be implemented in accordance with the contents of the specification, and to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following is a detailed description of the preferred embodiment of the present application in conjunction with the accompanying drawings.

[0029] Based on the detailed description of the specific embodiments of the present application in conjunction with the accompanying drawings below, those skilled in the art will become more aware of the above and other objects, advantages and features of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without inventive work. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0031] Figure 1 This is a flow chart of the method for detecting embankment leakage hazards using three-dimensional transient electromagnetic technology carried by an unmanned aerial vehicle (UAV).

[0032] Figure 2 A comparison diagram of positioning errors between the present invention and the prior art;

[0033] Figure 3 A depth error comparison diagram between the present invention and the prior art;

[0034] Figure 4 This is a comparison chart of the detection accuracy of the present invention and the prior art. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. In the following description, specific details such as specific configurations and components are provided only to help fully understand the embodiments of the present application. Therefore, it should be clear to those skilled in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. In addition, for clarity and brevity, the description of known functions and structures has been omitted in the embodiments.

[0036] It should be understood that references throughout this specification to "one embodiment" or "this embodiment" mean that a particular feature, structure, or characteristic associated with the embodiment is included in at least one embodiment of the present application. Therefore, the appearance of "one embodiment" or "this embodiment" throughout this specification does not necessarily refer to the same embodiment. Furthermore, these particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0037] In addition, the present application may repeat reference numerals and / or letters in different examples. This repetition is for the purpose of simplicity and clarity and does not in itself indicate the relationship between the various embodiments and / or settings discussed.

[0038] The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist at the same time. The term " / and" in this article describes another type of association object relationship, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the previous and subsequent associated objects are in an "or" relationship.

[0039] The term "at least one" in this article is merely a description of the association relationship between associated objects, indicating that three relationships may exist. For example, at least one of A and B can mean: A exists alone, A and B exist at the same time, and B exists alone.

[0040] It should also be noted that, in this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include," "comprises," or any other variations thereof are intended to cover non-exclusive inclusion.

[0041] Example 1

[0042] This embodiment mainly describes a method for detecting embankment leakage hazards based on transient electromagnetic three-dimensional detection by drones. Figure 1 Shown, including:

[0043] S1. Scan the detection area with detection equipment to obtain remote sensing images and point cloud image data of the embankment terrain, and set the flight trajectory of the drone according to the preset detection area;

[0044] S2. Perform mission flight according to the flight trajectory of the UAV. During the flight, transient electromagnetic pulse signals of different frequencies are emitted to the underground of the embankment through the transmitting coil according to the frequency range of the transmitting current;

[0045] S3, synchronously collecting secondary electromagnetic response signals induced by the underground medium using a receiving coil, discretizing the collected signals according to a sampling time interval to obtain a sampling data sequence;

[0046] S4. Preprocessing the sampled data sequence to remove noise interference and enhance effective signal characteristics to obtain preprocessed data, and constructing a three-dimensional resistivity model of the underground medium of the embankment through a three-dimensional inversion algorithm;

[0047] S5. Through the constructed three-dimensional resistivity model, the multi-dimensional characteristics of leakage hazards are extracted and the areas with leakage hazards in the embankment are identified.

[0048] Furthermore, the flight trajectory in S1 is preset by the ant colony-genetic hybrid optimization algorithm, and the levee length, shape complexity, and terrain relief are obtained through the three-dimensional spatial information model to determine the effective detection range of the detection equipment. As a basis, set the path planning fitness function , the formula is: ,in is the complexity of the embankment shape, is the length of the embankment, is the terrain undulation; through the path planning fitness function , dynamically adjust the search direction of the ant colony and the crossover mutation probability of the genetic algorithm in each iteration; dynamically determine the distance between adjacent track lines according to path planning , the formula is: :in It is an adjustment factor that forms a flight trajectory with adaptive variable density characteristics to ensure that the UAV flight trajectory fully covers the embankment area to be inspected.

