Overhead distribution line topology mapping method based on unmanned aerial vehicle inspection image density clustering

Abstract

The invention discloses an overhead distribution line topology mapping method and system based on unmanned aerial vehicle inspection image density clustering, and the method comprises the steps: shooting a visible light image of an overhead distribution line through an unmanned aerial vehicle; extracting image feature information of the visible light image, including detecting tower equipment and defect hidden dangers of the visible light image, and extracting coordinate position information of the visible light image; based on the image feature information of the visible light image, determining a density center point and a neighborhood point of the visible light image, and constructing a density cluster of spatial distribution of the visible light image; based on density clustering of spatial distribution of visible light images, mapping of overhead distribution line equipment and defects and a line topological structure is realized, and latest inspection data is obtained; according to the method, an original standing book is compared with latest inspection data, coordinate position deviation and equipment corresponding conditions are calculated, the original standing book is compared with a newly constructed standing book according to the type and number of equipment, and inconsistency is recognized and reminded.

Description

Technical Field

[0001] The present invention relates to the technical field of overhead distribution lines, and more specifically, to a method and system for topological mapping of overhead distribution lines based on density clustering of unmanned aerial vehicle inspection images. Background Art

[0002] With the continuous growth of the national economy, the scale of power grid construction has continued to expand, and the refined management of overhead distribution lines has become a key link in the operation of power systems. As an important basis for power grid management, the overhead distribution line ledger plays an indispensable role in line inspection, asset management, fault location, and line relocation. However, facing millions of kilometers of medium-voltage overhead distribution lines, the traditional ledger management method can no longer meet the needs of modern power grids.

[0003] Drone technology has been widely used in power grid operation and maintenance due to its high efficiency and convenience. Drones can quickly collect a large amount of image data of distribution towers, which is of great value for line status assessment and maintenance. However, there are also some problems in practical applications: drone operations often rely on ledgers to carry out work, and the existing distribution line ledgers are mostly completed by manual entry, involving a large amount of material information. If they are not strongly associated with specific towers, it is easy to cause confusion in material management and increase the difficulty of later maintenance. In addition, negligence or misunderstanding during manual entry will lead to inaccurate ledger information. After line changes or equipment updates, the ledger update lags and cannot reflect the latest line status in time. During the inspection process, the images taken need to be named according to the rules and correspond to the tower information in the ledger, which increases the complexity of the operation. Once the naming is wrong, it will affect the subsequent data analysis and fault location. In order to ensure that the image is consistent with the ledger, the inspection personnel need to limit the number of images taken each time, which limits the comprehensiveness and flexibility of information collection. The complex operating procedures require inspectors to have high professional skills, which increases the possibility of misoperation. During the drone shooting process, if problems such as blurred images or improper angles occur, the error correction process is complicated and time-consuming, reducing inspection efficiency. Summary of the invention

[0004] According to the present invention, a method and system for overhead distribution line topology mapping based on density clustering of drone inspection images are provided to solve the technical problems existing in existing drones in power grid operation and maintenance detection.

[0005] According to a first aspect of the present invention, there is provided a method for topological mapping of overhead distribution lines based on density clustering of drone inspection images, comprising:

[0006] Visible light images of overhead distribution lines captured by drones;

[0007] Extracting image feature information of the visible light image, including detecting equipment and defects on the tower in the visible light image, and extracting coordinate position information of the visible light image;

[0008] Based on the image feature information of the visible light image, determining the density center point and neighborhood points of the visible light image, and constructing density clustering of the visible light image in spatial distribution;

[0009] Based on the density clustering of visible light images in space, the mapping of overhead distribution line equipment and defects with line topology is achieved to obtain the latest inspection data;

[0010] By comparing the original ledger with the latest inspection data, calculating the coordinate position deviation and equipment correspondence, comparing the original ledger with the newly constructed ledger based on the equipment type and quantity, inconsistencies are identified and reminded.

[0011] Optionally, detecting equipment and hidden defects on the tower in the visible light image includes:

[0012] Trained on an object detection algorithm, a single neural network predicts bounding boxes and class probabilities in visible light images.

[0013] For tower equipment, key power equipment such as insulators, lightning arresters, primary and secondary fusion switches, circuit breakers, grounding switches, transformers and fusion terminals are identified and tested;

[0014] For hidden defects, we use the defect sample library to train common defects such as hanging foreign objects, bird nest construction, insulation damage, tower top damage, transformer oil leakage, etc. to identify and locate targets and defects;

[0015] The labels of the tower equipment or defect recognition results in the inspection image are associated with the corresponding images, which are used to associate the recognition results with the ledger information in the equipment or defect mapping module.

[0016] Optionally, extracting coordinate position information of the visible light image includes:

[0017] Extracting the exchangeable image file format in the visible light image data attribute file, and converting it into coordinate points in a rectangular coordinate system to obtain the longitude and latitude coordinate information of the tower;

[0018] Convert the extracted tower latitude and longitude coordinates into decimal format. The conversion formula is:

[0019]

[0020] Convert the longitude and latitude coordinates (B, L, H) to points in the spatial rectangular coordinate system (X, Y, Z), and use the Xi'an 80 coordinate system as the ellipsoid reference, where the major semi-axis a of the ellipsoid is 6378140±5 meters, the minor semi-axis b is 6356755.2882 meters, and the radius of curvature of the circle passing through the longitude and latitude coordinates is where e 2 is the second eccentricity of the ellipsoid, specifically

[0021] Therefore, the longitude and latitude coordinates (B, L, H) are converted to the spatial rectangular coordinate system (X, Y, Z)

[0022] X=(N+H)×cosB×cosL

[0023] Y=(N+H)×cosB×sinL

[0024] Z=(N×(1-e 2 )+H)×sinB.

[0025] Optionally, based on the image feature information of the visible light image, determining the density center point and neighborhood points of the visible light image, and constructing density clustering of the visible light image in spatial distribution, including:

[0026] Based on the image feature information of the visible light image, the tower head is identified, and when the center of the tower head is located in the center of the frame, it is the auxiliary clustering mark point, and the auxiliary clustering mark point of the visible light image is used as the density center point;

[0027] By calculating the distance between each point in the visible light image, the distance matrix D is represented by a matrix, where the rows and columns of the matrix represent different image positions, and the values ​​of the rows and columns are D i，j Indicates the distance between bright spots, and the main diagonal element is 0; when the density center point uses the tower head in the image recognition result, its main diagonal element is set to 1, otherwise it is set to 0;

[0028] Perform logical judgment on the elements of the matrix, construct the connection clue matrix C, calculate the connection clue matrix C, and judge whether each point is a neighborhood point and a density center point. Where n represents the total number of points in the image, S i It represents the number of points directly connected to point i, that is, the point density. The sum of each row S i Add 1 to get the adjusted point density S′ i =S i +1, that is, even if a point has no other directly connected points, it will be regarded as a member of its own neighborhood. For each point i, if the point density S′ i Greater than the density threshold Min p, then the point is considered to be the density center point. If the point density S′ i Less than the density threshold Min p But if it is greater than 1, then the point is a neighborhood point;

[0029] When there are multiple density center points, these density center points and their neighborhood points are integrated into a connected network based on the connection relationship between the points, forming a comprehensive set of all relevant shooting points to construct the tower drone inspection image clustering cluster;

[0030] Cascading clustering is performed on the unclustered density center points to ensure that each density center point and its neighborhood can be organized into a logically consistent and non-overlapping cluster set.

