Power transmission and distribution line inspection method and system

Abstract

The invention discloses a power transmission and distribution line inspection method and system, and solves the technical problem of poor inspection effect caused by incapability of effective complementation of multi-source information in the existing power transmission and distribution line inspection method. The method comprises the steps of obtaining an original point cloud data set of a complex scene of a power transmission and distribution line, and performing alignment through a point cloud registration algorithm to generate an aligned target point cloud data set; generating a visual image data set after interference filtering based on the point cloud data set by adopting Gaussian filtering and median filtering methods; fusing the two types of data sets to generate a preliminarily matched feature point set, and performing deep analysis to obtain a comprehensive feature description vector; and generating a preliminary positioning result through mode comparison, finally performing secondary positioning on the two types of data sets based on the result, and finally generating routing inspection information of the broken strand position of the wire.

Description

Technical Field

[0001] This invention relates to the field of power line inspection and identification technology, and in particular to a method and system for inspecting power transmission and distribution lines. Background Technology

[0002] In the field of modern power grid inspection and maintenance, accurately identifying conductors (transmission and distribution lines) is a crucial step in ensuring the safe operation of the power grid. Research in this area is directly related to the stability of the power system and public safety, and has undeniable strategic value.

[0003] With technological advancements, the demand for related recognition technologies in complex environments is increasing, becoming a key direction for driving the intelligent development of industries. However, current mainstream recognition methods often struggle to cope with changing environmental interference and the diversity of target features in complex scenarios. Especially in low light, inclement weather, or cluttered backgrounds, traditional sensing methods are easily limited, leading to decreased recognition accuracy and even misjudgments or missed detections. This limitation not only affects inspection efficiency but also poses potential risks to subsequent maintenance decisions.

[0004] Current power transmission and distribution line inspections mostly rely on single technologies such as lidar or visual imaging. However, in high-voltage line inspection scenarios, when the conductor is partially obscured by tree branches, the visual image cannot fully present the actual state of the conductor due to the obstruction. While lidar can provide spatial structure information, it is difficult to accurately distinguish the conductor from surrounding objects such as tree branches. The inherent shortcomings of the two technologies prevent multi-source information from effectively complementing each other, ultimately resulting in poor inspection results. Summary of the Invention

[0005] This invention provides a method and system for inspecting power transmission and distribution lines, which solves the technical problem that existing methods for inspecting power transmission and distribution lines cannot effectively complement information from multiple sources, ultimately leading to poor inspection results.

[0006] The first aspect of this invention provides a method for inspecting power transmission and distribution lines, comprising:

[0007] The original point cloud dataset of the complex scene of the power transmission and distribution line is obtained, and the point cloud registration algorithm is used to align the original point cloud dataset to generate the aligned target point cloud dataset.

[0008] Based on the aligned target point cloud dataset, Gaussian filtering and median filtering methods are used to generate a visual image dataset after filtering out interference.

[0009] Based on the visual image dataset after filtering out interference and the aligned target point cloud dataset, a preliminary matching set of feature points is generated.

[0010] A deep analysis is performed on the initially matched feature point set to generate a comprehensive feature description vector;

[0011] The comprehensive feature description vector is subjected to pattern comparison to generate preliminary localization results;

[0012] Based on the preliminary positioning results, secondary positioning is performed on the aligned target point cloud dataset and the interference-filtered visual image dataset to generate conductor strand breakage location inspection information.

[0013] Optionally, the step of aligning the original point cloud dataset using a point cloud registration algorithm to generate an aligned target point cloud dataset includes:

[0014] The point cloud registration algorithm is used to perform coordinate system alignment on the original point cloud dataset to determine the aligned initial point cloud dataset.

[0015] The aligned initial point cloud dataset is divided into regions to determine the conductor-related point cloud subset;

[0016] The broken strand locations are identified in the relevant point cloud subset of the conductor, and multiple broken strand spatial location information are output.

[0017] Based on the spatial location information of multiple broken strands, local point cloud reconstruction is performed, and the reconstructed local point cloud dataset is output.

[0018] Perform geometric shape analysis on the reconstructed local point cloud dataset and output the broken strand distribution information;

[0019] The fragmented strand distribution information is transformed using latitude and longitude coordinates and labeled with spatial coordinates to output an aligned target point cloud dataset.

[0020] Optionally, the step of generating a visual image dataset after filtering out interference based on the aligned target point cloud dataset using Gaussian filtering and median filtering methods includes:

[0021] Based on the aligned target point cloud dataset, an initial visual image dataset is collected;

[0022] The Gaussian filtering method is used to filter the initial visual image dataset to output a subset of visual images with enhanced edges.

[0023] The median filtering method is used to smooth the edge-enhanced visual image subset to generate a smoothed visual image subset.

[0024] Visual data fusion is performed on the smoothed visual image subset to generate a visual image dataset after filtering out interference.

[0025] Optionally, generating a preliminary matching feature point set based on the interference-filtered visual image dataset and the aligned target point cloud dataset includes:

[0026] Spatial continuity is extracted at gray-level abrupt changes in the visual image dataset after interference is removed, generating a spatially continuous feature subset;

[0027] Local texture features at locations of abrupt grayscale changes are extracted from the spatially continuous feature subset.

[0028] Spatial and image feature matching is performed on the local texture features at the gray-level abrupt change and the spatial feature data of the broken strand region point cloud in the aligned target point cloud dataset to output a preliminary set of matched feature points.

[0029] Optionally, the step of performing deep analysis on the initially matched feature point set to generate a comprehensive feature description vector includes:

[0030] Based on the initially matched feature point set, feature enhancement is performed, and an enhanced feature subset is output.

[0031] Based on the enhanced feature subset, calculate the geometric morphology analysis input data;

[0032] The density change of the geometric morphology analysis input data is monitored, and an extended matching set is output.

[0033] The extended matching set and the geometric morphology analysis input data are fused to generate a comprehensive feature description vector.

[0034] Optionally, the step of performing pattern comparison on the comprehensive feature description vector to generate preliminary localization results includes:

[0035] Perform pattern comparison on the comprehensive feature description vector and output stability features;

[0036] Based on the aforementioned stability characteristics, the target region is determined;

[0037] Extract the conductor strand breakage confirmation coordinate points in the target area, and perform edge sharpening processing on the conductor strand breakage confirmation coordinate points to determine the preliminary positioning coordinate set;

[0038] Based on the preliminary positioning coordinate set, the broken strand is monitored, and a preliminary positioning result is generated.

[0039] Optionally, the step of performing secondary localization on the aligned target point cloud dataset and the interference-filtered visual image dataset based on the preliminary localization result to generate conductor strand breakage location inspection information includes:

[0040] Based on the preliminary positioning results and the aligned target point cloud dataset, a set of boundary ranges is determined.

[0041] Based on the set of boundary ranges, perform visual representation matching of the conductors and output a set of coordinate points for secondary positioning;

[0042] Spatial coordinate mapping is performed on the set of coordinate points of the secondary positioning to generate an enhanced description vector;

[0043] Based on the enhanced description vector, a distribution feature map is generated;

[0044] The final information integration point is extracted from the distribution feature map, and the feature point is aggregated and verified to generate the conductor strand breakage location inspection information.

[0045] A second aspect of the present invention provides a power transmission and distribution line inspection system, comprising:

[0046] The acquisition module is used to acquire the original point cloud dataset of complex scenarios of power transmission and distribution lines, and to align the original point cloud dataset using a point cloud registration algorithm to generate the aligned target point cloud dataset.

[0047] A module is used to generate a visual image dataset after filtering out interference based on the aligned target point cloud dataset using Gaussian filtering and median filtering methods.

[0048] According to the module, it is used to generate a preliminary matching set of feature points based on the visual image dataset after filtering out interference and the aligned target point cloud dataset;

[0049] The analysis module is used to perform in-depth analysis on the initially matched feature point set and generate a comprehensive feature description vector;

[0050] The comparison module is used to perform pattern comparison on the comprehensive feature description vector and generate preliminary localization results;

[0051] The generation module is used to perform secondary positioning on the aligned target point cloud dataset and the interference-filtered visual image dataset based on the preliminary positioning results, and generate conductor strand breakage location inspection information.

[0052] A third aspect of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the power transmission and distribution line inspection method as described in any of the preceding claims.

[0053] The fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the transmission and distribution line inspection method as described in any of the preceding claims.

[0054] As can be seen from the above technical solutions, the present invention has the following advantages:

[0055] The above-mentioned technical solution of the present invention provides a method for inspecting power transmission and distribution lines. It acquires an original point cloud dataset of a complex scene of a power transmission and distribution line, and uses a point cloud registration algorithm to align the original point cloud dataset to generate an aligned target point cloud dataset. It then uses Gaussian filtering and median filtering methods to generate a visual image dataset after filtering out interference based on the aligned target point cloud dataset. Based on the visual image dataset after filtering out interference and the aligned target point cloud dataset, it generates a preliminary matching set of feature points. It performs in-depth analysis on the preliminary matching set of feature points to generate a comprehensive feature description vector. It performs pattern comparison on the comprehensive feature description vector to generate a preliminary positioning result. Based on the preliminary positioning result, it performs secondary positioning on the aligned target point cloud dataset and the visual image dataset after filtering out interference to generate inspection information on the location of broken conductor strands. Based on the above solution, the present invention weakens the influence of environmental interference through Gaussian filtering and median filtering methods, overcomes the bottleneck of visual information loss caused by tree branch occlusion by using multi-source data feature matching and in-depth analysis, and improves the differentiation accuracy between conductors and surrounding objects such as trees through pattern comparison of comprehensive feature vectors, thereby further improving the inspection effect. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 This is a flowchart of the steps of a power transmission and distribution line inspection method provided in Embodiment 1 of the present invention;

[0058] Figure 2 This is a flowchart illustrating the process of outputting a preliminary matched set of feature points as provided in Embodiment 1 of the present invention.

