An unmanned aerial vehicle autonomous inspection processing method based on a distribution network overhead line equipment

Abstract

This invention provides an autonomous UAV inspection method for overhead power distribution line equipment, belonging to the field of intelligent control technology. The method includes: generating an adaptive compensation factor based on envelope surface deformation characteristics; combining hovering state parameters; scanning and processing the line surface using a combination of lidar and dual-polarization radar to obtain corrected 3D topology reconstruction data and icing feature vectors; inputting the corrected 3D topology reconstruction data and icing feature vectors into a pre-trained ResNet deep learning model to identify icing thickness distribution, insulator damage, or suspended foreign object anomalies, outputting quantified anomaly type, spatial coordinates, and confidence scores; and generating an adaptive operation strategy including tool ID encoding and operation parameters based on the quantified anomaly type, spatial coordinates, and confidence scores. This invention ensures the stability and reliability of the inspection process.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, and in particular to a method for autonomous inspection and processing of overhead power distribution line equipment by unmanned aerial vehicles. Background Technology

[0002] In a scenario involving the inspection of overhead power distribution lines in a mountainous area, there are numerous winding 10kV overhead power distribution lines that cut through forests. The surrounding terrain is complex, with some areas exhibiting significant elevation changes, making manual inspections extremely difficult.

[0003] This inspection aimed to check for defects such as icing, damaged insulators, and hanging foreign objects on the power lines to ensure stable operation under adverse weather conditions. During the inspection, following established procedures, the drone first generated a three-dimensional flight trajectory and determined the inspection safety boundaries based on the geographical coordinates and topology data of the overhead power distribution line equipment. Subsequently, the drone successfully reached the target conductor coordinates, achieved centimeter-level dynamic hovering through multi-sensor fusion positioning, and output hovering status parameters.

[0004] During the data acquisition phase, a combination of lidar and dual-polarization radar was used to scan the line surface. However, due to the complex terrain in the mountainous area causing abnormal signal reflection, the envelope surface in some areas underwent irregular deformation due to terrain interference when generating adaptive compensation factors based on the envelope surface deformation characteristics. This prevented the generated adaptive compensation factors from accurately matching the actual situation, resulting in deviations in the correction of some point cloud data located in valleys or near tall trees when performing spatial coordinate transformation correction on the original point cloud dataset. For example, near a valley, the point cloud data of the conductor that should have been on the same plane showed a height difference of about 5 centimeters after correction. This deviation was further amplified when constructing the 3D topology reconstruction data. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method for autonomous inspection of overhead power distribution line equipment by unmanned aerial vehicles, which ensures the stability and reliability of the inspection process.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0007] A method for autonomous inspection and processing of overhead power distribution line equipment using unmanned aerial vehicles (UAVs), the method comprising:

[0008] Based on the geographic coordinates and topology data of overhead power distribution line equipment, a three-dimensional flight trajectory and inspection safety boundary of the UAV are generated.

[0009] Based on the three-dimensional flight trajectory and inspection safety boundary, the UAV is controlled to reach the target ground line coordinates. Centimeter-level dynamic hovering is achieved through multi-sensor fusion positioning. In the hovering state, three spatial reference coordinates are dynamically calibrated and hovering state parameters including pose accuracy, electromagnetic interference intensity and reference coordinate set are output.

[0010] Based on the reference coordinate set, a dynamic spatial feature envelope surface is constructed and the region is rasterized to form a multi-level scanning unit set; an adaptive compensation factor is generated based on the deformation characteristics of the envelope surface, and combined with the hovering state parameters, the line surface is scanned and processed by a combination of lidar and dual polarization radar to obtain corrected three-dimensional topology reconstruction data and icing feature vector.

[0011] The corrected 3D topology reconstruction data and icing feature vectors are input into a pre-trained ResNet deep learning model to identify icing thickness distribution, insulator damage or hanging foreign object anomalies, and output quantified anomaly type, spatial coordinates and confidence score.

[0012] Based on the quantified anomaly type, spatial coordinates, and confidence score, an adaptive job strategy is generated, which includes tool ID encoding and job parameters.

[0013] The above-described solution of the present invention has at least the following beneficial effects:

[0014] By generating 3D flight trajectories and safety boundaries based on geographic coordinates and topological data, and combining multi-sensor fusion positioning technology, centimeter-level dynamic hovering of UAVs was achieved under complex electromagnetic environments and airflow disturbances. The dynamically calibrated three spatial reference coordinates and output hovering state parameters provided accurate spatial positioning references for subsequent scanning operations, effectively solving the problems of missed detections and false detections caused by positioning deviations in traditional UAV inspections, and ensuring the stability and reliability of the inspection process. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating an embodiment of the present invention of an unmanned aerial vehicle (UAV) autonomous inspection and processing method for overhead power distribution line equipment.

[0016] Figure 2 This is an embodiment of the present invention. Figure 1 A flowchart illustrating step 1. Detailed Implementation

[0017] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0018] like Figure 1 As shown, an embodiment of the present invention proposes an autonomous inspection method for unmanned aerial vehicles (UAVs) based on overhead power distribution line equipment. The method includes the following steps:

[0019] Step 1: Based on the geographic coordinates and topology data of the overhead power distribution line equipment, generate the three-dimensional flight trajectory and inspection safety boundary of the UAV;

[0020] Step 2: Based on the three-dimensional flight trajectory and inspection safety boundary output in Step 1, control the UAV to reach the target ground line coordinates, achieve centimeter-level dynamic hovering through multi-sensor fusion positioning, dynamically calibrate three spatial reference coordinates in the hovering state and output hovering state parameters including pose accuracy, electromagnetic interference intensity and reference coordinate set.

[0021] Step 3: Based on the reference coordinate set output in Step 2, construct a dynamic spatial feature envelope surface and perform regional rasterization to form a multi-level scanning unit set; generate an adaptive compensation factor based on the deformation characteristics of the envelope surface, and combine it with the hovering state parameters to scan the line surface using a combination of lidar and dual-polarization radar and process it to obtain corrected three-dimensional topology reconstruction data and icing feature vector.

[0022] Step 4: Input the corrected 3D topology reconstruction data and icing feature vector into the pre-trained ResNet deep learning model to identify icing thickness distribution, insulator damage or hanging foreign object anomalies, and output the quantified anomaly type, spatial coordinates and confidence score.

[0023] Step 5: Based on the quantitative anomaly type, spatial coordinates, and confidence score output in Step 4, generate an adaptive job strategy that includes tool ID encoding and job parameters.

[0024] In this embodiment of the invention, dynamic spatial feature envelope construction and rasterization enable differentiated scanning of regions. Combined with adaptive compensation factors, geometric distortion, pose drift, and electromagnetic interference are corrected, making the three-dimensional topology reconstruction data and icing feature vectors more closely match the actual state of the equipment, providing high-precision data support for anomaly identification. A pre-trained ResNet deep learning model is used to fuse topology and icing features to accurately identify anomalies such as icing thickness distribution, insulator damage, and suspended foreign objects, outputting quantitative results and confidence levels, reducing human interpretation errors, and improving the automation and accuracy of defect identification. Based on anomaly type, coordinates, and confidence levels, operation tools, planning paths, and optimization parameters (such as laser power and robotic arm torque) are dynamically matched to form a set of operation instructions with spatiotemporal constraints, enabling the inspection to form a closed loop from data acquisition and anomaly identification to processing strategy generation, improving the overall intelligent level of distribution network line inspection.

[0025] like Figure 2As shown, in a preferred embodiment of the present invention, step 1, generating the three-dimensional flight trajectory and inspection safety boundary of the UAV based on the geographical coordinates and topology data of the overhead power distribution line equipment, includes:

[0026] Step 11: Obtain the GIS geographic coordinates and equipment topology data of the overhead distribution network lines, and extract the spatial location information of the towers, the spatial morphological parameters of the conductors and ground wires, and the topological relationship of adjacent equipment.

[0027] Step 12: Using the spatial morphology parameters of the conductor and ground wire output in Step 11 as input, generate the three-dimensional spatial trajectory of the conductor and ground wire through a spatial curve fitting algorithm, and calculate the dynamic vertical isolation threshold of the UAV based on electrical safety standards.

[0028] Step 13: Using the device topology extracted in Step 11 and the three-dimensional spatial trajectory of the conductor and ground wire generated in Step 12 as input, identify the distribution of vegetation coverage area, building shielding area and electromagnetic interference source, construct a rotationally symmetric dynamic envelope surface with the conductor and ground wire trajectory as the axis, and generate the inspection safety boundary.

[0029] Step 14: Integrate the three-dimensional spatial trajectory of the conductor ground line from Step 12 with the rotationally symmetric dynamic envelope from Step 13, and generate the waypoint sequence and attitude control command set of the UAV through a segmented trajectory optimization algorithm, outputting the three-dimensional flight trajectory and inspection safety boundary.

[0030] In this embodiment of the invention, the specific implementation process of step 11 is as follows:

[0031] Operators access the GIS platform module of the target overhead line area through the permission login interface of the power distribution network management system, enter the line number or name (such as "XX line #1-#50 tower section"), and initiate a data retrieval request. The data is output as a structured file (such as SHP format or CSV format), containing basic geographic information of the area through which the line passes. Specific fields include the unique identifier ID of each tower, latitude and longitude coordinates (accurate to 6 decimal places), tower top elevation, and coordinates and height data of surrounding obstacles (such as trees and buildings). Equipment Topology Database: Output in relational data table format, containing three sub-tables: Tower Information Table: Records tower type (e.g., "10kV straight tower", "35kV tension tower"), material, construction year, and GIS coordinate ID associated with the installation location; Conductor and Ground Wire Parameter Table: Records conductor and ground wire type (e.g., "JL / G1A-120 / 20 steel-cored aluminum stranded wire"), design sag curve parameters (including sag reference values ​​at different temperatures), tension section division (e.g., towers #1-#5 are one tension section), and the specific height of the conductor suspension point on the tower (e.g., 12 meters above the ground); Connection Relationship Table: Records the connection method between the tower and the conductor and ground wire (e.g., "#1 tower suspends A-phase conductor"), the spacing between adjacent towers (e.g., the spacing between towers #1 and #2 is 150 meters), the connection node between branch lines and the main line (e.g., "#10 tower T-connects branch line #1"), the connection position of the grounding device to the tower, and other topology logic.

[0032] All tower records were filtered from the GIS geographic coordinate dataset, sorted by tower ID, and the longitude, latitude, and altitude data of each record were extracted, removing invalid values ​​(such as missing coordinates or data outside the line range). The extracted three-dimensional coordinates (longitude, latitude, and altitude) were associated with the model and ID in the tower information table to generate a structured "tower spatial location list". The list format is as follows: [tower ID, longitude, latitude, altitude, model], ensuring that the spatial location of each tower can be accurately located. The towers were grouped by tension section, and the starting / ending tower ID, conductor and ground wire model, and corresponding sag curve parameters (such as span length and design maximum sag value) and conductor suspension point height (distinguishing different phase sequence conductors, such as the suspension height difference between phase A, phase B, and phase C) were recorded for each tension section. The azimuth angle of the line in each tension section was calculated using the coordinates of adjacent towers (e.g., the direction from tower #1 to tower #2 is 30° east of north). The actual shape parameters of the sag curve were corrected by combining the terrain undulation data (such as the tower height difference).

[0033] Based on the connection table, with the tower as the core node, the topological logic of "tower-conductor-peripheral equipment" is analyzed:

[0034] Determine the connection points between the towers and conductors: for example, the A-phase conductor is suspended on the left side of tower #3 and the B-phase conductor is suspended on the right side, with the conductors connected to the tower crossarm via insulator strings at specific locations; convert the longitude and latitude coordinates of the two towers into horizontal distances (e.g., the distance between towers #5 and #6 is 200 meters) as the basis for subsequent sag calculations; mark the spatial relationships of associated equipment: for example, there is a 110kV live line 50 meters away from tower #8 and a building 10 meters below tower #12, record the relative distances and orientations of these devices to the target line, providing a basis for safety boundary construction; cross-validate the extracted tower coordinates, conductor parameters, and topology relationships: for example, calculate the conductor-to-ground distance using tower height and suspension point height to ensure compliance with safety standards; verify the rationality of the conductor span using the spacing between adjacent towers; integrate the validated information into a structured dataset, outputting: a list of tower spatial locations, a table of conductor morphological parameters (including sag, suspension point, and direction angle), and a topology diagram of equipment relationships (including adjacent spacing and associated equipment locations), as input data for step 12.

