Unmanned aerial vehicle lead crossing intelligent detection method and system fused with real-time atmospheric compensation

Abstract

The invention relates to the field of power transmission line risk hidden danger monitoring, and provides an unmanned aerial vehicle wire crossing intelligent detection method and system fused with real-time atmospheric compensation, and the method comprises the steps: measuring an atmospheric extinction coefficient in real time, and obtaining a flight path parameter perpendicular to the axial direction of a wire; synchronously collecting multi-source sensing data to generate an observation data set; generating a lead initial point cloud with a credibility label through feature extraction and Bayesian confidence weight; generating a lead complete three-dimensional track based on a catenary equation and a locust optimization algorithm; generating three-dimensional wire space structure data with sub-meter precision by taking the track point as a control point; and calculating an actual vertical distance based on the elevation difference value and generating a crossing identification report. The system comprises a data acquisition module, a multi-source data acquisition module, a point cloud generation module, a catenary parameter optimization module, a three-dimensional structure optimization module and a safety evaluation report module. According to the invention, through real-time atmospheric compensation and dynamic weight fusion, the detection precision and robustness are improved, and the method is suitable for power transmission line inspection in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of power transmission line risk and hazard monitoring, and in particular to an intelligent detection method and system for conductor crossings using unmanned aerial vehicles (UAVs) that integrates real-time atmospheric compensation. Background Technology

[0002] As the power grid continues to expand, safety assessment of conductor crossings has become an important means of ensuring the stable operation of the power system. However, transmission lines are characterized by long spans and complex terrain along their routes, making manual inspections insufficient. Therefore, most power companies now use drones for conductor crossing identification.

[0003] Currently, conventional methods for identifying conductor crossings mainly employ multi-sensor collaborative acquisition technology, using millimeter-wave radar, visible light cameras, and GNSS / IMU to acquire spatial data of the conductors, and then using geometric feature matching to locate the crossing points.

[0004] Existing techniques typically process data based on static environment assumptions and rely on bundle adjustment to optimize the accuracy of 3D structures. Under ideal conditions, such methods can provide conductor trajectory data with sub-meter accuracy, offering fundamental support for safety assessments of crossings.

[0005] However, existing technologies have certain limitations in terms of dynamic environmental adaptability and data reliability quantification. For example, firstly, existing technologies mostly use a fixed threshold atmospheric extinction coefficient, which does not fully consider the impact of dynamic temperature and humidity changes on laser transmission attenuation, and may lead to deviations in the confidence assessment of point cloud data. Secondly, multi-source sensor data fusion usually relies on hard alignment strategies and lacks a dynamic confidence label generation mechanism, which will affect the robustness of subsequent catenary parameter optimization.

[0006] Therefore, this invention proposes an intelligent detection method and system for UAV conductor crossings that integrates real-time atmospheric compensation, in order to solve the problems existing in the prior art. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of existing technologies by providing an intelligent detection method and system for UAV conductor crossings that integrates real-time atmospheric compensation, thereby solving the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, this application provides an intelligent detection method for UAV conductor crossings that integrates real-time atmospheric compensation, comprising the following steps: S1. Measure the atmospheric extinction coefficient in real time and obtain the flight trajectory parameters perpendicular to the conductor axis; S2. Based on flight trajectory parameters, synchronously collect spatiotemporally synchronized multi-source sensing data to generate a multi-sensor observation dataset; S3. Obtain image edge points and radar peak points through feature extraction and recognition, and dynamically calculate Bayesian confidence weights by combining atmospheric extinction coefficients to generate initial point clouds of conductors with confidence labels. S4. Based on the initial point cloud of the traverse with credibility labels, generate the catenary equation parameters, and generate the complete three-dimensional trajectory of the traverse under the physical constraints of the catenary using the locust optimization algorithm. S5. Based on the complete three-dimensional trajectory of the traverse and the multi-sensor observation dataset, sub-meter level three-dimensional traverse spatial structure data are generated by using trajectory points as spatial control points. S6. Based on the three-dimensional traverse spatial structure data, the actual vertical distance at the intersection point is generated by the elevation difference between two different traverses, and compared with the safety regulation threshold to generate a traverse intersection identification report.