[0049] Furthermore, the three-dimensional spatial information model is extracted through the image feature matrix of satellite remote sensing image features. And the point cloud feature matrix after processing the lidar point cloud image data , constructed by spatial registration data fusion, the formula is: ,in is the spatial transformation matrix of satellite remote sensing images, is the spatial transformation matrix of the lidar point cloud data, ⊙ is the multiplication of the corresponding elements of the matrix, is the matrix norm, by fusion matrix Construct a three-dimensional spatial information model.

[0050] Furthermore, the frequency range of the emission current in S2 is determined according to the geological characteristics of the dike, the detection depth requirements and the performance parameters of the detection equipment, specifically:

[0051] Obtain the conductivity distribution of different depth layers of the embankment ,in The effective detection depth of the dike is combined with the detection equipment and minimum resolvable depth Determine the minimum and maximum frequencies of the emission current using the formula: , ,in and are the maximum and minimum resistivity within the dike, is the vacuum permeability, and the emission current frequency range is obtained The transmitting coil selects currents of different frequencies within the frequency range and transmits transient electromagnetic pulse signals to the underground of the embankment.

[0052] Furthermore, in S3, the receiving coil collects the secondary electromagnetic response signal through a phase-locked loop and multi-channel synchronous sampling technology, and phase-locks each channel of the receiving coil array with the trigger signal of the transmitting coil. The phase difference between the sampling clock and the trigger signal of each channel is controlled within ±100ps based on the time when the transmitting coil emits the transient electromagnetic pulse signal. The adaptive gain adjustment algorithm is used to adjust the gain of each channel in real time according to the preset detection depth-signal strength model. The formula is: ,in To detect depth, is the initial gain coefficient, The attenuation coefficient is set, and the signal acquisition window is dynamically divided according to the emission current frequency. In each frequency emission cycle, the optimal sampling time interval is automatically matched to collect the secondary electromagnetic response signal induced by the underground medium.

[0053] Furthermore, the sampling time interval in S3 is dynamically adjusted according to the signal change intensity, and the formula is: ,in is the minimum sampling period of the detection equipment, is the signal strength change rate, is the adjustment coefficient.

[0054] Furthermore, the signals collected in S3 are discretized and sampled through a dynamic adaptive sampling strategy. The frequency components and intensity change rates of the secondary electromagnetic response signals are analyzed in real time. Discrete sampling is performed through a multi-resolution sampling mechanism, and an oversampling mechanism is used for the high-frequency part of the signal. The formula is: ,in is the highest frequency component of the signal, is the conventional Nyquist sampling frequency, and the undersampling mechanism is used for the low-frequency stable part. The formula is: , The lowest frequency component of the signal is sampled in segments. The number of sampling points in each time window is dynamically determined according to the complexity of the signal to obtain a sampling data sequence containing different frequency and intensity characteristics.

[0055] Furthermore, in S4, a three-dimensional resistivity model of the underground medium of the embankment is constructed using a three-dimensional inversion algorithm, specifically:

[0056] The sampled data sequence is decomposed into multiple scales according to the frequency and time dimensions to obtain signal characteristic components of different resolutions. The structural tensor constraint term is introduced to construct the inversion objective function, which is formulated as follows: ,in is the measured data, Predict data for the model, is the data weight, is the total number of data, is the regularization parameter, The model smoothing constraint term of the structure tensor is used. The three-dimensional model space is divided into multiple sub-regions through a distributed parallel computing architecture. The Jacobian matrix of each sub-region is calculated in parallel using acceleration technology, and the model parameters are updated through an iterative optimization algorithm. During the iteration process, the model parameters are updated through the global search capability of the simulated annealing algorithm and the local optimization advantage of the conjugate gradient method. The formula is: ,in is the step length, The Hessian matrix is used, and the three-dimensional resistivity model of the underground medium of the embankment is constructed through iterative calculation until the error between the model response and the actual sampling data meets the preset threshold.