[0035] The specific implementation process of step 12 above is as follows:

[0036] From the conductor and ground wire spatial morphology parameters output in step 11, the data are divided into groups according to the tension section (e.g., towers #1-#5 are one tension section). Each group includes: the three-dimensional coordinates (longitude, latitude, conductor suspension point height) of the starting tower and the ending tower, the span (horizontal distance between the two towers), the sag reference value at different temperatures (e.g., sag corresponding to -10℃, 25℃, and 40℃), and the conductor and ground wire orientation angle. Combined with the real-time temperature on the day of inspection (e.g., obtained through a meteorological data interface), the sag parameters at the current temperature that match the sag reference value are selected as the key input for curve fitting.

[0037] For each tension section, the three-dimensional shape of the conductor and ground wire is constructed using the catenary equation fitting method:

[0038] Using the conductor suspension points of the starting and ending towers as the endpoints of the curve, the span, sag at the current temperature, and the directional angle are input to calculate the natural sag shape of the conductor in three-dimensional space (considering the tower height difference caused by terrain undulations). For line segments with complex terrain or long spans (such as those exceeding 300 meters), intermediate control points (such as the predicted sag position at the midpoint of the span) are added between the two endpoints, and cubic spline curve interpolation is used to supplement the fitting to ensure a smooth transition of the curve.

[0039] Based on the fitted continuous curve, uniform sampling is performed according to the inspection accuracy requirements (e.g., one sampling point every 5 meters). The three-dimensional coordinates (longitude, latitude, and height above the ground) of each sampling point are recorded. Abnormal points that exceed the range of the line connecting the tower suspension points are removed, and curve deviations caused by sudden changes in terrain are corrected (e.g., adjustment of the actual height of the conductor and ground wire near the hillside). All sampling points are spliced ​​together in the order of tension sections to form a complete three-dimensional reference trajectory of the conductor and ground wire from the starting tower to the ending tower. The trajectory data format is: [sampling point number, longitude, latitude, height, tension section ID].

[0040] Match the reference value according to the actual voltage level of the line: for example, the reference value for a 10kV line is 1.5 meters, and the reference value for a 35kV line is 2 meters; adjust the reference value according to the line's operating status (such as whether it is a live inspection): if it is a live inspection, add 0.5 meters of safety redundancy to the standard reference value; compensate for UAV flight attitude fluctuations: calculate the attitude compensation amount as 0.5 meters based on the performance parameters of the selected UAV (such as hovering accuracy ±0.3 meters, maximum vertical jitter ±0.2 meters). Based on the difference between the current temperature and historical extreme temperatures (such as a summer high of 40℃ and a winter low of -10℃), and according to the thermal expansion and contraction characteristics of the conductor and ground wire, the maximum sag change (e.g., ±0.4 meters) is calculated. Considering factors such as strong winds and airflow disturbances, an additional 0.1 meter compensation is added, resulting in a final total dynamic compensation of 0.5 + 0.4 + 0.1 = 1.0 meter. The baseline value and the dynamic compensation are then superimposed: for example, the baseline value of 1.5 meters for a 10kV live line + the dynamic compensation of 1.0 meter = 2.5 meters.

[0041] Set upper and lower thresholds: the lower limit is 2.5 meters (the drone must not approach the conductor below this distance), and the upper limit is set according to the inspection requirements (e.g., 5 meters, to ensure the lens can clearly capture the details of the conductor). The final output is the dynamic vertical isolation threshold range. Combine the surrounding equipment topology relationship extracted in step 11 (e.g., the height of nearby buildings, the distribution of trees) to verify the rationality of the threshold: for example, if there are 3-meter-high trees under the line, it is necessary to ensure that the lower threshold (2.5 meters) plus the height of the trees does not exceed the safe distance between the conductor and the ground. Adjust the threshold separately for special sections (e.g., conductor sections crossing highways or rivers): for example, for conductor sections crossing highways, the lower threshold is increased to 3 meters to avoid the safety risk caused by the drone flying too low. The three-dimensional reference trajectory of the conductor (including the coordinate sequence of sampling points) and the dynamic vertical isolation threshold of the drone (including the upper and lower limits) are used as the input data for step 13.

[0042] The specific implementation process of step 13 above is as follows:

[0043] Retrieve the vegetation layer around the line (including the distribution coordinates, tree species, and growth height records of trees and shrubs) from the GIS geographic data in step 11, and combine it with satellite remote sensing images from the three months prior to the inspection (updating recent vegetation growth). Determine the safe height threshold for vegetation based on the line voltage level (e.g., within 5 meters below and to both sides of a 10kV line, the tree height should not exceed 3 meters; within the corresponding range of a 35kV line, the tree height should not exceed 2 meters). Through spatial distance calculation (horizontal distance between tree coordinates and the three-dimensional trajectory of the conductor), filter out tree clusters with a horizontal distance ≤ 5 meters and a height exceeding the safe threshold, mark them as "vegetation cover danger zones," and record the boundary coordinates of the area and the height of the highest tree. Obtain information on buildings around the line from the GIS geographic data (coordinates, height, and structural type, such as residential buildings and factories), focusing on extracting building records with a horizontal distance ≤ 15 meters from the line. According to the "Design Code for Overhead Distribution Lines," the minimum horizontal safe distance between a 10kV line and a building is 1.5 meters, and for a 35kV line it is 3 meters. If the building height exceeds 10 meters (which may obstruct the drone's view), the safe distance needs to be increased by 20%.

[0044] Calculate the horizontal distance between the building and the conductor / ground wire trajectory. If it is less than the safe distance for the corresponding voltage level, or if the building height may obstruct the drone's flight path (e.g., the vertical distance between the building top and the conductor / ground wire trajectory is ≤3 meters), mark it as a "building obstruction hazard zone" and record the building boundary and its relative position to the line. Extract parameters of related equipment around the line (e.g., transformer location, capacity, switch station coordinates) from the equipment topology database, and retrieve historical electromagnetic interference records from the power dispatch system (e.g., a certain area was previously interrupted by a high-frequency signal tower for drone communication). Set the interference radius according to the equipment type (e.g., 5 meters for a 10kV transformer, 30 meters for a high-frequency signal tower), and adjust the interference range by combining historical records of areas where the interference intensity exceeds the drone communication threshold (e.g., signal attenuation ≥30%). Draw a circular area centered on the interference source according to the calculated interference radius, mark it as an "electromagnetic interference hazard zone," and label the interference intensity level (e.g., "strong interference zone" or "weak interference zone").

[0045] Using the three-dimensional spatial trajectory of the grounding wire generated in step 12 as the axis, a sampling point is taken every 10 meters along the trajectory. At each sampling point, a plane perpendicular to the direction of the grounding wire is established (e.g., if the direction of the grounding wire is 30° east of north, then the normal direction of the plane is 30° east of north). Within each plane, with the sampling point of the grounding wire trajectory as the center, the distance of the inner boundary (the side closer to the grounding wire) is strictly equal to the lower limit of the dynamic vertical isolation threshold calculated in step 12 (e.g., 2.5 meters), ensuring the minimum safe distance between the UAV and the grounding wire; Open area (no danger area): The inner and outer boundaries of the plane (the side away from the grounding wire) extend outward by 5 meters from the axis, forming a symmetrical cylindrical segment (inner... The total width is 7.5 meters (2.5 meters on the side + 5 meters on the outer side). If the plane overlaps with a vegetation hazard zone, the outer boundary extension distance is increased to 8 meters (3 meters more than the open area) to ensure that it avoids the possible swaying range of the treetops (calculated based on a tree swaying amplitude of 1.5 meters under the historical maximum wind speed, with redundancy). If the plane is close to a building hazard zone, the outer boundary extension distance is adjusted to 10 meters and must be 1.5 meters higher than the top of the building (to avoid the drone's view being blocked by the building during flight). If the plane is in an electromagnetic interference zone, the outer boundary must exceed the interference source range by 2 meters (if the interference radius is 30 meters, the outer boundary is extended to 32 meters) to ensure that the drone flies in a low-interference environment.

[0046] Following the sampling point sequence of the conductor trajectory, the inner and outer boundaries of each plane are connected to form a continuous rotationally symmetric envelope surface (similar to a "pipe" structure) along the trajectory. When the difference in the outer boundary expansion distance between adjacent sampling points exceeds 2 meters (such as a sudden change from 5 meters in an open area to 10 meters in a building area), 5 transition sampling points are added in the middle to make the expansion distance gradually change (such as 5 meters → 6 meters → 7 meters → 8 meters → 9 meters → 10 meters) to avoid acute angle turns in the envelope surface.

[0047] Check whether all inner boundaries of the envelope meet the dynamic vertical isolation threshold (no one is less than the lower limit); verify whether the outer boundary completely avoids all dangerous areas (the distance from vegetation, buildings, and interference sources is ≥0.5 meters redundancy), and perform separate verification for special sections (such as where the line crosses rivers or highways): the outer boundary of the envelope for sections crossing highways must be 5 meters higher than the road surface (to avoid the drone flying low and affecting traffic); convert the three-dimensional spatial range of the envelope into structured data: each sampling point corresponds to a set of boundary coordinates (three-dimensional coordinates of the inner boundary point and the outer boundary point), and label the section and dangerous area type; generate a visualized safety boundary model (such as a three-dimensional mesh model), overlay it on the GIS map, and intuitively display the space range in which the drone can fly (the inside of the envelope is the safe zone, and the outside is the no-fly zone), and finally output the inspection safety boundary dataset and visualization model containing the three-dimensional coordinate sequence as the constraint condition for generating the drone flight trajectory in step 14.

[0048] The specific implementation process of step 14 above is as follows:

[0049] Based on the three-dimensional spatial trajectory (sampling point sequence) of the conductor generated in step 12, verify one by one whether each sampling point is located inside the rotationally symmetric dynamic envelope generated in step 13: calculate the distance between the sampling point and the inner boundary of the envelope (≥0.3 meters safety redundancy) and the distance between the sampling point and the outer boundary (≥0.5 meters safety redundancy) by comparing spatial coordinates; if there is a trajectory point that exceeds the range of the envelope (e.g., a sampling point is only 0.1 meters away from the inner boundary), then make a fine adjustment to the point: offset it outward by 0.2 meters in a direction perpendicular to the direction of the conductor, to ensure that the adjusted trajectory is completely within the safety boundary and still conforms to the actual shape of the conductor after offset (deviation not exceeding 0.5 meters).

[0050] Associate the sampling point coordinates (longitude, latitude, altitude) of the conductor / ground wire trajectory with the envelope parameters of the safety boundary (inner / outer boundary coordinates corresponding to each sampling point) to generate a "trajectory-boundary association table" in the format: [sampling point number, conductor / ground wire trajectory coordinates, inner boundary coordinates, outer boundary coordinates, type of hazardous area]. Use two adjacent towers as a basic inspection segment (e.g., #1-#2 tower, #2-#3 tower). If the distance between two towers exceeds 300 meters (e.g., a long span section in a mountainous area), add a virtual segmentation point in the middle (e.g., 150 meters from #1 tower) to divide the long segment into two sub-segments, avoiding excessive difficulty in optimizing a single segment's trajectory. Mark the distribution of key equipment for each segment: extract key inspection targets within the segment from the equipment topology relationship in step 11, such as the three sets of insulators between #1 and #2 towers (located on the right side of #1 tower, the midpoint clamp of the span, and the left side of #2 tower), and two surge arresters, recording their precise spatial coordinates. Mark the optimal observation direction (e.g., insulators need to be observed from a 45° angle below); using the towers at both ends of the segment as the starting and ending points, generate an initial path along the centerline of the conductor-ground wire trajectory (the midpoint between the conductor-ground wire trajectory and the inner boundary of the envelope plane), ensuring a stable distance from the conductor-ground wire (e.g., the lower limit of the dynamic vertical isolation threshold + 0.5 meters of redundancy); search for the shortest path from the starting point to the ending point using the A* algorithm, prioritizing path nodes that are close to the conductor-ground wire but do not touch the inner boundary to avoid detours; for the location of critical equipment (e.g., insulators), adjust the path nodes to the optimal observation position of the equipment (e.g., passing 3 meters diagonally in front of the insulator), ensuring that the angle between the drone lens axis and the normal to the equipment surface is ≤30° (to avoid reflection or obstruction); if the segment contains vegetation-covered areas, the path needs to be offset by 1 meter away from the vegetation (within the envelope plane); if passing through areas with weak electromagnetic interference, the path needs to be close to the outer boundary of the envelope plane (away from the center of the interference source).