[0031] Optionally, based on the density clustering of the visible light image in space distribution, the mapping of overhead distribution line equipment and defects with line topology is achieved, including:

[0032] Based on the density clustering of visible light images in space, determine the effective shooting point set for each tower;

[0033] According to the spatial relationship between the effective shooting points, they are sorted and stored according to the nearest neighbor principle to obtain the tower clustering result;

[0034] The results of equipment identification or defect detection are associated with the tower clustering results to construct the topological structure of the line.

[0035] According to another aspect of the present invention, there is also provided an overhead distribution line topology mapping system based on drone inspection image density clustering, comprising:

[0036] A visible light image shooting module is used to shoot visible light images of overhead power distribution lines through a drone;

[0037] An image feature information extraction module is used to extract image feature information of the visible light image, including detecting equipment and defects on the tower of the visible light image, and extracting coordinate position information of the visible light image;

[0038] Constructing a density clustering module, which is used to determine the density center point and neighborhood points of the visible light image based on the image feature information of the visible light image, and construct density clustering of the visible light image in spatial distribution;

[0039] Implement a topological mapping module to map overhead power distribution line equipment and defects to line topology based on the density clustering of visible light images in space, and obtain the latest inspection data;

[0040] The comparison and identification module is used to compare the original ledger with the latest inspection data, calculate the coordinate position deviation and equipment correspondence, compare the original ledger with the newly constructed ledger based on the equipment type and quantity, and identify and alert inconsistencies.

[0041] Optionally, the module for extracting image feature information includes:

[0042] The predicted bounding box and class probability submodule is used to predict bounding boxes and class probabilities in visible light images through a single neural network trained based on the object detection algorithm;

[0043] The identification and detection equipment submodule is used to identify and detect key power equipment on the tower, including insulators, lightning arresters, primary and secondary fusion switches, circuit breakers, grounding switches, transformers and fusion terminals;

[0044] The defect identification and detection submodule is used to identify and locate targets and defects for potential defects such as hanging foreign objects, bird nest construction, insulation damage, tower top damage, transformer oil leakage, etc. It relies on the defect sample library for training;

[0045] The associated recognition result submodule is used to associate the labels of the tower equipment or defect recognition results in the inspection image with the corresponding image, and is used to associate the recognition results with the ledger information in the equipment or defect mapping module.

[0046] Optionally, the module for extracting image feature information includes:

[0047] The submodule for obtaining the longitude and latitude coordinate information of the tower is used to extract the exchangeable image file format in the visible light image data attribute file and convert it into coordinate points in a rectangular coordinate system to obtain the longitude and latitude coordinate information of the tower;

[0048] The submodule for converting decimal point format is used to convert the extracted tower latitude and longitude coordinate minute and second format into decimal point format. The conversion formula is:

[0049]

[0050] Convert the longitude and latitude coordinates (B, L, H) to points in the spatial rectangular coordinate system (X, Y, Z), and use the Xi'an 80 coordinate system as the ellipsoid reference, where the major semi-axis a of the ellipsoid is 6378140±5 meters, the minor semi-axis b is 6356755.2882 meters, and the radius of curvature of the circle passing through the longitude and latitude coordinates is where e 2 is the second eccentricity of the ellipsoid, specifically

[0051] The conversion module is used to convert the longitude and latitude coordinates (B, L, H) into the spatial rectangular coordinate system (X, Y, Z).

[0052] X=(N+H)×cosB×cosL

[0053] Y=(N+H)×cosB×sinL

[0054] Z=(N×(1-e 2 )+H)×sinB.

[0055] Optionally, build a density clustering module, including:

[0056] The density center point marking submodule is used to identify the tower head based on the image feature information of the visible light image, and when the center of the tower head is located in the center of the frame, it is an auxiliary clustering marking point, and the auxiliary clustering marking point of the visible light image is used as the density center point;

[0057] Determine the distance matrix submodule, which is used to calculate the distance between each point in the visible light image, and express it through a matrix, that is, the distance matrix D, where the rows and columns of the matrix represent different image positions, and the values ​​of the rows and columns are D i，j Indicates the distance between bright spots, and the main diagonal element is 0; when the density center point uses the tower head in the image recognition result, its main diagonal element is set to 1, otherwise it is set to 0;

[0058] Construct a connection clue matrix submodule, which is used to perform logical judgment on the elements of the distance matrix, construct a connection clue matrix C, calculate the connection clue matrix C, and determine whether each point is a neighborhood point and a density center point. Where n represents the total number of points in the image, S i It represents the number of points directly connected to point i, that is, the point density. The sum of each row S i Add 1 to get the adjusted point density S′ i =S i +1, that is, even if a point has no other directly connected points, it will be regarded as a member of its own neighborhood. For each point i, if the point density S′ i Greater than the density threshold Min p , then the point is considered to be the density center point. If the point density S′ i Less than the density threshold Min p But if it is greater than 1, then the point is a neighborhood point;

[0059] The image clustering submodule is used to integrate these density center points and their neighborhood points into a connected network when there are multiple density center points, based on the connection relationship between the points, to form a comprehensive set of all relevant shooting points and build the tower drone inspection image clustering cluster;

[0060] The clustering set organization submodule is used to perform cascade clustering on the unclustered density center points, ensuring that each density center point and its neighborhood can be organized into a logically consistent and non-overlapping cluster set.

[0061] Optionally, a topology mapping module is implemented, including:

[0062] Implement a topology mapping submodule for mapping overhead distribution line equipment and defects to line topology based on the density clustering of visible light images in space, including:

[0063] A submodule for determining a set of effective shooting points is used to determine a set of effective shooting points for each tower based on density clustering of visible light images in space distribution;

[0064] The tower clustering result obtaining submodule is used to sort and store the effective shooting points according to the spatial relationship between them and the nearest neighbor principle to obtain the tower clustering result;

[0065] The line topology structure submodule is used to associate the results of equipment identification or defect detection with the tower clustering results to construct the line topology structure method.

