A method for locating termite nests on embankments based on transient electromagnetics carried by unmanned aerial vehicles

Abstract

This invention discloses a method for locating termite holes on embankments based on transient electromagnetic (TEM) technology carried by unmanned aerial vehicles (UAVs). The method comprises: constructing a UAV-based transient electromagnetic (TEM) detection system; preprocessing the received secondary induced electromagnetic field signals to obtain a composite electromagnetic signal based on the response differences of the multi-frequency signals; inputting the signals into a pretrained deep learning neural network model, identifying characteristic signals associated with termite holes within the composite electromagnetic signal, and outputting the coordinates of the suspected termite holes; performing secondary verification based on thermal imaging images of the embankment surface acquired by a thermal imaging module; and transmitting the termite hole location information in real time to a ground control terminal via a UAV-mounted communication module. This method achieves efficient, accurate, and non-destructive positioning of termite holes on embankments, rapidly acquiring underground electromagnetic signals and surface thermal distribution information on the embankment, and outputting accurate termite hole location coordinates.

Description

Technical Field

[0001] The present invention relates to the technical field of water conservancy project safety monitoring, and in particular to a method for locating termite holes in embankments based on transient electromagnetics carried by an unmanned aerial vehicle. Background Art

[0002] In the field of water conservancy, dams, as key infrastructure for flood control and water storage, have a direct impact on the safety of people's lives, property, and the ecological environment. Termites pose a significant threat to dams. Termites build nests and tunnels within dams, creating hollow structures that weaken them and can even cause serious accidents such as pipe bursts and dam failures. Statistics show that the proportion of dam safety hazards caused by termites is increasing year by year. Therefore, developing efficient and accurate technology to locate termite nests on dams is crucial for ensuring their safe operation.

[0003] Traditional dam termite nest location technology relies primarily on manual inspections. Technicians determine the location of termite nests by observing obvious signs on the dam surface, such as termite trails and flight holes, and by listening for differences in sound by knocking on the surface. This method relies on the experience of inspectors and is highly subjective. Different individuals have varying standards and abilities, making it prone to misjudgments and omissions. Furthermore, manual inspections are extremely inefficient. Long-distance, large-scale dams require significant manpower and time, and are difficult to conduct effectively in complex and inaccessible areas, such as steep slopes and underwater areas. Furthermore, manual inspections can only detect surface or shallow termite activity and are unable to accurately detect deep-lying main nests and complex termite tunnel systems. While relatively inexpensive and easy to operate, manual inspections, due to their poor accuracy, low efficiency, and limited detection range, fall short of meeting the needs of modern dam safety monitoring.

[0004] With the advancement of science and technology, some existing technologies have begun to employ physical detection methods, such as ground-penetrating radar (GPR) and resistivity methods. GPR emits high-frequency electromagnetic waves, receives echoes reflected from the underground medium, and determines the target's location based on the echo's timing and waveform characteristics. This method offers high resolution, rapidly acquires information about underground structures, and is harmless to embankments. However, GPR is significantly affected by factors such as soil moisture content and the dielectric constant of the medium. In embankment areas with high moisture content, electromagnetic waves are severely attenuated, significantly reducing detection depth and accuracy. Furthermore, GPR is ineffective at identifying irregularly shaped and small termite holes, and can easily misidentify other geological anomalies as termite holes.

[0005] The resistivity method uses the resistivity differences between different underground media to detect targets. By placing electrodes on the ground, measuring the underground current and potential distribution, and inverting the underground resistivity distribution image, this method is suitable for large-scale detection and has a certain detection capability for deep targets. However, the resistivity method is susceptible to electromagnetic interference from the surrounding environment, data processing is complex, and the inversion results are highly multi-solution. It is difficult to accurately distinguish the resistivity differences between termite holes and other geological structures, resulting in low positioning accuracy. Moreover, this method requires the placement of a large number of electrodes on the ground, which is cumbersome to operate and has poor adaptability to complex terrain. It cannot meet the requirements for rapid and accurate detection of dams. Summary of the Invention

[0006] Based on the above technical problems, this application discloses a method for locating termite nests on embankments based on transient electromagnetic signals carried by drones, including:

[0007] Build a UAV transient electromagnetic detection system;

[0008] The received secondary induced electromagnetic field signal is pre-processed, and the weights of different frequency signals are calculated according to the response differences of multi-frequency signals, and the integrated electromagnetic signal is obtained by fusion;

[0009] The integrated electromagnetic signal is input into a pre-trained deep learning neural network model to identify characteristic signals related to termite nests in the integrated electromagnetic signal and output the location coordinates of the suspected termite nest;

[0010] Combined with the thermal imaging images of the dam surface obtained by the thermal imaging module, the coordinates of the suspected termite nest location output by the deep learning neural network model are verified again. If there is an abnormal temperature distribution area at the corresponding position in the thermal imaging image, the location is determined to be the termite nest location, and the termite nest location information is transmitted to the ground control terminal in real time through the communication module carried by the drone.