[0051] On the optimized segmented trajectory, basic waypoints are set at fixed intervals of 10 meters, and the three-dimensional coordinates of each point are recorded (longitude accurate to 0.00001°, latitude with the same accuracy, and height accurate to 0.1 meters). Near key equipment (such as insulators and clamps), the interval is shortened to 3 meters (e.g., one point is set 3 meters in front of the insulator, one point directly in front of it, and one point is set 3 meters behind it) to ensure multi-angle shooting. If the equipment is located near a tower (e.g., within 5 meters of the tower), an additional waypoint is set at the tower (1 meter above the top of the tower). This is used to observe the equipment at the top of the tower; it is numbered in the format of "segment ID-sequence number" (e.g., the 5th point of the #1-#2 tower segment is marked as "1-2-5"); based on the direction of adjacent waypoints (e.g., the azimuth from point A to point B is 20° east of north), the UAV's nose direction is calculated to be consistent with the path direction (yaw angle = azimuth angle) to ensure stable flight direction; if passing critical equipment, the yaw angle is adjusted to the azimuth of the equipment (e.g., if the equipment is 30° to the right of the path, the yaw angle is increased by 30°).

[0052] Pitch angle: Based on the vertical distance between the current waypoint and the ground guide line (e.g., the drone is 15 meters high and the ground guide line is 12 meters high), calculate the pitch angle as 15° downward (ensure the lens is aligned with the ground guide line); if the observation tower top equipment (18 meters high) and the drone height is 17 meters, then adjust the pitch angle to 5° upward.

[0053] Roll angle: Maintain 0° (horizontal state) during normal flight; if the path needs to bypass small obstacles (such as tree branches), adjust the roll angle according to the turning range (e.g., a 10° right turn corresponds to a roll angle of -5°, a left turn corresponds to +5°) to ensure smooth turning; calculate the flight time of adjacent waypoints based on the UAV's cruising speed (e.g., 5 m / s) (e.g., 10-meter interval corresponds to 2 seconds), assign a timestamp to each waypoint (e.g., the starting point is 0 seconds, the second point is 2 seconds, and so on), forming a three-dimensional trajectory containing time information: [timestamp, longitude, latitude, altitude]; convert the attitude parameters (yaw angle, pitch angle, roll angle) of each waypoint into a command format that the UAV can execute (e.g., a 30° yaw angle corresponds to the command "YAW+30"), and mark the command's effective time (synchronized with the waypoint timestamp); add a "take a picture" command when passing critical equipment (synchronized with the best observation point timestamp), and add a "decelerate" command before turning (speed reduced to 3 m / s) to ensure flight and observation coordination.

[0054] The final output and verification result in two types of core data:

[0055] 3D flight trajectory: A complete time series of waypoint coordinates, accompanied by a visual trajectory map (overlaid on a GIS map, with safety boundary range marked); attitude adjustment commands and auxiliary operation commands ordered by time, in a format compatible with the UAV control system (such as DJISDK compatible format); Simulate UAV flight along the trajectory to check whether all waypoints are within the safety boundary and whether attitude commands cause the camera to deviate from the target (deviation ≤5° is acceptable). If any problems are found, return to the optimization stage for readjustment.

[0056] The final output is a verified 3D flight trajectory and attitude control command set, which serves as the basis for the actual inspection of the UAV, and also includes inspection safety boundary data for real-time verification during flight.

[0057] In this embodiment of the invention, by accurately acquiring and extracting tower coordinates, conductor and ground wire morphology parameters, and equipment topology relationships, the data source is ensured to be highly matched with the actual equipment operating conditions, avoiding planning errors caused by deviations in basic data. A three-dimensional spatial trajectory that fits the actual shape of the conductor and ground wire is generated. Based on the vertical isolation threshold calculated using electrical standards and dynamic factors, the safety distance baseline between the UAV and the conductor and ground wire is determined from a technical perspective, effectively avoiding the risk of contact with energized equipment, while ensuring the trajectory's fit with the conductor and ground wire to improve the targeting of inspections. By identifying dangerous areas such as vegetation, buildings, and electromagnetic interference, a rotationally symmetric dynamic envelope surface is constructed to form an inspection safety boundary, achieving precise isolation of risks in complex environments. The dynamic adjustment characteristics of the boundary ensure that the UAV can maintain a safe flight space in different dangerous areas, eliminating potential hazards such as collisions and signal interference from a spatial perspective. Through the fusion and segmented optimization of the trajectory and boundary, the generated waypoint sequence and attitude command set not only strictly follow the safety boundary constraints but also maximize the observation effect of key equipment and reduce invalid flight mileage.

[0058] In a preferred embodiment of the present invention, step 2 involves controlling the UAV to reach the target ground line coordinates based on the three-dimensional flight trajectory and inspection safety boundary output in step 1. Centimeter-level dynamic hovering is achieved through multi-sensor fusion positioning. In the hovering state, three spatial reference coordinates are dynamically calibrated, and hovering state parameters including pose accuracy, electromagnetic interference intensity, and the reference coordinate set are output, including:

[0059] Step 21: Based on the three-dimensional flight trajectory and inspection safety boundary output in Step 1, generate multi-degree-of-freedom motion commands for the UAV;

[0060] Step 22: In response to the UAV arriving at the target ground line coordinates, real-time dynamic carrier phase differential positioning data, inertial measurement unit data and visual odometry data are fused to generate a real-time pose feedback stream with centimeter-level accuracy.

[0061] Step 23: Based on the real-time pose feedback stream, dynamically calibrate three spatial reference coordinates in the hovering state based on the spatial distribution characteristics of the conductor and ground wire, and simultaneously monitor the fluctuation of the environmental electromagnetic field intensity.

[0062] Step 24: Aggregate real-time pose feedback stream, spatial reference coordinate set, and electromagnetic field strength fluctuation data, and output hovering state parameters including pose accuracy, electromagnetic interference intensity quantification value, and reference coordinate set.

[0063] In this embodiment of the invention, the specific implementation process of step 21 is as follows:

[0064] Extract the waypoint sequence (including longitude, latitude, altitude, and timestamp) corresponding to the target guideline coordinates from the 3D flight trajectory output in step 1. Obtain the spatial constraint range of the area (such as the minimum distance of the inner boundary and the maximum distance of the outer boundary) from the inspection safety boundary data. Decompose the waypoint sequence into continuous position targets (one position point every 0.1 seconds) and attitude targets (corresponding to the pitch angle, yaw angle, and roll angle generated in step 14) according to the timestamp. Determine the spatial position and fuselage attitude that the UAV needs to reach at each moment. Based on the difference between the current position and the target position (such as the distance deviation in the X, Y, and Z axis directions), generate the horizontal (longitude, latitude) and vertical (longitude, latitude) coordinates. The vertical (altitude) speed control command (such as "X-axis +0.5m / s" and "Z-axis -0.2m / s") ensures that the movement speed does not exceed the safety threshold (such as the maximum horizontal speed of 5m / s and the vertical speed of 2m / s); based on the angular deviation between the current attitude and the target attitude (such as a yaw angle deviation of 10°), the rotational angular velocity command (such as "yaw +5° / s") is generated to control the drone's nose direction and camera angle. Before generating the command, it is verified whether each target position is within the inspection safety boundary (such as a distance ≥ 0.3 meters from the inner boundary). If it exceeds the boundary, the command is automatically adjusted (such as shifting to the safe area by 0.5 meters) to ensure that the movement does not touch the no-fly zone.

[0065] The specific implementation process of step 22 above is as follows:

[0066] Data is acquired via a GNSS receiver mounted on the UAV, with an output frequency of 10Hz (once every 0.1 seconds), including latitude, longitude, altitude, and positioning accuracy factor (such as PDOP value). Low-precision data with PDOP > 3 are discarded. Acceleration, angular velocity, and attitude angles (pitch, yaw, roll) of the UAV are collected, with an output frequency of 100Hz. High-frequency noise (such as interference caused by fuselage vibration) is removed using a low-pass filter. Images of the ground plane and surrounding environment are captured by the UAV's binocular camera. The displacement of feature points in adjacent frames is calculated, and the relative position change (such as a 0.1-meter movement along the X-axis) is output at a frequency of 30Hz. Abnormal frames with a feature point matching success rate < 80% are discarded. Using the timestamp of the GNSS data as a reference, IMU and visual odometry data are interpolated or resampled to ensure that the three types of data are aligned at the same time point. Time deviation ≤ 0.01 seconds); when the GNSS signal is stable (PDOP ≤ 2), GNSS data is used as the main reference, IMU data is used to fill the gaps in the GNSS sampling interval (the position at the intermediate moment is calculated by integration), and visual odometry data is used to correct the cumulative error of the IMU (such as calibration when the drift exceeds 0.05 meters); when the GNSS signal is blocked or interfered with (PDOP > 3), visual odometry data is used as the main reference (based on the relative displacement of the conductor ground wire feature points), and IMU data is used to help maintain short-term stability, while recording the period of signal instability; after fusion, real-time pose data with centimeter-level accuracy is output (position error ≤ 5 cm, attitude angle error ≤ 0.5°), sorted by timestamp to form a continuous stream, including the three-dimensional coordinates, attitude angle and data confidence level (such as "high confidence" and "medium confidence") at each moment.

[0067] The specific implementation process of step 23 above is as follows:

[0068] Based on the real-time pose feedback stream from step 22, when the UAV's position fluctuation is ≤3 cm and attitude angle fluctuation is ≤0.3° for 3 consecutive seconds, it is determined to be in a stable hovering state, triggering the reference coordinate calibration process. According to the spatial distribution characteristics of the conductor and ground wire (such as the midpoint of a straight segment, the connection point of a tension segment, and the suspension point of an insulator), three physical points with significant characteristics (such as the connection node between the conductor / ground wire and the insulator, and the position of the clamp) are identified by the visual camera. For each feature point, combined with the UAV's current pose (position and attitude angle) and camera intrinsic parameters (focal length, pixel size), the three-dimensional coordinates of the feature point in the global coordinate system (consistent with the GIS coordinates in step 1) are calculated. The spatial position is inferred from the image pixel coordinates, combined with the pose... The drone coordinates in the feedback stream are transformed; 10 sets of coordinate data are continuously collected for each reference point, the two sets with the largest deviation are removed, and the average of the remaining 8 sets is taken as the final reference coordinates to ensure calibration accuracy (error ≤ 2 cm); the electromagnetic field strength of the surrounding area is collected in real time by the electromagnetic sensor on the drone, with a sampling frequency of 1 Hz, and the electric field strength (unit: V / m) and magnetic field strength (unit: μT) are recorded; the maximum, minimum and standard deviation of the electromagnetic field strength within 30 consecutive seconds are calculated to determine the fluctuation range (e.g., "electric field strength fluctuation ±50V / m"), and compared with the safety threshold (e.g., the anti-interference critical value of the inspection equipment) to mark the interference level (e.g., "slight interference" "moderate interference").

[0069] The specific implementation process of step 24 above is as follows:

[0070] Integrate the real-time pose feedback stream from step 22 (taking stable data during hovering), the three spatial reference coordinate sets from step 23 (including confidence scores for each coordinate), and electromagnetic field strength fluctuation data (including average value and fluctuation range); check the spatial consistency between the reference coordinate sets and the three-dimensional trajectory of the conductor (e.g., the distance from the reference point to the trajectory should be ≤10 cm), and recalibrate if the deviation is too large; verify the sampling integrity of the electromagnetic field data (missing data ≤5%), otherwise supplement the data; based on the pose feedback stream during hovering, calculate the standard deviation of the position coordinates (e.g., X-axis standard deviation 2.1 cm, Y-axis 1.8 cm, Z-axis 2.3 cm), and take the maximum value as the pose accuracy index (e.g., "pose accuracy: 2.3 cm"); calculate the fluctuation range of the attitude angle (e.g., pitch angle ±0.2°), as the attitude stability parameter; average the electromagnetic field strength. The value is compared with a preset threshold (e.g., a safety threshold of 500V / m around a 10kV line) to calculate the relative intensity (e.g., "measured 320V / m, 64% of the threshold"). Combined with the fluctuation range, a quantized value is output (e.g., "Electromagnetic interference intensity: medium, fluctuation ±40V / m"). The pose accuracy, electromagnetic interference intensity quantized value, and three reference coordinate sets (including coordinate values ​​and confidence levels) are integrated into a structured parameter set. An example format is: Pose accuracy: position error ≤ 2.3cm, attitude angle fluctuation ≤ ±0.2°; Electromagnetic interference intensity: average electric field 320V / m (threshold 64%), fluctuation ±40V / m; Reference coordinate sets: P1 (longitude XXX, latitude XXX, altitude XXX, confidence level 98%), P2 (...), P3 (...). This parameter set is then output as the input data for step 3.