[0059] Figure 3 A flowchart illustrating the output of the comprehensive feature description vector provided in Embodiment 1 of the present invention;

[0060] Figure 4 This is a flowchart illustrating a method for inspecting power transmission and distribution lines according to Embodiment 1 of the present invention.

[0061] Figure 5 This is a structural block diagram of a power transmission and distribution line inspection system provided in Embodiment 2 of the present invention. Detailed Implementation

[0062] This invention provides a method and system for inspecting power transmission and distribution lines, which solves the technical problem that existing power transmission and distribution line inspection methods cannot effectively complement multi-source information, ultimately leading to poor inspection results.

[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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 embodiments of the present invention, not all embodiments. 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. It should be noted that in the optional embodiments of the present invention, the object information and other related data involved require the permission or consent of the object when the embodiments of the present invention are applied to specific products or technologies, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. That is to say, if the embodiments of the present invention involve data related to the object, it needs to be obtained with the authorization and consent of the object, the authorization and consent of the relevant departments, and in compliance with the relevant laws, regulations, and standards of the country and region. If personal information is involved in the embodiments, the acquisition of all personal information requires the consent of the individual. If sensitive information is involved, the separate consent of the information subject is required, and the embodiments also need to be implemented with the authorization and consent of the object.

[0064] Please see Figure 1 , Figure 1 The flowchart illustrates the steps of a power transmission and distribution line inspection method provided in Embodiment 1 of the present invention.

[0065] The present invention provides a method for inspecting power transmission and distribution lines, comprising:

[0066] Step 101: Obtain the original point cloud dataset of the complex scene of the power transmission and distribution line, and use the point cloud registration algorithm to align it according to the original point cloud dataset to generate the aligned target point cloud dataset.

[0067] It should be noted that this invention uses a lidar device to scan complex power transmission and distribution line scenarios, acquiring raw point cloud data, which is stored as an initial dataset. A point cloud registration algorithm is used to align the coordinate system of the initial dataset, generating an aligned initial point cloud dataset. For the aligned initial point cloud dataset, regions are divided according to the spatial distribution patterns of the conductor area, determining conductor-related point cloud subsets. If the point cloud density in the conductor-related point cloud subset is lower than a preset threshold, it is determined that a strand breakage location may exist in that region, and the corresponding spatial location information is extracted. Based on the extracted spatial location information, local point cloud reconstruction is performed on the strand breakage location, obtaining reconstructed local point cloud data. Geometric shape analysis is performed on the reconstructed local point cloud data to determine the specific distribution range of the strand breakage location, obtaining the final strand breakage distribution information. For the strand breakage distribution information, corresponding spatial coordinate labels are generated and stored as the final dataset of the strand breakage location (i.e., the aligned target point cloud dataset).

[0068] For example, in the field of power line inspection, LiDAR equipment can be used to scan complex scenes. In mountainous or urban environments surrounding high-voltage transmission lines, airborne LiDAR systems emit laser pulses and capture the reflected signals to generate 3D point cloud data. Each of these data points contains x, y, and z coordinates as well as intensity information, stored as an initial dataset, for example in PCD (Point Cloud Data) format, for easy subsequent processing. This scanning process is typically carried out on a drone to ensure full coverage of the power line. The initial dataset may contain millions of points, representing the geometry of the power line, towers, and surrounding vegetation.

[0069] Further, step 101 may include the following sub-steps:

[0070] S11. Use a point cloud registration algorithm to perform coordinate system alignment on the original point cloud dataset and determine the initial point cloud dataset after alignment.

[0071] S12. Divide the aligned initial point cloud dataset into regions to determine the point cloud subsets related to the traverse.

[0072] S13. Identify the broken strand locations of the relevant point cloud subset of the conductor and output multiple broken strand spatial location information.

[0073] S14. Reconstruct the local point cloud based on the spatial location information of multiple broken strands, and output the reconstructed local point cloud dataset.

[0074] S15. Perform geometric shape analysis on the reconstructed local point cloud dataset and output the broken strand distribution information;

[0075] S16. Perform latitude and longitude coordinate transformation and spatial coordinate labeling on the broken strand distribution information, and output the aligned target point cloud dataset.

[0076] Raw point cloud dataset: The dataset containing raw information such as three-dimensional spatial point coordinates and intensity is captured by scanning complex scenes of power transmission and distribution lines with LiDAR equipment. It is the foundation for all subsequent point cloud processing steps.

[0077] Point cloud registration algorithm: an algorithm used to unify raw point cloud data acquired from multi-view and multi-position scanning into the same coordinate system, eliminate spatial deviations, and achieve point cloud data alignment (such as ICP algorithm, Iterative Closest Point Algorithm).

[0078] The aligned initial point cloud dataset refers to the scene point cloud that has been processed by the point cloud registration algorithm ICP and has only completed the coordinate system one. At this time, the dataset may contain all objects such as wires, towers, ground, and vegetation.

[0079] Traverse-related point cloud subset: Based on the spatial pattern of linear extension and high-density distribution of the traverse, a traverse-specific point cloud subset is divided from the aligned initial point cloud dataset, excluding irrelevant interference such as ground and trees.

[0080] Point cloud density: The number of point cloud data points contained in a unit volume. It is the core indicator for determining whether there is a broken strand in the conductor (the point cloud density decreases because the broken strand reduces the number of laser reflection points).

[0081] Spatial location information of broken stock: suspected broken stock areas identified by comparing point cloud density with preset thresholds, including spatial location description information such as center coordinates and boundary range.

[0082] Reconstructed local point cloud dataset: Based on the spatial location information of the broken strand, a local point cloud dataset is formed by using point cloud reconstruction technology (such as Poisson surface reconstruction) to complete the details and geometrically restore the suspected broken strand area.

[0083] Broken strand distribution information: By performing geometric shape analysis on the reconstructed local point cloud dataset, the specific distribution range of broken strand locations is clearly identified.

[0084] Aligned target point cloud dataset: refers to the point cloud dataset rich in breakage information obtained after a series of fine processing for breakage detection (region division, density analysis, local reconstruction, geometric analysis, coordinate annotation) based on the initial aligned dataset.

[0085] It should be noted that, firstly, a comprehensive scan of the complex scene of power transmission and distribution lines is performed using LiDAR equipment to capture key information such as the coordinates and intensity of points in three-dimensional space to obtain raw point cloud data, which is then stored as the initial dataset. Next, a point cloud registration algorithm is used to align the coordinate system of the raw point cloud dataset, eliminating spatial deviations caused by multi-view scanning and determining the aligned initial point cloud dataset. Then, based on the spatial distribution pattern of the linear extension and high-density distribution of the conductor region, the aligned initial point cloud dataset is divided into regions, separating out conductor-related point cloud subsets that exclude interference from ground, trees, etc. Finally, the number of points per unit volume (i.e., point cloud density) of the conductor-related point cloud subset is calculated and compared with a preset density threshold; if it is lower than this threshold, a judgment is made. The corresponding region may contain broken strands. The system identifies the locations of broken strands in a subset of the relevant point cloud, outputting multiple spatial location information for the broken strands (including the coordinates of the region's center and boundary range). Based on this spatial location information, point cloud reconstruction technology is used to locally complete and restore details of the suspected broken strand regions, outputting a reconstructed local point cloud dataset. Furthermore, the reconstructed local point cloud dataset undergoes curvature calculation, principal component analysis, and other geometric shape analyses to clarify the specific distribution range of the broken strand locations, outputting broken strand distribution information. Finally, the broken strand distribution information is converted to latitude and longitude coordinates, and the spatial boundary coordinates of the broken strand regions are labeled, forming an aligned target point cloud dataset containing broken strand location labels. This provides fundamental data support for subsequent visual image matching and precise broken strand localization.

[0086] The point cloud registration algorithm is used to align the initial dataset to a single coordinate system. To understand this, point cloud registration involves unifying point cloud data from multiple scanning perspectives into a single coordinate system to eliminate deviations caused by device movement or multi-angle scanning. Specifically, this algorithm can be based on the Iterative Closest Point (ICP) method. It first selects a reference point cloud, then iteratively optimizes the rotation and translation matrices by calculating the distances between corresponding points in other point clouds and the reference point, until the error is minimized.

[0087] For example, when processing point clouds of two adjacent conductor segments, if one segment is scanned from east to west and the other from south to north, registration will calculate transformation parameters to align them and generate a uniform point cloud dataset, which helps avoid positional shifts in subsequent analysis.

[0088] For example, for an aligned point cloud dataset, regions can be divided according to the spatial distribution pattern of the traverse area. Clustering algorithms such as DBSCAN can be used to identify continuous linear structures based on the linear extension characteristics of the traverse.

