A high-voltage insulator string defect detection method and system based on unmanned aerial vehicle images

Abstract

The application provides a high-voltage insulator string defect detection method and system based on unmanned aerial vehicle images, and belongs to the technical field of image processing. The method comprises the following steps: obtaining an initial video image corresponding to an insulator string in a preset first period based on an unmanned aerial vehicle; determining motion data corresponding to each insulator in the insulator string based on the initial video image, wherein the motion data is motion displacement data of the insulator when the insulator is subjected to an acting force; and determining, based on the plurality of motion data, that an insulator corresponding to motion data that is different from the motion data of other insulators in the plurality of insulators is a target insulator. The application identifies a defective insulator by analyzing the motion displacement difference of each insulator in the same mechanical environment, does not depend on static apparent features such as the surface texture, color or contamination state of the insulator, effectively avoids the interference of uneven illumination, a complex background and surface contamination on traditional visual detection, and improves the robustness and accuracy of defect identification.

Description

Technical Field

[0001] This application relates to the field of image processing, and in particular to a method and system for detecting defects in high-voltage insulator strings based on UAV images. Background Technology

[0002] High-voltage insulator strings are critical components in transmission lines, serving both mechanical support and electrical insulation functions. Exposed to wind vibration, temperature differences, pollution, and strong electric fields in outdoor environments over extended periods, they are prone to defects such as skirt damage, loose steel caps, internal cracks, and material deterioration. If these defects are not detected promptly, they can develop into insulation breakdown or string failure, seriously threatening the safe operation of the power grid.

[0003] Currently, drone-based insulator string inspection mainly relies on manual interpretation of visible light images or deep learning methods. Its detection logic is based on the direct identification of the insulator string's surface texture, contour, and color. However, this approach has the following limitations: First, insulator surfaces are often covered with dirt, and outdoor inspections involve variable lighting conditions and complex transmission line backgrounds, significantly reducing the reliability of surface feature identification. Second, defects such as internal microcracks, adhesive debonding, and early loosening of hinge points produce almost no identifiable surface anomalies in static visible light images, making them difficult to detect effectively using conventional image recognition methods. Summary of the Invention

[0004] This application provides a method and system for detecting defects in high-voltage insulator strings from UAV images, in order to improve the above-mentioned problems.

[0005] In a first aspect, embodiments of this application provide a method for detecting defects in high-voltage insulator strings based on UAV images, comprising: acquiring initial video images of the insulator string corresponding to a preset first period based on the UAV; determining motion data corresponding to each insulator in the insulator string based on the initial video images, wherein the motion data is the displacement data of the insulator when it is subjected to a force; and determining, based on multiple motion data, the insulator whose corresponding motion data differs from the motion data of other insulators as the target insulator.

[0006] Therefore, it can be seen that defective insulators can be identified by analyzing the differences in motion displacement of each insulator in an insulator string under applied force. By utilizing the consistency of dynamic response of each insulator under the same mechanical environment, individuals whose motion displacement deviates from the overall distribution can be identified as target insulators. This effectively avoids the interference of uneven lighting, complex background and surface dirt on traditional visual inspection, and improves the robustness and accuracy of defect identification.

[0007] Optionally, in conjunction with the first aspect, based on the initial video image, determining the motion data corresponding to each insulator in the insulator string includes: determining the insulator string frame and the background frame in the initial video image; determining a reference point in the background frame, wherein the reference point is a fixed point in the actual scene corresponding to the background frame; performing motion compensation on the initial video image based on the fixed point and acquiring a second video image; and determining the motion data corresponding to each insulator in the insulator string based on the second video image.

[0008] Therefore, by determining a fixed point in the background image and performing motion compensation on the initial video image accordingly, the overall image displacement caused by the drone's unstable hovering can be effectively eliminated. This ensures that the pixel displacement of each insulator on the insulator string truly reflects its physical motion under the force in actual space, thus avoiding interference from the drone's own shaking on the accuracy of motion data measurement.

[0009] Optionally, in conjunction with the first aspect, based on multiple motion data, the target insulator is identified from among multiple insulators whose corresponding motion data differs from the motion data of other insulators, including: In the second video image, target points are acquired on each insulator of the insulator string. The target points are the structural corner points on the surface of the insulator. The displacement of each target point in a preset first cycle is acquired, and the maximum displacement is used as motion data. Based on the statistical distribution of all maximum displacements, the normal distribution range of the maximum displacement is determined. When there is an insulator whose maximum displacement exceeds the normal distribution range, the insulator is identified as the target insulator.

[0010] Therefore, it can be seen that using the natural structural corners of the insulator surface as target points for tracking eliminates the need to manually affix markers to the insulators, facilitating actual inspection operations. By determining the normal range and screening outliers based on the statistical distribution of the maximum displacement of all insulators, the system can adaptively determine the discrimination criteria based on the overall response level of each insulator on the same insulator string, avoiding the problem of poor adaptability of fixed thresholds under different inspection cycles or wind conditions.

[0011] Optionally, in conjunction with the first aspect, before acquiring the initial video image of the insulator string corresponding to the preset first period based on the UAV, the method further includes: controlling the UAV to hover near the insulator string and step-changing the rotor speed so that the insulator string is subjected to a momentary step excitation; and taking the natural decay vibration period after the step excitation ends as the preset first period.

[0012] Therefore, by controlling the UAV to change the rotor speed stepwise near the insulator string, and using the rotor downwash airflow pulse to apply an instantaneous step excitation to the insulator string, and using the natural decay vibration period after the excitation ends as the motion data acquisition period, each insulator can perform free decay vibration under uniform and controllable initial disturbance conditions. The excitation source is known and highly repeatable, and the motion data of each insulator has a uniform physical comparability benchmark. This effectively avoids misjudgment and missed detection caused by the strong randomness of environmental wind and the inconsistency of excitation conditions between different inspection frames.

[0013] Optionally, in conjunction with the first aspect, the method further includes: High-resolution images of insulator strings are acquired using drones; crack images on each insulator are acquired based on the sub-images corresponding to each insulator in the high-resolution images; multiple crack directions in each crack image are acquired, and the directional consistency constant of multiple crack directions is determined; if the directional consistency constant in a crack image is greater than a preset threshold, the insulator corresponding to the crack image is determined as the target insulator.

