Extra-high voltage insulator three-dimensional infrared model reconstruction method and system

Abstract

The invention discloses an extra-high voltage insulator three-dimensional infrared model reconstruction method and system. The method comprises the steps of obtaining aerial photography parameters and sending the aerial photography parameters to an unmanned aerial vehicle; the unmanned aerial vehicle receives the aerial photography parameters, flies according to the aerial photography parameters, collects an infrared image of the insulator in the flying process and sends the infrared image to the ground control station; receiving an infrared image of the insulator sent by the unmanned aerial vehicle; preprocessing each infrared image, calculating a camera pose of each infrared image, and generating a point cloud based on the plurality of preprocessed infrared images and the camera poses; fitting the point cloud by using a mixture model formed by K Gaussian components to obtain a 3D Gaussian mixture model; and mapping the infrared temperature of each infrared image to a 3D Gaussian mixture model to obtain an insulator three-dimensional infrared model. According to the insulator three-dimensional infrared model provided by the invention, infrared temperature information is fused, and the accuracy of defect identification can be improved.

Description

Technical Field

[0001] This invention relates to the field of visual inspection technology for power equipment, specifically to a method and system for reconstructing a three-dimensional infrared model of an ultra-high voltage insulator. Background Technology

[0002] Ultra-high voltage (UHV) post insulators are critical equipment in UHV transmission systems, serving the dual functions of supporting conductors and providing insulation. Their operational status directly affects the safety and stability of the power grid. During long-term operation, insulators are affected by natural environmental factors (such as atmospheric corrosion, pollution deposits, and temperature variations) and mechanical stresses (such as conductor tension and wind loads), making them prone to defects such as surface cracks, internal air gaps, and partial discharge. If these defects are not detected and addressed in a timely manner, they can lead to a continuous decline in insulation performance, triggering flashover, breakdown, and other faults, resulting in large-scale power outages.

[0003] Currently, the inspection of UHV post insulators mainly relies on manual inspection and traditional image detection technology, but there are obvious limitations: 1) Manual inspection: It requires inspectors to work at height, which is labor-intensive and inefficient (the inspection of a single insulator takes about 2-3 hours), and there is a risk of falling while working at height; 2) Two-dimensional image detection: It can only obtain local planar information of the insulator, and cannot determine the location and depth of defects in three-dimensional space, making it easy to miss hidden defects; 3) Traditional three-dimensional modeling: It is mostly based on visible light images to build morphological models, without integrating infrared temperature information, and cannot correlate temperature anomalies (such as hot spots generated by partial discharge) with structural defects; 4) Sampling trajectory defects: Existing UAV sampling mostly adopts fixed radius circumferential or straight line scanning trajectories, which easily form sampling blind spots at the top or bottom, resulting in insufficient modeling accuracy (average error > 0.1m). Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing UHV post insulator detection technologies by providing a new method and system for reconstructing a three-dimensional infrared model of UHV insulators. This method combines spiral trajectory sampling, Gaussian mixture modeling, and model fusion with infrared temperature information, thereby improving the detection level of UHV post insulators.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a method for reconstructing a three-dimensional infrared model of an ultra-high voltage insulator. The method includes the following steps: Step S10: The ground control station obtains aerial photography parameters indicating the aerial photography status of a drone based on user input information, and sends the aerial photography parameters to the drone. The aerial photography parameters include a spiral flight trajectory, flight speed, and shooting angle intervals; Step S20: The drone receives the aerial photography parameters sent by the ground control station, flies according to the aerial photography parameters, and controls the infrared camera mounted on the drone to collect infrared images of the insulator during flight, and sends the infrared images of the insulator to the ground control station in real time; Step S30: The ground control station receives the infrared images of the insulator sent by the drone in real time, and obtains an image dataset including multiple infrared images; Step S31: The ground control station receives the infrared images of the insulator sent by the drone in real time, and obtains an image dataset including multiple infrared images; Step S42: The ground control station receives the infrared images of the insulator sent by the drone in real time, and obtains an image dataset including multiple infrared images; Step S3 ... ground control station, and obtains an image dataset including multiple infrared images; Step S33: The ground control station receives S40. The ground control station preprocesses each infrared image and uses the structure-of-motion (SOG) algorithm to calculate the camera pose of each infrared image. Based on the preprocessed multiple infrared images and the corresponding camera pose of each infrared image, a point cloud is generated. The preprocessing of the infrared images includes denoising, distortion correction, and spatiotemporal alignment. S50. The ground control station fits the point cloud with a mixture model composed of K Gaussian components to obtain a 3D Gaussian mixture model, where K is a positive integer and 5000≦K≦8000. S60. Based on coordinate mapping processing, the ground control station uses a Gaussian kernel function to map the infrared temperature of each infrared image to the 3D Gaussian mixture model to reconstruct the three-dimensional infrared model of the insulator. The three-dimensional infrared model of the insulator is a model that fuses the geometric features and infrared temperature features of the insulator.