[0071] In this embodiment of the invention, multi-degree-of-freedom motion commands are generated based on preset trajectories and safety boundaries to ensure that the UAV can accurately and safely reach the target groundline coordinates. This strictly follows path planning constraints while ensuring flight stability through refined control of speed and attitude. Multi-sensor fusion positioning technology combines the advantages of different devices (global positioning of GNSS, dynamic response of IMU, and local accuracy of visual odometry), effectively overcoming the limitations of single sensors (such as GNSS signal obstruction and IMU drift), and generating centimeter-level real-time pose feedback. The three dynamically calibrated spatial reference coordinates provide reliable spatial anchor points for subsequent 3D reconstruction, ensuring data consistency and accuracy. Synchronous monitoring of electromagnetic field fluctuations allows for timely understanding of environmental interference. The aggregated hovering state parameters (pose accuracy, electromagnetic interference, and reference coordinates) comprehensively reflect the hovering quality of the UAV and environmental conditions, providing key correction basis for subsequent scanning and data processing, and also providing quantitative standards for judging abnormal situations (such as insufficient positioning accuracy and strong interference), ensuring the effectiveness of inspection data.

[0072] In a preferred embodiment of the present invention, step 3, based on the reference coordinate set output in step 2, constructs a dynamic spatial feature envelope surface and performs region rasterization to form a multi-level scanning unit set, including:

[0073] Step 31: Use the set of reference coordinates output in Step 2 as a sequence of spatial control points, and generate a dynamic spatial feature envelope surface using a non-uniform rational B-spline surface reconstruction algorithm.

[0074] Step 32: Perform adaptive region rasterization division on the dynamic spatial feature envelope surface, generate a multi-level scanning unit set based on the curvature change gradient, and extract the deformation feature parameters of the envelope surface at the same time.

[0075] Step 33: Calculate the spatial structure adaptive compensation factor based on the deformation characteristic parameters, and combine it with the pose accuracy and electromagnetic interference intensity quantization values ​​in the hovering state parameters output in Step 2, including:

[0076] Step 331: Based on the deformation feature parameters output in step 32, extract the principal curvature distribution gradient and normal vector offset of the envelope surface; calculate the curvature-induced spatial topological distortion intensity based on the principal curvature distribution gradient, and generate a local coordinate system rotation correction vector based on the normal vector offset; fuse the spatial topological distortion intensity and the local coordinate system rotation correction vector, and output the spatial geometric distortion compensation amount through vector synthesis operation.

[0077] Step 332: Extract the pose accuracy from the hovering state parameters output in Step 2, and generate the pose drift compensation amount through the positioning error transfer function.

[0078] Step 333: Extract the electromagnetic interference intensity quantization value from the hovering state parameters output in Step 2, and generate the signal propagation compensation amount based on the electromagnetic wave attenuation characteristic model.

[0079] Step 334: The spatial geometric distortion compensation, pose drift compensation, and signal propagation compensation are fused together, and an adaptive compensation factor for the spatial structure is generated through nonlinear weighted aggregation; the scanning unit set is then scanned collaboratively by lidar and dual polarization radar.

[0080] Step 34: The original point cloud data, ambient temperature and humidity, wind speed and electromagnetic field strength obtained by scanning are fused together, and wavelet threshold denoising and Kalman filtering are used for data enhancement processing. The output is three-dimensional topology reconstruction data and icing feature vector corrected by compensation factor.

[0081] In this embodiment of the invention, the specific implementation process of step 31 is as follows:

[0082] From the hovering state parameters output in step 2, locate the "Reference Coordinate Set" field and extract the complete 3D coordinate data of the 3 spatial control points, including longitude (accurate to 0.00001°), latitude (same accuracy), and altitude (accurate to 0.01 meters). Simultaneously extract the confidence score (e.g., "98%" or "85%)) corresponding to each coordinate. Set the confidence threshold to 90% and compare the confidence of each control point one by one. If the confidence of a point is <90% (e.g., 85%), mark it as an outlier and remove it from the control point sequence, recording the reason for removal ("insufficient confidence"). Count the number of remaining valid control points after removal. If there are still 3 points and their spatial distribution is spaced 5 intervals along the conductor's direction... If the distance is -10 meters (horizontal distance calculated by coordinates), no supplementation is needed; if the number of control points is less than 3 or the interval exceeds 10 meters due to elimination, interpolation supplementation is triggered; for missing areas, select two adjacent valid control points (such as P1 and P3), and calculate the coordinates of the missing middle point (P2) according to the coordinate difference and distance between the two points using linear interpolation: for example, if the distance between P1 (X1,Y1,Z1) and P3 (X3,Y3,Z3) is 8 meters, and the missing point P2 is located 4 meters in the middle, then the coordinates of P2 are ((X1+X3) / 2, (Y1+Y3) / 2, (Z1+Z3) / 2), ensuring that the control point sequence is evenly distributed along the conductor line after supplementation (interval of 5-10 meters).

[0083] Analyze the spatial morphology of conductors and ground wires: For straight segments (height difference between adjacent towers < 5 meters, azimuth angle change < 5°), set the NURBS surface order to 3; for sag segments (height difference ≥ 5 meters or azimuth angle change ≥ 5°), also use order 3 (balancing smoothness and shape fitting ability) to ensure the surface has no sharp angle transitions; construct a triangular basic grid with 3 control points as vertices; adjust the grid density according to the complexity of the conductors and ground wires: add 1 grid node every 10 meters for straight segments (generated by interpolation), and add 1 node every 5 meters for sag segments (densifying the grid to capture sag details), the grid node coordinates are determined by interpolating the control point coordinates; input the control point sequence (including supplementary points) and control grid parameters into the NURBS surface reconstruction algorithm, the algorithm generates an initial continuous surface based on the spatial relationship of the control points, and the algorithm automatically fits the actual distribution characteristics of the conductors and ground wires: for the natural sag area of ​​the conductor, the surface bends with the sag trend (e.g., the midpoint height is lower than two...). For the insulator area, the curved surface conforms to the concave-convex shape of the skirt (local curvature is increased), and a weight is assigned to each control point: the weight of points with a confidence level of 90%-100% is set to 1.0, and the weight of points with a confidence level of 80%-90% is set to 0.8 (points with a confidence level of <80% have been removed). During the fitting process, high-weight control points have a greater impact on the shape of the curved surface, reducing their deviation from the curved surface; for each control point, the vertical distance from its three-dimensional coordinates to the fitted curved surface (i.e., the shortest distance from the point to the curved surface) is calculated, and the deviation values ​​of all points are recorded (e.g., P1 deviation is 2 cm, P2 deviation is 4 cm). If the maximum deviation is ≤3 cm, the envelope surface is judged to be qualified; if the maximum deviation is >5 cm (e.g., a point deviation is 6 cm), an intermediate control point is added (an interpolation control point is added near the point with the maximum deviation), the control mesh is regenerated and the curved surface is fitted; the deviation is calculated repeatedly until the deviation of all control points is ≤3 cm, ensuring that the envelope surface accurately reflects the spatial characteristics of the conductor and surrounding equipment.

[0084] The specific implementation process of step 32 above is as follows:

[0085] From the dynamic spatial feature envelope generated in step 31, the curvature value (a quantitative index of the curvature of the surface) is calculated by sampling at 2-meter intervals. The absolute value of curvature at each sampling point is recorded (e.g., the curvature of a point on a straight section of the conductor / ground wire is 0.05 / m, and the curvature of a point at the insulator connection is 0.2 / m). When the absolute value of curvature at the sampling point is ≤0.1 / m, it is determined to be a "smooth area" (e.g., a straight section of the conductor / ground wire, a smooth surface of the tower), and the corresponding grid size is set to 50×50 cm (to reduce the amount of data and improve scanning efficiency). When the absolute value of curvature at the sampling point is >0.1 / m, it is determined to be a "complex area" (e.g., the edge of the insulator skirt, the connection between the conductor and the clamp), and the corresponding grid size is set to 20×20 cm (to densify sampling and capture detailed features).

[0086] Along the length direction (direction of the ground line) and width direction (perpendicular to the direction) of the dynamic spatial feature envelope, the grid is cut according to the set grid size: flat areas are divided into grids every 50 cm, covering an area of ​​50 cm (length) × 50 cm (width); complex areas are divided into grids every 20 cm, covering an area of ​​20 cm (length) × 20 cm (width); each cut grid cell is assigned a unique ID and records: the three-dimensional coordinates of the cell center (calculated by averaging the coordinates of the four corners of the grid), the area type label ("flat area" or "complex area", determined based on the curvature threshold of the area), and the curvature value corresponding to the cell center (extracted from the curvature data of the envelope). Traverse all basic scan units and group them by region type: flat area units are grouped into one group, and complex area units are grouped into another group; aggregate adjacent units (spatially continuous and without gaps) within each group: flat area: every three consecutive 50×50 cm coarse grids are aggregated into one secondary unit, covering an area of ​​150 cm (length) × 50 cm (width); complex area: every two consecutive 20×20 cm fine grids are aggregated into one secondary unit, covering an area of ​​40 cm (length) × 20 cm (width).

[0087] The basic scanning units are integrated with the aggregated secondary scanning units to form a multi-level scanning unit set of "basic unit + secondary unit". The basic unit ID and center coordinates of each secondary unit are labeled. For each grid unit, the maximum and minimum curvatures (reflecting the degree of curvature of the surface in different directions) at the unit center are calculated using the envelope surface equation; for example, the maximum curvature of a complex region unit is 0.3 / m, and the minimum curvature is 0.15 / m. The normal vector direction (three-dimensional vector) perpendicular to the envelope surface of the unit is calculated and represented by vector coordinates (e.g., X-axis component 0.2). The Y-axis component is 0.1 and the Z-axis component is 0.9, reflecting the tilt angle of the unit surface; the height difference of the four edges of the unit is measured, and the ratio of the slope difference of adjacent edges to the distance is calculated to obtain the slope change rate (e.g., if the edge slope increases from 0.1 to 0.3, the change rate is 0.2 / 0.5 meters = 0.4 / m), reflecting the rate of change of the steepness of the surface; the principal curvature value, normal vector direction, and slope change rate of each unit are associated according to the spatial position coordinates (coordinates of the unit center) and summarized to form a set of deformation characteristic parameters, ensuring that each parameter can be traced back to the specific grid unit.

[0088] The specific implementation process of step 331 above is as follows:

[0089] From the set of deformation feature parameters output in step 32, filter the "principal curvature values" (maximum curvature, minimum curvature) and "normal vector direction" parameters for each scanning unit; for each unit, identify its four adjacent units (front, back, left, and right), and extract the principal curvature values ​​of the adjacent units (taking the maximum curvature); calculate the difference in principal curvature between the current unit and each adjacent unit (e.g., if the current unit's curvature is 0.2 / m and the adjacent unit's curvature is 0.1 / m, the difference is 0.1 / m); divide the difference by the distance between adjacent units (e.g., in a flat area of ​​50 cm, the distance is 0.5 m) to obtain the rate of change of curvature (i.e., gradient), and take the maximum value of all adjacent gradients as the principal curvature distribution gradient of the unit (e.g., 0.1 / m ÷ 0.5m = 0.2 / m). 2 ).

[0090] Normal vector offset calculation:

[0091] Extract the angle between the element normal vector and the Z-axis (vertical direction) of the global coordinate system. This angle value is the normal vector offset (e.g., if the angle between the normal vector and the Z-axis is 15°, then the offset is 15°). Set the gradient threshold to 0.05 / m² and classify the gradient of the element's principal curvature distribution: gradient > 0.05 / m² (curvature abrupt change region, such as the edge of an insulator): assign a distortion weight of 1.0; gradient ≤ 0.05 / m² (gradient region, such as a straight section of a conductor): assign a distortion weight of 0.3. For each element, multiply its principal curvature distribution gradient by the corresponding weight (e.g., gradient 0.2 / m² × weight 1.0 = 0.2). Calculate the weighted average gradient of all elements as the overall spatial topological distortion intensity (quantifying the irregularity of the surface; a higher value indicates more severe distortion). Based on the normal vector offset (the angle between the element normal vector and the Z-axis), calculate the rotation angle: if the offset is 15°, then set the rotation angle to 15° (to align the Z-axis of the local coordinate system with the direction of the element normal vector).