[0089] Specifically, the spatial distribution pattern of the traverse region indicates that the traverse point cloud is usually distributed in a thin strip shape, with high density and extending in a certain direction. By setting the radius parameter and the minimum number of points threshold, the algorithm will divide the point cloud into multiple clusters, thereby determining the traverse-related point cloud subset. For example, it can extract a subset containing only the traverse part from the overall data, avoiding interference from ground or tree points.

[0090] In one possible implementation, if the point cloud density of a subset of the point cloud related to the conductor is lower than a preset threshold, it is determined that the area may have a broken strand. Here, point cloud density refers to the number of points per unit volume. For example, the preset threshold is 50 points per cubic meter. If the density of a subset of the conductor drops below 30 points, its spatial location information, such as the center coordinates and boundary range, is extracted. This is because a broken strand will cause the conductor to spread out locally, reducing the number of laser reflection points and thus lowering the density. Through this judgment, potential problem areas can be quickly located.

[0091] For example, based on the extracted spatial location information, local point cloud reconstruction can be performed at the location of the broken strand. The Poisson surface reconstruction method can be used to infer the continuous surface from the sparse point cloud.

[0092] Specifically, the reconstruction process involves calculating the normal vectors of the point cloud, then constructing an implicit function field, and generating a mesh model by solving the Poisson equation. For example, for a low-density conductor segment, the reconstruction yields more complete local point cloud data, which helps to restore the geometric details at the break point and improves the accuracy of subsequent analysis.

[0093] In one possible implementation, geometric shape analysis of the reconstructed local point cloud data can be performed to determine the specific distribution range of the broken strand location. Curvature can be calculated or principal component analysis can be used to detect abnormal shapes.

[0094] Specifically, geometric shape analysis involves extracting the local curvature of the point cloud. If a normal conductor is a smooth cylinder with uniform curvature, then abrupt changes or irregular protrusions may occur at the break point. By comparing the curvature values ​​with a standard threshold, the distribution range is determined. For example, a break region might extend from coordinates (100, 20, 50) to (105, 20, 50), thus obtaining the final break distribution information. This analysis provides accurate defect location, improving the safety of power maintenance.

[0095] For example, based on the distribution information of broken strands, corresponding spatial coordinate labels can be generated. The distribution range can be converted into latitude and longitude coordinates, labeled as vector graphics, and stored as the final dataset of broken strand locations, such as in GeoJSON format (Geography JavaScript Object Notation). This not only facilitates visualization but also supports integration into a GIS (Geographic Information System) for subsequent maintenance planning.

[0096] It is worth mentioning that this invention comprehensively captures the three-dimensional spatial information of complex power transmission and distribution line scenarios using lidar, and eliminates spatial deviations from multi-view scanning through point cloud registration algorithms, providing a unified standard basic dataset for subsequent data processing and effectively avoiding the problem of chaotic spatial data under a single technical path. Furthermore, based on the spatial distribution pattern of the conductors, regions are divided and interference is separated, accurately extracting relevant point cloud subsets of the conductors, solving the core pain point of lidar's difficulty in distinguishing conductors from surrounding objects such as the ground and trees. Suspected broken strand areas are initially identified through point cloud density threshold comparison, and occluded or sparse areas are filled in by combining local point cloud reconstruction. The geometric details compensate for the lack of spatial information in complex scenarios such as tree branch obstruction; the specific distribution range of broken strands is clarified through geometric shape analysis and coordinate labeling is completed. The resulting aligned target point cloud dataset not only achieves accurate preliminary positioning of suspected broken strand areas, but also builds a bridge between the spatial structure information of lidar and subsequent visual image data, laying a solid foundation for deep fusion of multi-source data. It solves the problem of poor inspection results caused by incomplete information and weak differentiation ability of single technology in existing inspections from the source, and significantly improves the accuracy and anti-interference ability of broken strand inspection of power transmission and distribution lines in complex scenarios.

[0097] In this embodiment, point cloud data in a complex scene is collected by a lidar sensor, and a registration algorithm is used to align the point cloud data to a coordinate system. The spatial distribution characteristics of the conductor area are initially screened to extract the spatial distribution information of possible broken strand locations, and the aligned target point cloud dataset is obtained.

[0098] Step 102: Using Gaussian filtering and median filtering methods, generate a visual image dataset after filtering out interference based on the aligned target point cloud dataset.

[0099] It should be noted that, based on the aligned point cloud dataset, corresponding image data is acquired from a synchronized visual camera. Gaussian filtering is used to process regions with abrupt grayscale changes to address illumination interference in the image data, resulting in a pre-processed image subset. For this pre-processed subset, boundary sharpness is calculated and adjusted using pixel gradient values. Boundary sharpness is obtained by averaging the grayscale differences between pixels to determine the edge positions of the interference regions, resulting in an edge-enhanced image subset. Illumination interference features are extracted from this edge-enhanced image subset. If the pixel gradient value exceeds a preset threshold, a mean square filtering method is applied to smooth the grayscale changes, resulting in a smoothed image subset. Based on this smoothed image subset, visual data fusion is performed. Visual data fusion determines the texture distribution of the guide wire region by overlaying point cloud coordinates with image pixels, resulting in a visual image dataset with interference removed.

[0100] Furthermore, step 102 may include the following sub-steps:

[0101] S21. Based on the aligned target point cloud dataset, collect the initial visual image dataset;

[0102] S22. Use Gaussian filtering to filter the initial visual image dataset and output a subset of visual images with enhanced edges.

[0103] S23. The median filtering method is used to smooth the visual image subset after edge enhancement to generate a smoothed visual image subset.

[0104] S24. Perform visual data fusion based on the smoothed visual image subset to generate a visual image dataset after filtering out interference.

[0105] Initial visual image dataset: A collection of original visual images of power transmission and distribution lines acquired by a visual camera, synchronized with the spatial coordinates of the aligned target point cloud dataset. It may contain noise such as illumination interference.

[0106] Gaussian filtering: a filtering technique used to smooth image noise. It processes image pixels by weighted averaging, focusing on weakening areas of gray-scale abrupt changes caused by illumination interference, while preserving core image details.

[0107] The visual image subset after edge enhancement: After Gaussian filtering, the location of interference edges is clarified by calculating the boundary sharpness and adjusting the pixel gradient value, making the boundary between the conductor and the background clearer.

[0108] Median filtering method: a nonlinear filtering technique that replaces the original pixel value with the median value in the neighborhood of the pixel, which can effectively smooth gray-level abrupt changes, eliminate residual illumination interference, and does not destroy the image edge features.

[0109] Smoothed visual image subset: After median filtering, gray-level abrupt changes are further suppressed, the overall image is more stable, and the details of the conductor region are fully preserved in the image subset.

[0110] Visual data fusion is a process that integrates geometric spatial features and visual texture features by superimposing the spatial coordinate information of point clouds with the pixel information of visual images.

[0111] Visual image dataset after interference removal: After multiple filtering and data fusion, the final visual image dataset with clear wire texture distribution and elimination of lighting interference is used for subsequent feature extraction and matching.

[0112] It should be noted that, based on the aligned target point cloud dataset, an initial visual image dataset corresponding to the spatial coordinates of the point cloud is acquired from a synchronously mounted visual camera. Since the initial visual image dataset may contain illumination interference, a Gaussian filtering method is used to specifically process regions with abrupt grayscale changes in the image. Then, the boundary sharpness of the pre-processed image is calculated (achieved by averaging the grayscale differences between pixels), and the edge positions of the illumination interference regions are clarified by adjusting the pixel gradient values, thus outputting a subset of visual images with enhanced edges. Illumination interference features are extracted from this subset. If the detected pixel gradient value exceeds a preset threshold, a median filtering method is used to further smooth the grayscale changes, generating a smoothed subset of visual images. Visual data fusion is performed based on this smoothed subset. By superimposing the correspondence between point cloud coordinates and image pixels, the texture distribution of the guide wire region is determined, ultimately generating a visual image dataset with interference filtered out, providing clear visual data support for subsequent feature matching.

[0113] In this embodiment, based on the aligned point cloud dataset, corresponding image data is acquired from a synchronized visual camera. To address potential lighting interference in the image, a filtering method is used to process the boundary sharpness and pixel gradient values ​​of grayscale abrupt change regions, resulting in a visual image dataset after interference removal.

[0114] Step 103: Generate a preliminary matching set of feature points based on the visual image dataset after filtering out interference and the aligned target point cloud dataset.

[0115] It should be noted that, by filtering out interference from the visual image dataset, spatial continuity at locations of abrupt gray-level changes is extracted, resulting in a spatially continuous feature subset. Based on this subset, local texture features at these gray-level abrupt changes are extracted. Combined with the geometric shape and local density changes of the broken regions in the point cloud data, spatial and image feature matching is performed to determine a preliminary set of matched feature points.

[0116] The "broken strand region" refers to the geometric gap area formed by the breakage of strands in the point cloud data. Using a preliminary matched set of feature points, the boundary curvature changes of the broken strand region are analyzed. A boundary enhancement descriptor is obtained by calculating the curvature difference between boundary points, where the curvature difference is based on the geometric coordinates of adjacent points. If the boundary enhancement descriptor aligns with local texture features, the matching threshold is adjusted by monitoring density changes to determine the expanded matching point set. The expanded matching point set is then obtained, and the geometric morphology quantization results are fused to determine the final feature point distribution map.