[0006] In one specific implementation, the reconstruction method further includes: calculating the temperature gradient using the finite difference method and calculating the surface curvature using the basic form of the surface, and identifying defects based on the temperature gradient and the surface curvature.

[0007] In one specific implementation, step S60 further includes: performing model optimization on the reconstructed three-dimensional infrared model, wherein the model optimization includes median filtering and outlier removal.

[0008] In one specific implementation, the median filtering includes smoothing the Gaussian center points using a 3×3×3 window to eliminate surface noise, and the isolated point removal includes removing isolated noise point clusters using the DBSCAN clustering algorithm, wherein the parameters of the DBSCAN clustering algorithm include a neighborhood radius of 0.02m and a minimum number of points of 5.

[0009] In one specific implementation, the probability density function of the hybrid model in step S50 is:

[0010] ,in, ,

[0011] Let be the mixing coefficient of the i-th Gaussian component. Let be the mean vector of the i-th Gaussian component. Let K be the covariance matrix of the i-th Gaussian component, and K be the total number of Gaussian components.

[0012] In one specific implementation, the denoising process employs nonlocal mean filtering; the distortion correction includes radial distortion correction and tangential distortion correction, and the distortion error of the corrected image is ≤1 pixel; the spatiotemporal alignment extracts feature points using the SIFT algorithm and removes mismatches using the RANSAC algorithm, so that the image sequence alignment error is ≤2 pixels, wherein each infrared image extracts more than or equal to 500 feature points.

[0013] In one specific implementation, the user input information includes the insulator's characteristic information, the pitch used to calculate the spiral flight trajectory equation, the flight speed, and the shooting angle interval. The insulator's characteristic information includes the insulator's total height, the insulator's bottom center coordinates, and the insulator's bottom maximum radius.

[0014] In one specific implementation, step S10 further includes the ground control station acquiring a spiral flight trajectory based on the input insulator feature information and the pitch. The step of acquiring the spiral flight trajectory includes: acquiring the polar coordinate trajectory equation of the spiral trajectory based on the insulator feature information and the pitch, and converting the polar coordinate trajectory equation into a trajectory equation in a Cartesian coordinate system.

[0015] The polar coordinate trajectory equation includes: ; ;

[0016] The trajectory equation in the rectangular coordinate system is: ; ; ;

[0017] Where r is the helix radius, h is the height, and R0 is the bottom helix radius, which is 1.2 to 1.5 times the maximum radius of the insulator's bottom. t R0 is the radius of the top helix, which is 0.8 to 0.9 times the radius of R0. H is the total height of the insulator, θ is the polar angle, θ0 is the initial polar angle, and P is the pitch. The coordinates of the center of the bottom of the insulator;

[0018] In one specific implementation, step S20 further includes the UAV recording the GPS coordinates of the sampling points and sending them to the ground control station. Step S30 further includes the ground control station receiving the GPS coordinates of the UAV sampling points in real time, and detecting whether the UAV's flight deviates from the spiral trajectory based on the real-time received GPS coordinates of the sampling points. If the trajectory deviation is >0.3m, the spiral trajectory is automatically corrected and sent to the UAV.