[0092] Based on the direction of the normal vector in the global coordinate system (e.g., biased towards the X-axis or Y-axis), the rotation angle is decomposed into rotation components around the X-axis and Y-axis: if the normal vector is biased towards the positive X-axis, the rotation component around the Y-axis is mainly allocated; if it is biased towards the Y-axis, the component around the X-axis is allocated; this is converted into a three-dimensional rotation vector (e.g., a 15° rotation around the Y-axis is represented as "X:0, Y:15°, Z:0"), used for subsequent coordinate system correction; the topological distortion intensity of the element (e.g., 0.2) is multiplied by the basic compensation coefficient (e.g., 5 cm, set according to historical data) to obtain the compensation amplitude (0.2 × 5 cm = 1 cm), the stronger the distortion, the larger the amplitude; the compensation direction is adjusted according to the direction of the rotation correction vector: if the rotation vector is 15° around the Y-axis, the compensation direction is along the axis of the local coordinate system after the rotation (e.g., the direction of tilting 15°); the amplitude and direction are fused to generate the three-dimensional geometric distortion compensation amount for each element (e.g., "X: +0.3 cm, Y: +0.8 cm, Z: -0.2 cm"), used to correct the geometric deviation of the scan data.

[0093] The specific implementation process of each sub-step in step 332 above is as follows:

[0094] From the hovering state parameters output in step 2, extract the specific data under the "pose accuracy" field: position error: X-axis standard deviation 2.1 cm, Y-axis 1.8 cm, Z-axis 2.3 cm (reflecting the positioning fluctuation range of the UAV in three-dimensional directions); attitude angle fluctuation: pitch angle ±0.2°, yaw angle ±0.3°, roll angle ±0.1° (reflecting the fluctuation range of the UAV's body tilt).

[0095] The error propagation function is established as follows:

[0096] The greater the relative distance between the drone and the scanning unit, the smaller the impact of the position error on the scanning point (the coefficient decreases as the distance increases). For example, the coefficient is 0.2 at a distance of 5 meters (an empirical value based on equipment accuracy testing). Multiply the position error by the attenuation coefficient. For example, the Z-axis position error is 2.3 cm × 0.2 = 0.46 cm (rounded to 0.5 cm), which means that the Z-axis scanning deviation is 0.5 cm at this distance.

[0097] Attitude angle fluctuation propagation calculation:

[0098] Based on trigonometric relationships (such as "deviation = distance × sin(angle)"), the angle fluctuation is converted into a linear deviation: if the relative distance between the UAV and the unit is 5 meters and the pitch angle fluctuates by ±0.2°, then the vertical deviation = 5 meters × sin(0.2°) ≈ 5 meters × 0.0035 ≈ 0.0175 meters = 1.75 centimeters, which is taken as 0.1 centimeters (actually scaled down proportionally because the angle fluctuation range is small).

[0099] The summary calculations for pose drift compensation are as follows:

[0100] Weights are assigned based on the relative distance between the scanning unit and the UAV: ​​close-range units ≤ 3 meters have a weight of 1.0, mid-range units 3-5 meters have a weight of 0.7, and far-range units > 5 meters have a weight of 0.4 (close-range units are more affected by pose errors); the position error propagation deviation is added to the attitude angle fluctuation propagation deviation to obtain the total original deviation of the unit (e.g., position deviation 0.5 cm + attitude deviation 0.1 cm = 0.6 cm); the total original deviation is multiplied by the unit weight to obtain the pose drift compensation amount for each scanning unit (e.g., close-range units 0.6 cm × 1.0 = 0.6 cm, far-range units 0.6 cm × 0.4 = 0.24 cm), which is used to correct scanning deviations caused by inaccurate UAV positioning.

[0101] The specific implementation process of step 333 above is as follows:

[0102] Extract the core data from the "Electromagnetic Interference Intensity" field from the hovering state parameters output in step 2:

[0103] Average electric field strength: 320V / m (reflecting the average level of interference); fluctuation range: ±40V / m (reflecting the stability of interference); interference level: "medium interference" (based on preset standards, such as 100-500V / m being considered medium).

[0104] Based on the interference level of "moderate interference," a preset "distance-intensity" attenuation model is matched (this model is suitable for scenarios with moderate and stable interference intensity). The attenuation is calculated using the reference coordinates and scanning unit coordinates from step 2, such as 3 meters (near-range unit) and 8 meters (far-range unit). Electric field strength value: Input the average electric field strength of 320V / m as the basic parameter for attenuation calculation. Based on the "distance-intensity" model, the attenuation deviation is calculated according to the following logic: Attenuation is directly proportional to electric field strength: the higher the strength, the greater the attenuation (320V / m corresponds to a base attenuation ratio of 10%); Attenuation is inversely proportional to the square of the distance: the farther the distance, the smaller the attenuation (3 meters corresponds to a distance coefficient of 1.0, and 8 meters corresponds to a coefficient of 0.2). Comprehensive calculation: Under moderate interference, the signal attenuation deviation at 3 meters = base attenuation ratio 10% × distance coefficient 1.0 = 10%; at 8 meters = 10% × 0.2 = 2% (in actual scenarios, fine-tuning is needed based on equipment characteristics, such as slightly lower attenuation for LiDAR optical signals and slightly higher attenuation for dual-polarization radar electromagnetic waves). The attenuation deviation is converted into a correction ratio for the signal amplitude: if the attenuation deviation is 10%, the compensation is "+10%" (i.e., the measured signal amplitude is increased by 10% to offset the attenuation effect). Unit compensation is correlated: adjusted according to the distance of the scanning unit and the interference fluctuation range: when the fluctuation range is ±40V / m, the compensation for close-range units (3 meters) is increased by 2% (total 12%), while the compensation for distant units (8 meters) remains at 2%, ultimately generating the signal propagation compensation for each scanning unit (e.g., "3-meter unit: +12%, 8-meter unit: +2%").

[0105] The specific implementation process of each sub-step in step 334 above is as follows:

[0106] Basic weight settings: Fixed weights are set according to the degree of deviation impact: Spatial geometric distortion compensation: 40% weight (dominantly affects device shape deviation, has the greatest impact); Pose drift compensation: 30% weight (dominantly affects positioning deviation); Signal propagation compensation: 30% weight (dominantly affects signal amplitude deviation). Nonlinear weighted adjustment: The weight of compensation amounts with deviation values ​​exceeding the threshold is dynamically increased: If the geometric distortion compensation amount of a certain unit is >5 cm (outside the normal range), its weight is increased from 40% to 50%; if the signal propagation compensation amount is >15% (strong attenuation), its weight is increased from 30% to 40%. Compensation factor calculation: The three compensation amounts are weighted and summed according to the final weight to generate the spatial structure adaptive compensation factor for each scanning unit (e.g., "Geometric compensation 3 cm × 50% + Pose compensation 2 cm × 30% + Signal compensation 1.5 cm × 20% = 2.4 cm").

[0107] The collaborative scanning control (path planning and parameter setting) is as follows:

[0108] The path planning is based on a multi-level scanning unit set (basic unit + secondary unit): Prioritize scanning fine grid units (20×20 cm) in complex areas, proceeding sequentially from the tower to the midpoint of the span; in flat areas, coarse grid units (50×50 cm) are scanned continuously in straight segments to reduce backtracking. The lidar uses high-frequency scanning (100 points / second, acquiring 1 point every 0.01 seconds to ensure detailed coverage), and the dual-polarization radar uses continuous wave detection (continuous signal transmission to improve the signal-to-noise ratio); the lidar also uses low-frequency scanning (50 points / second, acquiring 1 point every 0.02 seconds to reduce redundant data), and the dual-polarization radar uses pulse detection (transmitting a signal every 0.05 seconds to save power). A synchronization signal is set through the UAV control system: the scanning start time deviation between the lidar and the dual-polarization radar is ≤0.01 seconds, ensuring that the optical and electromagnetic wave data of the same scanning unit are aligned in time, allowing for direct correlation analysis later.

[0109] The specific implementation process of step 34 above is as follows:

[0110] The system collects raw point cloud data for each scanning unit, including 3D coordinates (X / Y / Z accurate to 0.01 meters) and reflection intensity values ​​(0-255 quantization values, reflecting the reflectivity of the object's surface); records the polarization degree (0-1, reflecting the crystal structure of the ice covering) and reflectivity (0-100%, reflecting the intensity of electromagnetic wave reflection) for each unit; it also collects real-time data on temperature and humidity (25℃, 60%RH), wind speed (3m / s), and electromagnetic field strength (320V / m±40V / m), with the sampling frequency consistent with the radar scanning frequency (10Hz); using the scanning unit ID as the core keyword, it aligns the lidar data, dual-polarization radar data, and environmental data according to timestamps (time deviation ≤0.01 seconds); and establishes a "scanning unit ID-multi-source data" association table to ensure that the 3D coordinates, reflection characteristics, and environmental parameters of each unit correspond one-to-one (e.g., "unit ID:101" is associated with its point cloud coordinates, polarization degree, wind speed, etc.).

[0111] Multi-scale wavelet decomposition (e.g., decomposition into 3 layers) is performed on the raw point cloud data of the lidar to divide the data into low-frequency approximate components (reflecting the main structure) and high-frequency detail components (containing noise); a threshold for the high-frequency components is set (based on the noise fluctuation range of historical normal data, such as twice the standard deviation of reflection intensity); signals in the high-frequency components that exceed the threshold (such as isolated points where the reflection intensity suddenly jumps from 50 to 200) are marked as noise points and directly removed; the low-frequency components and the high-frequency components that do not exceed the threshold are retained to reconstruct the denoised point cloud data.

[0112] Based on the reflectivity range (e.g., 20%-60%) and polarization range (e.g., 0.3-0.7) of normal equipment (such as clean conductors and insulators), set the filtering threshold:

[0113] Signals with reflectivity <10% or >80% (abnormally low values ​​or clutter caused by electromagnetic interference); signals with polarization degree <0.1 or >0.9 (abnormal values ​​not caused by icing or strong interference). Threshold filtering is performed: traversing dual-polarization radar data, signals meeting the above criteria are marked as clutter and removed, retaining valid signals within the normal threshold range. A coordinate smoothing model is established based on environmental parameters, the core content of which is:

[0114] The higher the wind speed (e.g., 3 m / s), the more severe the conductor vibration, and the higher the smoothing coefficient (e.g., setting the smoothing coefficient to 0.6, which is 0.3 in static conditions). Temperature and humidity affect conductor sag; sag is stable at 25℃. Smoothing should preserve the normal sag trend to avoid over-correction. Dynamic vibration error correction:

[0115] For continuously scanned point cloud coordinates (e.g., 10 points per second), a sliding window averaging method (window size 5 points) is used, with weighted calculation based on a smoothing coefficient: Current point coordinate = (coordinates of the previous 2 points × 0.2 + current point coordinate × 0.4 + coordinates of the next 2 points × 0.2) × smoothing coefficient; after correction, the coordinate fluctuation at a wind speed of 3 m / s is reduced from ±5 cm to ±2 cm, reducing positional deviation caused by jitter. Point cloud coordinate geometric deviation correction: Extract the spatial structure adaptive compensation factor for each scanning unit generated in step 334 (e.g., "X: +0.3 cm, Y: +0.8 cm, Z: -0.2 cm"); superimpose and correct the denoised point cloud coordinates: corrected coordinates = original coordinates + corresponding component of the compensation factor (e.g., original X = 100.5 cm, corrected = 100.5 + 0.3 = 100.8 cm). Radar reflectivity amplitude deviation adjustment: Based on the signal propagation compensation (e.g., +12%), the dual-polarization radar reflectivity is proportionally adjusted: corrected reflectivity = original reflectivity × (1 + compensation ratio) (e.g., original reflectivity 50%, corrected = 50% × 1.12 = 56%). The corrected point cloud coordinates are stitched together according to the scanning units to generate complete three-dimensional structural data of the equipment, including: the precise sag shape of the conductor and ground wire (one coordinate point per meter), the three-dimensional contour of the insulator skirt (edge ​​coordinate error ≤ 2 cm), and the spatial positional relationship of the tower connection parts.

[0116] Key parameters are extracted from the corrected dual-polarization radar data to form an icing feature vector:

[0117] Ice thickness: calculated by combining reflectivity and lidar point cloud thickness (higher reflectivity and greater point cloud thickness indicate thicker ice); Ice density: based on polarization degree (polarization degree 0.5-0.7 corresponds to medium density, <0.5 is low density); Polarization characteristics: recording the trend of polarization degree change with scanning angle (reflecting ice uniformity). Final output: outputting the above data in structured dataset form, labeling the confidence level of each parameter (based on noise reduction and correction effects, such as "high confidence" or "medium confidence").

[0118] This invention achieves comprehensive integration of equipment 3D morphology, electromagnetic reflection characteristics, and environmental influencing factors by associating data from lidar, dual-polarization radar, and environmental sensors. Wavelet threshold denoising effectively eliminates isolated noise in point clouds, and threshold filtering reduces radar clutter interference, significantly improving the purity of the original data and avoiding feature misjudgment caused by noise (such as misjudging clutter as icing). By combining environmental parameters such as wind speed to smooth coordinates, dynamic interference such as conductor jitter is specifically reduced, and position fluctuations are controlled at the centimeter level, ensuring the stability of 3D structure reconstruction. Adaptive compensation factors are applied to correct geometric and amplitude deviations, eliminating the effects of surface distortion, pose drift, and electromagnetic attenuation, making the equipment's 3D topology data highly consistent with the actual morphology (error ≤ 2 cm).