[0011] Preferably, the UAV transient electromagnetic detection system adopts multi-frequency time-sharing transmission technology to transmit at least three pulse current signals with different frequencies; a plurality of independent transmission waveform generation modules are set in the transient electromagnetic detection system, each module corresponds to a pulse current signal of a different frequency, and the timing switching circuit is controlled by a programmable logic controller to activate each transmission waveform generation module in sequence according to a preset time interval; in each transmission cycle, a low-frequency pulse current signal is first transmitted to detect the deep area of the dam, and a medium-frequency pulse current signal is transmitted after an interval of 5-10 milliseconds to detect the middle area, and a high-frequency pulse current signal is transmitted after an interval of 3-5 milliseconds to detect the shallow area. The transmission duration and current intensity of each frequency signal are adaptively adjusted according to the geological conditions of the dam and historical detection data to obtain effective excitation and detection of areas at different depths.

[0012] Preferably, the UAV transient electromagnetic detection system also includes an inertial navigation module, a lidar altimetry module and a thermal imaging module; the inertial navigation module is used to obtain the attitude and position information of the UAV in real time; the lidar altimetry module is used to monitor the distance between the UAV and the dam surface in real time; the thermal imaging module is used to assist in identifying abnormal heat source areas on the dam surface.

[0013] Preferably, the secondary induced electromagnetic field signal preprocessing is specifically: using a wavelet transform denoising algorithm to collect the original signal Process and decompose into high-frequency components of different scales and low-frequency components , the formula is: ,in, is the number of wavelet decomposition layers, for high frequency components Setting the adaptive threshold , for each layer of high frequency components The improved soft threshold function is used for denoising to generate the denoised high-frequency components. , the formula is:

[0014]

[0015] in, is a sign function, which converts the high-frequency components after denoising into With low frequency components The denoised signal is obtained by reconstructing the inverse wavelet transform .

[0016] Preferably, the weights of different frequency signals are calculated based on the denoised signal according to the response differences of the multi-frequency signals. , the formula is: ,in, For the The standard deviation of the frequency signal, is the number of frequency signals; Perform weighted summation on each frequency signal to obtain a comprehensive electromagnetic signal , the formula is: ,in, For the The denoised signal of frequency.

[0017] Preferably, the pre-trained deep learning neural network model adopts a dual-channel attention convolutional neural network structure, which includes a parallel frequency feature channel and a time series channel; the frequency feature channel adopts a multi-layer grouped convolution module, and the convolution kernel size and expansion rate of each group of convolution layers are adaptively adjusted according to the multi-frequency signal characteristics to extract the characteristics of electromagnetic signals of different frequencies; the time series channel adopts a structure combining a long short-term memory network and a self-attention mechanism to capture the sequence characteristics of electromagnetic signals that change over time; a cross-channel attention module is set after the two channels to generate an inter-channel attention weight matrix by calculating the mutual information between the frequency characteristics and the time series characteristics. , the formula is ,in is the frequency eigenvector, is the time series feature vector, and are linear transformation functions respectively; the weighted two channel features are cascaded and input into the global average pooling layer and the fully connected layer, and the focal loss function is used. Conduct training, including and is a hyperparameter, The probability of termite holes predicted by the model; the training data includes electromagnetic signal data of dams under different geological conditions and different degrees of termite damage, as well as the corresponding actual location annotation information of termite holes, to enhance the ability to recognize the characteristic signals of termite holes on dams.

[0018] Preferably, the process of identifying the characteristic signal related to the termite nest in the integrated electromagnetic signal by the deep learning neural network model and outputting the coordinates of the suspected termite nest position is as follows: Vectorize according to the time series and frequency dimensions and convert into a three-dimensional data tensor ,in is the number of time sampling points, is the number of frequency channels, is the number of data channels; the data tensor The input is fed into the deep learning neural network model, and the local electromagnetic features at different frequencies are extracted through the grouped convolution layer of the frequency feature channel. The LSTM-self-attention module of the time series channel captures the temporal variation of the electromagnetic signal. The cross-channel attention module then weights and fuses the features of the two channels. At the model output layer, the position decoding mechanism of the conditional random field is used to construct a joint probability model of the feature signal and the spatial position. The formula is: ,in is the position state sequence, is the normalization factor, is the transfer potential energy function between adjacent positions, is the characteristic potential energy function of the current position, For the time series The position state variable corresponding to the moment is searched for the optimal path of the joint probability model, and the probability value exceeds the preset threshold The position state is mapped to the coordinates in the geographic coordinate system, and the position coordinates of the suspected termite nest are output.

[0019] Preferably, the thermal imaging module acquires thermal imaging images of the dam surface, specifically: the thermal imaging module uses a dual-band infrared detector to simultaneously acquire thermal radiation signals in the 3-5μm medium-wave infrared band and the 8-14μm long-wave infrared band. During the acquisition process, the dam surface is scanned point by point using a spiral scanning path through a galvanometer scanning mechanism, and the scanning angle range covers a ±60° field of view angle directly below the drone; during the scanning process, the thermal imaging image is geometrically corrected for distortion in combination with the distance information acquired in real time by the laser radar altimeter module, and a background temperature dynamic model is established using the background suppression algorithm of the adaptive Kalman filter, and the formula is: ,in is the predicted background temperature at the current moment, is the predicted value at the previous moment, is the adaptive Kalman gain, To reduce the impact of ambient temperature changes on imaging, the dual-band thermal radiation signals are fused at the pixel level. The formula is: , generating the final thermal imaging image, where and are medium-wave and long-wave infrared images, 、 is the corresponding dynamic adjustment coefficient.