[0019] Secondly, the present invention also provides a three-dimensional infrared model reconstruction system for ultra-high voltage insulators. The system includes a ground control station and a drone. The ground control station is used to acquire aerial photography parameters indicating the drone's aerial photography status based on user input information, and sends these parameters to the drone. The aerial photography parameters include a spiral flight trajectory, flight speed, and shooting angle intervals. The drone receives the aerial photography parameters sent by the ground control station, flies according to these parameters, and controls an infrared camera mounted on the drone to acquire infrared images of the insulator during flight. The drone then sends these infrared images to the ground control station in real time. The ground control station also receives the infrared images of the insulator sent by the drone in real time, obtaining an image dataset including multiple infrared images. The ground control station preprocesses each infrared image and uses a motion recovery structure algorithm to calculate the camera pose of each infrared image. Based on the preprocessed multiple infrared images and the corresponding camera poses, a point cloud is generated. The preprocessing of the infrared images includes denoising, distortion correction, and spatiotemporal alignment. The ground control station also uses a K-axis... A mixture model composed of Gaussian components is fitted to the point cloud to obtain a 3D Gaussian mixture model, where K is a positive integer and 5000≦K≦8000; the ground control station is also used for coordinate mapping processing, using a Gaussian kernel function to map the infrared temperature of each infrared image to the 3D Gaussian mixture model, and reconstructing the three-dimensional infrared model of the insulator, which is a model that fuses the geometric features and infrared temperature features of the insulator.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] I. This invention provides a method for reconstructing a three-dimensional infrared model of an ultra-high voltage insulator. The method includes the following steps: Step S10: The ground control station obtains aerial photography parameters indicating the aerial photography status of a UAV based on user input information and sends the aerial photography parameters to the UAV. The aerial photography parameters include a spiral flight trajectory, flight speed, and shooting angle interval. Step S20: The UAV receives the aerial photography parameters sent by the ground control station, flies according to the aerial photography parameters, and controls the infrared camera mounted on the UAV to collect infrared images of the insulator during flight. The UAV then sends the infrared images of the insulator to the ground control station in real time. Step S30: The ground control station receives the infrared images of the insulator sent by the UAV in real time, obtaining an image dataset including multiple infrared images. Step S40: The ground control station preprocesses each infrared image and uses a motion recovery structure algorithm to calculate the camera pose of each infrared image. Based on the preprocessed multiple infrared images and the camera pose corresponding to each infrared image, a point cloud is generated. The preprocessing of the infrared images includes denoising, distortion correction, and spatiotemporal alignment. Step S50: The ground control station uses K... A mixture model composed of K Gaussian components is used to fit the point cloud to obtain a 3D Gaussian mixture model, where K is a positive integer and 5000≦K≦8000; in step S60, the ground control station uses coordinate mapping processing and a Gaussian kernel function to map the infrared temperature of each infrared image to the 3D Gaussian mixture model to reconstruct the three-dimensional infrared model of the insulator. The three-dimensional infrared model of the insulator is a model that fuses the geometric features and infrared temperature features of the insulator. In the method provided by this invention, a spiral trajectory is used for sampling, which improves the image coverage by 40% and eliminates sampling blind spots compared to traditional sampling trajectories; a mixture model composed of K Gaussian components is used to fit the point cloud, and the model parameters are optimized by the EM algorithm, resulting in high modeling accuracy; the reconstructed three-dimensional infrared model of the insulator incorporates infrared temperature information. When performing defect identification, the accuracy of defect identification can be improved by combining infrared temperature and geometric information (defect identification accuracy ≥90%).

[0022] Second, the method provided by this invention is fully automated, and the modeling time for a single insulator is ≤30 minutes, which is 80% more efficient than manual inspection and modeling.

[0023] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0024] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention. In the drawings:

[0025] Figure 1 This is a flowchart illustrating the steps of a method for reconstructing a three-dimensional infrared model of an ultra-high voltage insulator according to an embodiment of the present invention.

[0026] Figure 2 This is a schematic diagram of a spiral flight trajectory provided in an embodiment of the present invention;

[0027] Figure 3 This is a three-dimensional infrared model diagram constructed according to Embodiment 1 of the present invention. Detailed Implementation

[0028] The present invention will now be described in detail with reference to the embodiments shown in the accompanying drawings.

[0029] This invention provides a method for reconstructing a three-dimensional infrared model of an ultra-high voltage insulator, which is applicable to the reconstruction of a three-dimensional infrared model of an ultra-high voltage post insulator of 1000kV and above.