[0014] Therefore, by analyzing the direction consistency constant of crack orientation to distinguish between real cracks and surface interference, and by utilizing the characteristic difference that structural cracks usually have consistent orientations while interference such as stains and random scratches have disordered orientations, it is possible to effectively distinguish between real cracks with structured orientations and pseudo-defects without orientation, thereby reducing the false cracking rate caused by surface contamination.

[0015] Optionally, in conjunction with the first aspect, multiple crack orientations in each crack image are obtained, and the orientation consistency constants of the multiple crack orientations are determined, including: segmenting the crack image to form multiple crack sub-images; obtaining the structure tensor of the pixel gradient in each crack sub-image, and determining the main orientation of the crack sub-image based on the eigenvalues ​​and eigenvectors corresponding to the structure tensor; obtaining the main orientations of all crack sub-images in the crack image to form an orientation set; obtaining the circular variance of the orientation set, and using the circular variance as the orientation consistency constant.

[0016] Therefore, it can be seen that using circular variance to quantify the dispersion of the direction set can correctly handle the periodicity of angle data compared with linear variance. It avoids erroneous variance calculation results caused by cyclic jumps in angle when the direction crosses the 0° or 180° boundary, thus more accurately reflecting the consistency of crack direction and improving the reliability of crack judgment.

[0017] Optionally, in conjunction with the first aspect, obtaining the structure tensor of pixel gradients within each cracked sub-image, and determining the main direction of the cracked sub-image based on the eigenvalues ​​and eigenvectors corresponding to the structure tensor, includes: obtaining the horizontal and vertical gradient components of each pixel in the cracked sub-image, and determining the gradient vector of the pixel based on the horizontal and vertical gradient components; calculating the outer product of the gradient vectors and using it as the initial structure tensor of the pixel; summing and averaging the initial structure tensors of all pixels in the cracked sub-image to obtain the average value, which is used as the structure tensor of pixel gradients within the cracked sub-image; performing eigenvalue decomposition on the structure tensor to obtain a first eigenvalue and a second eigenvalue, as well as a first eigenvector corresponding to the first eigenvalue and a second eigenvector corresponding to the second eigenvalue, wherein the first eigenvalue is greater than or equal to the second eigenvalue; and determining the direction of the second eigenvector corresponding to the second eigenvalue as the main direction of the cracked sub-image.

[0018] Therefore, by constructing a structure tensor through the outer product of pixel gradient vectors, using eigenvalue decomposition to distinguish the difference in gradient changes along the crack direction and perpendicular to the crack direction, and determining the direction of the second eigenvector corresponding to the smaller eigenvalue as the main direction of the crack sub-image, it is possible to robustly estimate the structural extension direction of the local region of the image, and it is not affected by the overall shift of the image grayscale values.

[0019] Optionally, in conjunction with the first aspect, the method further includes: The system controls a drone to acquire low-light video images of insulators in energized operation within a preset second cycle, under ambient illumination conditions below a preset threshold. Based on these low-light video images, it acquires the maximum discharge spot size on the surface and edge regions of each insulator. It then acquires the spot area and peak brightness of each maximum discharge spot and determines the corresponding discharge intensity for each insulator. The discharge intensity is the product of the largest spot area and the largest peak brightness in any frame of the low-light video image. Based on the statistical distribution of all discharge intensities, it determines the normal distribution range of the discharge intensity. When an insulator's discharge intensity exceeds the normal distribution range, it is identified as the target insulator.

[0020] Therefore, it can be seen that acquiring low-light video images under low ambient illumination conditions can avoid the flooding effect of sunlight on the weak discharge spots on the surface of the insulator, and effectively capture the visible light signal of the local discharge caused by the defect. The product of the largest spot area and the largest brightness peak value that has appeared within the preset period is used as the discharge intensity, which comprehensively reflects the two dimensions of the spatial range and intensity of the discharge. It can more comprehensively characterize the degree of discharge anomaly caused by the defect than a single area or brightness index.

[0021] Optionally, in conjunction with the first aspect, the method further includes: A drone is controlled to fly between the insulator string and the sun, casting a moving shadow on the surface of the insulator string. This causes the surface of the insulator string to experience a temperature rise process from being shaded to being directly exposed to sunlight again within a preset third cycle. A thermal infrared camera on the drone simultaneously acquires thermal infrared images of the insulator string during the temperature rise process. From the thermal infrared images, the temperature rise time constant of each insulator surface after the shadow moves away, representing the temperature recovery from a low point to a steady state, is extracted. Based on the statistical distribution of all temperature rise time constants, the normal distribution range of the temperature rise time constant is determined. When an insulator has a temperature rise time constant exceeding the normal distribution range, that insulator is identified as the target insulator.

[0022] Therefore, it can be seen that by using the drone body to block sunlight and create a moving shadow on the surface of the insulator string to generate controllable temperature disturbance, active thermal excitation can be achieved without the need for additional heating devices. Instead of directly comparing static temperature values, the time constant of temperature rise from the low point to the steady state after the shadow is removed can be used as the criterion. This can eliminate the influence of uneven distribution of ambient temperature field and differences in the heat generated by the insulator itself on the test results, so that latent defects such as internal cracks and debonding that cause changes in local thermal diffusion properties can be effectively detected.

[0023] A second aspect of this invention provides a high-voltage insulator string defect detection system based on UAV images, used to perform a high-voltage insulator string defect detection method based on UAV images as described in the first aspect, comprising: The acquisition module is used to acquire the initial video image of the insulator string in the preset first cycle based on the UAV; The first determining module is used to determine the motion data corresponding to each insulator in the insulator string based on the initial video image, wherein the motion data is the displacement data of the insulator when it is subjected to a force. The second determining module is used to determine, based on multiple motion data, the target insulator from among multiple insulators whose corresponding motion data differs from the motion data of other insulators.

[0024] A third aspect of the present invention provides a processing apparatus, the processing apparatus comprising: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method proposed in the first aspect of the present invention.

[0025] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in the first aspect of the present invention. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the structure of a high-voltage insulator string defect detection system based on UAV images proposed in an embodiment of this application.

[0027] Figure 2 This is a schematic flowchart of a high-voltage insulator string defect detection method based on UAV images proposed in an embodiment of this application.