[0117] It is worth mentioning that the boundary curvature change of the broken strand region is analyzed by using the feature point set of the initial matching. The curvature difference between boundary points is obtained by calculating the geometric coordinates of adjacent points to obtain the boundary enhancement descriptor. This can accurately capture the geometric abrupt change characteristics of the broken strand edge and enhance the boundary distinction between the broken strand region and the normal conductor. If the boundary enhancement descriptor is aligned with the local texture features, it indicates the effectiveness of the feature matching. At this time, the matching threshold is dynamically adjusted by monitoring density changes, which can reasonably expand the matching point set, avoid missing broken strand-related feature points, and reduce the inclusion of irrelevant interference points. After obtaining the expanded matching point set, the geometric morphology quantization results are fused to determine the final feature point distribution map. This can integrate the geometric features and visual texture features of the broken strand region to form a comprehensive and accurate feature distribution visualization result. This not only improves the accuracy and comprehensiveness of feature matching, but also provides a high-quality feature foundation for subsequent comprehensive feature description vector generation and broken strand pattern comparison, effectively enhancing the robustness of conductor broken strand feature recognition in complex scenarios. Among them, the geometric morphology quantization results refer to the numerical results obtained after quantifying the geometry of the broken strand region (such as curvature and boundary range), which are used to supplement the feature dimensions.

[0118] Furthermore, step 103 may include the following sub-steps:

[0119] S31. Extract spatial continuity at gray-level abrupt changes in the visual image dataset after filtering out interference, and generate a spatially continuous feature subset.

[0120] S32. Extract local texture features at locations of abrupt gray-level changes in a spatially continuous feature subset;

[0121] S33. Perform spatial and image feature matching on the local texture features at the gray-scale abrupt change and the spatial feature data of the broken strand region point cloud in the aligned target point cloud dataset, and output a preliminary matching feature point set.

[0122] Spatial continuous feature subset: A set of features extracted from the visual image dataset after filtering out interference, which are continuously distributed at gray-level abrupt changes, reflecting the spatial coherence of the suspected broken strand region.

[0123] Local texture features at gray-level abrupt changes: Based on a subset of spatially continuous features, the pixel neighborhood texture pattern information of the gray-level abrupt change region is used to characterize the visual detail features of the abrupt change region.

[0124] Spatial feature data of point cloud in the broken strand region: In the aligned target point cloud dataset, the spatial features related to the broken strand, such as geometric shape (e.g., shape of geometric gap area) and local density changes, are collectively referred to.

[0125] Broken strand area: Geometric gaps in point cloud data caused by broken conductor strands. The point cloud density in this area is usually lower than that in the normal conductor area.

[0126] Spatial and image feature matching: The process of projecting texture and spatially continuous features in a visual image onto the spatial features of a point cloud, and then filtering out associated feature points by calculating feature similarity.

[0127] Preliminary matching feature point set: After spatial and image feature matching, the preliminary set of feature points related to the broken section region is selected, which includes visual and spatial correlation information.

[0128] Boundary enhancement descriptor: A descriptor obtained by calculating the curvature difference between boundary points of the broken strand region (based on the geometric coordinates of adjacent points), used to enhance the distinction between the edge of the broken strand and the normal conductor.

[0129] Expanding the matching point set: After aligning the boundary enhancement descriptor with the local texture features, a more comprehensive set of broken strand-related feature points is obtained by adjusting the matching threshold, thereby reducing feature omissions.

[0130] It should be noted that, as Figure 2 As shown, spatial continuity is extracted at gray-level abrupt changes in the visual image dataset after interference filtering to generate a spatially continuous feature subset. Local texture features at gray-level abrupt changes are extracted from the spatially continuous feature subset. Spatial and image feature matching is performed on the local texture features at gray-level abrupt changes and the spatial feature data of the broken strand region point cloud in the aligned target point cloud dataset (including the geometric shape and local density changes of the broken strand region, where the broken strand region refers to the geometric gap area formed by the break of the strand in the point cloud data), and a preliminary matching feature point set is output. This provides comprehensive and reliable feature support for subsequent accurate identification of broken strands and effectively solves the problem of misjudgment and missed judgment caused by incomplete feature extraction of single data.

[0131] Furthermore, the boundary curvature changes of the fractured region are analyzed using the initially matched feature point set. The curvature difference between boundary points is obtained by calculating the geometric coordinates of adjacent points to obtain a boundary enhancement descriptor. If this boundary enhancement descriptor aligns with local texture features, the effectiveness of the feature matching is confirmed. The matching threshold is then dynamically adjusted by monitoring density changes to filter out an expanded matching point set. Finally, this expanded matching point set is obtained and fused with the geometric morphology quantization results to determine the final feature point distribution map. The final feature point distribution map refers to a chart that visualizes the spatial distribution of fractured region-related feature points by fusing the expanded matching point set and the geometric morphology quantization results, providing intuitive support.

[0132] In one possible implementation, after filtering out interference from the visual image dataset, it is first necessary to extract the spatial continuity at gray-level abrupt changes. Here, spatial continuity refers to the coherent distribution of gray-level value change areas in three-dimensional space. For example, in power line inspection, the visual image dataset may come from images of power lines taken by drones. Gray-level abrupt changes often correspond to damage points on the surface of the conductor, such as shadow transition areas formed by rust or cracks. The extraction process can first convert the image into a grayscale image, and then use edge detection algorithms such as the Sobel operator to scan pixels to identify continuous gray-level gradient chains. Then, the spatial extension length of these chains in the point cloud coordinate system is calculated. If the chain length exceeds a preset value, such as 10 pixels, it is classified as a continuous feature, thus obtaining a subset of spatially continuous features. This subset helps to identify the risk of continuous conductor breakage, avoid isolated noise interference, and improve the accuracy of inspection.

[0133] For example, when extracting local texture features at gray-level abrupt changes based on spatially continuous feature subsets, it can be understood as analyzing the texture pattern of the abrupt change region. For instance, the LBP (Local Binary Pattern) method can be used to calculate the texture descriptor of the pixel neighborhood. When combining the geometric shape of the broken strand region and local density changes in the point cloud data, the broken strand region specifically refers to the geometric gap area formed by the breakage of the strand in the power conductor, such as the void caused by the breakage of a single strand in aluminum stranded wire. By matching the texture features with the geometric shape of the point cloud, such as curved surface gaps and areas of reduced density, the matching process involves projecting image feature points onto point cloud coordinates and calculating the Euclidean distance. If the distance is less than a threshold, it is considered a preliminary match, thereby determining the set of feature points for preliminary matching. This matching can effectively locate potential fault points in the conductor and provide accurate positioning for maintenance.

[0134] In point cloud data, a broken strand in a conductor refers to a localized area of ​​missing or sparse point cloud data, i.e., a geometric void or gap formed at the break point. Normal conductors are continuous cylinders with high point cloud density, while broken strands reduce laser reflection points and lower point cloud density, forming a void. In this scheme, the geometric void is a key indicator for strand break detection, identified through point cloud density analysis and geometric shape analysis. For example, in step 101, when the point cloud density is below a threshold, the extracted spatial location information corresponds to the geometric void.

[0135] In one possible implementation, a preliminary matching set of feature points is used to analyze the process of boundary curvature change in the broken section region. A boundary enhancement descriptor can be obtained by calculating the curvature difference between boundary points. The curvature difference is based on the geometric coordinates of adjacent points. For example, a sequence of points on the broken section boundary is selected in the point cloud model, and the local curve is fitted using the least squares method. Then, the curvature difference between adjacent points is calculated. If the difference is large, it indicates that the boundary is sharp. The enhancement descriptor integrates these differences to form a vector description, which helps to highlight the geometric features of the broken section edge and avoid misjudgment of smooth areas.

[0136] For example, if the boundary enhancement descriptor is aligned with the local texture features, the matching threshold is adjusted by density change monitoring to determine the expansion of the matching point set. The alignment check can compare the similarity of the descriptor vectors. If the cosine similarity is higher than 0.8, the alignment is confirmed. Density change monitoring involves statistically analyzing the rate of change of the number of local points in the point cloud. If the rate of change exceeds 20%, the threshold is lowered to expand the matching and thus obtain more relevant point sets.

[0137] In one possible implementation, after obtaining the extended matching point set, the geometric morphology quantization results are fused to determine the final feature point distribution map. For example, the quantization results include calculating the volume or area of ​​the broken strand region, which is then weighted and superimposed onto the matching point set to form a distribution map. This can generate a visual fault map in conductor inspection, improving detection efficiency.

[0138] In this embodiment, for the visual image dataset after filtering out interference, the spatial continuity and local texture features at the gray-level abrupt changes are extracted. At the same time, the geometric shape and local density changes of the broken strand region in the point cloud data are combined to perform preliminary spatial and image feature matching and determine the preliminary matching feature point set.

[0139] Step 104: Perform in-depth analysis on the initially matched feature point set to generate a comprehensive feature description vector.

[0140] It should be noted that, through the initial matching feature point set, the abrupt change characteristics of the broken edge are extracted, and the gray-scale abrupt change features are fused with the background contrast intensity to obtain an enhanced feature subset. The correlation value between spatial continuity and local texture extraction is calculated from the enhanced feature subset, where the correlation value is obtained by the distance difference between adjacent feature points, thus determining the input data for geometric morphology analysis. For the input data, density change monitoring is performed, and the matching threshold is adjusted to obtain an expanded matching set. From the expanded matching set, the geometric morphology analysis results are fused to generate a comprehensive descriptive vector.