[0020] Preferably, the process of secondary verification of the coordinates of the suspected termite hole position by combining the thermal imaging image of the dam surface obtained by the thermal imaging module is as follows: the coordinates of the suspected termite hole position output by the deep learning neural network model are converted into the pixel coordinate system of the thermal imaging image, a mapping relationship between the geographic coordinate system and the image coordinate system is established through the coordinate transformation matrix, a local area of pixel size is delineated with the converted pixel coordinates as the center, the temperature distribution data in the area is extracted, the local density and distance of each pixel point in the area are calculated through the abnormal temperature area recognition algorithm of density peak clustering, and the product of the local density and distance exceeds the set threshold. The pixel points are clustered into abnormal temperature areas. If the above abnormal temperature cluster area exists in the demarcated local area, the suspected location is determined to be the termite nest location; if not, the suspected location is excluded and the result is fed back to the ground control terminal.

[0021] Preferably, the communication module adopts ultra-wideband wireless transmission technology in time division duplex mode to build a local Internet of Things for the dam, encapsulates the location information data frame through the Beidou short message protocol, and transmits it directionally by the MIMO antenna array after channel coding, thereby realizing real-time and interference-resistant transmission of centimeter-level positioning data.

[0022] Compared with the prior art, the technical solution of this application has the following technical effects:

[0023] The present invention constructs a detection system by using a drone equipped with a transient electromagnetic device and multiple sensors, and combines it with automatic flight along a preset route to quickly cover a large area of dams. Compared with the traditional manual inspection that requires section-by-section inspection and is inefficient, the drone can complete the detection task of long-distance dams in a short time, greatly shortening the detection cycle; at the same time, multi-frequency time-sharing transmission technology combined with multi-sensor data collection can obtain multi-dimensional detection information at one time, reducing repetitive operations, and significantly improving the overall work efficiency of locating termite holes on dams, meeting the needs of modern water conservancy projects for efficient monitoring.

[0024] The present invention utilizes the transient electromagnetic method's sensitivity to differences in the conductivity of underground media, combined with a deep learning neural network model to identify the characteristics of comprehensive electromagnetic signals. It can accurately capture the differences in electromagnetic responses between termite nests and surrounding soil, effectively avoiding misjudgments caused by interference factors such as geological structures and underground pipelines. In addition, the thermal imaging module's secondary verification mechanism further confirms the suspected location from the perspective of temperature distribution. Through dual-band imaging, geometric correction and abnormal temperature area recognition algorithms, the reliability of the detection results is enhanced, and high-precision positioning of the termite nest is achieved, providing an accurate basis for subsequent prevention and control work.

[0025] This invention uses a non-contact detection method, eliminating the need for the drone to come into direct contact with the levee. This avoids the damage to the levee structure caused by excavation or drilling, as is the case with traditional detection methods, ensuring the levee's safety and integrity. Furthermore, the drone's maneuverability enables it to adapt to a variety of complex terrain environments, including steep mountain slopes, levees surrounded by water, and even in areas difficult to reach manually. Furthermore, the communication module enables real-time data transmission, facilitating remote monitoring and decision-making, making this technology suitable for a wide range of applications and environments.

[0026] 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.

[0027] 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

[0028] 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.

[0029] Figure 1 This is a flow chart of the method for locating termite holes on embankments based on transient electromagnetics using an unmanned aerial vehicle (UAV);

[0030] Figure 2 This is the structure diagram of the dual-channel attention convolutional neural network of the present invention;

[0031] Figure 3 This is a comparison chart of the accuracy of the transient electromagnetic of the present invention and the prior art;

[0032] Figure 4 A comparison chart of the error rate between the transient electromagnetic method of the present invention and the prior art;

[0033] Figure 5 This is a comparison chart of the convergence rate of the deep learning neural network of the present invention and the traditional convolutional neural network. DETAILED DESCRIPTION

[0034] 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.

[0035] 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.

[0036] 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.

[0037] 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.

[0038] 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.

[0039] 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.

[0040] Example 1

[0041] This embodiment mainly describes a method for locating termite holes on embankments based on transient electromagnetic signals carried by drones. Figure 1 As shown, specifically including:

[0042] Build a UAV transient electromagnetic detection system;

[0043] The received secondary induced electromagnetic field signal is pre-processed, and the weights of different frequency signals are calculated according to the response differences of multi-frequency signals, and the integrated electromagnetic signal is obtained by fusion;

[0044] The integrated electromagnetic signal is input into a pre-trained deep learning neural network model to identify characteristic signals related to termite nests in the integrated electromagnetic signal and output the location coordinates of the suspected termite nest;

[0045] Combined with the thermal imaging images of the dam surface obtained by the thermal imaging module, the coordinates of the suspected termite nest location output by the deep learning neural network model are verified again. If there is an abnormal temperature distribution area at the corresponding position in the thermal imaging image, the location is determined to be the termite nest location, and the termite nest location information is transmitted to the ground control terminal in real time through the communication module carried by the drone.