[0066] Therefore, by clustering the density of overhead distribution line images taken by drones, the mapping relationship between overhead distribution line equipment and defects and line topology is established, and the automatic classification of unsupervised and manually guided inspection images, the construction of new line topology, and the verification and correction of the original line topology are realized. The precise GNSS navigation and positioning coordinate information carried in the drone images is combined with the rich tower information in the images to reconstruct and correct the topology of the line through the density clustering algorithm. This method can not only reduce the workload of manual input and improve the accuracy of ledger information, but also enhance the level of autonomy of line ledger management, providing strong technical support for the refined management and maintenance of power grids. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] A more complete understanding of exemplary embodiments of the present invention may be obtained by referring to the following drawings:

[0068] Figure 1 A schematic diagram of an overhead distribution line topology mapping method based on density clustering of drone inspection images described in this embodiment;

[0069] Figure 2This is a schematic diagram of the basic principle of density clustering described in this implementation mode;

[0070] Figure 3 This is a schematic diagram of image acquisition for drone inspection of overhead power distribution lines according to this embodiment;

[0071] Figure 4 A schematic diagram of the technical route described in this implementation method;

[0072] Figure 5 It is a schematic diagram of the YOLO-based device and defect target detection principle framework described in this embodiment;

[0073] Figure 6 This is a schematic diagram of a typical material identification result described in this embodiment;

[0074] Figure 7 This is a schematic diagram of tower head target detection described in this implementation mode;

[0075] Figure 8 is a schematic diagram of the EXIFGPS coordinates of the image described in this embodiment;

[0076] Fig. 9 This is a schematic diagram of the principle of the density clustering algorithm based on drone inspection images described in this embodiment;

[0077] Fig.10 Schematic diagram of the calculation matrix conversion process described in this embodiment;

[0078] Fig.11 A schematic diagram of equipment or defect clustering and topology mapping according to this embodiment;

[0079] Fig.12 is a schematic diagram of an unnamed image according to this embodiment;

[0080] Fig.13 A topological diagram of the clustering result described in this implementation mode;

[0081] Fig.14 This is a schematic diagram of an overhead distribution line topology mapping system based on density clustering of drone inspection images described in this embodiment. DETAILED DESCRIPTION

[0082] Now, exemplary embodiments of the present invention are described with reference to the accompanying drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. These embodiments are provided to disclose the present invention in detail and completely and to fully convey the scope of the present invention to those skilled in the art. The terms used in the exemplary embodiments shown in the accompanying drawings are not intended to limit the present invention. In the accompanying drawings, the same units / elements are marked with the same reference numerals.

[0083] Unless otherwise specified, the terms (including technical terms) used herein have the commonly understood meanings to those skilled in the art. In addition, it is understood that the terms defined in commonly used dictionaries should be understood to have the same meanings as those in the context of the relevant fields, and should not be understood as idealized or overly formal meanings.

[0084] According to a first aspect of the present invention, a method 100 for topological mapping of overhead distribution lines based on density clustering of drone inspection images is provided, referring to Figure 1 As shown, the method 100 includes:

[0085] S101: photographing a visible light image of an overhead power distribution line using a drone;

[0086] S102: extracting image feature information of the visible light image, including detecting equipment and defects on the tower in the visible light image, and extracting coordinate position information of the visible light image;

[0087] S103: Based on the image feature information of the visible light image, determine the density center point and neighborhood points of the visible light image, and construct density clustering of the visible light image in spatial distribution;

[0088] S104: Based on the density clustering of the visible light image in the spatial distribution, the overhead distribution line equipment and defects are mapped with the line topology structure to obtain the latest inspection data;

[0089] S105: By comparing the original ledger with the latest inspection data, calculating the coordinate position deviation and equipment correspondence, comparing the original ledger with the newly constructed ledger based on the equipment type and quantity, identifying and reminding of inconsistencies.

[0090] Specifically, during the drone inspection of overhead distribution lines, the drone collects visible light images in accordance with relevant regulations and specifications, with the pole tower as the center. Each captured image will record the GNSS (Global Navigation Satellite System) coordinates of the shooting point in its EXIF ​​(Exchangeable Image File Format) data. These coordinates form a specific distribution pattern in space with the pole tower as the center.

[0091] In order to quantify this spatial distribution feature, the density reachable distance R can be used to identify it, which is defined as the Euclidean distance between two shooting points. According to the characteristics of the distribution relationship, the shooting points can be divided into three types, such as Figure 2 As shown, it specifically includes density center points, neighborhood points and noise points.

[0092] When the density around the shooting point is within the range of distance R, the number of shooting points is greater than the density threshold Min pWhen , the point is considered to be the density center, that is, it satisfies the following relationship:

[0093] |N R (P center )|≥Min p

[0094] Where P center is the density center point, N R (P center ) is the set of all shooting points within the radius R, including the density center point, with the density center point as the center, |N R (P center )| is the density center point P center The number of all captured points within the neighborhood.

[0095] When the density around the shooting point is within the range of distance R, the number of shooting points is less than the density threshold Min p , but it is not the only point, the point is considered to be a neighborhood point, that is, it satisfies the following relationship

[0096] Min p ≥|N R (P field )|≥2

[0097] Where P field is the neighborhood point, N R (P field ) is the set of all shooting points including the neighborhood point within the radius R with the neighborhood point as the center, |N R (P field )| is the neighborhood point P field The number of all captured points within the neighborhood.

[0098] When there are no other shooting points within the range of density R around the shooting point, the isolated point is considered to be a noise point.

[0099] According to the distance and connection relationship between the density center point and the neighborhood point, a connection clue matrix between each density center point and the neighborhood point is established. According to the connection clue matrix, a set of inspection images corresponding to the connection relationship matrix are deduplicated to form an inspection image cluster, which is the tower drone inspection image clustering cluster.

[0100] During the drone shooting process, the tower or channel is photographed according to the needs of on-site operations, such as Figure 3 As shown, to ensure the accuracy and practicality of image acquisition, there is no need to impose too many restrictions on the number of images taken or the naming of images. The relevant standards and specifications should be followed during the shooting process, and the distance should not be too large to ensure the clarity and detail quality of the image.

[0101] Specifically, long-distance or ultra-long-distance zoom shooting should be avoided as much as possible, as this shooting method may cause image blur or loss of details. At the same time, during the shooting process, the drone will record the GNSS (Global Navigation Satellite System) coordinate values ​​of each shooting point, which are stored in the EXIF ​​file of the image for subsequent analysis and processing.

[0102] By identifying the equipment and defects on the towers in the overhead distribution line images taken by drones, combined with the coordinate position information of the visible light image shooting points, and based on the density clustering of the drone inspection images in space, the mapping of overhead distribution line equipment and defects with line topology is realized, and the automatic classification of unsupervised and manually guided inspection images, the construction of new line topology and the verification and correction of the original line topology are realized. In terms of technical implementation, the following technical modules are included, such as Figure 4 shown.

[0103] The YOLO (You Only Look Once) algorithm is used for training. This algorithm uses a single neural network to simultaneously predict the bounding box and category probability in the image to achieve efficient and accurate target detection. The algorithm architecture principle is as follows Figure 5 shown.

[0104] In terms of material identification, the focus is on identifying key power equipment such as insulators, lightning arresters, primary and secondary fusion switches, circuit breakers, grounding switches, transformers and fusion terminals. The specific identification results are as follows: Figure 6 shown.

[0105] In particular, in order to enhance the clustering accuracy of the algorithm clustering module, the present invention identifies and detects the tower head as a key detection target, and the recognition effect is as follows: Figure 7 shown.

[0106] In terms of defect identification, the focus is on common defects such as hanging foreign objects, bird's nest construction, insulation damage, tower top damage, transformer oil leakage, etc., and training is carried out based on a million-level defect sample library. It can accurately identify and locate these targets and defects, significantly improving the efficiency and accuracy of drone inspections.

[0107] Through structured data processing, the labels of material targets or defect recognition results in the inspection images are associated with the corresponding images, which are used to associate the recognition results with the ledger information in the equipment or defect mapping module.