[0119] In a preferred embodiment of the present invention, an adaptive compensation factor is generated based on the deformation characteristics of the envelope surface. Combined with hovering state parameters, the track surface is scanned and processed by a combination of lidar and dual-polarization radar to obtain corrected three-dimensional topology reconstruction data and icing feature vectors, including:

[0120] Step 335: In response to the coordinated scanning of the scanning unit set by the lidar and the dual-polarization radar, the original point cloud dataset and dual-polarization echo signal stream are acquired, and the time series data of ambient temperature and humidity, wind speed and electromagnetic field intensity are collected simultaneously.

[0121] Step 336: Based on the original point cloud dataset obtained in step 335, perform spatial coordinate transformation correction using the spatial structure adaptive compensation factor generated in step 334 to generate a geometrically consistent point cloud set.

[0122] Step 337: Perform wavelet threshold denoising on the geometrically consistent point cloud set output in step 336 to eliminate high-frequency vibration noise; at the same time, based on the fusion of the dual-polarization echo signal stream and environmental time series data obtained in step 335 using Kalman filtering, generate an enhanced radar reflection feature matrix.

[0123] Step 338: Extract the dielectric constant distribution spectrum and surface scattering characteristic vector from the enhanced radar reflection feature matrix output in step 337, and combine them with the electromagnetic interference intensity quantization value in the hovering state parameters output in step 2 to construct the icing feature vector.

[0124] Step 339: Perform 3D surface topology reconstruction on the denoised geometrically consistent point cloud set output in step 337, and output 3D topology reconstruction data corrected by compensation factor.

[0125] In this embodiment of the invention, the specific implementation process of step 335 is as follows:

[0126] When the lidar and dual-polarization radar begin to scan the scanning unit set in a coordinated manner, the system simultaneously starts data acquisition: collecting the raw point cloud data output by the lidar (including the three-dimensional coordinates and reflection intensity information of each scanning point); collecting the echo signal stream of the dual-polarization radar (recording the raw signal data of electromagnetic wave reflection); and simultaneously triggering environmental sensors to continuously collect time-series data of temperature and humidity (such as real-time recording of temperature and relative humidity values), wind speed (real-time wind speed magnitude), and electromagnetic field intensity (recording the change value of electromagnetic field intensity in chronological order), ensuring that all data are aligned and correlated in the time dimension.

[0127] The specific implementation process of step 336 above is as follows:

[0128] Based on the original point cloud dataset obtained in step 335, the spatial structure adaptive compensation factor generated in step 334 is called to adjust the three-dimensional coordinates of each point in the original point cloud: according to the coordinate correction rules in the compensation factor, the X, Y, and Z axis coordinates of the original point cloud are transformed and corrected respectively to eliminate coordinate deviations caused by geometric distortion, pose drift, etc.; after correction, all point cloud data are kept consistent in spatial geometric relationship to form a geometrically consistent point cloud set.

[0129] The specific implementation process of step 337 above is as follows:

[0130] For the geometrically consistent point cloud set output in step 336, a wavelet threshold denoising method is adopted: by setting a reasonable threshold, noise points (such as isolated points with abnormal reflection intensity) caused by high-frequency vibration (such as equipment shaking, environmental interference) are screened and removed from the point cloud, while retaining the effective point cloud data; at the same time, the dual-polarization echo signal stream and environmental time series data (temperature, humidity, wind speed, electromagnetic field strength, etc.) obtained in step 335 are fused using the Kalman filter algorithm: by dynamically adjusting the signal weights through the filtering model, the influence of noise such as electromagnetic interference on the echo signal is suppressed, the effective signal characteristics are enhanced, and finally an enhanced radar reflection feature matrix containing clear reflection characteristics is generated.

[0131] The specific implementation process of step 338 above is as follows:

[0132] The enhanced radar reflection feature matrix output in step 337 is analyzed. The matrix region is divided according to the scanning unit, and the feature parameters related to icing are extracted unit by unit: Spatial distribution spectrum of dielectric constant: The dielectric constant of different scanning positions is calculated by using the electromagnetic wave reflection amplitude and phase change data in the matrix (reflecting the insulation properties of the material; the dielectric constant of icing is significantly different from that of air and conductors), and the distribution spectrum is formed by arranging them according to spatial coordinates (e.g., the dielectric constant of a certain area on the conductor surface is 3.5, and the dielectric constant of the icing area is 6.0); Surface scattering characteristic vector: The reflection direction and energy distribution law of radar waves on the scanning surface are analyzed, and parameters such as scattering intensity and scattering angle distribution are extracted to form vector data (e.g., the scattering intensity of the icing surface is 20% higher than that of the clean conductor, and the scattering angle range is wider).

[0133] Retrieve the electromagnetic interference intensity quantification value from the hovering state parameters in step 2: average electric field strength (e.g., 320V / m) and fluctuation range (±40V / m) to determine the degree of interference's impact on the radar signal (e.g., moderate interference causing a ±5% deviation in the reflected signal amplitude); correct the extracted dielectric constant distribution spectrum and scattering characteristic vector: subtract the system error value caused by interference (e.g., 0.2) from the dielectric constant calculation result, and adjust the scattering intensity according to the fluctuation range ratio (e.g., +5% correction) to eliminate the influence of electromagnetic interference on the deviation of characteristic parameters; based on the corrected dielectric constant distribution spectrum, identify the icing region (regions with dielectric constant > 5.0), and calculate the region thickness (the higher the dielectric constant, the greater the thickness, e.g., a dielectric constant of 6.0 corresponds to a thickness of 5 mm); combine with the surface scattering characteristic vector to analyze the surface roughness of the icing (the larger the scattering angle range, the rougher the icing), and infer the icing density (rough surfaces correspond to low-density, fluffy icing, and smooth surfaces correspond to high-density, hard icing); integrate the icing thickness, density, average dielectric constant, scattering characteristic parameters, etc., and arrange them in the order of scanning units to form a complete icing characteristic vector.

[0134] The specific implementation process of step 339 above is as follows:

[0135] The denoised geometrically consistent point cloud set output in step 337 is preprocessed as follows: a small number of residual noise points (such as isolated points with a distance deviation of >3 cm from surrounding points) are removed, and the points are grouped according to scanning units. The spatial coordinate range of each group of point clouds is marked (such as 10-15 meters on the X-axis and 20-25 meters on the Y-axis). The point cloud data of adjacent scanning units are stitched together. By matching feature points in the overlapping areas (such as continuous points on the conductor surface), the coordinate deviation between units is eliminated to form an overall point cloud model. A surface fitting algorithm (such as moving least squares method) is used to smooth the stitched point cloud and construct a continuous three-dimensional surface model: the conductor part is fitted as a smooth cylindrical surface, and the insulator part is fitted as a curved surface with the concave and convex structure of the umbrella skirt to restore the actual shape of the equipment.

[0136] The spatial structure adaptive compensation factor generated in step 334 is used to locally correct the fitted 3D surface model: for areas with curvature distortion (such as the edge of the insulator skirt), the surface coordinates are adjusted according to the compensation factor (e.g., offset outward by 0.3 cm); for areas affected by pose drift (such as point clouds near the tower), the height direction deviation is corrected (e.g., adjusted upward by 0.2 cm) to ensure that the model is consistent with the spatial topology of the actual equipment; the final generated 3D topology reconstruction data includes: the accurate 3D coordinates of each component of the equipment (error ≤ 2 cm), the surface curvature distribution, and the connection relationship between components (e.g., the connection position between the conductor and the insulator), output in a structured format, and labeled with the data confidence level (e.g., "high confidence area: straight section of conductor" "medium confidence area: insulator skirt").

[0137] This invention extracts the dielectric constant distribution spectrum and surface scattering characteristic vector from the enhanced radar reflection feature matrix, directly capturing the differences in physical properties (such as dielectric constant and scattering patterns) between iced and non-iced regions. It then corrects the feature parameters by combining the electromagnetic interference quantization value in the hovering state parameters, effectively offsetting electromagnetic interference to the radar signal and avoiding misjudgments of icing characteristics due to signal deviations (such as misjudging interference signals as icing). By integrating the corrected dielectric properties, scattering patterns, and other parameters, a feature vector containing key indicators such as icing thickness and density is formed, achieving a multi-dimensional quantitative description of the icing state. Finally, it uses point cloud stitching... By fitting the surface, discrete point cloud data is transformed into a continuous three-dimensional surface model, fully restoring the spatial structural morphology of line equipment (such as conductors and insulators). During the reconstruction process, an adaptive compensation factor is applied to correct geometric deviations and eliminate the influence of factors such as surface distortion and pose drift, ensuring that the three-dimensional topology reconstruction data is highly consistent with the actual spatial topology of the equipment (error controlled within the centimeter level). The final output three-dimensional topology reconstruction data includes the precise coordinates of the equipment, surface curvature, and component connection relationships, providing a reliable spatial reference for structural defect identification (such as conductor deformation and insulator damage) and dimensional measurement needs in line inspection.

[0138] In a preferred embodiment of the present invention, step 4 involves inputting the corrected 3D topology reconstruction data and icing feature vectors into a pre-trained ResNet deep learning model to identify icing thickness distribution, insulator damage, or suspended foreign object anomalies, and outputting quantified anomaly type, spatial coordinates, and confidence score, including:

[0139] Step 41: Perform spatial grid normalization on the 3D topology reconstruction data corrected by the compensation factor output in step 339 to generate a topology tensor; at the same time, embed the ice accretion feature vector output in step 338 into the complex domain feature space to generate an enhanced ice accretion feature matrix.

[0140] Step 42: Combine the topological tensor generated in step 41 with the enhanced icing feature matrix, and construct a multimodal input data cube using a spatial feature alignment algorithm;

[0141] Step 43: Input the multimodal input data cube into the pre-trained ResNet-101 deep learning model, extract the spatial topological features and dielectric properties correlation mapping through three layers of residual convolutional blocks, and output the anomaly probability distribution heatmap.

[0142] Step 44: Based on the anomaly probability distribution heatmap, the spatial regions of ice thickness distribution, insulator damage and suspended foreign objects are identified by Gaussian mixture model clustering. Combined with the spatial coordinates in the three-dimensional topology reconstruction data in step 339, the quantitative anomaly type, the coordinates of the center point of the anomaly region and the confidence score are generated.

[0143] In this embodiment of the invention, the specific implementation process of step 41 is as follows:

[0144] Set uniform spatial grid parameters (e.g., grid size of 10×10×5 cm, covering the entire inspection area), and map the spatial coordinates of the 3D topology reconstruction data output in step 339 to this grid to ensure the standardization of coordinate ranges for different equipment areas; extract the topological features of each grid cell (e.g., surface curvature, structure type label: conductor / insulator / tower), arrange the feature values ​​in order of grid position, and form a topological tensor with dimension [grid length × grid width × grid height × number of features]; perform complex domain mapping on the icing feature vector (including parameters such as thickness, density, and dielectric constant) output in step 338: decompose each feature value into a real part (e.g., absolute value of icing thickness) and an imaginary part (e.g., thickness change rate) to enhance the dimension of feature expression; arrange the complex domain features into a matrix according to the spatial coordinate order of the scanning cells, with the matrix rows / columns corresponding to spatial positions and the elements being complex feature values, to generate an enhanced icing feature matrix.

[0145] The specific implementation process of step 42 above is as follows:

[0146] Using the spatial coordinates of the 3D topology reconstruction data as a reference, coordinate matching is performed on the topology tensor and the enhanced icing feature matrix: through the coordinate mapping algorithm, it is ensured that the topology features (from the tensor) and the icing features (from the matrix) at the same physical location correspond one-to-one in the spatial dimension, eliminating positional deviation; the aligned topology tensor (spatial topology features) and the enhanced icing feature matrix (icing physical properties) are fused according to the channel dimension: the topology tensor is used as the "structure channel" and the icing matrix is ​​used as the "icing channel", which are combined to form a multimodal input data cube with dimensions of [spatial length × spatial width × spatial height × number of modes (2 types)], which completely preserves the spatial location and multi-feature association.