[0046] Furthermore, the UAV transient electromagnetic detection system adopts multi-frequency time-sharing transmission technology to transmit at least three pulse current signals of different frequencies; multiple independent transmission waveform generation modules are set in the transient electromagnetic detection system, each module corresponds to a pulse current signal of a different frequency, and the timing switching circuit is controlled by a programmable logic controller to activate each transmission waveform generation module in sequence according to a preset time interval; in each transmission cycle, a low-frequency pulse current signal is first transmitted to detect the deep area of the dam, and a medium-frequency pulse current signal is transmitted after an interval of 5-10 milliseconds to detect the middle area, and then a high-frequency pulse current signal is transmitted after an interval of 3-5 milliseconds to detect the shallow area. The transmission duration and current intensity of each frequency signal are adaptively adjusted according to the geological conditions of the dam and historical detection data to obtain effective excitation and detection of areas at different depths.

[0047] Furthermore, the UAV transient electromagnetic detection system also includes an inertial navigation module, a lidar altimeter module and a thermal imaging module; the inertial navigation module is used to obtain the UAV's attitude and position information in real time; the lidar altimeter module is used to monitor the distance between the UAV and the dam surface in real time; the thermal imaging module is used to assist in identifying abnormal heat source areas on the dam surface.

[0048] Furthermore, the secondary induced electromagnetic field signal is preprocessed, specifically: the original signal collected is denoised by wavelet transform algorithm Process and decompose into high-frequency components of different scales and low-frequency components , the formula is: ,in, is the number of wavelet decomposition layers, for high frequency components Setting the adaptive threshold , for each layer of high frequency components The improved soft threshold function is used for denoising to generate the denoised high-frequency components. , the formula is:

[0049]

[0050] in, is a sign function, which converts the high-frequency components after denoising into With low frequency components The denoised signal is obtained by reconstructing the inverse wavelet transform .

[0051] Furthermore, based on the denoised signal, the weights of different frequency signals are calculated according to the response differences of the multi-frequency signals. , the formula is: ,in, For the The standard deviation of the frequency signal, is the number of frequency signals; Perform weighted summation on each frequency signal to obtain a comprehensive electromagnetic signal , the formula is: ,in, For the The denoised signal of frequency.

[0052] Further, if Figure 2 As shown in the figure, the pre-trained deep learning neural network model adopts a dual-channel attention convolutional neural network structure, which includes parallel frequency feature channels and time series channels; the frequency feature channel adopts a multi-layer grouped convolution module, and the convolution kernel size and expansion rate of each convolution layer are adaptively adjusted according to the characteristics of the multi-frequency signal to extract the characteristics of electromagnetic signals of different frequencies; the time series channel adopts a structure that combines the long short-term memory network with the self-attention mechanism to capture the sequence characteristics of the electromagnetic signal that change over time; a cross-channel attention module is set after the two channels to generate an inter-channel attention weight matrix by calculating the mutual information between the frequency feature and the time series feature. , the formula is ,in is the frequency eigenvector, is the time series feature vector, and are linear transformation functions respectively; the weighted two channel features are cascaded and input into the global average pooling layer and the fully connected layer, and the focal loss function is used. Conduct training, including and is a hyperparameter, The probability of termite holes predicted by the model; the training data includes electromagnetic signal data of dams under different geological conditions and different degrees of termite damage, as well as the corresponding actual location annotation information of termite holes, to enhance the ability to recognize the characteristic signals of termite holes on dams.

[0053] Furthermore, the process of identifying the characteristic signals related to termite nests in the integrated electromagnetic signals through the deep learning neural network model and outputting the coordinates of the suspected termite nests is as follows: Vectorize according to the time series and frequency dimensions and convert into a three-dimensional data tensor ,in is the number of time sampling points, is the number of frequency channels, is the number of data channels; the data tensor The input is fed into the deep learning neural network model, and the local electromagnetic features at different frequencies are extracted through the grouped convolution layer of the frequency feature channel. The LSTM-self-attention module of the time series channel captures the temporal variation of the electromagnetic signal. The cross-channel attention module then weights and fuses the features of the two channels. At the model output layer, the position decoding mechanism of the conditional random field is used to construct a joint probability model of the feature signal and the spatial position. The formula is: ,in is the position state sequence, is the normalization factor, is the transfer potential energy function between adjacent positions, is the characteristic potential energy function of the current position, For the time series The position state variable corresponding to the moment is searched for the optimal path of the joint probability model, and the probability value exceeds the preset threshold The position state is mapped to the coordinates in the geographic coordinate system, and the position coordinates of the suspected termite nest are output.

[0054] Furthermore, the thermal imaging module acquires thermal imaging images of the dam surface. Specifically, the thermal imaging module uses a dual-band infrared detector to simultaneously collect thermal radiation signals in the 3-5μm medium-wave infrared band and the 8-14μm long-wave infrared band. During the acquisition process, the dam surface is scanned point by point using a spiral scanning path through a galvanometer scanning mechanism. The scanning angle range covers a ±60° field of view directly below the drone. During the scanning process, the thermal imaging image is geometrically corrected for distortion in combination with the distance information obtained in real time by the lidar altimeter module. A background temperature dynamic model is established using the background suppression algorithm of the adaptive Kalman filter. The formula is: ,in is the predicted background temperature at the current moment, is the predicted value at the previous moment, is the adaptive Kalman gain, To reduce the impact of ambient temperature changes on imaging, the dual-band thermal radiation signals are fused at the pixel level. The formula is: , generating the final thermal imaging image, where and are medium-wave and long-wave infrared images, 、 is the corresponding dynamic adjustment coefficient.