[0108] The EXIF ​​in the attribute file of the image data of the drone inspection is extracted and converted into coordinate points in the rectangular coordinate system. The tower coordinate information of the inspection image taken by the drone is as follows: Figure 8 As shown, the extracted tower latitude and longitude coordinates are converted from minute and second format to decimal format. The conversion formula is:

[0109]

[0110] In order to calculate the distance between bright spots, the longitude and latitude coordinates (B, L, H) are converted to points in the spatial rectangular coordinate system (X, Y, Z), and the Xi'an 80 coordinate system is used as the ellipsoid reference, where the major semi-axis a of the ellipsoid is 6378140±5 meters, the minor semi-axis b is 6356755.2882 meters, and the radius of curvature of the circle passing through the longitude and latitude coordinates is where e 2 is the second eccentricity of the ellipsoid, specifically

[0111] Therefore, the longitude and latitude coordinates (B, L, H) are converted to the spatial rectangular coordinate system (X, Y, Z)

[0112] X=(N+H)cosB×cosL

[0113] Y=(N+H)×cosB×sinL

[0114] Z=(N×(1-e 2 )+H)×sinB

[0115] Based on the basic principle of density clustering, the improved application is carried out according to the operation characteristics of the overhead distribution line drone inspection to form a cluster analysis module. The technical schematic diagram of this module is shown in the figure. Fig. 9 shown.

[0116] The density-reachable distance R is determined based on the standard operating specifications for drone inspection operations and combined with the average span of line towers. The specific value is generally set between 5 and 20 meters. When the distance is less than 5 meters, it is difficult to establish an effective connection between the shooting points, and it is easy to form multiple separate clusters, resulting in complex and inaccurate data processing; when the distance is greater than 20 meters, since the average span of the towers is usually between 40 and 70 meters, and even reaches 30 meters in some areas, coupled with the shooting points between the two spans during channel inspection, it is difficult to accurately distinguish and cluster the towers. In order to ensure the continuity of image acquisition and the accuracy of clustering, the density-reachable distance R is set to 10 meters in the present invention. This setting can not only ensure the effective connection between the shooting points and avoid the formation of too many separate clusters, but also ensure accurate identification and distinction of the towers at different spans.

[0117] Density threshold Min pThe determination is also based on the standard operating specifications of drone inspection operations. In daily inspections, the number of shots of straight towers is relatively small, and usually only images need to be collected at the top and both sides of the tower. For complex tower types such as corner towers, tension towers, and terminal towers, especially double-circuit lines on the same tower, it is often necessary to take 6 or even more images to ensure full coverage. When inspecting a channel, it is generally only necessary to take 2 images on each side of the line. Based on the above inspection requirements and actual operating experience, the present invention sets the density threshold Min p Set to 3. This setting ensures that images taken under different tower types and inspection scenarios can effectively establish connections and form reasonable clusters. This setting can ensure the consistency and integrity of image data, improve the efficiency and accuracy of inspection work, and avoid the complexity of data processing caused by too many or too few shooting points.

[0118] The density center point, neighborhood point and noise point are determined according to the basic principle of the algorithm in Section 4.1. Fig. 9 As shown, ①, ②, ③, ⑤ and ⑥ are density center points, ④ and ⑧ are neighborhood points, and ⑦ is a noise point. In the application of the present invention, a data processing flow combining image feature information during power grid inspection is designed, that is, through the equipment or defect detection module such as Figure 7 As shown in , the tower head is identified, and when the center of the tower head is located in the center of the picture, it is the auxiliary clustering mark point, such as Fig. 9 As shown in point ① in the figure, the image is taken as the density center point.

[0119] In particular, when the coordinates of the auxiliary clustering marker point of the image with the tower head are used as the density center point, the density can be expanded to a distance R, and the coordinates of this point as the density center point are extended to 20m.

[0120] By calculating the distance between each point in the image, the distance matrix D is represented by a matrix, such as Fig.10 As shown. The rows and columns of the matrix represent different image positions, and the values ​​of the rows and columns are D i，j It represents the distance between bright spots, so the main diagonal elements are 0; when the density center point uses the tower head marker in the image recognition result, its main diagonal elements are set to 1, otherwise they are set to 0.

[0121] The distance matrix D is used to quantify the relative position relationship between bright spots at different positions. The rows and columns represent different image shooting points, and its element value D i，j Represents the distance between the i-th point and the j-th point, such as Fig.10As shown in the figure. Since the distance from any point to itself is zero, the main diagonal elements of the matrix are usually set to 0. When the coordinates of the auxiliary clustering marker points of the image with tower heads are used as the density center points, the corresponding main diagonal elements in the distance matrix D are set to 1. For other common positions that are not designated as density center points, their main diagonal elements are kept unchanged at 0.

[0122] At the same time, the elements of the matrix are logically judged to construct the connection clue matrix C. For each point, different density reachable distances are set according to whether it is marked as a tower head (that is, whether it is used as an auxiliary clustering marker point). If a point is an auxiliary clustering marker point determined by image recognition results, the main diagonal element corresponding to the point is set to 1 in the distance matrix D, and its density reachable distance R is set to 20 meters; for all other points (points with main diagonal elements of 0), their density reachable distance R is set to 10 meters at the same time. If D i，j ≤R, that is, the distance between the two points is less than or equal to the density reachable distance, then C i，j Set to 1, indicating that there is a direct connection between the two points. Otherwise, D i，j >R, then C i，j Setting it to 0 means that there is no direct connection between the two points and they are not within the direct reach of each other. Since the connection values ​​between most non-adjacent points are 0, the connection clue matrix C is a sparse matrix.

[0123] The connection clue matrix C is calculated to determine whether each point is a neighborhood point and a density center point. Where n represents the total number of points in the image, S i represents the number of points directly connected to point i, that is, the point density. In order to ensure that each point includes at least itself as one of the neighborhood members, we sum the result of each row S i Add 1, such as Fig.10 As shown, the adjusted point density is S′ i =S i +1, that is, even if a point has no other directly connected points, it will be considered as a member of its own neighborhood. For each point i, according to the density clustering principle in 4.1, if the point density S′ i Greater than the density threshold Min p , then the point is considered to be the density center point. If the point density S′ i Less than the density threshold Min p But if it is greater than 1, then the point is a neighborhood point.

[0124] The chain structure is used to construct the tower drone inspection image clustering cluster. When there are multiple density center points, these center points and their neighborhood points are integrated into a connected network according to the connection relationship between each point, forming a comprehensive set containing all relevant shooting points, ensuring that each point in the set is regarded as a valid shooting point of the tower. First, the search starts from the auxiliary clustering mark point (that is, the point where the main diagonal element of the distance matrix is ​​1) as the starting point, and the directly related points are found in its row to build the initial clustering cluster. Then, for each point in the clustering cluster, other element points are searched in its corresponding row, and the existing clustering points are checked for duplicates during the search process, and the newly discovered and unincluded points are added to the current clustering cluster. This process is extended step by step using a chain structure, and a deduplication check is performed after each new point is added to ensure that each point is only classified into one cluster. When no new points are added to a clustering cluster, the construction of the clustering cluster is completed. In this way, a set of tower shooting points with the density center point as the core can be effectively formed to ensure that all relevant shooting points are accurately included in the corresponding clusters.