[0028] Figure 3 This is a schematic diagram of the processing device provided in an embodiment of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] To facilitate understanding of the embodiments of the present invention, a high-voltage insulator string defect detection method based on UAV images, applicable to the embodiments of the present invention, will first be described in detail using a high-voltage insulator string defect detection system based on UAV images as an example. Exemplarily, this high-voltage insulator string defect detection system based on UAV images is configured to perform the following operations: The initial video image of the insulator string is acquired by the UAV in a preset first cycle. Based on the initial video image, the motion data corresponding to each insulator in the insulator string is determined, wherein the motion data is the displacement data of the insulator when it is subjected to force. Based on multiple motion data, the insulator whose corresponding motion data differs from the motion data of other insulators is identified as the target insulator.

[0031] Specifically, Figure 1 This diagram illustrates the structure of a high-voltage insulator string defect detection system based on UAV imagery. To achieve the aforementioned operations, as follows... Figure 1 As shown, the high-voltage insulator string defect detection system based on UAV images includes: an acquisition module, a first determination module, and a second determination module.

[0032] Specifically, the acquisition module is used to acquire initial video images of the insulator string within a preset first cycle based on the UAV. This acquisition module can communicate with a visible light camera mounted on the UAV to receive video images continuously acquired by the camera from the target high-voltage insulator string within the preset first cycle. The insulator string can be a porcelain insulator string, a glass insulator string, or a composite insulator string, and the preset first cycle is a time period that covers several reciprocating motion cycles of the insulator string under applied force.

[0033] The first determining module is used to determine the motion data corresponding to each insulator in the insulator string based on the initial video image. The motion data refers to the displacement data of the insulator when subjected to a force. The force can be natural wind, downwash airflow from the UAV rotor, or vibration differential forces caused by structural constraint changes due to defects, etc. The first determining module performs image separation, background reference point extraction, and motion compensation processing on the initial video image to eliminate the influence of the UAV's own shaking on the image, and then extracts the actual displacement data of each insulator from the compensated video image.

[0034] The second determination module is used to identify target insulators from a pool of insulators whose motion data differs from that of other insulators, based on multiple motion data sets. In a healthy insulator string, all insulators are under the same mechanical environment and connection constraints, and their displacement under the same force should exhibit overall consistency. When an insulator has defects such as breakage, loose connections, or internal cracks, its local constraint stiffness decreases, and it will exhibit a significantly different motion displacement response compared to other normal insulators under the same force. The second determination module analyzes the statistical distribution of all insulator motion data to establish a normal distribution range reflecting the overall response level, and identifies insulators whose motion displacement deviates from this normal range as target insulators with defects.

[0035] Figure 2 This is a flowchart illustrating a method for detecting defects in high-voltage insulator strings based on UAV images, provided in an embodiment of the present invention. This method is applicable to the aforementioned system.

[0036] The specific process is as follows: S201: Acquire the initial video image of the insulator string in the preset first cycle based on the UAV.

[0037] The insulator string can be a porcelain insulator string, a glass insulator string, or a composite insulator string. The first cycle is preset to cover the time period during which the insulator string completes several reciprocating motion cycles when subjected to force.

[0038] S202: Based on the initial video image, determine the motion data corresponding to each insulator in the insulator string, wherein the motion data is the displacement data of the insulator when it is subjected to a force.

[0039] Understandably, force refers to external or internal mechanical factors that cause the insulator string to move. Examples include aerodynamic loads exerted on the insulator string by natural wind, and the impact force of the downwash airflow from a drone's rotor on the insulator string during hovering. Motion displacement data can be represented as the pixel distance or physical distance of the insulator from its initial position in the image coordinate system or a transformed world coordinate system. For example, in a video image, a healthy insulator may oscillate back and forth along the wind direction under wind force, with a maximum deviation of 5 pixels from its equilibrium position. Its motion displacement data would then be the curve showing the change of that pixel distance over time.

[0040] For example, step S202 may include the following steps: S2021: Determine the insulator string image and the background image in the initial video image.

[0041] Understandably, this division can be achieved through image semantic segmentation algorithms, or through manual labeling or traditional methods based on edge detection and morphological processing. For example, using the outline of the catenary and the periodic texture of the insulator string as features, pixel regions in the image that exhibit a string-like structure are labeled as the insulator string image, while other areas such as the sky, towers, and ground are labeled as the background image. The purpose of image separation is to distinguish the object of analysis from the reference benchmark.

[0042] S2022: Determine a reference point in the background image, where the reference point is a fixed point in the actual scene corresponding to the background image.

[0043] Understandably, fixed points can be connection nodes of transmission tower materials, corner points of signs on the tower, edges of ground-based fixed facilities, etc. The selection criterion is that the point has sufficient texture contrast in the image and can be stably tracked in each frame of the video. Since the fixed point is stationary in actual space, any pixel displacement it exhibits in the video image is caused by the drone's own shaking or drifting during hovering, not by the movement of the fixed point itself. For example, a bolt hole on a tower with clear texture against an open background is selected as a reference point, and its physical position remains unchanged throughout the video.

[0044] S2023: Perform motion compensation on the initial video image based on a fixed point and acquire the second video image.

[0045] Understandably, when a change in the pixel coordinates of a reference point is detected over time, it indicates that the drone itself is shaking, causing the entire image to shift or rotate. By calculating the positional offset of the reference point relative to the initial or previous frame in each frame, reverse translation compensation or affine transformation compensation is performed on each frame to keep the pixel positions of the reference point aligned across frames, thereby eliminating the global image motion caused by the drone's shaking. The compensated video image is the second video image. For example, if the reference point in frame 5 is shifted 3 pixels to the right and 2 pixels down relative to frame 1, then frame 5 is shifted 3 pixels to the left and 2 pixels up to align the reference point between the two frames.

[0046] S2024 determines the motion data corresponding to each insulator in the insulator string based on the second video image.

[0047] The second video image has eliminated the global displacement introduced by the drone's own shaking. The pixel position changes of each insulator on the insulator string in the image are caused only by the physical motion of the insulator when it is subjected to forces in actual space. Therefore, by tracking the feature position on each insulator in the second video image and extracting its displacement change within a preset first period, it can be used as the true motion data of each insulator.

[0048] S203: Based on multiple motion data, identify the target insulator from among multiple insulators whose corresponding motion data differs from the motion data of other insulators.