[0030] Please see Figure 1 According to a first aspect of the invention, the method includes the following steps:

[0031] Step S10: The ground control station obtains aerial photography parameters indicating the aerial photography status of the UAV based on user input information, and sends the aerial photography parameters to the UAV. The aerial photography parameters include a spiral flight trajectory, flight speed, and shooting angle interval.

[0032] In this invention, the ground control station is an industrial computer equipped with trajectory planning and data processing software, wherein the industrial computer has a CPU ≥ 3.0 GHz and memory ≥ 16 GB.

[0033] In an optional implementation, the user input information includes the insulator's characteristic information, the pitch used to calculate the spiral flight trajectory equation, the flight speed, and the shooting angle interval. The insulator's characteristic information includes the insulator's total height, the insulator's bottom center coordinates, and the insulator's bottom maximum radius.

[0034] In this invention, the flight speed of the UAV is 0.8 to 1.2 m / s.

[0035] In this invention, the shooting angle interval is 30°, that is, the drone collects one infrared image every 30° of rotation.

[0036] In an optional implementation, step S10 further includes the ground control station obtaining a spiral flight trajectory based on the input insulator feature information and the pitch. The step of obtaining the spiral flight trajectory includes: obtaining the polar coordinate trajectory equation of the spiral trajectory based on the insulator feature information and the pitch, and converting the polar coordinate trajectory equation into a trajectory equation in a rectangular coordinate system.

[0037] The polar coordinate trajectory equation includes: . .

[0038] The trajectory equation in the rectangular coordinate system is: . . .

[0039] Where r is the helix radius, h is the height, and R0 is the bottom helix radius, which is 1.2 to 1.5 times the maximum radius of the insulator's bottom. t R0 is the radius of the top helix, which is 0.8 to 0.9 times the radius of R0. H is the total height of the insulator, θ is the polar angle, θ0 is the initial polar angle, and P is the pitch. These are the coordinates of the center of the bottom of the insulator.

[0040] In one optional implementation, the pitch is dynamically set according to the total height of the insulator to ensure that the number of sampling turns is 3 to 5 turns, preferably, H / 5≦P≦H / 3.

[0041] In this invention, the pitch is dynamically adjusted according to the total height of the insulator to ensure that the number of sampling turns is 3 to 5 turns, so as to avoid blind spots.

[0042] Please refer to section 2. Figure 2 It shows a spiral flight trajectory.

[0043] This invention uses a spiral trajectory for sampling, which improves image coverage by 40% compared to traditional sampling trajectories and eliminates sampling blind spots.

[0044] Step S20: The UAV receives the aerial photography parameters sent by the ground control station, flies according to the aerial photography parameters, and controls the infrared camera mounted on the UAV to collect infrared images of the insulator during the flight, and sends the infrared images of the insulator to the ground control station in real time.

[0045] In this invention, the drone is a quadcopter drone with a flight time of ≥30 minutes and a wind resistance level of ≥6; GPS + Beidou dual-mode positioning with a positioning accuracy of ≤0.5m; attitude sensor with a heading angle of ±0.1° and a pitch / roll angle of ±0.1°.

[0046] In this invention, the infrared camera carried by the UAV is a 640×512 resolution infrared thermal imager with a temperature measurement range of -20~150℃ and an accuracy of ±2℃.

[0047] In this invention, the UAV flies along a spiral trajectory, acquiring one infrared image every 30° of rotation, and simultaneously recording GPS coordinates (accuracy ≤ 0.5m) and attitude data (heading angle, pitch angle, roll angle ±0.1). ° ).

[0048] Step S30: The ground control station receives the infrared images of the insulators sent by the UAV in real time, and obtains an image dataset including multiple infrared images.

[0049] For ease of understanding, if the drone collects infrared images at a total of 60 sampling points, then the image dataset includes a total of 60 infrared images.

[0050] Step S20 further includes the UAV recording the GPS coordinates of the sampling points and sending them to the ground control station. Step S30 further includes the ground control station receiving the GPS coordinates of the UAV sampling points in real time, and detecting whether the UAV's flight deviates from the spiral trajectory based on the real-time received GPS coordinates of the sampling points. When the trajectory deviation is >0.3m, the spiral trajectory is automatically corrected and sent to the UAV.