[0141] The comprehensive feature description vector is a multi-dimensional feature vector that integrates multiple features extracted from point cloud data and image data to comprehensively describe the characteristics of the broken strand region. These features include: abrupt changes in the broken strand edges, such as sharp changes in point cloud curvature or gradient jumps at image edges; contrast intensity between grayscale abrupt changes and the background, the difference in pixel brightness between the broken strand region and the surrounding background; spatial continuity, the degree of coherent distribution of feature points in three-dimensional space; and local texture features, the texture pattern of the broken strand region in the image (such as LBP features). Geometric morphology analysis results include, for example, the volume, curvature, or principal component analysis results of the broken strand region. The comprehensive feature description vector is used for pattern comparison (such as similarity calculation with a preset broken strand pattern) to confirm the existence of the broken strand. For example, in step 105, this vector is used to calculate the similarity with the broken strand pattern; if it exceeds a threshold, the broken strand is confirmed.

[0142] Furthermore, step 104 may include the following sub-steps:

[0143] S41. Perform feature enhancement based on the initially matched feature point set, and output the enhanced feature subset;

[0144] S42. Based on the enhanced feature subset, perform geometric morphology analysis on the input data;

[0145] S43. Monitor density changes in the geometric morphology analysis input data and output an extended matching set;

[0146] S44. The extended matching set and geometric morphology analysis input data are fused to generate a comprehensive feature description vector.

[0147] Feature enhancement: By extracting the characteristics of abrupt changes at the edge of the broken strand and fusing the contrast intensity of gray-scale changes with the background, the processing of strengthening effective features and weakening interfering features improves feature recognition.

[0148] Enhanced feature subset: The core feature set obtained after feature enhancement processing highlights key information such as edge abrupt changes and grayscale contrast in the broken strand region, and reduces the influence of irrelevant noise.

[0149] Correlation value: A numerical value used to characterize the correlation between spatial continuity and local texture extraction. It is obtained by calculating the distance difference between adjacent feature points and is used to screen effective data for geometric morphology analysis.

[0150] Input data for geometric morphology analysis: Based on the correlation values ​​of the enhanced feature subset, the data is used as the basis for subsequent analysis of the geometry (such as curvature and boundary range) of the broken section area.

[0151] Density change monitoring: Dynamically monitor the point cloud density or feature point distribution density in the area where the feature points are located to determine whether there is a process of abnormal density change caused by the break in the strand.

[0152] Expanded matching set: After monitoring density changes and adjusting the matching threshold, a more comprehensive set of feature points related to broken stock is obtained, which avoids feature omissions and eliminates invalid interference points.

[0153] Geometric morphology analysis results: Numerical results obtained after quantitative analysis of the geometry, boundary curvature, and spatial distribution range of the broken strand area, supplementing feature dimension information.

[0154] It should be noted that, as Figure 3 As shown, feature enhancement is performed based on the initially matched feature point set. By extracting the abrupt change characteristics of the broken strand edge, the gray-scale abrupt change features and background contrast intensity are fused to output an enhanced feature subset. Based on the enhanced feature subset, the correlation value between spatial continuity and local texture extraction is calculated (the correlation value is obtained by the distance difference between adjacent feature points) to determine the geometric morphology analysis input data. Density change is monitored on the geometric morphology analysis input data, and the matching threshold is dynamically adjusted to include more effective feature points, outputting an expanded matching set. The expanded matching set and the geometric morphology analysis results are deeply fused to integrate the geometric, texture, and density features of the broken strand region, generating a comprehensive feature description vector. This provides comprehensive feature support for subsequent broken strand pattern comparison and localization, helping to solve the problem of insufficient broken strand recognition accuracy in complex scenarios.

[0155] In addition, the feature distribution is generated based on the comprehensive feature description vector. This comprehensive feature description vector refers to a vector that integrates the extended matching set and the geometric morphology analysis results, combining multi-dimensional features such as geometry, texture, and density, and is used to characterize the core characteristics of the broken strand region.

[0156] In one possible implementation, when extracting the abrupt change characteristics of the broken strand edge through a pre-matched set of feature points, abrupt change points in the edge region can be identified from these feature points. For example, in a power line inspection scenario, the pre-matched set of feature points may come from the fusion of images collected by drones and point cloud data. These points correspond to potential broken strand locations on the surface of the conductor. The extraction process involves scanning the boundary of the point set and calculating abrupt changes in edge curvature. For example, the curvature difference between adjacent points can be calculated using a difference method. If the difference exceeds a set threshold, it is marked as an abrupt change characteristic, thus highlighting the geometric discontinuity of the broken strand edge. Next, the gray-level abrupt change features are fused with the background contrast intensity. Gray-level abrupt change features refer to the regions where pixel gray values ​​jump in the image, obtained through the Canny edge detection algorithm. The background contrast intensity is quantified by calculating the brightness difference between the abrupt change point and the surrounding background pixels. For example, the difference is normalized and weighted and fused into the abrupt change characteristics to form an enhanced feature subset. This subset integrates visual and geometric information, facilitating subsequent analysis.

[0157] For example, when calculating the correlation value between spatial continuity and local texture extraction after obtaining an enhanced feature subset, spatial continuity can be understood as the coherent distribution of feature points in three-dimensional space. Local texture extraction involves using methods such as Haralick texture feature analysis to analyze the statistical patterns of pixel neighborhoods. The correlation value is obtained by the distance difference between adjacent feature points. Specifically, point pairs in the subset are selected, their Euclidean distance difference is calculated, and the differences are accumulated to form a correlation index. If the index is higher than the threshold, it indicates a strong correlation, thereby determining the input data for geometric morphology analysis. This input data filters out noise points and ensures the reliability of the analysis.

[0158] In one possible implementation, density change monitoring for input data refers to the fluctuation of the number of points in a local area of ​​the statistical point cloud. For example, grids are divided near the broken strand area, and the average and standard deviation of the point density of each grid are calculated. If the fluctuation exceeds 15%, it is considered a significant change. The matching threshold is adjusted accordingly, such as lowering the threshold to include more points, thereby obtaining an extended matching set. This set expands the scope of the initial matching and covers the potential extended broken strand area.

[0159] For example, when generating a comprehensive descriptive vector by fusing geometric morphology analysis results from an extended matching set, the geometric morphology analysis results include the shape quantification of the broken region, such as calculating the convex hull volume of the region. First, the principal axis of the shape is extracted through principal component analysis, and then these results are concatenated with the feature vectors in the set to form a multidimensional comprehensive descriptive vector that integrates density and morphological information.

[0160] The expanded matching set is a more comprehensive set of feature points obtained after preliminary feature matching by adjusting the matching threshold (e.g., based on density changes), containing more potential break points. Geometric morphology analysis results are then fused: this includes quantifying the shape of the break region, such as calculating curvature, volume, and principal component analysis. These analysis results (e.g., convex hull volume or curvature values) are combined with feature points in the expanded matching set, for example, through vector concatenation or weighted fusion. The comprehensive description vector generated in the above steps represents the fused features as a multi-dimensional vector, which integrates geometric, texture, and density information for subsequent pattern comparison. Based on the comprehensive description vector, feature distribution generation is monitored: a feature distribution map is generated by performing clustering analysis (e.g., K-means algorithm) or spatial distribution analysis on the comprehensive description vector. This distribution map visualizes the clustering of break feature points, used to identify high-risk areas or verify the stability of break locations. In step 104, this process helps generate the final feature point distribution map, providing a basis for precise localization.

[0161] In one possible implementation, when generating the feature distribution based on the comprehensive description vector monitoring, the vectors can be grouped using a clustering algorithm such as K-means to generate a distribution map. For example, in conductor inspection, the grouping results can be mapped to a coordinate system to highlight the distribution of high-risk broken strands, which helps in overall fault monitoring.

[0162] In this embodiment, feature extraction technology is used to perform in-depth analysis on the initially matched feature point set, and the abrupt change characteristics of the broken edge and the contrast intensity between the gray-scale abrupt change and the background are integrated to generate a comprehensive feature description vector.

[0163] Step 105: Perform pattern comparison on the comprehensive feature description vector to generate preliminary localization results.

[0164] It should be noted that by performing pattern comparison on the comprehensive feature description vector, the stability features of surface roughness and gray-level abrupt changes at the broken strand location are obtained in multiple frames of images. The continuity of the stability features is judged to obtain the pattern similarity value. Based on the pattern similarity value, the matching degree between the preset broken strand pattern and the vector is calculated after fusing the image sequence compensated for illumination changes, and regions where the similarity calculation value exceeds a predetermined threshold are identified. The coordinate points for confirming the broken strand of the conductor are extracted from the region, and edge sharpening is used to enhance the boundary clarity to obtain a preliminary location coordinate set. Based on the preliminary location coordinate set, the dynamic stability in the multi-frame image sequence is monitored to determine the existence of the broken strand target of the conductor, and a preliminary location result is obtained.