[0125] When the clusters of all density center points are built, if the distance between the neighborhood points is close, these neighborhood points can be further combined into image clusters. However, image clusters composed only of neighborhood points cannot be directly used in the line topology correction module because they lack clear density center points as references. Such clusters need to be manually confirmed before they can be used in equipment or defect mapping modules. Through manual confirmation, the validity and accuracy of these neighborhood points can be ensured. This not only improves the flexibility of data processing, but also ensures the accuracy and practicality of the final results.

[0126] After completing the cluster indexing for the auxiliary clustering marker points (i.e., the points whose main diagonal elements are 1), the other density center points that have not been clustered are then cascade clustered. The specific method is the same as that for processing the auxiliary clustering marker points: starting from each density center point that has not been clustered, directly related points are retrieved in its row to build a new clustering cluster. Then, for each point in the newly formed clustering cluster, other related points are continued to be found in its corresponding row, and duplicates are checked to avoid repeated inclusion of existing clustering points. Through this chain expansion method, new points are added step by step and duplicates are removed until there are no new points, at which point the clustering cluster is completed. This process will continue until all density center points and their related neighborhood points are classified into the corresponding clustering clusters, thereby fully completing the clustering of all density center points. This method ensures that each density center point and its neighborhood can be effectively organized into a logically consistent and non-overlapping cluster set.

[0127] like Fig. 9As shown in the figure, the first tower UAV inspection image cluster formed is a point cluster composed of the density center point ① and neighborhood points ②③④⑤⑥⑧ within the envelope.

[0128] In the calculation process, due to the large number of daily shooting points, in order to ensure the calculation efficiency, the block matrix calculation method can be used. Specifically, the data is processed in sections according to the shooting time, and the matrix size of each block does not exceed 10000×10000. This not only reduces the amount of data calculated in a single time and improves the processing speed, but also effectively manages memory usage. For each block, cluster indexing and cascade clustering are performed according to the method described previously, that is, starting from the auxiliary clustering marker point to build the initial cluster cluster, and then gradually expanding to other density center points and their neighborhood points until the clustering within the entire block is completed. Through this block processing strategy, it is possible to efficiently process large-scale data sets and ensure the accuracy and completeness of the clustering results.

[0129] The overall distribution of the lines is radial. In order to optimize data management and analysis, we use a cluster analysis module to process the calculated coordinate points and store these points by sorting by nearest connection. Specifically, cluster analysis is first used to determine the set of valid shooting points for each tower. Then, according to the spatial relationship between these points, they are sorted and stored according to the nearest neighbor principle. In addition, the results of the equipment identification or defect detection module are associated with the tower clustering results to construct the topological structure of the line. This not only clearly reflects the actual layout of the line, but also facilitates subsequent maintenance and management, ensuring that all key information can be presented in an intuitive and orderly manner, providing strong support for the overall monitoring and troubleshooting of the line. Fig.11 shown.

[0130] The line topology correction module is mainly used for the verification and correction of existing ledgers. By comparing the original ledger with the latest inspection data, the coordinate position deviation and the corresponding equipment are calculated. For the coordinate position of the tower, if the deviation is found to be more than 20 meters, the tower will be marked and manual secondary confirmation will be required; for the tower with a deviation of less than 20 meters, the coordinate position of this inspection will be recorded, and its position information will be gradually corrected through multiple iterations. In addition, for the type and quantity of equipment, especially those key equipment with unique quantity and must be photographed during inspection (such as primary and secondary fusion switches, circuit breakers, grounding switches, transformers and fusion terminals, etc.), the module will compare the original ledger with the newly constructed ledger to identify and remind any inconsistencies. These inconsistencies need to be corrected through on-site confirmation or other means to ensure the accuracy and timeliness of the ledger information. Through this series of automated comparison and manual confirmation processes, the quality of ledger management can be effectively improved, providing reliable data support for subsequent operation and maintenance work.

[0131] Therefore, during the line acceptance stage, the coordinate position, equipment type and quantity, and installation and construction process defects can be automatically identified and found without manually entering any ledger information. Through advanced image recognition technology and data analysis algorithms, the system can accurately detect the position of each tower and compare it with the design drawings to ensure that it meets the planning requirements. At the same time, the module can also accurately identify the type and quantity of key equipment (such as primary and secondary fusion switches, circuit breakers, grounding switches, transformers, and fusion terminals, etc.) and check them with the design specifications. In addition, the system can also detect potential defects in the construction process, such as irregular installation or equipment damage, and automatically generate detailed reports. This automated and high-precision acceptance process not only greatly simplifies the workload of completion acceptance, but also significantly improves the accuracy and efficiency of acceptance, providing a solid foundation for subsequent operation and maintenance management.

[0132] In daily line inspections, the efficiency and convenience of image management can be significantly improved. Through automated cluster analysis and nearest connection sorting, inspectors only need to upload the captured images, and the system can automatically classify them into the corresponding towers and equipment and record their coordinate information. This not only simplifies image management and retrieval, but also ensures that the data of each inspection can be accurately archived and stored. At the same time, these ordered image data also provide an intuitive reference for subsequent troubleshooting and maintenance, improving overall work efficiency. Inspectors no longer need to do tedious manual renaming and classification work, and can focus more on actual inspection tasks, thereby improving the quality and efficiency of inspections.

[0133] It can effectively improve the accuracy and consistency of ledger data. By comparing the latest inspection data with the existing ledger, the system can automatically calculate the deviation of the tower coordinate position, mark the tower with large deviation, and remind relevant personnel to conduct on-site confirmation and correction. For key equipment, the system will also compare the number and type of equipment, identify and prompt any inconsistencies. This automated correction process not only reduces the errors caused by manual operation, but also greatly improves the update speed and accuracy of ledger data. Through multiple iterations and continuous corrections, it is ultimately ensured that the ledger information is completely consistent with the actual line conditions, providing reliable data support for daily operation and management.

[0134] A drone inspection was carried out on the Chengnan 10 line of an overhead power distribution line in a certain place, and the collected images were randomly named in batches as follows: Fig.12 As shown, the image naming does not reflect the tower and ledger information. The method proposed in the present invention is used for processing, and the image EXIF ​​image coordinate information is extracted.

[0135] Clustering is performed using the clustering analysis module. The clustering results are shown in Table 1. There are 15 towers in total. The topological diagram of the clustering results is shown in Fig.13 shown.

[0136] Table 1 Tower coordinate clustering results

[0137]

[0138]

[0139] The distance deviation between the clustering results and the true results was analyzed and calculated, and compared with the manually collected coordinate values. The average error was 3.77 meters and the maximum error was 8 meters. The calculation results are shown in Table 2.

[0140] Table 2 Clustering deviation table

[0141]

[0142]

[0143] The acceptance topology building module is used to calculate the distance between adjacent towers of the line, that is, the line tower span and total length are 616.37 meters, see Table 3.