[0057] Furthermore, the multi-dimensional characteristics of leakage hazards in S5 include the gradient change rate of resistivity and the spatial distribution variance of resistivity; the formula for the gradient change rate of resistivity is: ,in is the resistivity, is the three-dimensional spatial coordinate; the spatial distribution variance formula of resistivity is: ,in is the number of leakage risk data points, is the average leakage potential resistivity, For the The resistivity of a leakage hazard.

[0058] Furthermore, S5 identifies areas of the embankment with potential leakage risks, specifically:

[0059] A deep convolutional neural network model is constructed, taking multi-dimensional features as input. The deep convolutional neural network model contains multiple convolutional layers, pooling layers, and fully connected layers. Convolution kernels of different scales are used in the convolutional layers to capture resistivity feature information in different spatial ranges. The pooling layer is used to reduce the feature dimension and reduce the amount of calculation. The fully connected layer maps the features to the leakage hazard probability space. The deep convolutional neural network model is trained using a large number of three-dimensional resistivity model samples of embankments with known leakage hazards and those without leakage hazards. The network parameters are optimized using the cross-entropy loss function. The formula is: ,in is the number of samples, is the true label, To predict the probability; after the training is completed, the three-dimensional resistivity model to be tested is input into the trained network, and the leakage hidden danger probability is output according to the output , set the probability threshold , when the leakage probability of the area When the leakage risk is detected, the area is judged to have leakage risk and the area with leakage risk in the embankment is accurately identified.

[0060] This embodiment describes in detail how to plan a flight trajectory to comprehensively cover the detection area and efficiently acquire data. In signal acquisition and processing, multiple technologies work together to ensure accurate sampling and noise reduction, build a precise three-dimensional resistivity model, and accurately identify areas with leakage risks through multi-dimensional feature extraction and deep convolutional neural networks, greatly improving the efficiency, accuracy and reliability of detection and effectively ensuring the safety of levees.

[0061] Based on Example 1, this implementation describes in detail the specific implementation effects and comparative verification of this application, specifically:

[0062] By comparing the application of the present invention's technology with existing technologies in actual levee leakage detection, the superiority of the present invention's technology in detection efficiency, detection accuracy, and leakage hazard identification accuracy was verified in detail. A 3.5-kilometer-long levee in East China was selected as the experimental area. The levee has typical complexity, and its geological structure includes clay and sand layers. Preliminary investigations have confirmed the presence of multiple leakage risk points of varying degrees in the area. These risk points are widely distributed and at varying depths, covering different locations from shallow surface layers to deeper layers.

[0063] Using detection equipment, we conducted a full-scale scan of the dikes in the experimental area, successfully acquiring remote sensing images and point cloud image data of the dike terrain in the area and constructing a three-dimensional spatial information model.

[0064] In the process of building the model, the shape complexity of the embankment is obtained through calculation of the data. The value is 1.273, which is obtained by quantitatively analyzing the complexity of the levee profile, taking into account the factors such as the curvature of the levee, the number of turning points, and the geometric shape changes at different parts of the levee; the levee length The measured length is 3478.56m, which provides important basic information for subsequent flight trajectory planning; the terrain undulation The calculated value is 4.382m, which reflects the height variation of the terrain in the experimental area and the effective detection range of the detection equipment. The maximum distance the device can effectively detect under ideal conditions is 50.23m.

[0065] Based on the above data, the ant colony-genetic hybrid optimization algorithm is used to plan the flight trajectory of the UAV. Based on the shape complexity, length and terrain undulation of the embankment, the path planning fitness function is set , after calculation, It is 0.0144, reflecting the rationality and effectiveness of flight trajectory planning. The larger the value, the more the flight trajectory can make full use of the effective detection range of the detection equipment under the current conditions and achieve more efficient detection coverage.