[0051] Step S40: The ground control station preprocesses each infrared image and uses the structure-of-motion-recovery algorithm to calculate the camera pose of each infrared image. Based on the preprocessed multiple infrared images and the camera pose corresponding to each infrared image, a point cloud is generated. The preprocessing of the infrared images includes denoising, distortion correction, and spatiotemporal alignment.

[0052] In this invention, the density of the point cloud is ≥100 points / m. 2 .

[0053] In one optional implementation, the denoising process uses nonlocal mean filtering to remove noise; the distortion correction includes radial distortion correction and tangential distortion correction, and the image distortion error after correction is ≤1 pixel; the spatiotemporal alignment extracts feature points using the SIFT algorithm and uses the RANSAC algorithm to remove mismatches, so that the image sequence alignment error is ≤2 pixels, wherein the feature points extracted for each infrared image are greater than or equal to 500.

[0054] In this invention, nonlocal mean filtering is used to remove noise, and a 7×7 neighborhood window is set to improve the image signal-to-noise ratio by more than 30%.

[0055] In this invention, distortion correction is performed based on camera intrinsic parameters, and the image distortion error after correction is ≤1 pixel, ensuring the accuracy of subsequent modeling.

[0056] It should be noted that this invention does not improve the denoising technology, distortion correction technology, or spatiotemporal alignment technology; it merely applies existing image processing methods.

[0057] Step S50: The ground control station uses a mixture model composed of K Gaussian components to fit the point cloud to obtain a 3D Gaussian mixture model, where K is a positive integer and 5000≦K≦8000.

[0058] This step can be understood as "continuous modeling and optimization" of the initial point cloud through mathematical models, solving the problems of discreteness, uneven density, and isolated information in the initial point cloud.

[0059] In an optional implementation, the overall probability density function of the hybrid model in step S50 is: ,in, ,

[0060] Let be the mixing coefficient of the i-th Gaussian component and satisfy... , Let be the mean vector of the i-th Gaussian component. Let K be the covariance matrix of the i-th Gaussian component, and K be the total number of Gaussian components.

[0061] In this invention, the model parameters are optimized using the EM algorithm until the parameters converge, and the average distance error between the 3D model and the point cloud is ≤0.05m.

[0062] In this invention, a parameter change of <0.001 is considered parameter convergence.

[0063] Step S60: The ground control station uses coordinate mapping processing and a Gaussian kernel function to map the infrared temperature of each infrared image to the 3D Gaussian mixture model to reconstruct the three-dimensional infrared model of the insulator. The three-dimensional infrared model of the insulator is a model that fuses the geometric features and infrared temperature features of the insulator.

[0064] In this invention, the three-dimensional infrared model of the insulator is equipped with temperature markings and supports local magnification (up to 10 times) and arbitrary axis temperature profile analysis.

[0065] In one optional implementation, the infrared temperature of each infrared image in step S60 is the corrected infrared temperature obtained by correcting the absolute temperature of the target infrared image as a reference, wherein the target infrared image is any one of the multiple infrared images.

[0066] It is understandable that infrared thermal imagers measure apparent temperature, which is affected by emissivity, which in turn changes with the viewing angle. Additionally, reflection interference can also affect the accuracy of temperature measurements.

[0067] In this invention, the infrared temperature of other infrared images is corrected based on the absolute temperature of the target infrared image, which is a prior art technique. The correction can be based on the emissivity of the observation angle, such as the reflection temperature compensation method.

[0068] In an optional implementation, step S60 further includes: performing model optimization on the reconstructed three-dimensional infrared model, wherein the model optimization includes median filtering and outlier removal.

[0069] In this invention, the median filtering includes smoothing the Gaussian center point using a 3×3×3 window to eliminate surface noise, and the isolated point removal includes removing three isolated noise point clusters using the DBSCAN clustering algorithm, wherein the parameters of the DBSCAN clustering algorithm include a neighborhood radius of 0.02m and a minimum number of points of 5.

[0070] In an optional implementation, the reconstruction method further includes: the ground control station using the finite difference method to calculate the temperature gradient and the surface curvature through the basic form of the surface, identifying defects based on the temperature gradient and the surface curvature, and generating a defect report.