[0055] Furthermore, the process of secondary verification of the coordinates of the suspected termite hole position in combination with the thermal imaging image of the dam surface obtained by the thermal imaging module is as follows: the coordinates of the suspected termite hole position output by the deep learning neural network model are converted into the pixel coordinate system of the thermal imaging image, and the mapping relationship between the geographic coordinate system and the image coordinate system is established through the coordinate transformation matrix. The local area of pixel size is delineated with the converted pixel coordinates as the center, and the temperature distribution data in the area is extracted. The local density and distance of each pixel point in the area are calculated through the abnormal temperature area recognition algorithm of density peak clustering, and the product of local density and distance exceeds the set threshold. The pixel points are clustered into abnormal temperature areas. If the above abnormal temperature cluster area exists in the demarcated local area, the suspected location is determined to be the termite nest location; if not, the suspected location is excluded and the result is fed back to the ground control terminal.

[0056] Furthermore, the communication module adopts ultra-wideband wireless transmission technology in time-division duplex mode to build a local Internet of Things for the dam. The location information data frame is encapsulated through the Beidou short message protocol, and after channel coding, it is directionally transmitted by the MIMO antenna array, realizing real-time and interference-resistant transmission of centimeter-level positioning data.

[0057] This embodiment describes in detail how to construct a detection system using multiple modules such as transient electromagnetic sensors equipped on drones to achieve efficient and accurate positioning of termite nests on dams. The drones quickly cover large areas, and multi-frequency transmission and multiple sensors work together to improve detection efficiency. Deep learning and thermal imaging provide secondary verification to accurately identify termite nest characteristics and reduce misjudgments. Non-contact detection avoids damage to the dam structure and can adapt to complex terrain. The communication module enables real-time data transmission, providing reliable technical support for dam termite control.

[0058] Based on Example 1, this example describes in detail the inertial navigation module, lidar altimeter module, and thermal imaging module of the UAV transient electromagnetic detection system, specifically:

[0059] In the method for locating termite nests on embankments using a drone equipped with transient electromagnetic sensors, the inertial navigation module, through its built-in accelerometer and gyroscope inertial sensors, can collect the drone's acceleration and angular velocity data in three-dimensional space in real time and at high frequency. The accelerometer can accurately measure the drone's acceleration along each axis during flight, thereby inferring the drone's speed and displacement changes; the gyroscope focuses on detecting changes in the drone's attitude angles, including pitch, roll, and yaw, providing key information for the drone's flight attitude control. During embankment detection, the complex flight environment, such as airflow disturbances and terrain undulations, can cause the drone to experience attitude deviations. The inertial navigation module can quickly perceive these changes and transmit the collected data to the drone's flight control system. The flight control system adjusts the drone's flight attitude and trajectory in real time based on this data, ensuring that the drone always flies stably along the preset detection route, thereby ensuring that the transient electromagnetic transmitting and receiving devices are at the optimal position and angle for signal acquisition. In addition, the precise position and attitude information recorded by the inertial navigation module can be accurately matched with the transient electromagnetic signal acquisition data, making subsequent data processing and termite nest positioning results more accurate and reliable, avoiding detection deviations caused by errors in the drone's position and attitude.

[0060] The LiDAR altimeter module accurately monitors the distance between the drone and the dam surface in real time. It emits a laser beam toward the dam surface, receives the reflected laser signal, and calculates the distance based on the laser's flight time. In dam detection scenarios, the surface terrain is complex, with varying slopes and uneven surfaces. Improper drone altitude control can affect the quality of transient electromagnetic signal acquisition and pose a safety risk of collision. The LiDAR altimeter module continuously acquires distance data at an extremely high sampling frequency, enabling rapid response to terrain changes. When it detects that the terrain ahead is rising, the module quickly transmits the distance change information to the drone control system, which then adjusts the drone's flight altitude to keep it within the preset safe detection altitude range. Conversely, when the terrain drops, the flight altitude can be lowered in time to ensure that the distance between the detection device and the dam surface is always in the optimal state, thereby ensuring the stable acquisition of transient electromagnetic signals. The altitude data obtained by the lidar altimetry module can also be used to perform elevation correction on the collected electromagnetic signals. Combined with the position information provided by the inertial navigation module, a more accurate three-dimensional model of the dam's underground structure can be constructed, providing a richer and more accurate spatial information basis for the positioning of termite holes.

[0061] The thermal imaging module provides multi-dimensional detection information. It uses infrared detection technology to capture infrared radiation emitted by objects on the surface of the dam and convert it into a visual thermal image. Since termites generate metabolic heat during their nest activities, and the material and structure of termite nests are different from those of the surrounding soil, the heat conduction and heat radiation characteristics of the nest area are different from those of the surrounding normal areas, which appear as abnormal temperature areas on the thermal imaging image. The thermal imaging module uses a dual-band infrared detector to simultaneously collect 3-5 Mid-wave infrared band and 8-14 Thermal radiation signals in the long-wave infrared band. These two bands each have advantages in sensitivity and penetration for objects of varying temperatures. Pixel-level fusion technology enables the acquisition of clearer and more accurate thermal images. During detection, the thermal imaging module, aligned with the drone's flight path, comprehensively scans the dam surface, quickly identifying potential areas of temperature anomalies. This information is cross-validated with the locations of suspected termite nests obtained through transient electromagnetic detection. When the temperature anomalies in the thermal image match those in the electromagnetic signal, the accuracy and reliability of termite nest location are greatly improved. Furthermore, the thermal imaging module is unrestricted by lighting conditions, enabling continuous and stable monitoring of dam surface temperature distribution day or night, providing robust support for all-weather detection of termite nests on the dam.