[0066] After determining the value of the fitness function, according to the pre-set adjustment factor =0.5, calculate the distance between adjacent track lines of the path , the calculated result is approximately 49.94m. This spacing ensures that adjacent tracklines fully cover the detection area during flight without excessive overlap, thereby improving detection efficiency while ensuring comprehensiveness. Through continuous iterative optimization, a drone flight trajectory with adaptive variable density characteristics was generated. This trajectory automatically adjusts the flight path and trackline spacing based on the actual topography and geology of the experimental area, ensuring comprehensive and complete coverage of the area to be investigated.

[0067] According to the geological survey report of the experimental area, the conductivity distribution of different depth layers of the embankment was obtained in detail. Through in-depth analysis of these data, combined with the effective detection depth of the detection equipment, and minimum resolvable depth , accurately calculate the lowest frequency of the emission current and the highest frequency , after calculation, About 12.56Hz, It is about 156.38 Hz, and the frequency range of the emission current is determined to be [12.56 Hz, 156.38 Hz];

[0068] As the drone flies along its predetermined trajectory, the transmitting coil continuously emits transient electromagnetic pulse signals of varying frequencies within the aforementioned frequency range. These signals propagate underground at specific frequencies and intensities, interacting with the subsurface medium and inducing secondary electromagnetic response signals.

[0069] The receiving coils utilize advanced phase-locked loops and multi-channel synchronous sampling technology to acquire these secondary electromagnetic response signals. During the acquisition process, each channel of the receiving coil array is precisely phase-locked to the trigger signal of the transmitting coil. Using the instant when the transmitting coil emits the transient electromagnetic pulse signal as a reference, a high-precision clock synchronization system strictly controls the phase difference between each channel's sampling clock and the trigger signal to within ±95.3ps. This high-precision phase-locking technology ensures that each channel accurately acquires signals at the same time, avoiding signal acquisition errors caused by phase differences, thereby improving the accuracy and reliability of signal acquisition.

[0070] At the same time, in order to further optimize the signal acquisition effect, an adaptive gain adjustment algorithm is used to adjust the gain of each channel in real time according to the pre-established detection depth-signal strength model. Set to 1.53, the attenuation coefficient Set to 0.12, in the actual detection process, as the detection depth Changes in the gain of each channel It will automatically adjust. Through this adaptive gain adjustment mechanism, it can ensure that stable and appropriately strong signals can be collected at different detection depths, avoiding information loss or distortion caused by signals that are too strong or too weak.

[0071] In terms of sampling time interval, this experiment dynamically adjusts the minimum sampling period of the detection equipment according to the signal change intensity. 0.001s, adjustment coefficient Set to 0.8, and monitor the signal intensity change rate in real time during the experiment The method of dynamically adjusting the sampling time interval can flexibly adjust the sampling frequency according to the real-time changes of the signal, ensuring more intensive sampling when the signal changes drastically and capturing more detailed information; when the signal is relatively stable, the sampling frequency is appropriately reduced to reduce data redundancy and improve data acquisition efficiency.

[0072] In addition, to obtain signal characteristics more comprehensively and accurately, the experiment adopted a dynamic adaptive sampling strategy to discretize the signal sampling. By analyzing the frequency components and intensity change rate of the secondary electromagnetic response signal in real time, a multi-resolution sampling mechanism was used for sampling. For the high-frequency portion of the signal, an oversampling mechanism was used to obtain more accurate high-frequency information; for the low-frequency stable portion, an undersampling mechanism was used to reduce the amount of data and improve processing efficiency. By segmenting the signal, the number of sampling points in each time window was dynamically determined according to the signal complexity, thereby obtaining a sampled data sequence containing different frequency and intensity characteristics. This sampling strategy can effectively reduce the amount of data collected while ensuring the acquisition of sufficient signal information, alleviating the burden of subsequent data processing.

[0073] After acquiring a large amount of sampled data, the data is first preprocessed. The preprocessing process mainly includes two steps: denoising and signal enhancement. The wavelet filtering algorithm is used to effectively remove noise from the data. The wavelet filtering algorithm can accurately separate and remove the noise component based on the characteristic differences between the signal and noise at different frequency scales, while maximally retaining the characteristics of the effective signal. In terms of signal enhancement, the adaptive enhancement algorithm is used to perform targeted enhancement processing on the effective signal based on the statistical characteristics of the signal and prior knowledge, improving the signal clarity and recognizability.