[0165] Furthermore, step 105 may include the following sub-steps:

[0166] S51. Perform pattern comparison on the comprehensive feature description vector and output stability features;

[0167] S52. Determine the target area based on stability characteristics;

[0168] S53. Extract the coordinate points for confirming broken strands of the conductor in the target area, and perform edge sharpening on the coordinate points for confirming broken strands of the conductor to determine the preliminary positioning coordinate set.

[0169] S54. Based on the preliminary positioning coordinate set, perform strand breakage monitoring and generate preliminary positioning results.

[0170] Pattern comparison: The process of calculating the similarity between the comprehensive feature description vector and the preset wire strand break pattern (including typical strand break texture and geometric feature template) to select target features.

[0171] Stability characteristics: The surface roughness (local surface variance quantization of point cloud) and gray-level abrupt changes of the broken section remain continuous and without significant fluctuations in multiple frames of images, which is a key indicator for judging the authenticity of the broken section.

[0172] Pattern similarity value: Calculated using algorithms such as normalized cross-correlation, it represents the degree of matching between the comprehensive feature description vector and the preset stock breaking pattern. The higher the value, the stronger the matching degree.

[0173] Image sequence after illumination change compensation: A standardized image set that uses methods such as histogram equalization to adjust the brightness distribution of multiple frames of images and eliminate the influence of illumination fluctuations on feature recognition.

[0174] Preset conductor strand breakage pattern: Based on common conductor strand breakage samples, a pre-constructed standard template containing features such as strand breakage texture and geometric shape is used for pattern comparison reference.

[0175] Target area: The area where the pattern similarity value exceeds a predetermined threshold, i.e., the area where there is a suspected broken strand in the conductor.

[0176] Conductor strand breakage confirmation coordinates: Specific pixel coordinates extracted from the target area, corresponding to the boundary and core area of ​​the broken strand.

[0177] Edge sharpening: The image is convolved using algorithms such as the Laplacian operator to enhance the boundary distinction between the broken area and the surrounding background, making the coordinate point extraction process more accurate.

[0178] Preliminary location coordinate set: After edge sharpening, the set of valid broken stock coordinate points is selected, which preliminarily represents the spatial location of the broken stock.

[0179] Dynamic stability monitoring: The process of tracking the displacement changes of the initial positioning coordinate set in multiple frames of images and determining whether the broken strand target is real (not noise or interference) by the centroid offset.

[0180] Preliminary location results: After dynamic stability verification, preliminary spatial location information of the broken strand of the conductor was confirmed, providing a basis for secondary precise location.

[0181] It should be noted that pattern comparison is performed on the comprehensive feature description vector to extract the stability features of surface roughness and gray-level abrupt changes in the broken strand in multiple frames of images. The continuity of this stability feature is judged to obtain the pattern similarity value, and the stability feature is output. Based on the stability feature and the corresponding pattern similarity value, the image sequence after illumination change compensation (the brightness distribution is adjusted by histogram equalization to eliminate the influence of illumination fluctuation) is fused, and its matching degree with the preset wire broken strand pattern is calculated. The region with the similarity calculation value exceeding the predetermined threshold is determined as the target region. The wire broken strand confirmation coordinate points are extracted in the target region, and edge sharpening processing (such as Laplacian operator convolution) is used to enhance the clarity of the broken strand boundary. Valid coordinate points are further screened to determine the preliminary positioning coordinate set. Based on the preliminary positioning coordinate set, the dynamic stability in the multi-frame image sequence is monitored (tracking the displacement change of coordinate points and judging whether the centroid offset is within the preset range) to confirm the authenticity of the wire broken strand target. Finally, the preliminary positioning result is generated, which provides a reliable basis for subsequent secondary precise positioning and effectively improves the accuracy of broken strand recognition in complex scenes.

[0182] In one possible implementation, when obtaining the stability features of surface roughness and gray-level abrupt changes in the broken section across multiple frames of images by performing pattern comparison on the comprehensive feature description vector, the data in the vector can first be decomposed into surface roughness components and gray-level abrupt change components. For example, in the scenario of power line inspection, surface roughness is quantified by calculating the variance of the local surface in the point cloud data. The specific process involves selecting coordinate points in the vector, constructing a neighborhood with a radius of 5 mm around each point, and calculating the distance deviation from the points in the neighborhood to the fitting plane. The larger the deviation value, the higher the roughness. Gray-level abrupt changes are extracted from the pixel gradient in the image sequence, and the edge intensity is detected using the Sobel operator. These components are compared across multiple frames of images to determine the continuity of the stability features. For example, by calculating the mean of the roughness deviation between adjacent frames, if the deviation is less than 0.1, it is considered stable, thus obtaining the pattern similarity value. This value is obtained using the cosine similarity formula, where the smaller the angle between vectors, the higher the similarity.

[0183] For example, when fusing pattern similarity values ​​with image sequences after illumination change compensation, illumination compensation is first applied to the image sequences.

[0184] Specifically, histogram equalization is used to adjust the brightness distribution of each frame of the image to keep it consistent under different lighting conditions. Then, the matching degree between the pre-set broken strand pattern and the vector is calculated. For example, the pre-set pattern includes typical broken strand texture templates, such as spiral crack patterns. The matching degree is calculated by template matching algorithms such as normalized cross-correlation. Regions with similarity calculation values ​​exceeding a predetermined threshold, such as 0.8, are identified. These regions are marked as potential broken strand areas and further connected to the subsequent extraction process.

[0185] In one possible implementation, when extracting the coordinate points for confirming the broken strands of the conductor from the region, edge sharpening is used to enhance the clarity of the boundaries. For example, the Laplacian operator is used to perform a convolution operation on the region image to highlight edge details and obtain a preliminary set of positioning coordinates. This set is fitted to the coordinate points using the least squares method to form the broken strand boundary curve, ensuring the accuracy of the positioning.

[0186] For example, when monitoring the dynamic stability of a multi-frame image sequence based on a preliminary location coordinate set, the displacement changes of coordinate points in the sequence can be tracked to determine the existence of a broken wire target. For example, the centroid offset of the point set can be calculated. If the offset is less than 2 pixels, the target is confirmed to be stable, and a preliminary location result is obtained. This result provides basic data for subsequent inspections.

[0187] In this embodiment, by performing pattern comparison on the comprehensive feature description vector, it is determined whether it contains the stability features of surface roughness and gray-scale abrupt changes of the broken strand in multiple frames of images. If the similarity between the comprehensive feature description vector and the preset wire broken strand pattern exceeds a predetermined threshold, the existence of the wire broken strand target is confirmed, and a preliminary positioning result is obtained.

[0188] Step 106: Based on the preliminary positioning results, perform secondary positioning on the aligned target point cloud dataset and the visual image dataset after filtering out interference to generate conductor strand breakage location inspection information.

[0189] It should be noted that, based on preliminary results, the contrast and gray-level abrupt changes of the broken strands are obtained from the point cloud data. These gray-level abrupt changes are calculated using pixel gradient differences. By fusing shape features from the image data, the boundary range of the abrupt change region is determined, resulting in a set of boundary ranges. For this set of boundary ranges, visual representations around the conductor are used for matching. These visual representations are based on color distribution contrast to determine the continuity of the matching, obtaining a set of coordinate points for secondary localization. This set of coordinate points is then combined with the mapping relationship of spatial coordinates.

[0190] The mapping relationship is established through projection transformation from point cloud to image. If the mapping relationship meets a preset threshold, the enhanced description of the visual representation is determined, resulting in an enhanced description vector. The stability of gray-level abrupt changes in the enhanced description vector is obtained, calculated using the variance of multiple frame sequences. The correspondence between point cloud data and image data is fused to determine the distribution of stability features, resulting in a distribution feature map. The integration points of the final information are extracted from the distribution feature map. These integration points are obtained through feature point aggregation and verified using the shape features of the abrupt change region to determine the specific spatial coordinates of the broken strand in the conductor (transmission and distribution line), thus obtaining the final broken strand location information.

[0191] Furthermore, step 106 may include the following sub-steps:

[0192] S61. Based on the preliminary positioning results and the aligned target point cloud dataset, determine the set of boundary ranges;

[0193] S62. Perform visual representation matching of the traverse based on the boundary range set, and output the coordinate point set for secondary positioning;

[0194] S63. Perform spatial coordinate mapping on the set of coordinate points for secondary positioning to generate an enhanced description vector;

[0195] S64. Generate a distribution feature map based on the enhanced description vector;

[0196] S65. Extract the final information integration point from the distribution feature map, and perform feature point aggregation and verification on the final information integration point to generate conductor strand breakage location inspection information.

[0197] Boundary range set: Based on the preliminary positioning results, combined with the contrast of broken strands in the point cloud data, gray-level abrupt changes (pixel gradient difference calculation), and shape features in the image data, the boundary range set of the abrupt change region is determined.

[0198] Visual representation matching of conductors: This is a process of matching the boundary range set with the visual features around the conductor, with color distribution contrast as the core, to verify the correlation of features.

[0199] The set of coordinate points for secondary positioning: a more accurate set of coordinate points related to the broken strand obtained after matching the visual representation of the traverse and satisfying the continuity judgment.

[0200] Spatial coordinate mapping relationship: The correspondence between the three-dimensional point cloud coordinates and the two-dimensional image pixel coordinates established through the projection transformation from point cloud to image.

[0201] Enhanced description vector: When the spatial coordinate mapping relationship meets the preset threshold, the feature vector formed by integrating visual performance enhancement description contains key information such as gray-scale change stability.