[0147] The specific implementation process of step 43 above is as follows:

[0148] A cube of multimodal input data is fed into a pre-trained ResNet-101 model. Basic spatial features (such as equipment edges and contours) are extracted through a first residual convolutional block (containing 3×3 convolution, batch normalization, and ReLU activation). A second residual block enhances local detail features (such as insulator skirt texture and icing surface undulations). A third residual block fuses global features and establishes a mapping between spatial topological features (such as conductor sag shape) and dielectric properties (such as the distribution of icing dielectric constant). After processing by the three residual blocks, the anomaly probability (between 0 and 1, with higher values ​​indicating a greater likelihood of anomaly) of each spatial grid cell is calculated through the model output layer (containing 1×1 convolution and sigmoid activation function). A two-dimensional anomaly probability distribution heatmap is generated according to the grid position to visually display potential anomaly areas.

[0149] The specific construction process of the above pre-trained ResNet-101 model is as follows:

[0150] Collect multi-source labeled data from distribution network line inspections, including: 3D topology reconstruction data (containing the 3D structure of conductors and insulators in normal and abnormal states), icing feature vectors (labeling parameters such as icing thickness and density), and abnormal label data (manually labeled areas with excessive icing, insulator damage locations, types of suspended foreign objects, and coordinates), ensuring a balanced ratio of normal to abnormal samples (e.g., 1:1); perform enhancement processing on the original data: randomly rotate (±10°), scale (0.8-1.2 times), and locally crop the 3D topology data (focusing on key areas such as insulator skirts); add slight noise (±5%) to the icing feature vectors to simulate measurement errors and increase dataset diversity; normalize the spatial coordinates of the 3D topology data to a uniform range. (e.g., [-1,1]), the icing feature vector is standardized by mean-standard deviation to ensure consistent input data scale; the training set (70%), validation set (20%), and test set (10%) are divided; the input layer is set to receive multimodal data cubes, and the dimensions are matched to the structure output in step 42 (e.g., [64×64×32×2], where 64×64×32 is the spatial grid size, and 2 is the number of modalities: topology + icing features), and the input channels are mapped to 64-dimensional feature channels through convolutional layers; a 7×7 convolutional layer (stride 2) is added to perform preliminary feature extraction on the input data to capture basic edge and contour information; a 3×3 max pooling layer (stride 2) is connected to compress the spatial dimension, reduce the amount of computation, and the output feature map size is 1 / 4 of the input.

[0151] Residual block structure: Two types of residual blocks are designed:

[0152] Ordinary residual block: contains two 3×3 convolutional layers (both with batch normalization and ReLU activation), with the same number of input and output channels; skip connections directly add the input features to the convolutional output, alleviating the gradient vanishing problem.

[0153] Downsampling residual block: The stride of the first 3×3 convolutional layer is set to 2 (to achieve downsampling), and the number of input channels is half of the output; the skip connection adjusts the number of channels through a 1×1 convolutional layer and adds it to the output to ensure dimension matching.

[0154] Network stage division: ResNet-101 contains 4 residual convolutional stages, with the number of residual blocks set sequentially as 3, 4, 23, and 3.

[0155] Phase 1: 3 ordinary residual blocks, 256 output feature channels;

[0156] Phase 2: 4 residual blocks (including 1 downsampling block), output feature channels 512;

[0157] Phase 3: 23 residual blocks (including 1 downsampling block), output feature channels 1024 (core feature extraction stage, capturing complex topology and icing-related features);

[0158] Stage 4: 3 residual blocks (including 1 downsampling block), output feature channels number 2048.

[0159] A global average pooling layer is added after four residual stages to compress the high-dimensional feature map into a 2048-dimensional global feature vector, fusing global correlation information between spatial topology and dielectric properties.

[0160] Output layer construction: connecting the fully connected layer and the classification head:

[0161] The first fully connected layer maps 2048-dimensional features to 1024-dimensional features, followed by ReLU activation and dropout (probability 0.5) to prevent overfitting; the output layer adopts a multi-label classification design, and outputs through three parallel convolutional layers: an abnormal ice thickness probability map, an insulator damage probability map, and a suspended foreign object probability map, with each map dimension corresponding to the input space grid (e.g., 64×64).

[0162] Pre-training process implementation:

[0163] Loss function selection: A weighted cross-entropy loss function is used, assigning higher weights to anomalous samples (ice accumulation, damage, foreign objects) (e.g., normal samples weight 1.0, anomalous samples weight 2.0) to address class imbalance. The Adam optimizer is used with an initial learning rate of 0.001, employing cosine annealing scheduling (the learning rate decays to half its current value every 10 epochs). The batch size is set to 32, and the total training epochs are 100. After each training epoch, performance is evaluated on the validation set (accuracy and recall for anomaly detection are calculated), and the model with the lowest validation set loss is saved as the pre-trained base model. If the validation set loss is consistently low for 10 epochs... If there is no performance improvement, training is terminated early. The pre-trained model is then fine-tuned on a specific dataset for distribution network inspection: the parameters of the first two residual stages are frozen (to retain the ability to extract general features), and only the last two stages and the output layer are trained; the learning rate is reduced to 0.0001, and the model is trained for 20 rounds to enhance its ability to specifically identify abnormal features of line equipment. The performance of the fine-tuned model is verified on the test set. If the accuracy of identifying a certain type of abnormality (such as small foreign objects) is low, the training weights of that type of sample are increased accordingly. The fine-tuning is repeated until the accuracy on the test set is ≥90%. Finally, the pre-trained ResNet-101 model is determined to be used for abnormal identification in step 4.

[0164] The specific implementation process of step 44 above is as follows:

[0165] Based on the anomaly probability distribution heatmap, a probability threshold is set (e.g., areas with a probability > 0.6 are considered candidate anomaly areas). A Gaussian mixture model is used to cluster the candidate areas: they are divided into 3 categories according to feature differences (icing anomaly, insulator damage, and suspended foreign objects). Icing anomaly areas correspond to high dielectric constant + continuous spatial distribution characteristics, insulator damage corresponds to local structural abrupt changes + low reflectivity characteristics, and suspended foreign objects correspond to isolated high-probability points + non-equipment structural characteristics. For each clustered area, the anomaly type is determined (matching a preset feature template: e.g., "high icing thickness" corresponds to the icing type). The coordinates of the center point of the area are calculated (taking the average of all grid coordinates in the area). The average anomaly probability in the area is used as the confidence score (e.g., an average probability of 0.85 corresponds to a confidence score of 85%). Combined with the precise coordinates in the 3D topology reconstruction data in step 339, the structured results are output: quantified anomaly type (e.g., "excessive icing thickness" or "insulator damage"), 3D coordinates (longitude / latitude / altitude) of the center point of the anomaly area, and confidence score.

[0166] In a preferred embodiment of the present invention, step 5, based on the quantified anomaly type, spatial coordinates, and confidence score output in step 4, generates an adaptive job strategy including tool ID encoding and job parameters, including:

[0167] Step 51: Analyze the quantitative anomaly type, anomaly region center point coordinates, and confidence score output in Step 44, and match the tool ID encoding using the preset anomaly-tool mapping rule base;

[0168] Step 52: Based on the coordinates of the center point of the abnormal area and combined with the three-dimensional flight trajectory of Step 1, generate the drone approach path, and dynamically adjust the operation safety distance threshold according to the confidence score.

[0169] Step 53: Combine the anomaly type from step 44 with the electromagnetic interference intensity quantification value from the hovering state parameters in step 2, and calculate the laser power, robotic arm operating torque, and de-icing time using the operation parameter optimization function;

[0170] Step 54: Aggregate tool ID encoding, UAV approach path, operational safety distance threshold, and operational parameters to generate an adaptive operational strategy instruction set containing spatiotemporal constraints; calculate laser power, robotic arm operating torque, and de-icing time using operational parameter optimization functions, including:

[0171] The quantitative anomaly type output in step 44 is analyzed, and the baseline value of the physical action parameter of the target tool is determined based on the preset anomaly physical characteristic mapping rule. The quantified value of electromagnetic interference intensity in the hovering state parameter output in step 2 is extracted, and the baseline value of the physical action parameter is dynamically corrected through the electromagnetic attenuation compensation model to generate intermediate values ​​of anti-interference operation parameters. According to the confidence score output in step 44, the dynamic scaling coefficient of the operation parameter is calculated through the safety margin adjustment function. The intermediate value of the anti-interference operation parameter and the dynamic scaling coefficient are weighted and fused: for the icing anomaly type, the laser power parameter and de-icing time are output by combining the dielectric constant distribution spectrum in the icing feature vector in step 338; for the insulator damage or hanging foreign object type, the robotic arm operating torque parameter is output by combining the surface curvature feature in the three-dimensional topology reconstruction data in step 339. The laser power, robotic arm operating torque and de-icing time parameters are aggregated to generate the operation parameter optimization result.

[0172] In this embodiment of the invention, step 51 is specifically implemented as follows:

[0173] Extract key information from the output of step 44: quantify the anomaly type (e.g., "excessive ice thickness", "damaged insulator", "suspended foreign object"), coordinates of the center point of the anomaly area (3D coordinate value), and confidence score (e.g., 85%), and classify them by type (e.g., group all "ice-related anomalies" into one category); call the preset "anomaly-tool mapping rule base", which stores the association between anomaly types and corresponding operating tools (e.g., "ice-related anomaly" corresponds to a laser de-icing tool, coded "TOOL-ICE-001"; "suspended foreign object" corresponds to a robotic arm grasping tool, coded "TOOL-ARM-002"); accurately match the corresponding tool ID code according to the parsed anomaly type, and record the associated tool code for each anomaly type.

[0174] The specific implementation process of step 52 above is as follows:

[0175] Using the coordinates of the center point of the abnormal area output in step 44 as the target point, the three-dimensional flight trajectory generated in step 1 is retrieved as the basic path framework. A path planning algorithm (such as the A* algorithm) is used to plan the shortest approach path from the current hovering point to the target point within the trajectory framework. The path must avoid the inspection safety boundaries in step 1 (such as dangerous areas far from towers and conductors), ensuring a smooth path without sharp turns. A basic safety distance threshold is set (e.g., 5 meters under normal conditions). The threshold is adjusted based on the confidence score in step 44: when the confidence score is ≥90%, the threshold is lowered by 10% (e.g., 4.5 meters, to improve operational accuracy); when the score is <70%, the threshold is increased by 20% (e.g., 6 meters, to increase safety redundancy); when the score is between 70% and 90%, the basic threshold is maintained. Finally, the operational safety distance threshold for each abnormal area is determined.

[0176] The specific implementation process of step 53 above is as follows:

[0177] Determine the anomaly type in step 44 (e.g., "icing anomaly" or "insulator damage"), and retrieve the quantified value of electromagnetic interference intensity from the hovering state parameters in step 2 (e.g., electric field strength 320V / m, fluctuation range ±40V / m). Using the operation parameter optimization function, combine the icing thickness (from the icing feature vector in step 338) and electromagnetic interference intensity to calculate the laser power (the stronger the interference, the higher the power should be to compensate for attenuation) and the de-icing time (the greater the thickness, the longer the time). Combine the structural strength of the abnormal area (from the 3D topology reconstruction data in step 339) and the electromagnetic interference intensity to calculate the robotic arm operating torque (avoiding excessive torque under strong interference that could damage the equipment) to ensure smooth operation.

[0178] The specific implementation process of step 54 above is as follows:

[0179] Collect the tool ID code from step 51, the drone approach path and safe operating distance threshold from step 52, and the operating parameters (laser power, robotic arm torque, de-icing time) from step 53, and group them according to "anomaly type-tool-path-parameter". Add spatiotemporal constraints to each group of information: in terms of time, sort the operation priority according to the anomaly confidence (high confidence anomalies are executed first); in terms of space, determine the start and end times of the approach path and the effective range of the safe distance. Finally, integrate them into a structured adaptive operation strategy instruction set, which includes executable information such as tool call order, path execution nodes, and parameter adjustment timing.

[0180] The specific implementation process of the "job parameter optimization function" in step 54 is as follows:

[0181] The process involves analyzing the quantified anomaly type from step 44 and calling the preset "anomaly physical characteristic mapping rules." Based on the icing thickness level (e.g., "5-10mm thick icing"), a laser power reference value (e.g., 30W) and a de-icing time reference value (e.g., 60 seconds) are set. Based on the anomaly size (e.g., "5cm diameter foreign object"), a robotic arm operating torque reference value (e.g., 5N·m) is set. The electromagnetic interference intensity quantification value from step 2 is extracted, and the reference value is corrected using an electromagnetic attenuation compensation model: for every 100V / m increase in electromagnetic interference intensity, the laser power reference value increases by 5% (to offset signal attenuation), and the robotic arm torque reference value decreases by 3% (to avoid misoperation due to electromagnetic interference), generating intermediate values ​​for anti-interference operation parameters. The process also involves calculating the reduction based on the confidence score from step 44. The coefficient for performance evaluation is 1.1 for scores above 90% (improving work efficiency), 1.0 for scores between 70% and 90% (maintaining standards), and 0.9 for scores below 70% (reducing intensity to ensure safety). The final laser power is determined by multiplying the median anti-interference laser power by the scaling factor and combining it with the dielectric constant distribution spectrum in the icing feature vector from step 338 (the higher the dielectric constant, the more power fine-tuning is increased by 2%). The de-icing time is calculated and output as "thickness × scaling factor". The final operating torque is determined by multiplying the median anti-interference robotic arm torque by the scaling factor and combining it with the surface curvature of the 3D topology reconstruction data from step 339 (the greater the curvature, the less torque fine-tuning is decreased by 1%). The laser power, robotic arm torque, and de-icing time are integrated to generate optimized work parameters.