[0074] After preprocessing, the data is decomposed into multiple scales according to the frequency and time dimensions. Through multi-scale analysis methods such as wavelet transform, the signal is decomposed into signal characteristic components of different resolutions. These components of different resolutions contain information about the signal at different frequency and time scales.

[0075] When constructing the 3D resistivity model, the structural tensor constraint term was introduced and the inversion objective function was constructed. , data weight The weight is determined dynamically based on the signal quality. Data points with good signal quality and high stability are given higher weights to highlight their importance in model construction. For data points with poor signal quality and certain noise interference, the weight is appropriately reduced to reduce their impact on model accuracy. The regularization parameter Set to 0.08, it plays a balance between the smoothness and fitting accuracy of the model; by adjusting The value of can control the degree of fit of the model to the data and avoid overfitting or underfitting.

[0076] Using a distributed parallel computing architecture, the three-dimensional model space is divided into multiple sub-regions. In each sub-region, the Jacobian matrix is calculated in parallel using acceleration technology. The Jacobian matrix reflects the degree to which a small change in the model parameters affects the model response. By calculating the Jacobian matrix, it can provide an important basis for the subsequent iterative optimization of the model parameters. During the iterative optimization process, the model parameters are updated by combining the simulated annealing algorithm with the conjugate gradient method. The simulated annealing algorithm has strong global search capabilities, capable of finding the optimal solution within a large parameter space and preventing the model from falling into a local optimum. The conjugate gradient rule, on the other hand, excels at local optimization and can quickly converge to a near-optimal solution. By combining these two algorithms, iterative calculations were performed until the error between the model response and the actual sampled data met a preset threshold (set to 5% in this experiment). After multiple iterations, a three-dimensional resistivity model was successfully constructed that accurately reflects the resistivity distribution of the levee's underground medium.

[0077] After constructing a three-dimensional resistivity model, multi-dimensional features of potential leakage hazards are extracted from the model, including the resistivity gradient change rate and the spatial distribution variance of the resistivity. A deep convolutional neural network model is used to identify potential leakage areas, and the extracted multi-dimensional features serve as input data for the model. The deep convolutional neural network model consists of multiple convolutional layers, pooling layers, and fully connected layers. In the convolutional layers, kernels of varying sizes are used to capture resistivity characteristics across different spatial ranges. Small kernels capture subtle local features, while larger kernels capture broader spatial features. The pooling layers reduce feature dimensionality and computational complexity while preserving important feature information. Max pooling or average pooling operations compress the feature map in the spatial dimension, improving the model's computational efficiency without losing critical information. The fully connected layers map the convolutional and pooled features into a potential leakage probability space, outputting the probability of a potential leakage in each area.

[0078] During the model training phase, a large number of 3D resistivity model samples were collected from embankments with known and no known leakage risks. These samples covered a wide range of geological conditions and leakage levels to ensure the model's broad adaptability and accuracy. The network parameters were optimized using a cross-entropy loss function, and the model's weights and biases were continuously adjusted to ensure that the model's predictions were close to the true labels. Stochastic gradient descent was employed during training to accelerate model convergence and improve training efficiency. After multiple rounds of training, the model concluded when performance on the validation set reached stability and met the preset accuracy requirements.

[0079] After the training is completed, the three-dimensional resistivity model to be tested is input into the trained network, and the leakage hidden danger probability output by the network is calculated. , set the probability threshold =0.6. When the probability of leakage risk in a region, P, is greater than 0.6, the region is considered to have leakage risk. This method can accurately identify areas of embankment with leakage risks, providing precise guidance for subsequent repair and reinforcement work.