[0049] Understandably, on the same insulator string, all insulators are in the same mechanical environment, subjected to the same forces, and connected sequentially via hinge points. If the insulator string is intact as a whole, the structural stiffness and connection constraint state of each insulator are basically consistent, and their motion displacement under the same force should exhibit overall consistency, meaning the motion data of each insulator are close to each other. Conversely, when an insulator has defects such as damage, loose connections, or internal cracks, the local structural constraint stiffness of that insulator decreases, and its motion response under force will be significantly different from other normal insulators. For example, an insulator with a damaged skirt may have a smaller sway amplitude under the same wind force due to its reduced wind-receiving area; an insulator with a loose hinge may have a significantly larger sway amplitude due to its weakened constraint. Based on this principle, the motion data of each insulator is compared, and the insulator whose motion data differs significantly from that of most other insulators is identified as the target insulator, which is the insulator that may have defects.

[0050] Specifically, step S203 may include the following steps: S2031: In the second video image, obtain the target point on each insulator of the insulator string. The target point is the structural corner point on the surface of the insulator.

[0051] Understandably, to accurately quantify the movement of each insulator, a fixed position needs to be determined on each insulator for frame-by-frame video tracking—the target point. Here, natural structural corner points on the insulator surface are chosen as target points, such as the edges of the insulator skirts where they meet the steel cap, and the inflection points of the lower edge of the skirts. These structural corner points are feature points formed by the insulator's own geometric shape, possessing sufficient grayscale gradient contrast in the video image. They can be stably extracted and tracked frame-by-frame by corner point detection algorithms, and no artificial markings need to be affixed to the insulator surface, facilitating actual inspection operations.

[0052] S2032: Obtain the displacement of each target point in the preset first cycle, and use the maximum displacement as motion data.

[0053] After identifying the target point on each insulator, the target point is tracked frame by frame, recording its pixel coordinates in each frame within a preset first period. Using the position of the target point in the starting frame of the preset first period as a reference position, the pixel offset distance of the target point relative to the reference position in subsequent frames is calculated, forming a displacement curve over time. The maximum displacement value is extracted from this curve, representing the farthest pixel distance the target point deviates from the reference position within the preset first period. This maximum displacement is used as the maximum displacement of the corresponding insulator, and it is then used as the insulator's motion data. The maximum displacement reflects the amplitude of the insulator's swaying or vibration when subjected to a force.

[0054] For example, if the displacement curve of a target point on an insulator within a preset first period shows that its maximum pixel distance from the reference position is 8 pixels, then the maximum displacement of the insulator is 8 pixels.

[0055] S2033: Based on the statistical distribution of all maximum displacements, determine the normal distribution range of the maximum displacement.

[0056] After obtaining the maximum displacement of all insulators on the insulator string, statistical analysis is performed on these data to obtain statistical characteristic parameters reflecting the overall motion response level of the insulator string, and the normal distribution interval is determined accordingly. For example, the mean μ and standard deviation σ of all maximum displacements are calculated, and [μ-2σ, μ+2σ] or [μ-3σ, μ+3σ] is used as the normal distribution interval. The normal distribution interval represents the normal motion displacement range exhibited by most insulators of the insulator string under the current mechanical environment. Continuing the previous example, suppose there is an insulator string with 10 insulators, and the maximum displacements of each insulator are 8, 7, 9, 8, 8, 7, 8, 15, 8, 8 pixels. The mean μ is calculated to be 8.6 pixels, and the standard deviation σ is approximately 2.3 pixels. The normal distribution interval is [μ-2σ, μ+2σ] = [4.0, 13.2] pixels.

[0057] S2034: When there is an insulator whose maximum displacement exceeds the normal distribution range, the insulator is identified as the target insulator.

[0058] Understandably, the maximum displacement of a normal insulator should fall within the normal distribution range. When the maximum displacement of an insulator exceeds the upper or lower limit of the range, it indicates that the motion response of the insulator deviates significantly from the overall level of the insulator string, that is, there is a significant difference in motion data with other insulators. Based on this, it is identified as a target insulator, that is, an insulator that may have defects such as damage, loose connection or reduced structural stiffness.

[0059] Continuing with the previous example, the maximum displacement of the 8th insulator is 15 pixels, exceeding the upper limit of the normal distribution range of 13.2 pixels. Therefore, this insulator is identified as the target insulator. The reason for its large displacement may be that the hinge connecting this insulator to the adjacent insulator is loose, resulting in weakened constraint and a significantly increased sway amplitude under the same wind force.

[0060] Optionally, in this embodiment, before acquiring the initial video image of the insulator string corresponding to a preset first period based on the UAV, the process may further include: The drone is controlled to hover near the insulator string and its rotor speed is changed in a step manner to subject the insulator string to a momentary step excitation. The natural decay vibration period after the step excitation ends is taken as the preset first cycle.

[0061] Specifically, the drone is first controlled to fly to the vicinity of the target insulator string and hover. This hovering position is typically chosen on the windward or lateral side of the insulator string, allowing the downwash airflow generated by the drone's rotor to effectively act on the insulator string. The rotor speed is then rapidly and significantly altered via the flight control system, for example, by instantaneously increasing the throttle to cause a step increase in speed or instantaneously decreasing the throttle to cause a step decrease in speed. This instantaneous change in rotor speed will cause a corresponding instantaneous change in the thrust of the generated downwash airflow. This abrupt change in airflow impacts the insulator string in a pulse form, equivalent to applying a brief step excitation force to the insulator string. Under this step excitation, the insulator string experiences an initial disturbance and subsequently deviates from its original equilibrium position.

[0062] Understandably, after the step excitation ends, due to the disappearance of the airflow pulse, the insulator string, under the action of its own gravity and structural damping, begins to oscillate freely back and forth with a gradually decreasing amplitude from its deviated state after excitation, thus entering a state of natural decay vibration. Taking the moment when the step excitation ends as the starting moment, a segment of the subsequent natural decay vibration process is extracted as the preset first period for the initial video image acquisition in step S201.

[0063] This active excitation method ensures that the vibration response of the insulator string within the first preset cycle is entirely triggered by a known and controllable step excitation. Furthermore, the free decay vibration after excitation ends depends solely on the structural parameters of the insulator string itself, unaffected by continuously changing random natural wind loads. Therefore, regardless of the number of inspections, geographical location, or weather conditions, the testing of different insulator strings or the same insulator string at different times can be conducted under uniform initial excitation conditions, significantly enhancing the physical comparability of the motion data between different insulators.