[0144] Table 3 Clustering span results

[0145] distance The distance between the first and second towers 44.20435 The distance between the 2nd and 3rd towers 61.81463 The 3rd to 4th tower spacing 34.94769 The 4th to 5th tower spacing 35.78711 The 5th to 6th tower spacing 43.44169 The 6th to 7th tower spacing 40.68049 The 7th to 8th tower spacing 49.74881 The 8th to 9th tower spacing 47.6908 The 9th to 10th tower spacing 37.60404 The 10th to 11th tower spacing 39.21713 The span between the 11th and 12th towers 55.49309 The 12th to 13th tower spacing 20.52624 The 13th to 14th tower spacing 55.52366 The 14th to 14th tower spacing 49.69508 Total line length 616.3748

[0146] The defective equipment detection module and defect mapping module are used to analyze and identify image defects. The defects found in this line inspection are broken insulation layer of insulated wire and loose fixation (tilt), see Table 4.

[0147] Table 4 Defect identification result report

[0148]

[0149] The equipment detection module and equipment mapping module are used to obtain the ledger information shown in Table 5, which includes information such as tower material, tower nature, tower height, coordinate position, conductor model, number of insulators and number of cross arms.

[0150] Table 5 Clustering ledger table

[0151]

[0152]

[0153] Optionally, detecting equipment and hidden defects on the tower in the visible light image includes:

[0154] Trained on an object detection algorithm, a single neural network predicts bounding boxes and class probabilities in visible light images.

[0155] For tower equipment, key power equipment such as insulators, lightning arresters, primary and secondary fusion switches, circuit breakers, grounding switches, transformers and fusion terminals are identified and tested;

[0156] For hidden defects, we use the defect sample library to train common defects such as hanging foreign objects, bird nest construction, insulation damage, tower top damage, transformer oil leakage, etc. to identify and locate targets and defects;

[0157] The labels of the tower equipment or defect recognition results in the inspection image are associated with the corresponding images, which are used to associate the recognition results with the ledger information in the equipment or defect mapping module.

[0158] Optionally, extracting coordinate position information of the visible light image includes:

[0159] Extracting the exchangeable image file format in the visible light image data attribute file, and converting it into coordinate points in a rectangular coordinate system to obtain the longitude and latitude coordinate information of the tower;

[0160] Convert the extracted tower latitude and longitude coordinates into decimal format. The conversion formula is:

[0161]

[0162] Convert the longitude and latitude coordinates (B, L, H) to points in the spatial rectangular coordinate system (X, Y, Z), and use the Xi'an 80 coordinate system as the ellipsoid reference, where the major semi-axis a of the ellipsoid is 6378140±5 meters, the minor semi-axis b is 6356755.2882 meters, and the radius of curvature of the circle passing through the longitude and latitude coordinates is where e 2 is the second eccentricity of the ellipsoid, specifically

[0163] Therefore, the longitude and latitude coordinates (B, L, H) are converted to the spatial rectangular coordinate system (X, Y, Z)

[0164] X=(N+H)×cosB×cosL

[0165] Y=(N+H)×cosB×sinL

[0166] Z=(N×(1-e 2 )+H×sinB.

[0167] Optionally, based on the image feature information of the visible light image, determining the density center point and neighborhood points of the visible light image, and constructing density clustering of the visible light image in spatial distribution, including:

[0168] Based on the image feature information of the visible light image, the tower head is identified, and when the center of the tower head is located in the center of the frame, it is the auxiliary clustering mark point, and the auxiliary clustering mark point of the visible light image is used as the density center point;

[0169] By calculating the distance between each point in the visible light image, the distance matrix D is represented by a matrix, where the rows and columns of the matrix represent different image positions, and the values ​​of the rows and columns are D i，j Indicates the distance between bright spots, and the main diagonal element is 0; when the density center point uses the tower head in the image recognition result, its main diagonal element is set to 1, otherwise it is set to 0;

[0170] Perform logical judgment on the elements of the matrix, construct the connection clue matrix C, calculate the connection clue matrix C, and judge whether each point is a neighborhood point and a density center point. Where n represents the total number of points in the image, S i It represents the number of points directly connected to point i, that is, the point density. The sum of each row S i Add 1 to get the adjusted point density S′ i =S i +1, that is, even if a point has no other directly connected points, it will be regarded as a member of its own neighborhood. For each point i, if the point density S′ i Greater than the density threshold Min p , then the point is considered to be the density center point. If the point density S′ i Less than the density threshold Min p But if it is greater than 1, then the point is a neighborhood point;

[0171] When there are multiple density center points, these density center points and their neighborhood points are integrated into a connected network based on the connection relationship between the points, forming a comprehensive set of all relevant shooting points to construct the tower drone inspection image clustering cluster;

[0172] Cascading clustering is performed on the unclustered density center points to ensure that each density center point and its neighborhood can be organized into a logically consistent and non-overlapping cluster set.

[0173] Optionally, based on the density clustering of the visible light image in space distribution, the mapping of overhead distribution line equipment and defects with line topology is achieved, including:

[0174] Based on the density clustering of visible light images in space, determine the effective shooting point set for each tower;

[0175] According to the spatial relationship between the effective shooting points, they are sorted and stored according to the nearest neighbor principle to obtain the tower clustering result;

[0176] The results of equipment identification or defect detection are associated with the tower clustering results to construct the topological structure of the line.

[0177] Therefore, by clustering the density of overhead distribution line images taken by drones, the mapping relationship between overhead distribution line equipment and defects and line topology is established, and the automatic classification of unsupervised and manually guided inspection images, the construction of new line topology, and the verification and correction of the original line topology are realized. The precise GNSS navigation and positioning coordinate information carried in the drone images is combined with the rich tower information in the images to reconstruct and correct the topology of the line through the density clustering algorithm. This method can not only reduce the workload of manual input and improve the accuracy of ledger information, but also enhance the level of autonomy of line ledger management, providing strong technical support for the refined management and maintenance of power grids.

[0178] According to another aspect of the present invention, there is also provided an overhead distribution line topology mapping system 1400 based on drone inspection image density clustering, comprising:

[0179] A visible light image shooting module 1410 is used to shoot a visible light image of an overhead power distribution line through a drone;

[0180] An image feature information extraction module 1420 is used to extract image feature information of the visible light image, including detecting equipment and defects on the tower of the visible light image, and extracting coordinate position information of the visible light image;

[0181] A density clustering module 1430 is used to determine the density center point and neighborhood points of the visible light image based on the image feature information of the visible light image, and to construct density clustering of the visible light image in spatial distribution;

[0182] Implementing a topology mapping module 1440 for mapping overhead power distribution line equipment and defects with line topology structures based on density clustering of visible light images in space distribution, and obtaining the latest inspection data;

[0183] The comparison and identification module 1450 is used to compare the original ledger with the latest inspection data, calculate the coordinate position deviation and equipment correspondence, compare the original ledger with the newly constructed ledger based on the equipment type and quantity, and identify and remind inconsistencies.