[0080] The existing technology used in this experiment was electrical depth sounding, a comparative technique. Probe points were arranged at regular intervals (average intervals of 50m) along the levee in the experimental area. At each probe point, specialized electrical depth sounding equipment was used to obtain resistivity data at different depths by varying the spacing between the power supply electrodes. During operation, technicians, based on experience and relevant standards, observed the changing trends in the resistivity data to determine whether there was a risk of leakage. However, electrical depth sounding has significant limitations. Due to the undulating terrain of the experimental area and the presence of certain electromagnetic interference sources in the surrounding area, these factors significantly affect the detection results of electrical depth sounding. Under complex geological conditions, such as when encountering drastic structural changes or the presence of multiple strata with significantly different conductivity levels, the detection accuracy of electrical depth sounding is severely restricted, making misjudgments and missed detections prone to occur.

[0081] It can be seen intuitively from the table below that the technology of this application has a huge advantage in detection efficiency with the help of drone-mounted equipment. The detection task of 3.5km of embankment was completed in only about 4.56 hours, while the electrical depth sounding method required 12.34 hours. The detection time of the technology of this application is only about one-third of that of the existing technology. At the same time, the technology of this application covered about 850 points through the flight track of the drone. In comparison, the existing technology only arranged about 70 detection points. This shows that the technology of this application can obtain richer detection data in a shorter time, greatly improving the detection efficiency and providing a more efficient solution for large-scale embankment detection.

[0082] Detection technology Time required to complete 3.5km embankment detection (h) Flight / Detection Points This application technology 4.56 850 (number of points covered by the drone flight track) Existing technology 12.34 70 (number of detection points)

[0083] Randomly select 10 detection points to compare positioning and depth errors; Figure 2 As shown, in terms of positioning error, the error range of the technology of the present application is between 1.08m-1.52m, while the positioning error of the existing technology is between 2.98m-4.31m. The positioning error of the technology of the present application is significantly smaller, and the specific location of the leakage hazard can be determined more accurately. Figure 3 As shown, in terms of depth detection error, the error range of the proposed technology is between 0.29m and 0.45m, while the depth detection error of the existing technology is between 1.15m and 1.61m. The proposed technology also shows higher accuracy. This means that the proposed technology can more accurately grasp the actual location and depth of the leakage risk underground, providing a reliable basis for subsequent targeted remediation measures.

[0084] from Figure 4 It can be seen that the recognition accuracy of this application is as high as 92.86%. The recognition accuracy of the existing technology is only 67.86%. This shows that the technology of this application can more effectively screen out areas with real leakage risks from complex geological data through multi-dimensional feature extraction and the powerful recognition ability of deep convolutional neural networks, greatly reducing the probability of misjudgment and missed judgment, and providing more reliable protection for embankment safety detection. Through this comprehensive experiment on the detection of embankment leakage risks under complex geological conditions, the transient electromagnetic three-dimensional detection technology based on drones in this application was deeply compared with the traditional electrical depth sounding method. The experimental results fully demonstrate that the technology of this application is significantly superior to the existing technology in key performance indicators such as detection efficiency, detection accuracy and leakage risk identification accuracy.

[0085] This embodiment verifies the technical solution of the present invention through specific implementation comparison. In terms of detection efficiency, the technology of the present application can complete large-scale embankment detection tasks in a short period of time with the help of the efficient mobility of drones and optimized flight trajectory planning; in terms of detection accuracy, the technology of the present application uses advanced signal acquisition and processing technology, combined with precise model construction and optimization algorithms, to more accurately locate the position and depth of leakage hazards, reducing the treatment deviation caused by errors; in terms of leakage hazard identification accuracy, the technology of the present application uses multi-dimensional feature analysis and the intelligent recognition capabilities of deep convolutional neural networks to accurately distinguish areas with leakage hazards from complex geological data, effectively avoiding misjudgment and missed judgment, and providing a more reliable basis for embankment safety assessment.