[0071] In specific embodiments, temperature gradient thresholds and surface curvature thresholds can be set for judgment. These thresholds can be set empirically, specifically greater than or equal to 19℃ / m and less than or equal to 22℃ / m. For example, the temperature gradient threshold could be 19℃ / m, 19.5℃ / m, 20℃ / m, etc., and the surface curvature threshold could be set to an absolute Gaussian curvature greater than 0.5m. -2 wait.

[0072] In this invention, the temperature gradient and surface curvature identification of defects can greatly improve the accuracy and reliability of detection and reduce false detections.

[0073] In this invention, the three-dimensional infrared model of the insulator and the defect report can be exported and connected to the power operation and maintenance system to achieve data linkage.

[0074] In this invention, the principal curvature, Gaussian curvature, or mean curvature of each point on the surface is calculated using a differential geometry algorithm, and whether it is a defect is determined based on the principal curvature, Gaussian curvature, or mean curvature of each point.

[0075] It should be noted that the calculation formulas provided by the prior art used in this invention to calculate the principal curvature, Gaussian curvature and mean curvature will not be repeated here.

[0076] According to a second aspect of the present invention, the present invention provides a three-dimensional infrared model reconstruction system for ultra-high voltage insulators. The system includes a ground control station and a drone, wherein: the ground control station is used to acquire aerial photography parameters indicating the drone's aerial photography status based on user input information, and sends the aerial photography parameters to the drone, the aerial photography parameters including a spiral flight trajectory, flight speed, and shooting angle interval; the drone is used to receive the aerial photography parameters sent by the ground control station, fly according to the aerial photography parameters, and control the infrared camera mounted on the drone to acquire infrared images of the insulator during flight, and send the infrared images of the insulator to the ground control station in real time; the ground control station is also used to receive the infrared images of the insulator sent by the drone in real time, and obtain an image dataset including multiple infrared images; the ground control station is also used to preprocess each infrared image and calculate the camera pose of each infrared image using a motion recovery structure algorithm, and generate a point cloud based on the preprocessed multiple infrared images and the camera pose corresponding to each infrared image, the infrared image preprocessing including denoising, distortion correction, and spatiotemporal alignment; the ground control station is also used to use K... The point cloud is fitted with a mixture model composed of Gaussian components to obtain a 3D Gaussian mixture model. The ground control station is also used for coordinate mapping processing, using a Gaussian kernel function to map the infrared temperature of each infrared image to the 3D Gaussian mixture model, and reconstructing the three-dimensional infrared model of the insulator. The three-dimensional infrared model of the insulator is a model that fuses the geometric features and infrared temperature features of the insulator.

[0077] Example 1

[0078] This embodiment uses a 1000kV UHV substation post insulator as the test object, and its core parameters are as follows:

[0079] 1. The maximum outer diameter of the insulator bottom, D0, is 1.2m, which means the maximum radius of the insulator bottom is 0.6m; the total height of the insulator, H, is 12m, and the outer diameter of the insulator top, Dt, is 0.8m.

[0080] 2. Environmental conditions: wind speed 2.5 m / s, temperature 25℃, humidity 60%;

[0081] 3. Equipment parameters: Quadcopter drone (35-minute flight time), equipped with a 640×512 resolution infrared thermal imager (25mm focal length, temperature measurement range -20~150℃).

[0082] (II) Detailed Explanation of Implementation Steps

[0083] 1. Trajectory Planning and Sampling

[0084] (1) Calculate the bottom spiral radius R0 = 1.2 × D0 / 2 = 1.2 × 0.6 = 0.72 m (since D0 ≥ 1 m, take a coefficient of 1.2).

[0085] (2) Top spiral radius R t =0.9×R0=0.9×0.72=0.65m.

[0086] (3) Pitch: Based on H=12m, set P=H / 5=2.4m (to ensure that the number of sampling turns = 12 / 2.4=5 turns, with no blind zone, and meets the setting requirements of H / 5~H / 3).

[0087] (4) Flight parameters: speed v = 1.0 m / s, tangential velocity vθ ≈ 0.88 m / s, axial velocity vz ≈ 0.49 m / s.

[0088] (5) Sampling process: 1 frame is collected for every 30° rotation, 12 frames are collected per revolution (360° / 30°), and a total of 60 frames are collected for 5 revolutions, covering the full height of the insulator.