[0184] Optionally, the module for extracting image feature information includes:

[0185] The predicted bounding box and class probability submodule is used to predict bounding boxes and class probabilities in visible light images through a single neural network trained based on the object detection algorithm;

[0186] The identification and detection equipment submodule is used to identify and detect key power equipment on the tower, including insulators, lightning arresters, primary and secondary fusion switches, circuit breakers, grounding switches, transformers and fusion terminals;

[0187] The defect identification and detection submodule is used to identify and locate targets and defects for potential defects such as hanging foreign objects, bird nest construction, insulation damage, tower top damage, transformer oil leakage, etc. It relies on the defect sample library for training;

[0188] The associated recognition result submodule is used to associate the labels of the tower equipment or defect recognition results in the inspection image with the corresponding image, and is used to associate the recognition results with the ledger information in the equipment or defect mapping module.

[0189] Optionally, the module for extracting image feature information includes:

[0190] The submodule for obtaining the longitude and latitude coordinate information of the tower is used to extract the exchangeable image file format in the visible light image data attribute file and convert it into coordinate points in a rectangular coordinate system to obtain the longitude and latitude coordinate information of the tower;

[0191] The submodule for converting decimal point format is used to convert the extracted tower latitude and longitude coordinate minute and second format into decimal point format. The conversion formula is:

[0192]

[0193] Convert the longitude and latitude coordinates (B, L, H) to points in the spatial rectangular coordinate system (X, Y, Z), and use the Xi'an 80 coordinate system as the ellipsoid reference, where the major semi-axis a of the ellipsoid is 6378140±5 meters, the minor semi-axis b is 6356755.2882 meters, and the radius of curvature of the circle passing through the longitude and latitude coordinates is where e 2 is the second eccentricity of the ellipsoid, specifically

[0194] The conversion module is used to convert the longitude and latitude coordinates (B, L, H) into the spatial rectangular coordinate system (X, Y, Z).

[0195] X=(N+H)×cosB×cosL

[0196] Y=(N+H)×cosB×sinL

[0197] Z=(N×(1-e 2 )+H)×sinB.

[0198] Optionally, build a density clustering module, including:

[0199] The density center point marking submodule is used to identify the tower head based on the image feature information of the visible light image, and when the center of the tower head is located in the center of the frame, it is an auxiliary clustering marking point, and the auxiliary clustering marking point of the visible light image is used as the density center point;

[0200] Determine the distance matrix submodule, which is used to calculate the distance between each point in the visible light image, and express it through a matrix, that is, the distance matrix D, where the rows and columns of the matrix represent different image positions, and the values ​​of the rows and columns are D i，j Indicates the distance between bright spots, and the main diagonal element is 0; when the density center point uses the tower head in the image recognition result, its main diagonal element is set to 1, otherwise it is set to 0;

[0201] Construct a connection clue matrix submodule, which is used to perform logical judgment on the elements of the distance matrix, construct a connection clue matrix C, calculate the connection clue matrix C, and determine whether each point is a neighborhood point and a density center point. Where n represents the total number of points in the image, S i It represents the number of points directly connected to point i, that is, the point density. The sum of each row S i Add 1 to get the adjusted point density S′ i =S i +1, that is, even if a point has no other directly connected points, it will be regarded as a member of its own neighborhood. For each point i, if the point density S′ i Greater than the density threshold Min p , then the point is considered to be the density center point. If the point density S′ i Less than the density threshold Min p But if it is greater than 1, then the point is a neighborhood point;

[0202] The image clustering submodule is used to integrate these density center points and their neighborhood points into a connected network when there are multiple density center points, based on the connection relationship between the points, to form a comprehensive set of all relevant shooting points and build the tower drone inspection image clustering cluster;

[0203] The clustering set organization submodule is used to perform cascade clustering on the unclustered density center points, ensuring that each density center point and its neighborhood can be organized into a logically consistent and non-overlapping cluster set.

[0204] Optionally, a topology mapping module is implemented, including:

[0205] Implement a topology mapping submodule for mapping overhead distribution line equipment and defects to line topology based on the density clustering of visible light images in space, including:

[0206] A submodule for determining a set of effective shooting points is used to determine a set of effective shooting points for each tower based on density clustering of visible light images in space distribution;

[0207] The tower clustering result obtaining submodule is used to sort and store the effective shooting points according to the spatial relationship between them and the nearest neighbor principle to obtain the tower clustering result;

[0208] The line topology structure submodule is used to associate the results of equipment identification or defect detection with the tower clustering results to construct the line topology structure method.

[0209] An overhead distribution line topology mapping system 1400 based on drone inspection image density clustering of an embodiment of the present invention corresponds to an overhead distribution line topology mapping method 100 based on drone inspection image density clustering of another embodiment of the present invention, which will not be repeated here.

[0210] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0211] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0212] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0213] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0214] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0215] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for topological mapping of overhead distribution lines based on density clustering of drone inspection images, characterized in that: include: Visible light images of overhead distribution lines captured by drones; Extracting image feature information of the visible light image, including detecting equipment and defects on the tower in the visible light image, and extracting coordinate position information of the visible light image; Based on the image feature information of the visible light image, determining the density center point and neighborhood points of the visible light image, and constructing density clustering of the visible light image in spatial distribution; Based on the density clustering of visible light images in space, the mapping of overhead distribution line equipment and defects with line topology is achieved to obtain the latest inspection data; By comparing the original ledger with the latest inspection data, calculating the coordinate position deviation and equipment correspondence, comparing the original ledger with the newly constructed ledger based on the equipment type and quantity, inconsistencies are identified and reminded.

2. The method according to claim 1, characterized in that Detection of equipment and defects on the tower using the visible light image includes: Trained on an object detection algorithm, a single neural network predicts bounding boxes and class probabilities in visible light images. For tower equipment, key power equipment such as insulators, lightning arresters, primary and secondary fusion switches, circuit breakers, grounding switches, transformers and fusion terminals are identified and tested; For hidden defects, we use the defect sample library to train common defects such as hanging foreign objects, bird nest construction, insulation damage, tower top damage, transformer oil leakage, etc. to identify and locate targets and defects; The labels of the tower equipment or defect recognition results in the inspection image are associated with the corresponding images, which are used to associate the recognition results with the ledger information in the equipment or defect mapping module.