[0086] The above are only preferred embodiments of the present invention, which do not limit the scope of protection of the present invention. For those skilled in the art, the present invention can be modified and varied in various ways. Any changes, modifications, replacements, integrations and parameter changes to these embodiments through conventional substitutions or that can achieve the same functions without departing from the principles and spirit of the present invention fall within the scope of protection of the present invention.

Claims

1. A method for detecting embankment leakage hazards based on transient electromagnetic three-dimensional detection by drones, characterized by: include: S1. Scan the detection area with detection equipment to obtain remote sensing images and point cloud image data of the embankment terrain, and set the flight trajectory of the drone according to the preset detection area; S2. Perform mission flight according to the flight trajectory of the UAV. During the flight, transient electromagnetic pulse signals of different frequencies are emitted to the underground of the embankment through the transmitting coil according to the frequency range of the transmitting current; S3, synchronously collecting secondary electromagnetic response signals induced by the underground medium using a receiving coil, discretizing the collected signals according to a sampling time interval to obtain a sampling data sequence; S4. Preprocessing the sampled data sequence to remove noise interference and enhance effective signal characteristics to obtain preprocessed data, and constructing a three-dimensional resistivity model of the underground medium of the embankment through a three-dimensional inversion algorithm; S5. By constructing a three-dimensional resistivity model, we extract the multi-dimensional characteristics of leakage hazards and identify areas with leakage hazards in the embankment. In S3, the receiving coil collects the secondary electromagnetic response signal through a phase-locked loop and multi-channel synchronous sampling technology, and phase-locks each channel of the receiving coil array with the trigger signal of the transmitting coil. The phase difference between the sampling clock and the trigger signal of each channel is controlled within ±100ps based on the time when the transmitting coil emits the transient electromagnetic pulse signal. The adaptive gain adjustment algorithm is used to adjust the gain of each channel in real time according to the preset detection depth-signal strength model. The formula is: ,in To detect depth, is the initial gain coefficient, The attenuation coefficient is used to dynamically divide the signal acquisition window according to the frequency of the transmitted current. In each frequency transmission cycle, the optimal sampling time interval is automatically matched to collect the secondary electromagnetic response signal induced by the underground medium. In S4, a three-dimensional resistivity model of the underground medium of the embankment is constructed by a three-dimensional inversion algorithm. The sampling data sequence is multi-scale decomposed according to the frequency and time dimensions to obtain signal characteristic components of different resolutions. The structural tensor constraint term is introduced to construct the inversion objective function. The formula is: ,in is the measured data, Predict data for the model, is the data weight, is the total number of data, is the regularization parameter, The model smoothing constraint term of the structure tensor is used. The three-dimensional model space is divided into multiple sub-regions through a distributed parallel computing architecture. The Jacobian matrix of each sub-region is calculated in parallel using acceleration technology, and the model parameters are updated through an iterative optimization algorithm. During the iteration process, the model parameters are updated through the global search capability of the simulated annealing algorithm and the local optimization advantage of the conjugate gradient method. The formula is: ,in is the step length, The Hessian matrix is calculated iteratively until the error between the model response and the actual sampling data meets the preset threshold, and the three-dimensional resistivity model of the underground medium of the embankment is constructed; The multi-dimensional characteristics of leakage hazards in S5 include the gradient change rate of resistivity and the spatial distribution variance of resistivity; the formula for the gradient change rate of resistivity is: ,in is the resistivity, is the three-dimensional spatial coordinate; the spatial distribution variance formula of resistivity is: ,in is the number of leakage risk data points, is the average leakage potential resistivity, For the Resistivity of leakage hazards; In S5, the areas with leakage risks in the embankment are identified, and a deep convolutional neural network model is constructed. The multi-dimensional features are used as input. The deep convolutional neural network model includes multiple convolutional layers, pooling layers, and fully connected layers. Convolution kernels of different scales are used in the convolutional layers to capture resistivity feature information in different spatial ranges. The pooling layer is used to reduce the feature dimension and reduce the amount of calculation. The fully connected layer maps the features to the leakage risk probability space. The deep convolutional neural network model is trained by a large number of three-dimensional resistivity model samples of embankments with known leakage risks and without leakage risks. The network parameters are optimized by the cross entropy loss function. The formula is: ,in is the number of samples, is the true label, To predict the probability; after the training is completed, the three-dimensional resistivity model to be tested is input into the trained network, and the leakage hidden danger probability is output according to the output , set the probability threshold , when the leakage probability of the area When the leakage risk is detected, the area is judged to have leakage risk and the area with leakage risk in the embankment is accurately identified.