[0202] Gray-level mutation stability: This is determined by calculating the variance of gray-level mutations in a multi-frame image sequence, and it characterizes the stability of gray-level mutation features in the time dimension.

[0203] Distribution feature map: A chart that integrates the correspondence between point cloud and image data to visualize the spatial distribution of stability features.

[0204] Final information integration point: a key point extracted from the distribution feature map that can comprehensively reflect the core characteristics of the stock breakup.

[0205] Feature point aggregation: The process of clustering scattered final information points according to spatial correlation to form a set of points that centrally represent the broken strand region.

[0206] Information on the location of broken conductor strands: After feature point aggregation and shape feature verification, the specific spatial coordinates of the broken conductor strands are determined, which is the core result of the inspection.

[0207] It should be noted that, based on the preliminary localization results and the aligned target point cloud dataset, the contrast and gray-level abrupt changes (calculated by pixel gradient differences) of the broken strands are obtained from the point cloud data, and shape features from the visual image dataset after filtering out interference are fused to jointly determine the boundary range of the abrupt change region, resulting in a boundary range set. Based on this boundary range set, visual representation matching of the conductor is performed, where visual representation is based on color distribution contrast. Valid matching results are filtered by judging the continuity of this matching, and a set of coordinate points for secondary localization is output. Spatial coordinate mapping is performed on the secondary localization coordinate point set, combined with the spatial coordinate mapping relationship established through the projection transformation from point cloud to image. If the mapping relationship meets a preset threshold, the enhanced description of the visual performance is determined, and an enhanced description vector is generated. Based on the enhanced description vector, the gray-scale mutation stability is extracted (where gray-scale mutation stability is calculated by multi-frame sequence variance). The correspondence between point cloud data and image data is fused to determine the distribution of stability features and generate a distribution feature map. The final information integration point is extracted from the distribution feature map. This final information integration point is obtained by feature point aggregation. Then, the shape features of the mutation region are used for verification to accurately determine the specific spatial coordinates of the conductor strand breakage. Finally, the conductor strand breakage location inspection information is generated, providing accurate location basis for the maintenance of conductor strand breakage in power transmission and distribution lines.

[0208] In one possible implementation, when obtaining the strand breakage contrast and grayscale abrupt changes from point cloud data based on preliminary results, the point cloud data can be preprocessed first. For example, in power line inspection, the point cloud can be converted into a grayscale image, and the contrast can be quantified by calculating the brightness difference between pixels. The specific process involves selecting a set of points in the conductor region, calculating the neighborhood average grayscale value of each point, and then calculating the difference with the overall average value to form a contrast index. The grayscale abrupt changes are calculated using pixel gradient differences. For example, the Prewitt operator can be used to scan the image and detect gradient changes in the horizontal and vertical directions. Regions with high gradient values ​​indicate obvious abrupt changes. In this way, shape features in the image data can be integrated, such as using contour extraction algorithms to identify irregular edges and determine the boundary range of the abrupt change region. Finally, a set of boundary ranges is obtained. This set contains the coordinate boundaries of possible strand breakage locations in the conductor, providing a basis for subsequent matching.

[0209] For example, when matching the set of boundary ranges using the visual representation around the conductor, the visual representation is based on the color distribution comparison. In the inspection scenario, the color histogram of the conductor surface is first analyzed, and the RGB distribution difference between the broken strand area and the normal area is compared. If the difference exceeds a preset value, such as 20%, it is considered that the match is initially established. Then, the continuity of the match is judged by checking the continuous frame changes of color distribution in multiple frames of images to ensure that the match is not an isolated phenomenon, thereby obtaining a set of coordinate points for secondary positioning. This set of points further refines the broken strand location.

[0210] The broken strand location information includes the node location and boundary range of the broken strand area. During the point cloud identification of the conductor, the conductor point cloud image is simultaneously divided into multiple unit-length node numbers according to length. The broken strand location information refers to the specific spatial location information of the broken strand (i.e., the broken strand) of the conductor determined through the fusion perception of LiDAR and visual images during the conductor identification process. It includes geometric descriptions such as the node location (i.e., node number) and boundary range (i.e., the difference in node numbers spanned by the broken strand area) of the broken strand area. Simultaneously, the three-dimensional coordinates (e.g., x, y, z coordinates) of the broken strand area need to be determined based on the three-dimensional point cloud data of the conductor point cloud image. In this scheme, the broken strand location information is the final output, used for fault location and maintenance decisions in power line inspection. For example, it may be stored in GeoJSON format, containing the center coordinates of the broken strand area, facilitating visualization in a GIS system.

[0211] In one possible implementation, when combining the coordinate point set with the spatial coordinate mapping relationship, the mapping relationship is established through the projection transformation from point cloud to image. For example, the camera intrinsic and extrinsic parameter matrices are used for projection to map the 3D point cloud points to the 2D image plane. If the coordinate deviation after mapping is less than 1 pixel, that is, the preset threshold is met, then the enhanced description of the visual performance is determined, and the enhanced description vector is obtained. This vector integrates color and shape information, which improves the robustness of detection.

[0212] For example, when obtaining the stability of gray-scale abrupt changes in the enhanced descriptor vector, it is calculated by multi-frame sequence variance.

[0213] Specifically, grayscale value sequences of the same area are selected from multiple video frames, and their variance is calculated. If the variance is less than 0.05, it indicates stability. The correspondence between point cloud data and image data is fused, such as by associating the two through a feature point matching algorithm, to determine the distribution of stability features and obtain a distribution feature map. This map visualizes the stability of the broken strand area, which helps to quickly identify potential risks in business.

[0214] In one possible implementation, when extracting the integration point of the final information from the distribution feature map, the integration point is obtained by aggregating feature points. For example, the K-means clustering method is used to aggregate similar feature points into groups, and then the shape features of the abrupt change region are used for verification, such as checking whether the ellipticity of the aggregation point conforms to the spiral shape of the broken strand, determining the specific spatial coordinates of the broken strand of the conductor, and obtaining the final broken strand location information. This information can provide accurate coordinates in the inspection business, support maintenance decisions, avoid the inefficiency of manual inspection, and thus improve the reliability of the overall power system.

[0215] In this embodiment, based on the preliminary positioning results, combined with the contrast between the broken strand and the surrounding conductors and the shape characteristics of the gray-scale abrupt change area, the point cloud data and image data are used for secondary precise positioning to determine the specific spatial coordinates of the broken conductor strand, thus obtaining the final broken strand location information (i.e., conductor broken strand location inspection information).

[0216] As a comparison of technical effects, existing technologies can be used as a reference. Currently, power transmission and distribution line inspections mostly rely on single technical paths such as lidar or visual imaging, or simply perform data overlay, without achieving deep integration. Lidar can accurately capture three-dimensional spatial structure and distance information, providing reliable geometric support for conductor positioning, but it has shortcomings in the perception of visual features such as texture and color. When faced with complex backgrounds such as dense vegetation and buildings, it is difficult to accurately distinguish conductors from surrounding interfering objects. Visual imaging can capture rich details, textures, and color information, helping to identify surface defects of conductors, but it is easily affected by factors such as changes in lighting, weather conditions, and occlusion—such as light spot interference in backlit scenes and image blurring in rainy weather. Especially in high-voltage line inspections, when conductors are partially obscured by tree branches, visual information will be missing, resulting in the incomplete presentation of defect features.

[0217] How to effectively integrate information from different sources to improve the robustness of recognition? Deep fusion of LiDAR and visual images is one of the core challenges. LiDAR can provide accurate distance and spatial structure information, but its ability to perceive texture and color features is weak; while visual images can capture rich visual details, they are easily affected by changes in lighting and occlusion. The complementarity of these two at the information level has not been fully utilized, making it difficult to form a unified perception result in complex scenes, thus affecting the accurate positioning of power line targets. For example, in high-voltage power line inspection, when the power line is partially obscured by tree branches, visual information may not be fully presented, and LiDAR data may not be able to accurately distinguish the power line from other objects, ultimately leading to recognition failure. Therefore, how to achieve deep fusion based on multi-source information and ensure high-precision recognition of power line targets in complex environments has become a key problem that this research urgently needs to overcome. Solving this problem is not only about technological breakthroughs, but also directly affects the actual efficiency and safety of power line inspection.

[0218] For the above issues, please refer to Figure 4 This invention proposes a method for inspecting power transmission and distribution lines. Based on the aligned point cloud dataset, corresponding image data is acquired from a synchronized visual camera. To address potential illumination interference in the images, a filtering method is used to process the boundary clarity and pixel gradient values ​​of gray-scale abrupt change regions, resulting in a visual image dataset with interference removed. Feature extraction technology is employed to perform in-depth analysis on the initially matched feature point set, fusing the abrupt change characteristics of the broken strand edge and the contrast intensity between the gray-scale abrupt change and the background to generate a comprehensive feature description vector. Based on the preliminary positioning results, combined with the contrast between the broken strand and the surrounding conductors and the shape features of the gray-scale abrupt change region, a secondary precise positioning is performed on the point cloud data and image data to determine the specific spatial coordinates of the broken conductor strand, thus obtaining the final broken strand location information.