[0064] Detection schemes based on motion displacement data are highly sensitive to defects affecting dynamic response, such as loose connections and decreased structural stiffness. However, some defects, such as micro-cracks on the surface of insulator skirts, have minimal impact on the structural stiffness and dynamic response of the insulator in the initial stage and may not be effectively detected by differences in motion displacement.

[0065] Furthermore, in this application, to reduce the probability of misjudgment, the method may also include the following steps: S301: High-resolution images of insulator strings acquired using drones.

[0066] Specifically, the drone is controlled to approach the insulator string and, under good lighting conditions, hovers and takes close-up photos to obtain high-resolution visible light images. High resolution here means that the pixel size of a single insulator in the image is sufficient to distinguish the fine textures and crack features on the surface of the sheds. During shooting, the insulator string can be in a non-energized state to avoid corona discharge interfering with image quality.

[0067] S302: Based on the sub-image corresponding to each insulator in the high-resolution image, obtain the crack image on each insulator.

[0068] First, the sub-image region corresponding to each insulator is segmented from the high-resolution image. Then, the sub-image of each insulator is preprocessed and suspected crack regions are extracted. For example, pixel regions exhibiting dark linear features in the sub-image can be extracted through methods such as local contrast enhancement, grayscale thresholding, or edge detection to form the crack image of that insulator. The crack image is a binary image or a grayscale image, where non-zero pixels correspond to the locations of suspected cracks.

[0069] S303: Obtain multiple crack orientations in each crack image and determine the orientation consistency constant of the multiple crack orientations.

[0070] Understandably, the extracted crack image is divided into multiple local sub-regions, and the local propagation direction of the crack, i.e., the crack orientation, is estimated within each sub-region. Once the local orientations of all sub-regions within the entire crack image have been estimated, a set of orientation data is formed. Based on this set of orientation data, its orientation consistency constant is calculated. This constant is used to quantify the degree of consistency among all local orientations: if all local orientations in the crack image are highly consistent, the orientation consistency constant is large; if the local orientations are disordered, the orientation consistency constant is small.

[0071] Specifically, step S303 may include the following steps: S3031: Segment the crack image to form multiple crack sub-images.

[0072] Understandably, the purpose of segmentation is to decompose the multiple crack segments with different orientations that may exist in the whole crack image into several sub-regions with a single dominant direction in the local area, so as to independently estimate the local crack orientation of each sub-region.

[0073] S3032: Obtain the structure tensor of the pixel gradient within each crack sub-image, and determine the main direction of the crack sub-image based on the eigenvalues ​​and eigenvectors corresponding to the structure tensor.

[0074] Understandably, the gradient reflects the rate and direction of change of image gray values ​​in space: at the crack edge, the gray values ​​change drastically and the gradient magnitude is large; in smooth areas, the gray values ​​change gently and the gradient magnitude is close to zero.

[0075] Specifically, the horizontal and vertical gradient components of each pixel in the crack sub-image can be obtained, and the gradient vector of the pixel can be determined based on the horizontal and vertical gradient components.

[0076] For example, for each pixel in the cracked sub-image, the rate of gray-level change along the horizontal direction (x-direction) of the image is calculated to obtain the horizontal gradient component gx, and the rate of gray-level change along the vertical direction (y-direction) of the image is calculated to obtain the vertical gradient component gy. The gradient components can be obtained by convolving the image with edge detection convolution kernels such as the Sobel operator or the Scharr operator. The two-dimensional gradient vector g=[gx,gy] of the pixel is composed of gx and gy. This vector points in the direction of the fastest increase in gray-level value at the pixel position, and its length reflects the drasticness of the gray-level change.

[0077] Then, the outer product of the gradient vectors is calculated and used as the initial structure tensor of the pixel. The outer product operation multiplies a 2×1 column vector by its own 1×2 transpose row vector, resulting in a 2×2 symmetric matrix. This matrix is ​​the initial structure tensor of the pixel. The top-left element of the matrix is ​​gx. 2 The bottom right element is gy 2 The two anti-diagonal elements are both gx·gy. This matrix records the complete distribution information of the gradient energy at this pixel in the horizontal and vertical directions and their coupling relationships.

[0078] The average value of the initial structure tensors of all pixels within the cracked sub-image is obtained by summing and averaging the values, and this average value is used as the structure tensor of the pixel gradients within the cracked sub-image. Specifically, the initial structure tensors corresponding to each pixel within the cracked sub-image are summed element-wise, and then divided by the total number of pixels to obtain the average matrix of all initial structure tensors. This average matrix is ​​then used as the structure tensor of the pixel gradients within the cracked sub-image. This summation and averaging operation integrates the gradient information across the entire sub-image region, ensuring that the final structure tensor represents the overall texture direction characteristics of the sub-image region while effectively suppressing noise interference from individual pixels.

[0079] The structure tensor is decomposed into eigenvalues ​​to obtain the first and second eigenvalues, as well as the first eigenvector corresponding to the first eigenvalue and the second eigenvector corresponding to the second eigenvalue, where the first eigenvalue is greater than or equal to the second eigenvalue. Since the structure tensor is a 2×2 real symmetric matrix, it necessarily has two real eigenvalues ​​and two mutually orthogonal real eigenvectors. Let the first eigenvalue obtained from the decomposition be λ1, and the second eigenvalue be λ2, where λ1 ≥ λ2; the eigenvector corresponding to the first eigenvalue λ1 is called the first eigenvector e1, and the eigenvector corresponding to the second eigenvalue λ2 is called the second eigenvector e2.

[0080] The eigenvalues ​​and eigenvectors contain key information about the texture directionality of the sub-image region: the first eigenvalue λ1 reflects the gradient energy perpendicular to the crack direction, where the grayscale change is most drastic; the second eigenvalue λ2 reflects the gradient energy along the crack direction, where the grayscale change is most gradual. If there are obvious linear structures (such as cracks) in the sub-image region, λ1 will be significantly greater than λ2; if the region contains uniform texture or noise, λ1 and λ2 will be approximately equal.

[0081] The direction of the second eigenvector corresponding to the second eigenvalue is determined as the main direction of the crack sub-image. Since e2 points to the direction of minimum gradient energy, i.e., the slowest gray-level change, and the gray-level change of the crack in its extension direction is exactly the most gradual, the direction of e2 is the best estimate of the local crack orientation in this sub-image.