[0182] This invention combines three-dimensional trajectory generation to create the optimal approach path, while dynamically adjusting the safety distance based on confidence level. At high confidence levels, the distance is shortened to improve accuracy, while at low confidence levels, the distance is increased to reduce risk, balancing efficiency and safety. It integrates electromagnetic interference intensity, icing characteristics, and topological features to optimize parameters, with laser power offsetting electromagnetic attenuation and robotic arm torque adapting to surface curvature, ensuring stable and reliable operation in complex environments. Furthermore, by correcting parameter baseline values ​​through an electromagnetic attenuation compensation model, it effectively counteracts the impact of electromagnetic interference on laser power and robotic arm torque, avoiding operational failures caused by environmental interference.

[0183] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for autonomous inspection and processing of overhead power distribution line equipment using unmanned aerial vehicles (UAVs), characterized in that, The method includes: Based on the geographic coordinates and topology data of overhead power distribution line equipment, a three-dimensional flight trajectory and inspection safety boundary of the UAV are generated. Based on the three-dimensional flight trajectory and inspection safety boundary, the UAV is controlled to reach the target ground line coordinates. Centimeter-level dynamic hovering is achieved through multi-sensor fusion positioning. In the hovering state, three spatial reference coordinates are dynamically calibrated and hovering state parameters are output, including pose accuracy, electromagnetic interference intensity and reference coordinate set. Based on the reference coordinate set, a dynamic spatial feature envelope surface is constructed and the region is rasterized to form a multi-level scanning unit set; based on the deformation characteristics of the envelope surface, an adaptive compensation factor for the spatial structure is generated, and combined with the hovering state parameters, the line surface is scanned and processed by a combination of lidar and dual polarization radar to obtain corrected three-dimensional topology reconstruction data and icing feature vector. The corrected 3D topology reconstruction data and icing feature vectors are input into a pre-trained ResNet deep learning model to identify icing thickness distribution, insulator damage or hanging foreign object anomalies, and output quantified anomaly type, spatial coordinates and confidence score. Based on the quantified anomaly type, spatial coordinates, and confidence score, an adaptive job strategy is generated, which includes tool ID encoding and job parameters.

2. The method for autonomous inspection and processing of overhead power distribution line equipment by unmanned aerial vehicles according to claim 1, characterized in that, Based on the geographic coordinates and topology data of overhead power distribution line equipment, a three-dimensional flight trajectory and inspection safety boundary for the UAV are generated, including: Obtain GIS geographic coordinates and equipment topology data of overhead power distribution lines, and extract tower spatial location information, conductor and ground wire spatial morphology parameters, and topological relationships of adjacent equipment; Using the spatial morphology parameters of the conductor and ground wire as input, a three-dimensional spatial trajectory of the conductor and ground wire is generated through a spatial curve fitting algorithm, and the dynamic vertical isolation threshold of the UAV is calculated based on electrical safety standards. Using the equipment topology and the three-dimensional spatial trajectory of the conductor and ground wire as input, the distribution of vegetation coverage areas, building shielding areas and electromagnetic interference sources is identified, and a rotationally symmetric dynamic envelope surface with the conductor and ground wire trajectory as the axis is constructed to generate the inspection safety boundary. By integrating the three-dimensional spatial trajectory of the ground wire with the rotationally symmetric dynamic envelope, a segmented trajectory optimization algorithm is used to generate the waypoint sequence and attitude control command set of the UAV, and output the three-dimensional flight trajectory and inspection safety boundary.

3. The method for autonomous inspection and processing of overhead power distribution line equipment by unmanned aerial vehicles according to claim 2, characterized in that, Based on the three-dimensional flight trajectory and inspection safety boundary, the UAV is controlled to reach the target ground line coordinates. Centimeter-level dynamic hovering is achieved through multi-sensor fusion positioning. In the hovering state, three spatial reference coordinates are dynamically calibrated, and hovering state parameters including pose accuracy, electromagnetic interference intensity, and the reference coordinate set are output, including: Based on the three-dimensional flight trajectory and inspection safety boundary, multi-degree-of-freedom motion commands for the UAV are generated. In response to the drone's arrival at the target ground line coordinates, real-time dynamic carrier phase differential positioning data, inertial measurement unit data, and visual odometry data are integrated to generate a real-time pose feedback stream with centimeter-level accuracy. Based on the real-time pose feedback stream, three spatial reference coordinates are dynamically calibrated in the hovering state based on the spatial distribution characteristics of the conductor and ground wire, and the fluctuation of the environmental electromagnetic field intensity is monitored simultaneously. It aggregates real-time pose feedback stream, spatial reference coordinate set, and electromagnetic field strength fluctuation data, and outputs hovering state parameters including pose accuracy, electromagnetic interference intensity quantification value, and reference coordinate set.

4. The method for autonomous inspection and processing of overhead power distribution line equipment by unmanned aerial vehicles according to claim 3, characterized in that, Based on the reference coordinate set, a dynamic spatial feature envelope is constructed and the region is rasterized to form a multi-level scanning unit set, including: The reference coordinate set is used as a sequence of spatial control points, and a dynamic spatial feature envelope surface is generated by a non-uniform rational B-spline surface reconstruction algorithm. Adaptive region rasterization is performed on the dynamic spatial feature envelope surface, and a multi-level scanning unit set is generated based on the curvature change gradient. At the same time, the deformation feature parameters of the envelope surface are extracted. Based on the deformation characteristic parameters, the spatial structure adaptive compensation factor is calculated. Combined with the pose accuracy and electromagnetic interference intensity quantization value in the hovering state parameters, the scanning unit set is scanned collaboratively by lidar and dual polarization radar. The raw point cloud data, ambient temperature and humidity, wind speed and electromagnetic field intensity obtained by fusion scanning are combined, and wavelet threshold denoising and Kalman filtering are used for data enhancement processing. The output is three-dimensional topology reconstruction data and icing feature vector corrected by spatial structure adaptive compensation factor.

5. The method for autonomous inspection and processing of overhead power distribution line equipment by unmanned aerial vehicles according to claim 4, characterized in that, The adaptive compensation factor for the spatial structure is calculated based on the deformation characteristic parameters, and combined with the pose accuracy and electromagnetic interference intensity quantification values ​​in the hovering state parameters, including: Based on the deformation characteristic parameters, the gradient of the principal curvature distribution of the envelope surface and the offset of the normal vector are extracted; the intensity of the spatial topological distortion induced by curvature is calculated based on the gradient of the principal curvature distribution, and a local coordinate system rotation correction vector is generated based on the normal vector offset; the spatial topological distortion intensity and the local coordinate system rotation correction vector are fused, and the spatial geometric distortion compensation amount is output through vector synthesis operation. Extract the pose accuracy from the hovering state parameters and generate the pose drift compensation amount through the positioning error transfer function; Extract the quantized value of electromagnetic interference intensity from the hovering state parameters, and generate the signal propagation compensation amount based on the electromagnetic wave attenuation characteristic model; By integrating spatial geometric distortion compensation, pose drift compensation, and signal propagation compensation, an adaptive compensation factor for spatial structure is generated through nonlinear weighted aggregation.

6. The method for autonomous inspection and processing of overhead power distribution line equipment by unmanned aerial vehicles according to claim 5, characterized in that, Based on the deformation characteristics of the envelope surface, an adaptive compensation factor for the spatial structure is generated. Combined with hovering state parameters, the track surface is scanned and processed using a combination of lidar and dual-polarization radar to obtain corrected three-dimensional topology reconstruction data and icing feature vectors, including: In response to the coordinated scanning of the scanning unit set by lidar and dual-polarization radar, the original point cloud dataset and dual-polarization echo signal stream are acquired, and the time series data of ambient temperature and humidity, wind speed and electromagnetic field intensity are collected simultaneously. Based on the original point cloud dataset, spatial coordinate transformation correction is performed through spatial structure adaptive compensation factor to generate a geometrically consistent point cloud set. Wavelet threshold denoising is performed on the geometrically consistent point cloud set to eliminate high-frequency vibration noise; at the same time, an enhanced radar reflection feature matrix is ​​generated by fusing dual-polarization echo signal stream and environmental time series data based on Kalman filtering. The dielectric constant distribution spectrum and surface scattering characteristic vector are extracted from the enhanced radar reflection feature matrix, and combined with the electromagnetic interference intensity quantization value in the hovering state parameters to construct the icing feature vector; Perform 3D surface topology reconstruction on the denoised geometrically consistent point cloud set, and output 3D topology reconstruction data corrected by spatial structure adaptive compensation factor.

7. The method for autonomous inspection and processing of overhead power distribution line equipment by unmanned aerial vehicles according to claim 6, characterized in that, The corrected 3D topology reconstruction data and icing feature vectors are input into a pre-trained ResNet deep learning model to identify icing thickness distribution, insulator damage, or suspended foreign object anomalies. The model outputs quantified anomaly type, spatial coordinates, and confidence scores, including: Spatial grid normalization is performed on the 3D topology reconstruction data corrected by the spatial structure adaptive compensation factor to generate a topology tensor; at the same time, the icing feature vector is embedded into the complex domain feature space to generate an enhanced icing feature matrix. By fusing the topological tensor and the enhanced icing feature matrix, a multimodal input data cube is constructed using a spatial feature alignment algorithm; A multimodal input data cube is input into a pre-trained ResNet-101 deep learning model. Spatial topological features and dielectric properties are mapped through three layers of residual convolutional blocks, and an anomaly probability distribution heatmap is output. Based on the anomaly probability distribution heatmap, Gaussian mixture model clustering is used to identify spatial regions of ice thickness distribution, insulator damage, and suspended foreign objects. Combined with spatial coordinates in the three-dimensional topology reconstruction data, quantitative anomaly types, coordinates of the center point of the anomaly region, and confidence scores are generated.

8. The method for autonomous inspection and processing of overhead power distribution line equipment by unmanned aerial vehicles according to claim 7, characterized in that, Based on the quantified anomaly type, spatial coordinates, and confidence score, an adaptive job strategy is generated, including tool ID encoding and job parameters, including: The system analyzes and quantifies the anomaly type, the coordinates of the center point of the anomaly region, and the confidence score, and matches the tool ID encoding through a preset anomaly-tool mapping rule base. Based on the coordinates of the center point of the abnormal area, the drone approach path is generated by combining the three-dimensional flight trajectory, and the operation safety distance threshold is dynamically adjusted according to the confidence score. By integrating the electromagnetic interference intensity quantification values ​​in the anomaly type and hovering state parameters, the laser power, robotic arm operating torque, and de-icing time are calculated through the operation parameter optimization function. By aggregating tool ID encoding, drone approach path, operational safety distance threshold, and operational parameters, an adaptive operational strategy instruction set containing spatiotemporal constraints is generated.

9. A method for autonomous inspection and processing of overhead power distribution line equipment using unmanned aerial vehicles (UAVs) according to claim 8, characterized in that, The laser power, robotic arm operating torque, and de-icing time are calculated using an operation parameter optimization function, including: The anomaly type is analyzed and quantified, and the baseline value of the physical action parameter of the target tool is determined based on the preset anomaly physical characteristic mapping rule. The electromagnetic interference intensity quantification value is extracted from the hovering state parameters, and the reference value of the physical action parameter is dynamically corrected through the electromagnetic attenuation compensation model to generate intermediate values ​​of anti-interference operation parameters. Based on the confidence score, the dynamic scaling factor of the operation parameters is calculated using the safety margin adjustment function; The intermediate values ​​of the anti-interference operation parameters are weighted and fused with the dynamic scaling factor: Based on the type of icing anomaly, the laser power parameters and de-icing time are output by combining the dielectric constant distribution spectrum in the icing feature vector; For insulator damage or suspended foreign objects, the operating torque parameters of the robotic arm are output by combining the surface curvature features in the three-dimensional topology reconstruction data. The parameters of laser power, robotic arm operating torque, and de-icing time are combined to generate optimized operation parameters.

Images