[0062] This embodiment describes in detail the inertial navigation module of the UAV transient electromagnetic detection system in this application, which calibrates the UAV's attitude and trajectory in real time to ensure the accuracy of the detection route; the lidar altimeter module dynamically adjusts the flight altitude to ensure that the detection device is at the optimal working distance; the thermal imaging module uses infrared radiation to capture temperature anomalies, combined with electromagnetic signal cross-verification. The three work together to greatly improve the efficiency, accuracy and reliability of locating termite holes on the dam.

[0063] Based on Example 1, this example describes in detail how to perform secondary verification of the coordinates of the suspected termite nest using the thermal imaging image of the dam surface acquired by the thermal imaging module, specifically:

[0064] The coordinates of the suspected termite nest position output by the deep learning neural network model are converted into the pixel coordinates of the thermal imaging image through the coordinate transformation matrix To build a connection between the two, the formula is , Represents geographic coordinates, obtained by the high-precision positioning system onboard the drone combined with the inertial navigation module, which can accurately determine the actual physical location on the embankment; is the pixel coordinate after conversion, corresponding to the specific position point in the thermal imaging image, the coordinate transformation matrix It is obtained by pre-calibrating the thermal imaging module, taking into account the flight altitude, attitude angle and distortion parameter factors of the thermal imaging lens of the drone. In actual operation, the attitude of the drone will continue to change during flight, and the distance information monitored in real time by the lidar height measurement module will also be fed back to the coordinate conversion process. Make dynamic corrections to ensure the accuracy of conversion and to define For example, if accurate coordinate transformation is not performed on a pixel-sized local area, the extracted area will be misaligned with the actual suspected termite nest location, thus affecting the subsequent verification results. Only based on accurate coordinate mapping can it be guaranteed that the local area extracted from the thermal imaging image actually covers the suspected location determined by the deep learning model.

[0065] After completing the coordinate transformation and demarcating the local area, the abnormal temperature area recognition algorithm of density peak clustering is used to process the temperature distribution data in the area and calculate the local density of each pixel in the area. and distance ,in, Pixel and The Euclidean distance of temperature reflects the size of the temperature difference between two points. The cutoff distance is a threshold value pre-set based on a large amount of experimental data and the temperature distribution characteristics of the dam surface, which is used to define the distance between pixels. is the indicator function, when When it is less than 0, The value is 1, otherwise it is 0. The indicator function can be used to count the number of pixels. Distance less than The number of pixels, and then the local density ,The higher the local density, the denser the pixels with similar temperature around the point are; It represents the pixel The minimum distance to the pixel point with a larger local density than the minimum distance to the pixel point. This parameter is used to measure the relative isolation of the pixel point in the temperature distribution. The product of the local density and the distance exceeds the threshold. The pixels are clustered into abnormal temperature areas, and the threshold It is obtained through machine learning algorithms, combined with historical dam termite nest thermal imaging data and normal area data training. It can effectively distinguish between temperature anomalies caused by termite activity and temperature fluctuations caused by environmental factors. The surface of the dam is affected by environmental factors such as sunlight and wind direction, which will produce certain temperature changes, but these changes are usually continuous and uniform. The temperature distribution in the termite nest area will show local aggregation and obvious differences from the surrounding area due to the biological activities and special structures inside the nest. The density peak clustering algorithm can accurately capture this abnormal temperature distribution. If there is an abnormal temperature clustering area that meets the conditions in the demarcated local area, the suspected location is determined to be the termite nest location. If not, the suspected location is excluded, thereby achieving effective verification of the output results of the deep learning model and improving the reliability and accuracy of termite nest positioning.

[0066] This embodiment describes in detail how to associate the suspected termite nest locations output by deep learning with thermal imaging images through precise coordinate mapping and density peak clustering algorithms. The coordinate transformation matrix is dynamically corrected to ensure accurate regional correspondence. The clustering algorithm combines temperature distribution characteristics to accurately identify anomalies, effectively eliminate environmental interference, achieve double verification, and significantly improve the accuracy and reliability of termite nest positioning.

[0067] Based on Example 1, this example describes in detail the specific implementation process of this application, specifically:

[0068] In terms of detection efficiency, the drone in this application has a shorter detection time and a more convenient detection method than ground penetrating radar and resistivity meter, which shortens the detection cycle and has significant advantages in large-scale embankment detection scenarios.

[0069] like Figure 3-4 As shown, positioning accuracy is the core indicator for measuring embankment termite hole detection technology. The technology of this application achieves a positioning accuracy of 94.29% and a missed detection rate of only 5.71% through multi-frequency transient electromagnetic signal acquisition, wavelet transform denoising and deep learning neural network recognition; while the accuracy of ground penetrating radar is 59.52% and the missed detection rate is 28.57%; the accuracy of resistivity meter is 57.89% and the missed detection rate is 39.47%; the technology of this application accurately identifies 8 more locations than ground penetrating radar and 11 more than resistivity meter, and the number of false positives is 10 and 8 less than the two respectively. The secondary verification mechanism of the thermal imaging module further improves the reliability. For the 35 suspected locations output by the deep learning model, the verification success rate reaches 80%, effectively reducing the risk of false positives.