[0082] S3033: Obtain the main directions of all crack sub-images within the crack image and form a direction set.

[0083] S3034: Obtain the circular variance of the direction set and use the circular variance as the direction consistency constant.

[0084] Directional data exhibits periodicity. For example, 0° and 180° actually represent the same straight line direction, but in linear numerical calculations, the difference between them is 180°. Directly using linear variance will produce erroneous results. Circular variance is a dispersion index specifically used for statistical analysis of directional data. Its calculation is based on the statistics of the sine and cosine values ​​of each directional angle, and it can correctly handle the cyclic jump problem when the angle crosses the 0° / 180° boundary.

[0085] The circular variance typically ranges from [0,1]: when all principal directions in the direction set are of uniform height, the circular variance is close to 0, indicating that the crack directions are highly consistent; when the principal directions are uniformly distributed across various angles, the circular variance is close to 1, indicating that the direction distribution is completely disordered. The calculated circular variance is used as the direction consistency constant. It is worth noting that in other embodiments of the present invention, the circular standard deviation or other concentration indicators based on direction data statistics can also be used to characterize direction consistency, and this is not limited here.

[0086] S304: If the directional consistency constant in a crack image is greater than a preset threshold, then the insulator corresponding to the crack image is determined to be the target insulator.

[0087] Understandably, the calculated directional consistency constant is compared with a preset threshold. The core mechanism of this step is that real structural cracks are usually caused by mechanical stress or electric field concentration, and their extension direction has obvious regularity and consistency; while the directional distribution of stains, water stains, random scratches or image noise on the surface of the insulator is usually disordered and does not have a consistent directionality.

[0088] Therefore, if the orientation consistency constant of a crack image is higher than a preset threshold, it indicates that the suspected crack pixel region in the image has significant orientation consistency, conforming to the topological characteristics of a real crack, and the corresponding insulator is determined to be a target insulator with a crack defect. Conversely, if the orientation consistency constant is lower than or equal to the preset threshold, its orientation distribution is random, and it is more likely to be surface stains, water stains, or noise interference, and is not determined to be a defect.

[0089] This crack orientation consistency analysis can effectively detect surface crack-type defects in the static high-resolution image dimension, complementing the aforementioned motion displacement analysis dimension. The combination of the two can cover a more comprehensive range of defect types and mutually verify each other to reduce the probability of misjudgment based on a single criterion.

[0090] Optionally, in low-light environments, the following steps may also be included: S404: Control the drone to acquire low-light video images of insulators in a energized operating state within a preset second cycle when the ambient light level is lower than the preset light threshold.

[0091] In addition to the aforementioned motion displacement analysis and crack detection scheme based on daytime visible light, this step further introduces the dimension of dark light discharge detection as a supplementary detection method for abnormal electrical performance of insulators under energized operation.

[0092] Specifically, the drone is controlled to fly near the target insulator string under conditions where the ambient light is below a preset threshold, such as dusk, night, or cloudy days. At this time, due to the weak ambient light, the faint discharge spots on the insulator surface caused by local defects will not be obscured by strong sunlight and can be effectively captured by the camera. The drone is equipped with a high-sensitivity visible light camera to continuously capture video of the energized insulator string, obtaining dark-light video images within a preset second period. The duration of the preset second period should be sufficient to cover several flicker cycles of the discharge spots.

[0093] S405: Based on low-light video images, obtain the maximum discharge spot on the surface and edge areas of each insulator.

[0094] After the bright spots are extracted, they are assigned to the corresponding insulators according to their spatial position in the image. For each insulator, all discharge spots that have appeared in all frames within a preset second period are recorded on its surface and in the edge area of ​​the skirt. The spot with the largest area is selected as the largest discharge spot of that insulator.

[0095] Understandably, the area of ​​the light spot is the number of pixels occupied by the light spot, reflecting the spatial extent of the discharge; the peak brightness is the maximum gray value of the pixels within the light spot area, reflecting the intensity of the discharge.

[0096] Here, the discharge intensity is defined as the product of the largest spot area and the largest brightness peak in any given frame of a low-light video image. That is, by traversing all video frames within a preset second period, the historical maximum area and the historical maximum brightness peak are independently found, and the product of these two maximum values ​​is taken as the discharge intensity of the insulator. This definition method integrates both the spatial extent and local intensity of the discharge, providing a more comprehensive reflection of the degree of abnormal discharge caused by defects than using only area or brightness.

[0097] S406: Obtain the spot area and brightness peak value of each maximum discharge spot, and determine the discharge intensity corresponding to each insulator. The discharge intensity is the product of the largest spot area and the largest brightness peak value that have appeared in any frame of the low-light video image.

[0098] S407: Determine the normal distribution range of discharge intensity based on the statistical distribution of all discharge intensities.

[0099] Calculate the mean and standard deviation of all discharge intensities, and use these as a basis to determine the normal distribution range. The normal distribution range represents the range of discharge intensities exhibited by most normal insulators under the current operating voltage and environmental conditions.

[0100] S408: When there is an insulator with a discharge intensity exceeding the normal distribution range, the insulator is identified as the target insulator.

[0101] Understandably, the discharge intensity of a normal insulator should be within the normal distribution range. When the discharge intensity of an insulator exceeds the upper limit of this range, it indicates that the surface discharge activity of that insulator is significantly stronger than that of other insulators in the same string. This may indicate defects such as surface damage, cracks, or severe contamination leading to localized electric field concentration, thus identifying it as the target insulator. This dark-light discharge detection scheme forms a complementary detection system with the aforementioned mechanical and apparent crack dimensions in the electrical dimension. Especially under energized operation, it can effectively detect insulation defects that cause partial discharge, further reducing the probability of missed detection.

[0102] In order to discover latent defects in insulators that alter local thermal conductivity properties, such as internal cracks, adhesive debonding, or material deterioration.

[0103] In some implementations, the following steps may also be included: S501: Control the drone to fly between the insulator string and the sun, and use the drone's body to project a moving shadow on the surface of the insulator string, so that the surface of the insulator string experiences a temperature rise process from being shaded to being directly exposed to sunlight again within a preset third cycle.