[0009] Preferably, the specific steps of S1 are as follows: S1.1 Real-time acquisition of atmospheric transmitted light intensity data, and calculation of atmospheric extinction coefficient using Beer-Lambert law; S1.2 Based on the extinction coefficient, temperature and humidity compensation is performed through a dynamic temperature and humidity compensation mechanism to generate an atmospheric extinction coefficient sequence; S1.3 Based on the atmospheric extinction coefficient sequence, millimeter-wave radar activation markers are generated through static activation rules, and the initial three-dimensional trajectory of the conductor is reconstructed through an improved density clustering algorithm using point cloud data collected by millimeter-wave radar. S1.4 Based on the initial traverse 3D trajectory, the original traverse direction vector is calculated by the spatial coordinate difference between the first and last points, and the traverse axial unit vector is generated by normalizing the modulus. S1.5. Based on the unit vector of the conductor axis and the constant gravitational acceleration vector, a unit vector of the horizontal and vertical flight direction is generated through vector cross product operation; S1.6. Based on the unit vectors of horizontal and vertical flight directions, a basic waypoint sequence is generated through the waypoint linear offset algorithm, and then the coordinates of each basic waypoint are calculated through the safe distance radial offset algorithm to generate flight trajectory parameters perpendicular to the axis of the guide.

[0010] Preferably, the specific steps of S2 are as follows: S2.1 Based on flight trajectory parameters, a timestamp synchronization instruction set is generated through the sensor's built-in time counter and fiber optic transmission delay compensation; S2.2 Acquire millimeter-wave radar point cloud data, uncompressed visible light RAW images, and positioning and attitude data to generate raw sensor datasets; S2.3. By converting millimeter-wave radar point cloud data to a local geographic rectangular coordinate system and combining it with positioning and attitude data and translation vectors, a complete three-dimensional trajectory of the traverse is generated. S2.4. Using simultaneous data frames, encapsulate spatiotemporally synchronized multi-source sub-item data into a structured set to generate a multi-sensor observation dataset.

[0011] Preferably, the specific steps of S3 are as follows: S3.1 Based on the multi-sensor observation dataset, the millimeter-wave radar point cloud data is processed by the constant false alarm rate detection algorithm to generate a set of millimeter-wave radar reflection peak points; S3.2. Match data points at the same time through spatiotemporal alignment conditions to generate a set of spatiotemporal matching point pairs; S3.3. Generate a set of spatiotemporal matching point pairs with Bayesian confidence weights by dynamically updating the weight values ​​of each data point pair. S3.4. Generate the initial point cloud of the traverse line with confidence labels using the confidence label annotation rules.

[0012] Preferably, step S4 is as follows: S4.1. Obtain the reference elevation parameters by combining the ratio of the conductor span to the maximum allowable sag with the offset of the lowest elevation point, and generate the initial parameters of the catenary equation. S4.2 Based on the initial parameters of the catenary equation, generate feasible solutions for solution vectors that violate the physical constraint of sag coefficient and are greater than zero, and obtain the optimized catenary equation parameter set; S4.3 Based on the optimized catenary equation parameter set, the two-dimensional trajectory is mapped to three-dimensional space through three-dimensional spatial rotation and translation transformation to generate a complete three-dimensional trajectory of the conductor.

[0013] Preferably, S4.2 specifically includes: Based on the initial parameters of the catenary equation, individual solution vectors are generated by initializing the locust population. The solution vectors are then driven to move towards the current optimal solution by distance-weighted position updates. A geometric series penalty function is applied to solution vectors that violate the sag coefficient and are greater than zero physical constraints to force the generation of feasible solutions, thereby obtaining the optimized catenary equation parameter set.

[0014] Preferably, the specific steps of S5 are as follows: S5.1 Obtain the key positions of the complete three-dimensional trajectory of the conductor, encapsulate the three-dimensional coordinates of the key positions, and generate a set of spatial control points; S5.2. Generate the optimized set of spatial control point coordinates and camera attitude parameters using the hybrid gradient trust region optimization algorithm. S5.3. Based on the optimized spatial control point coordinates and camera attitude parameter set, generate a physically constrained optimized spatial control point coordinate and camera attitude parameter set. S5.4. Through the spatial structure accuracy verification protocol, perform collaborative verification of the accuracy of the three-dimensional coordinates of spatial control points to generate three-dimensional traverse spatial structure data with sub-meter accuracy.