[0070] As shown in Figure 5, the improved deep learning neural network proposed in this application utilizes a dual-channel attention convolutional neural network architecture. During training, it converges 37% faster than traditional convolutional neural networks. For example, training 100 epochs requires approximately 0.23 hours for a traditional convolutional neural network to reduce its loss to 0.35, while the improved model in this application only takes 0.18 hours. In terms of test set performance, the proposed model excels at extracting complex electromagnetic signal features, achieving an accuracy of 92.34%, while the traditional convolutional neural network achieves only 78.65%. In contrast, existing ground-penetrating radar technology experiences a 40% drop in resolution, with detection depth dropping from 3m to 1.2m in areas with 35% soil moisture. Furthermore, resistivity meters experience a data misclassification rate of up to 45% in electromagnetic interference environments. Both methods struggle to reliably and accurately detect termite nests in embankments. However, the proposed improved model maintains efficient feature extraction and localization capabilities in complex electromagnetic environments, demonstrating a significant performance advantage.

[0071] In terms of anti-environmental interference performance, the technology of this application also performs well. In an environment with a wind speed of 5.6m / s, the inertial navigation module and the lidar altimeter module work together to ensure that the fluctuation of the drone's flight altitude is controlled within ±0.3m, and the signal acquisition stability reaches 98.7%. In the same environment, the signal distortion rate of the ground penetrating radar is as high as 23%, and the resistivity meter data efficiency drops to 65%. Overall, whether it is detection efficiency, positioning accuracy, or data processing capabilities and environmental adaptability, the various values of the technology of this application are significantly better than existing technologies, fully demonstrating its innovation and practicality.

[0072] This embodiment describes in detail that the technology of the present application demonstrates excellent performance in the detection of termite holes in dams by virtue of multi-module collaboration and advanced algorithms. The detection efficiency, positioning accuracy, and missed detection rate far exceed those of existing technologies. The data processing is efficient, the anti-interference ability is strong, and it still maintains high stability in complex environments, which is comprehensively superior to traditional detection methods.

[0073] 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 locating termite holes on embankments based on transient electromagnetic signals carried by drones, characterized in that: include: Build a UAV transient electromagnetic detection system; The received secondary induced electromagnetic field signal is pre-processed, and the weights of different frequency signals are calculated according to the response differences of multi-frequency signals, and the integrated electromagnetic signal is obtained by fusion; The integrated electromagnetic signal is input into a pre-trained deep learning neural network model to identify characteristic signals related to termite nests in the integrated electromagnetic signal and output the location coordinates of the suspected termite nest; Combined with thermal imaging images of the dam surface acquired by the thermal imaging module, the coordinates of suspected termite nest locations output by the deep learning neural network model are re-verified. If an abnormal temperature distribution area exists at the corresponding location in the thermal imaging image, the location is determined to be a termite nest. The termite nest location information is then transmitted in real time to the ground control terminal via the communication module onboard the drone. The UAV transient electromagnetic detection system adopts multi-frequency time-sharing transmission technology to transmit pulse current signals of at least three different frequencies; a plurality of independent transmission waveform generation modules are set in the transient electromagnetic detection system, each module corresponds to a pulse current signal of a different frequency, and the timing switching circuit is controlled by a programmable logic controller, and each transmission waveform generation module is activated in sequence according to a preset time interval; in each transmission cycle, a low-frequency pulse current signal is first transmitted to detect the deep area of the dam, and a medium-frequency pulse current signal is transmitted after an interval of 5-10 milliseconds to detect the middle area, and a high-frequency pulse current signal is transmitted after an interval of 3-5 milliseconds to detect the shallow area. The transmission duration and current intensity of each frequency signal are adaptively adjusted according to the geological conditions of the dam and historical detection data to obtain effective excitation and detection of areas at different depths.

2. The method for locating termite holes on embankments based on transient electromagnetics using an unmanned aerial vehicle according to claim 1, characterized in that: The UAV transient electromagnetic detection system also includes an inertial navigation module, a lidar altimeter module and a thermal imaging module; the inertial navigation module is used to obtain the UAV's attitude and position information in real time; the lidar altimeter module is used to monitor the distance between the UAV and the dam surface in real time; and the thermal imaging module is used to assist in identifying abnormal heat source areas on the dam surface.

3. The method for locating termite holes on embankments based on transient electromagnetics using an unmanned aerial vehicle according to claim 1, characterized in that: The secondary induced electromagnetic field signal preprocessing is specifically: the original signal collected is processed by wavelet transform denoising algorithm Process and decompose into high-frequency components of different scales and low-frequency components , the formula is: ,in, is the number of wavelet decomposition layers, for high frequency components Setting the adaptive threshold , for each layer of high frequency components The improved soft threshold function is used for denoising to generate the denoised high-frequency components. , the formula is: ； in, is a sign function, which converts the high-frequency components after denoising into With low frequency components The denoised signal is obtained by reconstructing the inverse wavelet transform .

4. The method for locating termite holes on embankments based on transient electromagnetics using an unmanned aerial vehicle according to claim 1 or 3, characterized in that: The weights of different frequency signals are calculated based on the denoised signal and the response differences of the multi-frequency signals. , the formula is: ,in, For the The standard deviation of the frequency signal, is the number of frequency signals; Perform weighted summation on each frequency signal to obtain a comprehensive electromagnetic signal , the formula is: ,in, For the The denoised signal of frequency.