[0104] Specifically, the drone is controlled to fly along the line connecting the sun and the target insulator string, blocking some sunlight and creating a moving shadow area on the insulator string surface. The drone flies slowly along the insulator string's extension or moves along a preset path, sweeping its shadow from one end of the string to the other. For a fixed location on the insulator string, as the shadow moves away and sunlight returns to direct illumination, the surface undergoes a complete temperature recovery process, gradually rising from the lower temperature under shadow to the steady-state temperature under sunlight. The time period of this process is the preset third cycle. By using the drone as a shader and the sun as a natural heat source, a controllable temperature step excitation can be applied to the insulator string surface without the need for additional active heating devices. The movement speed of the shadow can be adjusted by the drone's flight speed to accommodate the thermal response characteristics of insulators of different materials and sizes.

[0105] S502: The thermal infrared image of the insulator string is collected synchronously during the temperature rise using a thermal infrared camera mounted on a drone.

[0106] An infrared camera records the entire process of temperature change at various locations on the surface of the insulator string after the shadow moves away at a fixed frame rate, and obtains thermal infrared images within a preset third cycle.

[0107] S503: Extract the temperature rise time constant of each insulator surface from the thermal infrared image, showing the temperature recovery from a low point to a steady state after the shadow is removed.

[0108] For example, the curve of the average surface temperature changing over time can be extracted from the pixel region corresponding to each insulator in the thermal infrared image sequence. This curve records the complete process of the insulator surface gradually rising from the lowest temperature point to reaching a steady state of thermal equilibrium under sunlight after the shadow is removed. The temperature rise time constant is extracted from this temperature curve. The temperature rise time constant is a characteristic parameter describing how quickly an object responds to a temperature step excitation; its physical meaning is the time it takes for the object's surface temperature to change from its initial value to its steady-state value after being subjected to a temperature step excitation. The temperature rise time constant is determined by the thermal diffusivity, heat capacity, and internal structural continuity of the insulator material. A well-functioning insulator with uniform material and internal continuity has a definite temperature rise time constant; insulators with internal cracks, debonding, or material deterioration have impeded heat conduction at the defect interface, resulting in a reduced equivalent thermal diffusivity and a significantly larger temperature rise time constant.

[0109] S504: Determine the normal distribution range of the temperature rise time constant based on the statistical distribution of all temperature rise time constants.

[0110] Understandably, this range represents the thermal response range of a normal insulator in the insulator string to the same step excitation of sunlight temperature.

[0111] S505: When there is an insulator whose temperature rise time constant exceeds the normal distribution range, the insulator shall be identified as the target insulator.

[0112] Understandably, if the temperature rise time constant of a certain insulator exceeds the upper limit of the normal distribution range, it indicates that the temperature recovery rate of the insulator is significantly slower than that of other insulators in the same string, and there is an abnormality in its internal heat conduction path. Based on this, the insulator is determined to be a target insulator with latent defects such as internal cracks, adhesive debonding, or material deterioration.

[0113] This active thermal excitation detection scheme uses the temperature rise time constant rather than the static temperature value as the criterion, which completely eliminates the interference of uneven distribution of ambient temperature field and differences in the heat generated by the insulator itself on the detection results. It has high sensitivity to the detection of internal latent defects and forms a complementary detection system with motion displacement analysis and dark light discharge detection schemes in the three physical dimensions of mechanics, electricity and heat.

[0114] Based on the same inventive concept, embodiments of this application also propose a processing apparatus, which includes: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the high-voltage insulator string defect detection method based on UAV images according to the embodiments of this application.

[0115] Furthermore, to achieve the above objectives, embodiments of this application also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the high-voltage insulator string defect detection method based on UAV images according to embodiments of this application.

[0116] The following is a detailed introduction to each component of the processing equipment: Figure 3 This is a schematic diagram of the structure of a processing device provided in an embodiment of the present invention. Exemplarily, this processing device may be a network device, or a chip (system) or other component or assembly that can be disposed in a network device. Figure 3 As shown, the processing device 300 may include a processor 301. Optionally, the processing device 300 may also include a memory 302 and / or a transceiver 303. The processor 301 is coupled to the memory 302 and the transceiver 303, for example, via a communication bus.

[0117] The following is combined with Figure 3 A detailed description of each component of the processing equipment 300 is provided below: The processor 301 is the control center of the processing device 300. It can be a single processor or a collective term for multiple processing elements. For example, the processor 301 can be one or more central processing units (CPUs), application-specific integrated circuits (ASICs), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs).

[0118] Alternatively, the processor 301 can perform various functions of the processing device 300 by running or executing software programs stored in the memory 302 and by calling data stored in the memory 302.

[0119] In a specific implementation, as one example, processor 301 may include one or more CPUs, for example... Figure 3 CPU0 and CPU1 are shown in the diagram.

[0120] In a specific implementation, as one embodiment, the processing device 300 may also include multiple processors. Each of these processors may be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). Here, a processor may refer to one or more devices, circuits, and / or processing cores used for processing data (e.g., computer program instructions).

[0121] The memory 302 is used to store the software program that executes the solution of the present invention, and the processor 301 controls the execution. The specific implementation method can be referred to the above method embodiment, and will not be repeated here.

[0122] Optionally, the memory 302 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 302 may be integrated with the processor 301 or exist independently, and may be connected via the interface circuit of the processing device 300. Figure 3 (Not shown in the image) is coupled to processor 301, and this embodiment of the invention does not specifically limit this.

[0123] Transceiver 303 is used for communication with other processing devices. For example, if processing device 300 is a terminal, transceiver 303 can be used to communicate with a network device or with another terminal device. As another example, if processing device 300 is a network device, transceiver 303 can be used to communicate with a terminal or with another network device.

[0124] Alternatively, transceiver 303 may include a receiver and a transmitter. Figure 3 (Not shown separately). The receiver is used to implement the receiving function, and the transmitter is used to implement the transmitting function.

[0125] Optionally, the transceiver 303 can be integrated with the processor 301, or it can exist independently and be connected via the interface circuit of the processing device 300. Figure 3(Not shown in the image) is coupled to processor 301, and this embodiment of the invention does not specifically limit this.

[0126] Understandable Figure 3 The structure of the processing device 300 shown does not constitute a limitation on the processing device. The actual processing device may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0127] Furthermore, the technical effects of the processing device 300 can be referred to the technical effects of the methods in the above-described method embodiments, and will not be repeated here.

[0128] It should be understood that the processor in the embodiments of the present invention can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0129] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DRRAM). The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuitry), firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to embodiments of the present invention is produced. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium can be any available medium that a computer can access, or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives (SSDs).