[0015] Preferably, step S6 is as follows: S6.1 Based on the three-dimensional traverse spatial structure data, select point pairs whose horizontal projection distance does not exceed the safety regulation threshold and whose elevation difference is greater than the safety margin threshold, and generate an intersection point set; S6.2. Based on the set of intersection points, perform dynamic compensation to generate the actual vertical distance after dynamic compensation; S6.3. Based on the actual vertical distance, identify the minimum allowable vertical distance and make dynamic corrections according to the working environment; S6.4. Generate a conductor crossing identification report based on safety margin calculation and early warning classification rules.

[0016] Preferably, S6.2 specifically includes: Based on the intersection point set, the absolute elevation difference between two different conductors is obtained through the vertical distance formula. Dynamic compensation is then performed by combining the conductor's thermal expansion coefficient with the measured temperature difference on site to generate the actual vertical distance after dynamic compensation.

[0017] Secondly, this application also provides an intelligent detection system for UAV conductor crossings that integrates real-time atmospheric compensation, comprising: The data acquisition module is used to measure the atmospheric extinction coefficient in real time and obtain flight trajectory parameters perpendicular to the axis of the conductor. The multi-source data acquisition module is used to synchronously acquire spatiotemporally synchronized multi-source sensing data based on flight trajectory parameters, and generate a multi-sensor observation dataset. The point cloud generation module is used to obtain image edge points and radar peak points through feature extraction and recognition, and dynamically calculate Bayesian confidence weights by combining atmospheric extinction coefficients to generate an initial point cloud of the conductor with confidence labels. The catenary parameter optimization module is used to generate catenary equation parameters based on the initial point cloud of the traverse with confidence labels, and generate the complete 3D trajectory of the traverse under the physical constraints of the catenary through the locust optimization algorithm. The 3D structure optimization module is used to generate 3D traverse spatial structure data with sub-meter accuracy based on the complete 3D trajectory of the traverse and the multi-sensor observation dataset, by using trajectory points as spatial control points. The safety assessment report module is used to generate the actual vertical distance at the crossing point based on the elevation difference between two different conductors, using three-dimensional conductor spatial structure data, and compare it with the safety regulation threshold to generate a conductor crossing identification report.

[0018] The present invention discloses an intelligent detection method for UAV conductor crossings that integrates real-time atmospheric compensation, which has the following beneficial effects.

[0019] This invention compensates for atmospheric transmission attenuation by measuring the atmospheric extinction coefficient in real time and combining it with dynamic temperature and humidity compensation, thereby improving the accuracy of data acquisition and point cloud confidence in dynamic and complex environments and avoiding the evaluation bias caused by fixed thresholds in existing technologies.

[0020] This invention employs dynamic calculation of Bayesian confidence weights and credibility labeling to enhance the robustness of multi-source sensor data fusion, ensure the reliability of the initial point cloud of the conductor, and provide a solid foundation for subsequent catenary parameter optimization.

[0021] This invention introduces a locust optimization algorithm to generate a complete three-dimensional trajectory of a conductor under the physical constraints of a catenary. It optimizes the parameter solving process and improves the efficiency and accuracy of trajectory reconstruction, making it suitable for the inspection of long-span transmission lines in complex terrain.

[0022] This invention uses trajectory points as spatial control points and generates three-dimensional traverse spatial structure data with sub-meter accuracy through a hybrid gradient trust region optimization algorithm, further improving the accuracy of intersection point positioning.

[0023] This invention calculates the actual vertical distance based on elevation difference and dynamic compensation, and generates an identification report by comparing it with the threshold of safety regulations. This achieves automated early warning classification, reduces the need for manual intervention, and improves the overall efficiency and safety of monitoring potential risks in transmission lines. Attached Figure Description

[0024] Figure 1 This is an overall flowchart of an intelligent detection method for UAV conductor crossings that incorporates real-time atmospheric compensation, as described in this application.

[0025] Figure 2 This is a schematic diagram of a wire crossing recognition system based on drone images.

[0026] Figure 3 This is a flowchart of multi-source data synchronous acquisition and processing.

[0027] Figure 4 This is a flowchart of catenary parameter optimization and 3D trajectory generation. Detailed Implementation

[0028] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0029] In this document, the term "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The term "embodiment" appearing in various places throughout the specification does not necessarily refer to the same embodiment, nor does it specifically limit its independence or connection with other embodiments. In principle, in this application, as long as there are no technical contradictions or conflicts, the technical features mentioned in each embodiment can be combined in any way to form corresponding implementable technical solutions.

[0030] Unless otherwise defined, the technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the use of related terms herein is merely for the purpose of describing particular embodiments and is not intended to limit this application.