5. The method for locating termite holes on embankments based on transient electromagnetics using an unmanned aerial vehicle according to claim 1, characterized in that: The pre-trained deep learning neural network model adopts a dual-channel attention convolutional neural network structure, including parallel frequency feature channels and time series channels; The frequency feature channel uses a multi-layer group convolution module. The convolution kernel size and expansion rate of each convolution layer are adaptively adjusted according to the characteristics of the multi-frequency signal to extract the characteristics of electromagnetic signals at different frequencies. The time series channel uses a structure that combines long short-term memory networks with self-attention mechanisms to capture the time-varying sequence characteristics of electromagnetic signals. A cross-channel attention module is set after the two channels to generate the inter-channel attention weight matrix by calculating the mutual information between the frequency feature and the time series feature. , the formula is ,in is the frequency eigenvector, is the time series feature vector, and are linear transformation functions respectively; the weighted two channel features are cascaded and input into the global average pooling layer and the fully connected layer, and the focal loss function is used. Conduct training, including and is a hyperparameter, The probability of termite holes predicted by the model; the training data includes electromagnetic signal data of dams under different geological conditions and different degrees of termite damage, as well as the corresponding actual location annotation information of termite holes, to enhance the ability to recognize the characteristic signals of termite holes on dams.

6. The method for locating termite holes on embankments based on transient electromagnetics using an unmanned aerial vehicle according to claim 1 or 5, characterized in that: The process of identifying the characteristic signal related to the termite nest in the integrated electromagnetic signal by the deep learning neural network model and outputting the coordinates of the suspected termite nest position is as follows: Vectorize according to the time series and frequency dimensions and convert into a three-dimensional data tensor ,in is the number of time sampling points, is the number of frequency channels, is the number of data channels; the data tensor The input is fed into the deep learning neural network model, and the local electromagnetic features at different frequencies are extracted through the grouped convolution layer of the frequency feature channel. The LSTM-self-attention module of the time series channel captures the temporal variation of the electromagnetic signal. The cross-channel attention module then weights and fuses the features of the two channels. At the model output layer, the position decoding mechanism of the conditional random field is used to construct a joint probability model of the feature signal and the spatial position. The formula is: ,in is the position state sequence, is the normalization factor, is the transfer potential energy function between adjacent positions, is the characteristic potential energy function of the current position, For the time series The position state variable corresponding to the moment is searched for the optimal path of the joint probability model, and the probability value exceeds the preset threshold The position state is mapped to the coordinates in the geographic coordinate system, and the position coordinates of the suspected termite nest are output.

7. The method for locating termite holes on embankments based on transient electromagnetics using an unmanned aerial vehicle according to claim 1, characterized in that: The thermal imaging module acquires thermal imaging images of the dam surface. Specifically, the thermal imaging module uses a dual-band infrared detector to simultaneously collect thermal radiation signals in the 3-5μm medium-wave infrared band and the 8-14μm long-wave infrared band. During the acquisition process, the dam surface is scanned point by point using a spiral scanning path through a galvanometer scanning mechanism, and the scanning angle range covers a ±60° field of view angle directly below the drone. During the scanning process, the thermal imaging image is geometrically corrected for distortion in combination with the distance information obtained in real time by the laser radar altimeter module. A background temperature dynamic model is established using the background suppression algorithm of the adaptive Kalman filter. The formula is: ,in is the predicted background temperature at the current moment, is the predicted value at the previous moment, is the adaptive Kalman gain, To reduce the impact of ambient temperature changes on imaging, the dual-band thermal radiation signals are fused at the pixel level. The formula is: , generating the final thermal imaging image, where and are medium-wave and long-wave infrared images, 、 is the corresponding dynamic adjustment coefficient.

8. The method for locating termite holes on embankments based on transient electromagnetics using an unmanned aerial vehicle according to claim 1 or 7, characterized in that: The process of secondary verification of the coordinates of the suspected termite hole position by combining the thermal imaging image of the dam surface obtained by the thermal imaging module is as follows: the coordinates of the suspected termite hole position output by the deep learning neural network model are converted into the pixel coordinate system of the thermal imaging image, and the mapping relationship between the geographic coordinate system and the image coordinate system is established through the coordinate transformation matrix. The local area of pixel size is delineated with the converted pixel coordinates as the center, and the temperature distribution data in the area is extracted. The local density and distance of each pixel point in the area are calculated through the abnormal temperature area recognition algorithm of density peak clustering, and the local density and distance product exceeds the set threshold. The pixel points are clustered into abnormal temperature areas. If the above abnormal temperature cluster area exists in the demarcated local area, the suspected location is determined to be the termite nest location; if not, the suspected location is excluded and the result is fed back to the ground control terminal.

9. The method for locating termite holes on embankments based on transient electromagnetics using an unmanned aerial vehicle according to claim 1, characterized in that: The communication module adopts ultra-wideband wireless transmission technology in time division duplex mode to build a local Internet of Things for the dam. The location information data frame is encapsulated through the Beidou short message protocol, and is directionally transmitted by the MIMO antenna array after channel coding, thereby realizing real-time and interference-resistant transmission of centimeter-level positioning data.

Images