3. The method according to claim 1, characterized in that: Extracting coordinate position information of the visible light image includes: Extracting the exchangeable image file format in the visible light image data attribute file, and converting it into coordinate points in a rectangular coordinate system to obtain the longitude and latitude coordinate information of the tower; Convert the extracted tower latitude and longitude coordinates into decimal format. The conversion formula is: Convert the longitude and latitude coordinates (B, L, H) to points in the spatial rectangular coordinate system (X, Y, Z), and use the Xi'an 80 coordinate system as the ellipsoid reference, where the major semi-axis a of the ellipsoid is 6378140±5 meters, the minor semi-axis b is 6356755.2882 meters, and the radius of curvature of the circle passing through the longitude and latitude coordinates is where e 2 is the second eccentricity of the ellipsoid, specifically Therefore, the longitude and latitude coordinates (B, L, H) are converted to the spatial rectangular coordinate system (X, Y, Z) X=(N+H)×cosB×cosL Y=(N+H)×cosB×sinL Z=(N×(1-e 2 )+H)×sinB。 4. The method according to claim 1, characterized in that Based on the image feature information of the visible light image, a density center point and a neighborhood point of the visible light image are determined, and a density clustering of the visible light image in spatial distribution is constructed, including: Based on the image feature information of the visible light image, the tower head is identified, and when the center of the tower head is located in the center of the frame, it is the auxiliary clustering mark point, and the auxiliary clustering mark point of the visible light image is used as the density center point; By calculating the distance between each point in the visible light image, the distance matrix D is represented by a matrix, where the rows and columns of the matrix represent different image positions, and the values ​​of the rows and columns are D i, j represent the distance between bright spots, and the main diagonal elements are 0; when the density center point uses the tower head in the image recognition result, its main diagonal elements are set to 1, otherwise they are set to 0; Perform logical judgment on the elements of the matrix, construct the connection clue matrix C, calculate the connection clue matrix C, and judge whether each point is a neighborhood point and a density center point. Where n represents the total number of points in the image, S i It represents the number of points directly connected to point i, that is, the point density. The sum of each row S i Add 1 to get the adjusted point density S′ i =S i +1, that is, even if a point has no other directly connected points, it will be regarded as a member of its own neighborhood. For each point i, if the point density S′ i Greater than the density threshold Min p , then the point is considered to be the density center point. If the point density S′ i Less than the density threshold Min p But if it is greater than 1, then the point is a neighborhood point; When there are multiple density center points, these density center points and their neighborhood points are integrated into a connected network based on the connection relationship between the points to form a comprehensive set of all relevant shooting points and construct the tower drone inspection image cluster: Cascading clustering is performed on the unclustered density center points to ensure that each density center point and its neighborhood can be organized into a logically consistent and non-overlapping cluster set.

5. The method according to claim 1, characterized in that Based on the density clustering of visible light images in space, the mapping of overhead distribution line equipment and defects with line topology is realized, including: Based on the density clustering of visible light images in space, determine the effective shooting point set for each tower; According to the spatial relationship between the effective shooting points, they are sorted and stored according to the nearest neighbor principle to obtain the tower clustering result; The results of equipment identification or defect detection are associated with the tower clustering results to construct the topological structure of the line.

6. An overhead distribution line topology mapping system based on drone inspection image density clustering, characterized in that: include: A visible light image shooting module is used to shoot visible light images of overhead power distribution lines through a drone; An image feature information extraction module is used to extract image feature information of the visible light image, including detecting equipment and defects on the tower of the visible light image, and extracting coordinate position information of the visible light image; Constructing a density clustering module, which is used to determine the density center point and neighborhood points of the visible light image based on the image feature information of the visible light image, and construct density clustering of the visible light image in spatial distribution; Implement a topological mapping module to map overhead power distribution line equipment and defects to line topology based on the density clustering of visible light images in space, and obtain the latest inspection data; The comparison and identification module is used to compare the original ledger with the latest inspection data, calculate the coordinate position deviation and equipment correspondence, compare the original ledger with the newly constructed ledger based on the equipment type and quantity, and identify and alert inconsistencies.

7. The system according to claim 6, characterized in that The module for extracting image feature information includes: The predicted bounding box and class probability submodule is used to predict bounding boxes and class probabilities in visible light images through a single neural network trained based on the object detection algorithm; The identification and detection equipment submodule is used to identify and detect key power equipment on the tower, including insulators, lightning arresters, primary and secondary fusion switches, circuit breakers, grounding switches, transformers and fusion terminals; The defect identification and detection submodule is used to identify and locate targets and defects for potential defects such as hanging foreign objects, bird nest construction, insulation damage, tower top damage, transformer oil leakage, etc. It relies on the defect sample library for training; The associated recognition result submodule is used to associate the labels of the tower equipment or defect recognition results in the inspection image with the corresponding image, and is used to associate the recognition results with the ledger information in the equipment or defect mapping module.

8. The system according to claim 6, characterized in that Extract image feature information module, including: The submodule for obtaining the longitude and latitude coordinate information of the tower is used to extract the exchangeable image file format in the visible light image data attribute file and convert it into coordinate points in a rectangular coordinate system to obtain the longitude and latitude coordinate information of the tower; The submodule for converting decimal point format is used to convert the extracted tower latitude and longitude coordinate minute and second format into decimal point format. The conversion formula is: Convert the longitude and latitude coordinates (B, L, H) to points in the spatial rectangular coordinate system (X, Y, Z), and use the Xi'an 80 coordinate system as the ellipsoid reference, where the major semi-axis a of the ellipsoid is 6378140±5 meters, the minor semi-axis b is 6356755.2882 meters, and the radius of curvature of the circle passing through the longitude and latitude coordinates is where e 2 is the second eccentricity of the ellipsoid, specifically The conversion space rectangular coordinate system submodule is used to convert the longitude and latitude coordinates (B, L, H) into the space rectangular coordinate system (X, Y, Z). X=(N+H)×cosB×cosL Y=(N+H)×cosB×sinL Z=(N×(1-e 2 )+H)×sinB。 9. The system according to claim 6, characterized in that Build a density clustering module, including: The density center point marking submodule is used to identify the tower head based on the image feature information of the visible light image, and when the center of the tower head is located in the center of the frame, it is an auxiliary clustering marking point, and the auxiliary clustering marking point of the visible light image is used as the density center point; Determine the distance matrix submodule, which is used to calculate the distance between each point in the visible light image, and express it through a matrix, that is, the distance matrix D, where the rows and columns of the matrix represent different image positions, and the values ​​of the rows and columns are D i，j Indicates the distance between bright spots, and the main diagonal element is 0; when the density center point uses the tower head in the image recognition result, its main diagonal element is set to 1, otherwise it is set to 0; Construct a connection clue matrix submodule, which is used to perform logical judgment on the elements of the distance matrix, construct a connection clue matrix C, calculate the connection clue matrix C, and determine whether each point is a neighborhood point and a density center point. Where n represents the total number of points in the image, S i It represents the number of points directly connected to point i, that is, the point density. The sum of each row S i Add 1 to get the adjusted point density S′ i =S i +1, that is, even if a point has no other directly connected points, it will be regarded as a member of its own neighborhood. For each point i, if the point density S′ i Greater than the density threshold Min p , then the point is considered to be the density center point. If the point density S′ i Less than the density threshold Min p But if it is greater than 1, then the point is a neighborhood point; The image clustering submodule is used to integrate these density center points and their neighborhood points into a connected network when there are multiple density center points, based on the connection relationship between the points, to form a comprehensive set of all relevant shooting points and build the tower drone inspection image clustering cluster; The clustering set organization submodule is used to perform cascade clustering on the unclustered density center points, ensuring that each density center point and its neighborhood can be organized into a logically consistent and non-overlapping cluster set.

10. The system according to claim 6, characterized in that Implement the topology mapping module, including: Implement a topology mapping submodule for mapping overhead distribution line equipment and defects to line topology based on the density clustering of visible light images in space, including: A submodule for determining a set of effective shooting points is used to determine a set of effective shooting points for each tower based on density clustering of visible light images in spatial distribution; The tower clustering result obtaining submodule is used to sort and store the effective shooting points according to the spatial relationship between them and the nearest neighbor principle to obtain the tower clustering result; The line topology construction submodule is used to associate the results of equipment identification or defect detection with the tower clustering results to construct the line topology.

Images