[0219] In summary, this invention addresses the interconnected operational problems in complex scenarios, such as misaligned point cloud data coordinates, blurred boundaries of grayscale abrupt change areas caused by image illumination interference, and inaccurate extraction of broken strand features. These problems collectively restrict the accuracy and efficiency of power line maintenance. This invention collects point cloud data using LiDAR and employs a registration algorithm for coordinate system alignment, initially screening the spatial distribution information of conductor areas. Simultaneously, it filters illumination interference from synchronized visual images, extracts continuity and texture features at grayscale abrupt changes, matches them with the geometric shape and density changes of the point cloud, generates a comprehensive feature description vector, and judges the roughness and stability of the broken strand location through pattern comparison. If the similarity exceeds a threshold, the target is confirmed. Finally, it combines contrast and shape features for secondary precise positioning, thereby solving the problem of accurately identifying and locating broken conductor strands in interference environments. Ultimately, it achieves high-precision automated detection, improving the safety and maintenance efficiency of power infrastructure.

[0220] In this embodiment of the invention, an original point cloud dataset of a complex scenario of a power transmission and distribution line is acquired, and a point cloud registration algorithm is used to align the original point cloud dataset to generate an aligned target point cloud dataset. Gaussian filtering and median filtering methods are used to generate a visual image dataset after filtering out interference based on the aligned target point cloud dataset. A preliminary matching feature point set is generated based on the visual image dataset after filtering out interference and the aligned target point cloud dataset. Deep analysis is performed on the preliminary matching feature point set to generate a comprehensive feature description vector. Pattern comparison is performed on the comprehensive feature description vector to generate a preliminary positioning result. Secondary positioning is performed on the aligned target point cloud dataset and the visual image dataset after filtering out interference based on the preliminary positioning result to generate conductor breakage location inspection information. Based on the above scheme, this invention weakens the influence of environmental interference through Gaussian filtering and median filtering methods, overcomes the bottleneck of visual information loss caused by tree branch occlusion by using multi-source data feature matching and deep analysis, and improves the differentiation accuracy between conductors and surrounding objects such as trees through pattern comparison of comprehensive feature vectors, thereby further improving the inspection effect.

[0221] Please see Figure 5 , Figure 5 This is a structural block diagram of a power transmission and distribution line inspection system provided in Embodiment 2 of the present invention.

[0222] The power transmission and distribution line inspection system provided by this invention includes:

[0223] The acquisition module 501 is used to acquire the original point cloud dataset of the complex scene of the power transmission and distribution line, and to align it with the original point cloud dataset using a point cloud registration algorithm to generate the aligned target point cloud dataset.

[0224] Module 502 is used to generate a visual image dataset after filtering out interference based on the aligned target point cloud dataset using Gaussian filtering and median filtering methods.

[0225] According to module 503, a preliminary matching set of feature points is generated based on the visual image dataset after filtering out interference and the aligned target point cloud dataset.

[0226] Analysis module 504 is used to perform in-depth analysis on the initially matched feature point set and generate a comprehensive feature description vector;

[0227] The comparison module 505 is used to perform pattern comparison on the comprehensive feature description vector and generate preliminary localization results.

[0228] The generation module 506 is used to perform secondary positioning on the aligned target point cloud dataset and the visual image dataset after filtering out interference based on the preliminary positioning results, and generate the conductor strand breakage location inspection information.

[0229] This invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program; when the computer program is executed by the processor, the processor performs the steps of the power transmission and distribution line inspection method as described in any of the above embodiments.

[0230] This invention also provides a computer-readable storage medium storing a computer program / instructions thereon, which, when executed by a processor, implements the steps of the power transmission and distribution line inspection method as described in any of the above embodiments.

[0231] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system and modules described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0232] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for inspecting power transmission and distribution lines, characterized in that, include: The original point cloud dataset of the complex scene of the power transmission and distribution line is obtained, and the point cloud registration algorithm is used to align the original point cloud dataset to generate the aligned target point cloud dataset. Based on the aligned target point cloud dataset, Gaussian filtering and median filtering methods are used to generate a visual image dataset after filtering out interference. Based on the visual image dataset after filtering out interference and the aligned target point cloud dataset, a preliminary matching set of feature points is generated. A deep analysis is performed on the initially matched feature point set to generate a comprehensive feature description vector; The comprehensive feature description vector is subjected to pattern comparison to generate preliminary localization results; Based on the preliminary positioning results, secondary positioning is performed on the aligned target point cloud dataset and the interference-filtered visual image dataset to generate conductor strand breakage location inspection information.

2. The method for inspecting power transmission and distribution lines according to claim 1, characterized in that, The step of aligning the original point cloud dataset using a point cloud registration algorithm to generate an aligned target point cloud dataset includes: The point cloud registration algorithm is used to perform coordinate system alignment on the original point cloud dataset to determine the aligned initial point cloud dataset. The aligned initial point cloud dataset is divided into regions to determine the conductor-related point cloud subset; The broken strand locations are identified in the relevant point cloud subset of the conductor, and multiple broken strand spatial location information are output. Based on the spatial location information of multiple broken strands, local point cloud reconstruction is performed, and the reconstructed local point cloud dataset is output. Perform geometric shape analysis on the reconstructed local point cloud dataset and output the broken strand distribution information; The broken strand distribution information is transformed into latitude and longitude coordinates and labeled with spatial coordinates to output an aligned target point cloud dataset.

3. The method for inspecting power transmission and distribution lines according to claim 1, characterized in that, The step of generating a visual image dataset after filtering out interference based on the aligned target point cloud dataset using Gaussian filtering and median filtering methods includes: Based on the aligned target point cloud dataset, an initial visual image dataset is collected; The Gaussian filtering method is used to filter the initial visual image dataset to output a subset of visual images with enhanced edges. The median filtering method is used to smooth the edge-enhanced visual image subset to generate a smoothed visual image subset. Visual data fusion is performed on the smoothed visual image subset to generate a visual image dataset after filtering out interference.

4. The method for inspecting power transmission and distribution lines according to claim 1, characterized in that, The step of generating a preliminary matching feature point set based on the interference-filtered visual image dataset and the aligned target point cloud dataset includes: Spatial continuity is extracted at gray-level abrupt changes in the visual image dataset after interference is removed, generating a spatially continuous feature subset; Local texture features at locations of abrupt grayscale changes are extracted from the spatially continuous feature subset. Spatial and image feature matching is performed on the local texture features at the gray-level abrupt change and the spatial feature data of the broken strand region point cloud in the aligned target point cloud dataset to output a preliminary set of matched feature points.

5. The method for inspecting power transmission and distribution lines according to claim 1, characterized in that, The step of performing in-depth analysis on the initially matched feature point set to generate a comprehensive feature description vector includes: Based on the initially matched feature point set, feature enhancement is performed, and an enhanced feature subset is output. Based on the enhanced feature subset, calculate the geometric morphology analysis input data; The density change of the geometric morphology analysis input data is monitored, and an extended matching set is output. The extended matching set and the geometric morphology analysis input data are fused to generate a comprehensive feature description vector.

6. The method for inspecting power transmission and distribution lines according to claim 1, characterized in that, The step of performing pattern comparison on the comprehensive feature description vector to generate preliminary localization results includes: Perform pattern comparison on the comprehensive feature description vector and output stability features; Based on the aforementioned stability characteristics, the target region is determined; Extract the conductor strand breakage confirmation coordinate points in the target area, and perform edge sharpening processing on the conductor strand breakage confirmation coordinate points to determine the preliminary positioning coordinate set; Based on the preliminary positioning coordinate set, the broken strand is monitored, and a preliminary positioning result is generated.

7. The method for inspecting power transmission and distribution lines according to claim 1, characterized in that, The step of performing secondary localization on the aligned target point cloud dataset and the interference-filtered visual image dataset based on the preliminary localization results to generate conductor breakage location inspection information includes: Based on the preliminary positioning results and the aligned target point cloud dataset, a set of boundary ranges is determined. Based on the set of boundary ranges, perform visual representation matching of the conductors and output a set of coordinate points for secondary positioning; Spatial coordinate mapping is performed on the set of coordinate points of the secondary positioning to generate an enhanced description vector; Based on the enhanced description vector, a distribution feature map is generated; The final information integration point is extracted from the distribution feature map, and the feature point is aggregated and verified to generate the conductor strand breakage location inspection information.

8. A power transmission and distribution line inspection system, characterized in that, include: The acquisition module is used to acquire the original point cloud dataset of complex scenarios of power transmission and distribution lines, and to align the original point cloud dataset using a point cloud registration algorithm to generate the aligned target point cloud dataset. A module is used to generate a visual image dataset after filtering out interference based on the aligned target point cloud dataset using Gaussian filtering and median filtering methods. According to the module, it is used to generate a preliminary matching set of feature points based on the visual image dataset after filtering out interference and the aligned target point cloud dataset; The analysis module is used to perform in-depth analysis on the initially matched feature point set and generate a comprehensive feature description vector; The comparison module is used to perform pattern comparison on the comprehensive feature description vector and generate preliminary localization results; The generation module is used to perform secondary positioning on the aligned target point cloud dataset and the interference-filtered visual image dataset based on the preliminary positioning results, and generate conductor strand breakage location inspection information.

9. An electronic device, characterized in that, The method includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor causes the processor to perform the steps of the power transmission and distribution line inspection method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the power transmission and distribution line inspection method as described in any one of claims 1-7.

Images