[0089] 2. Image preprocessing

[0090] (1) Non-local mean filtering: Using a 7×7 window and a smoothing parameter h=10, the signal-to-noise ratio of the image after denoising is improved from 25dB to 33dB.

[0091] (2) Distortion correction: Using the coefficients obtained from calibration, k1=-0.012, k2=0.003, p1=0.001, p2=-0.0005, the edge error after correction is ≤1 pixel.

[0092] (3) Spatiotemporal alignment: 620 SIFT feature points were extracted per frame, with a matching accuracy of 92%, and the offset between adjacent frames after alignment was ≤2 pixels.

[0093] 3. 3D Gaussian Reconstruction

[0094] (1) Point cloud generation: 18,600 three-dimensional points were obtained by the SfM algorithm, with a density of 125 points / m in the bottom region. 2 .

[0095] (2) Function fitting: Initialize 6000 Gaussian components (kernel function), EM algorithm converges after 65 iterations, and the average distance error between the model and the point cloud is 0.042m.

[0096] (3) Information fusion: Temperature mapping error ±0.4℃.

[0097] 4. Model Optimization and Output

[0098] (1) Median filtering: Remove 3 isolated noise points.

[0099] (2) Output results: Generate a three-dimensional infrared model of the insulator, support local magnification to view defect details, and output a defect report.

[0100] Please see Figure 3 , Figure 3 This is the three-dimensional infrared model of the insulator output in Example 1.

[0101] As can be seen from this embodiment, the application of this method on 12m high UHV post insulators is very effective. The total time for the entire process is about 28 minutes, which includes: UAV flight sampling (24.6m trajectory ÷ 1.0m / s ≈ 24 seconds), image preprocessing (60 frames × 8 seconds / frame ≈ 8 minutes), point cloud generation (SfM algorithm processing ≈ 6 minutes), 3D Gaussian model fitting (EM iteration 65 times ≈ 10 minutes), and model optimization (median filtering + outlier removal ≈ 3.5 minutes). The modeling time for a single insulator is ≤ 30 minutes, which is 80% more efficient than manual inspection.

[0102] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for reconstructing a three-dimensional infrared model of an ultra-high voltage insulator, characterized in that, The method includes the following steps: Step S10: The ground control station obtains aerial photography parameters indicating the aerial photography status of the UAV based on user input information, and sends the aerial photography parameters to the UAV. The aerial photography parameters include a spiral flight trajectory, flight speed, and shooting angle interval. Step S20: The UAV receives the aerial photography parameters sent by the ground control station, flies according to the aerial photography parameters, and controls the infrared camera mounted on the UAV to collect infrared images of the insulator during the flight, and sends the infrared images of the insulator to the ground control station in real time. Step S30: The ground control station receives the infrared images of the insulators sent by the UAV in real time, and obtains an image dataset including multiple infrared images; Step S40: The ground control station preprocesses each infrared image and uses the structure-of-motion-recovery algorithm to calculate the camera pose of each infrared image. Based on the preprocessed multiple infrared images and the camera pose corresponding to each infrared image, a point cloud is generated. The preprocessing of the infrared images includes denoising, distortion correction, and spatiotemporal alignment. Step S50: The ground control station uses a mixture model composed of K Gaussian components to fit the point cloud to obtain a 3D Gaussian mixture model, where K is a positive integer and 5000≦K≦8000; Step S60: The ground control station uses coordinate mapping processing and a Gaussian kernel function to map the infrared temperature of each infrared image to the 3D Gaussian mixture model to reconstruct the three-dimensional infrared model of the insulator. The three-dimensional infrared model of the insulator is a model that fuses the geometric features and infrared temperature features of the insulator.

2. The reconstruction method according to claim 1, characterized in that, The reconstruction method further includes: The temperature gradient is calculated using the finite difference method, and the surface curvature is calculated using the basic form of the surface. Based on the temperature gradient and the surface curvature, defects are identified.

3. The reconstruction method according to claim 1, characterized in that, Step S60 further includes: performing model optimization on the reconstructed three-dimensional infrared model, wherein the model optimization includes median filtering and outlier removal.