2. The method for detecting embankment leakage hazards using transient electromagnetic three-dimensional detection based on an unmanned aerial vehicle according to claim 1 is characterized in that: The flight trajectory in S1 is preset by the ant colony-genetic hybrid optimization algorithm, and the levee length, shape complexity, and terrain undulation are obtained through the three-dimensional spatial information model to determine the effective detection range of the detection equipment. As a basis, set the path planning fitness function , the formula is: ,in is the complexity of the embankment shape, is the length of the embankment, is the terrain undulation; through the path planning fitness function ,dynamically adjust the search direction of the ant colony and the crossover and mutation probability of the genetic algorithm in each iteration; Dynamically determine the distance between adjacent track lines based on path planning , the formula is: :in It is an adjustment factor that forms a flight trajectory with adaptive variable density characteristics to ensure that the UAV flight trajectory fully covers the embankment area to be inspected.

3. The method for detecting embankment leakage hazards using transient electromagnetic three-dimensional detection based on an unmanned aerial vehicle according to claim 2 is characterized in that: The three-dimensional spatial information model is an image feature matrix extracted by satellite remote sensing image features. And the point cloud feature matrix after processing the lidar point cloud image data , constructed by spatial registration data fusion, the formula is: ,in is the spatial transformation matrix of satellite remote sensing images, is the spatial transformation matrix of the lidar point cloud data, ⊙ is the multiplication of the corresponding elements of the matrix, is the matrix norm, by fusion matrix Construct a three-dimensional spatial information model.

4. The method for detecting embankment leakage hazards using transient electromagnetic three-dimensional detection based on an unmanned aerial vehicle according to claim 1 is characterized in that: The frequency range of the emission current in S2 is determined according to the geological characteristics of the dike, the detection depth requirements and the performance parameters of the detection equipment, specifically: Obtain the conductivity distribution of different depth layers of the embankment ,in The effective detection depth of the dike is combined with the detection equipment and minimum resolvable depth Determine the minimum and maximum frequencies of the emission current using the formula: , ,in and are the maximum and minimum resistivity within the dike, is the vacuum permeability, and the emission current frequency range is obtained The transmitting coil selects currents of different frequencies within the frequency range and transmits transient electromagnetic pulse signals to the underground of the embankment.

5. The method for detecting embankment leakage hazards using transient electromagnetic three-dimensional detection based on an unmanned aerial vehicle according to claim 1 is characterized in that: The sampling time interval in S3 is dynamically adjusted according to the signal change intensity, and the formula is: ,in is the minimum sampling period of the detection equipment, is the signal strength change rate, is the adjustment coefficient.

6. The method for detecting embankment leakage hazards using three-dimensional transient electromagnetic technology based on an unmanned aerial vehicle according to claim 1 is characterized in that: The signal collected in S3 is discretized and sampled through a dynamic adaptive sampling strategy. The frequency component and intensity change rate of the secondary electromagnetic response signal are analyzed in real time. Discrete sampling is performed through a multi-resolution sampling mechanism, and an oversampling mechanism is used for the high-frequency part of the signal. The formula is: ,in is the highest frequency component of the signal, is the conventional Nyquist sampling frequency, and the undersampling mechanism is used for the low-frequency stable part. The formula is: , The lowest frequency component of the signal is sampled in segments. The number of sampling points in each time window is dynamically determined according to the complexity of the signal to obtain a sampling data sequence containing different frequency and intensity characteristics.

Images