[0031] In the description of this application, the term "and / or" is used to describe the logical relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A exists, B exists, and A and B exist simultaneously. Additionally, the character " / " in this document generally indicates that the preceding and following objects have an "or" logical relationship.

[0032] Unless otherwise expressly specified or limited, the terms "installation," "connection," "linking," "fixing," and "setting," as used in the description of the embodiments of this application, should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral setting; it can be a mechanical connection, an electrical connection, or a communication connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be the internal connection of two components or the interaction between two components. For those skilled in the art to which this application pertains, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0033] As disclosed in the background section, existing technologies have certain limitations in terms of dynamic environmental adaptability and data reliability quantification. For example, firstly, existing technologies mostly use a fixed threshold atmospheric extinction coefficient, which does not fully consider the impact of dynamic temperature and humidity changes on laser transmission attenuation, and may lead to bias in the confidence assessment of point cloud data. Secondly, multi-source sensor data fusion usually relies on hard alignment strategies and lacks a dynamic confidence label generation mechanism, which will affect the robustness of subsequent catenary parameter optimization.

[0034] To address the aforementioned problems, this invention proposes an intelligent detection method for UAV conductor crossings that integrates real-time atmospheric compensation. This method includes the following steps: S1. Measure the atmospheric extinction coefficient in real time and obtain the flight trajectory parameters perpendicular to the conductor axis; S2. Based on flight trajectory parameters, synchronously collect spatiotemporally synchronized multi-source sensing data to generate a multi-sensor observation dataset; S3. Obtain image edge points and radar peak points through feature extraction and recognition, and dynamically calculate Bayesian confidence weights by combining atmospheric extinction coefficients to generate initial point clouds of conductors with confidence labels. S4. Based on the initial point cloud of the traverse with credibility labels, generate the catenary equation parameters, and generate the complete three-dimensional trajectory of the traverse under the physical constraints of the catenary using the locust optimization algorithm. S5. Based on the complete three-dimensional trajectory of the traverse and the multi-sensor observation dataset, sub-meter level three-dimensional traverse spatial structure data are generated by using trajectory points as spatial control points. S6. Based on the three-dimensional traverse spatial structure data, the actual vertical distance at the intersection point is generated by the elevation difference between two different traverses, and compared with the safety regulation threshold to generate a traverse intersection identification report.

[0035] The technical solution provided by this invention generates an initial point cloud of a conductor with a confidence label by dynamically calculating Bayesian confidence weights, thereby achieving probabilistic fusion of multi-source heterogeneous data and improving anti-interference capabilities; and generates a complete three-dimensional trajectory of the conductor under the physical constraints of the catenary by using the locust optimization algorithm, thereby ensuring the physical rationality of the conductor's spatial morphology and achieving global optimization.

[0036] The above plan will be explained in detail below.

[0037] Please see the appendix Figure 1 , Figure 1 This is a schematic flowchart illustrating the main steps of an intelligent detection method for UAV conductor crossings that integrates real-time atmospheric compensation, according to the present invention. Figure 1 As shown, an intelligent detection method for UAV conductor crossings integrating real-time atmospheric compensation in an embodiment of the present invention mainly includes the following steps: like Figure 1 As shown, S1, the atmospheric extinction coefficient is measured in real time, and the flight trajectory parameters perpendicular to the conductor axis are obtained; In this embodiment, the atmospheric transmitted light intensity data includes the initial laser emission intensity, the received light intensity, and the transmission path length.

[0038] Preferably, in this embodiment, the specific steps of S1 are as follows: S1.1 Real-time acquisition of atmospheric transmitted light intensity data, and calculation of atmospheric extinction coefficient using Beer-Lambert law; Specifically, in this embodiment, a laser beam is emitted using a laser emitter of a specific wavelength, maintaining a stable laser emission intensity. Simultaneously, a high-sensitivity photodetector receives the transmitted light intensity signal after atmospheric attenuation, and the laser transmission path length is precisely recorded. Based on Beer-Lambert's law, the measured initial emitted light intensity, received light intensity, and transmission path length are substituted into a formula to calculate the initial extinction coefficient.

[0039] The initial extinction coefficient expression is: ; in, This represents the initial extinction coefficient (typically ranging from 0.001 to 0.05m). -1 (The specific value depends on environmental conditions and measurement system parameters.) Indicates the length of the laser transmission path. It represents the natural logarithm (used to accurately describe the physical property that light intensity decreases exponentially with transmission distance). This indicates the initial intensity of the laser emission. This indicates the received light intensity after atmospheric attenuation.