4. The reconstruction method according to claim 3, characterized in that, The median filtering includes smoothing the Gaussian center points using a 3×3×3 window to eliminate surface noise. The isolated point removal includes removing isolated noise point clusters using the DBSCAN clustering algorithm, wherein the parameters of the DBSCAN clustering algorithm include a neighborhood radius of 0.02m and a minimum number of points of 5.

5. The reconstruction method according to claim 1, characterized in that, The probability density function of the mixture model described in step S50 is: ,in , Let be the mixing coefficient of the i-th Gaussian component and , Let be the mean vector of the i-th Gaussian component. Let K be the covariance matrix of the i-th Gaussian component, and K be the total number of Gaussian components.

6. The reconstruction method according to claim 1, characterized in that, The denoising process employs nonlocal mean filtering; the distortion correction includes radial distortion correction and tangential distortion correction, and the distortion error of the corrected image is ≤1 pixel; the spatiotemporal alignment extracts feature points using the SIFT algorithm and removes mismatches using the RANSAC algorithm, so that the image sequence alignment error is ≤2 pixels, wherein each infrared image extracts more than or equal to 500 feature points.

7. The reconstruction method according to claim 1, characterized in that, The user input information includes the insulator's characteristic information, the pitch used to calculate the spiral flight trajectory equation, the flight speed, and the shooting angle interval. The insulator's characteristic information includes the insulator's total height, the insulator's bottom center coordinates, and the insulator's bottom maximum radius.

8. The reconstruction method according to claim 7, characterized in that, Step S10 further includes the ground control station obtaining a helical flight trajectory based on the input insulator feature information and the pitch. The step of obtaining the helical flight trajectory includes: obtaining the polar coordinate trajectory equation of the helical trajectory based on the insulator feature information and the pitch, and converting the polar coordinate trajectory equation into a trajectory equation in a rectangular coordinate system. The polar coordinate trajectory equation includes: ； ； The trajectory equation in the rectangular coordinate system is: ； ； ； Where r is the helix radius, h is the height, and R0 is the bottom helix radius, which is 1.2 to 1.5 times the maximum radius of the insulator's bottom. t R0 is the radius of the top helix, which is 0.8 to 0.9 times the radius of R0. H is the total height of the insulator, θ is the polar angle, θ0 is the initial polar angle, and P is the pitch. These are the coordinates of the center of the bottom of the insulator.

9. The reconstruction method according to claim 8, characterized in that, Step S20 further includes the UAV recording the GPS coordinates of the sampling points and sending them to the ground control station. Step S30 further includes the ground control station receiving the GPS coordinates of the UAV sampling points in real time, and detecting whether the UAV's flight deviates from the spiral trajectory based on the real-time received GPS coordinates of the sampling points. When the trajectory deviation is >0.3m, the spiral trajectory is automatically corrected and sent to the UAV.

10. A three-dimensional infrared model reconstruction system for ultra-high voltage insulators, characterized in that, The system includes a ground control station and a drone, wherein: The ground control station is used to obtain aerial photography parameters that indicate the aerial photography status of the UAV based on user input information, and send the aerial photography parameters to the UAV. The aerial photography parameters include a spiral flight trajectory, flight speed, and shooting angle interval. The UAV is used to receive the aerial photography parameters sent by the ground control station, fly according to the aerial photography parameters, and control the infrared camera mounted on the UAV to collect infrared images of the insulator during the flight, and send the infrared images of the insulator to the ground control station in real time. The ground control station is also used to receive infrared images of insulators sent by the UAV in real time, and obtain an image dataset including multiple infrared images; The ground control station is also used to preprocess each infrared image and calculate the camera pose of each infrared image using the structure of motion recovery algorithm. Based on the preprocessed multiple infrared images and the camera pose corresponding to each infrared image, a point cloud is generated. The preprocessing of the infrared images includes denoising, distortion correction and spatiotemporal alignment. The ground control station is also used to fit the point cloud with a mixture model composed of K Gaussian components to obtain a 3D Gaussian mixture model, where K is a positive integer and 5000≦K≦8000; The ground control station is also used for coordinate mapping processing, using a Gaussian kernel function to map the infrared temperature of each infrared image to the 3D Gaussian mixture model, and reconstructing the three-dimensional infrared model of the insulator. The three-dimensional infrared model of the insulator is a model that fuses the geometric features and infrared temperature features of the insulator.

Images