[0130] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0131] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0132] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0133] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

Claims

1. A method for detecting defects in high-voltage insulator strings based on UAV images, characterized in that, include: The initial video image of the insulator string in the preset first cycle is obtained by using a drone; Based on the initial video image, determine the motion data corresponding to each insulator in the insulator string, wherein the motion data is the displacement data of the insulator when it is subjected to a force; Based on multiple motion data, the insulator whose corresponding motion data differs from the motion data of other insulators is identified as the target insulator.

2. The method for detecting defects in high-voltage insulator strings based on UAV images according to claim 1, characterized in that, Based on the initial video image, determine the motion data corresponding to each insulator in the insulator string, including: Determine the insulator string image and the background image in the initial video image; A reference point is determined in the background image, wherein the reference point is a fixed point in the actual scene corresponding to the background image; Motion compensation is performed on the initial video image based on the fixed point, and a second video image is obtained; Based on the second video image, motion data corresponding to each insulator in the insulator string is determined.

3. The method for detecting defects in high-voltage insulator strings based on UAV images according to claim 2, characterized in that, Based on multiple motion data, the insulator whose corresponding motion data differs from the motion data of other insulators is identified as the target insulator, including: In the second video image, target points are obtained on each insulator of the insulator string, and the target points are structural corner points on the surface of the insulator. The displacement of each target point in the preset first cycle is obtained, and the maximum displacement is used as the motion data. Based on the statistical distribution of all the maximum displacements, the normal distribution range of the maximum displacements is determined; When there is an insulator whose maximum displacement exceeds the normal distribution range, the insulator is identified as the target insulator.

4. A method for detecting defects in high-voltage insulator strings based on UAV images according to any one of claims 1-3, characterized in that, Before acquiring the initial video image of the insulator string in the preset first period based on the drone, the following is also included: The drone is controlled to hover near the insulator string and its rotor speed is changed in a step manner so that the insulator string is subjected to a momentary step excitation. The natural decay vibration period after the step excitation ends is taken as the preset first period.

5. The method for detecting defects in high-voltage insulator strings based on UAV images according to claim 1, characterized in that, The method further includes: High-resolution images of the insulator string are obtained using the drone; Based on the sub-image corresponding to each insulator in the high-resolution image, obtain the crack image on each insulator; Obtain multiple crack orientations in each crack image and determine the orientation consistency constant of the multiple crack orientations; If the directional consistency constant in a crack image is greater than a preset threshold, then the insulator corresponding to the crack image is determined to be the target insulator.

6. The method for detecting defects in high-voltage insulator strings based on UAV images according to claim 5, characterized in that, Obtain multiple crack orientations in each crack image and determine the orientation consistency constants of the multiple crack orientations, including: The crack image is segmented to form multiple crack sub-images; Obtain the structural tensor of the pixel gradient within each crack sub-image, and determine the main direction of the crack sub-image based on the eigenvalues ​​and eigenvectors corresponding to the structural tensor; Obtain the main directions of all the crack sub-images within the crack image to form a direction set; Obtain the circular variance of the direction set and use the circular variance as the direction consistency constant.

7. The method for detecting defects in high-voltage insulator strings based on UAV images according to claim 6, characterized in that, Obtain the structure tensor of pixel gradients within each crack sub-image, and determine the main direction of the crack sub-image based on the eigenvalues ​​and eigenvectors corresponding to the structure tensor, including: Obtain the horizontal and vertical gradient components of each pixel in the crack sub-image, and determine the gradient vector of the pixel based on the horizontal and vertical gradient components; Calculate the outer product of the gradient vectors and use it as the initial structure tensor of the pixel; The average value is obtained by summing and averaging the initial structure tensors of all pixels in the cracked sub-image, and is used as the structure tensor of the pixel gradient in the cracked sub-image. The structural tensor is subjected to eigenvalue decomposition to obtain a first eigenvalue and a second eigenvalue, as well as a first eigenvector corresponding to the first eigenvalue and a second eigenvector corresponding to the second eigenvalue, wherein the first eigenvalue is greater than or equal to the second eigenvalue; The direction of the second feature vector corresponding to the second feature value is determined as the main direction of the crack sub-image.

8. The method for detecting defects in high-voltage insulator strings based on UAV images according to claim 1, characterized in that, The method further includes: The drone is controlled to acquire low-light video images of the insulator in a energized operating state within a preset second cycle under the condition that the ambient illuminance is lower than a preset illuminance threshold. Based on the low-light video image, the maximum discharge spot size of each insulator surface and edge region is obtained; The spot area and brightness peak value of each maximum discharge spot are obtained, and the discharge intensity corresponding to each insulator is determined, wherein the discharge intensity is the product of the largest spot area that appears in any frame of the low-light video image and the largest brightness peak value that appears in any frame. Based on the statistical distribution of all the discharge intensities, the normal distribution range of the discharge intensity is determined; When an insulator has a discharge intensity that exceeds the normal distribution range, the insulator is identified as the target insulator.

9. The method for detecting defects in high-voltage insulator strings based on UAV images according to claim 1, characterized in that, The method further includes: The drone is controlled to fly between the insulator string and the sun, and the drone's body is used to cast a moving shadow on the surface of the insulator string, so that the surface of the insulator string experiences a temperature rise process from being shaded to being directly exposed to sunlight again within a preset third cycle. The thermal infrared image of the insulator string is simultaneously acquired by the thermal infrared camera mounted on the drone during the temperature rise process. From the thermal infrared image, extract the temperature rise time constant of each insulator surface as the temperature recovers from the low point to the steady state after the shadow is removed; Based on the statistical distribution of all the temperature rise time constants, the normal distribution range of the temperature rise time constants is determined; When there is an insulator whose temperature rise time constant exceeds the normal distribution range, the insulator is identified as the target insulator.

10. A high-voltage insulator string defect detection system based on UAV imagery, characterized in that, A method for performing high-voltage insulator string defect detection based on UAV images as described in claim 1 includes: The acquisition module is used to acquire the initial video image of the insulator string in a preset first period based on the UAV; The first determining module is used to determine the motion data corresponding to each insulator in the insulator string based on the initial video image, wherein the motion data is the displacement data of the insulator when it is subjected to a force; The second determining module is used to determine, based on multiple motion data, the target insulator among the multiple insulators whose corresponding motion data differs from the motion data of other insulators.

Images