[0040] To eliminate environmental interference factors, ambient temperature and humidity data are collected simultaneously, and the initial extinction coefficient is dynamically corrected by a compensation coefficient, finally outputting a calibrated atmospheric extinction coefficient sequence.

[0041] S1.2 Based on the extinction coefficient, temperature and humidity compensation is performed through a dynamic temperature and humidity compensation mechanism to generate an atmospheric extinction coefficient sequence; Ambient temperature and relative humidity data are acquired in real time using a high-precision digital temperature and humidity sensor. A dynamic weighted fusion algorithm is used to condition and eliminate noise in the ambient temperature and relative humidity data, followed by dynamic correction using a compensation algorithm. The temperature and humidity compensation coefficients in the compensation algorithm are optimized and determined through comparative experiments simulating different temperature and humidity ranges in a standard climate laboratory. The compensated extinction coefficient eliminates measurement bias caused by temperature and humidity fluctuations and is further processed using a time-series smoothing algorithm to ultimately generate an atmospheric extinction coefficient sequence with temporal continuity and spatial consistency.

[0042] The mathematical expression for the temperature and humidity compensation model is: ; in, This represents the compensated atmospheric extinction coefficient (typically ranging from 0.001m). -1 -0.05m -1 ); This represents the original parameter values ​​under reference temperature and humidity conditions; This represents the temperature compensation coefficient (typically ranging from -0.1 % / °C to +0.02 % / °C). Indicates the current ambient temperature; This indicates the reference temperature (typically ranging from 20°C to 25°C). This represents the humidity compensation factor (typically ranging from 0.01% / %RH to 0.5% / %RH). Indicates the current ambient relative humidity; This indicates the reference relative humidity (values ​​range from 50%RH to 60%RH).

[0043] S1.3 Based on the atmospheric extinction coefficient sequence, millimeter-wave radar activation markers are generated through static activation rules, and the initial three-dimensional trajectory of the conductor is reconstructed through an improved density clustering algorithm using point cloud data collected by millimeter-wave radar. Specifically, in this embodiment, when the atmospheric extinction coefficient reaches the activation threshold for conductor safety inspection, the static activation rule automatically triggers the millimeter-wave radar to operate. This rule comprehensively considers factors such as current environmental visibility, target distance, and equipment power consumption. The millimeter-wave radar performs high-precision three-dimensional point cloud acquisition at a specific operating frequency and scanning rate. Each scanning cycle acquires dense spatial point data. The raw point cloud data is first preprocessed using an improved density clustering algorithm to effectively filter out noise points and irrelevant background, retaining a valid set of conductor feature points. The improved density clustering algorithm expression is as follows: ; in, Point The overall influence or importance within its neighborhood (weighted local density). This indicates that for all points located at... Adaptive neighborhood radius Points within the range Perform a traversal and summation; Point The adaptive neighborhood radius; Indicates the point used for measurement With point The degree of proximity in terms of spatial distance; Indicates the point used for measurement With point The degree of similarity in millimeter-wave radar reflection intensity values; For example, Represented as: ; in, >0 represents the distance scale parameter; Represented as: ; in, , >0 represents the intensity similarity scale parameter.

[0044] It should be noted that the improved density clustering algorithm aims to enhance its ability to preserve low-density conductor point clouds and its noise robustness, ensuring the accuracy of subsequent feature extraction, while reducing computational overhead to meet the real-time processing requirements of embedded devices. It should also be noted that the conductor safety inspection activation threshold is defined as a dynamic parameter based on a comprehensive consideration of multiple factors, including sensor performance characteristics, target detection requirements, and operational efficiency, and typically ranges from 0.003 to 0.015m. -1 .

[0045] In the conductor identification stage, a multi-scale geometric feature matching algorithm is adopted to extract candidate points through local curvature analysis and normal vector consistency test, and to accurately identify conductor features by combining a feature extraction network based on deep learning (e.g., dynamic graph convolutional neural network). The expression for the multi-scale geometric feature matching algorithm is: ; in, Point It was identified as a candidate point for the traverse. Indicates the analysis point Curvature values ​​at multiple scales (such as small, medium, and large neighborhood radii); Indicates the curvature consistency threshold; The normal consistency score is calculated as a point. The sum of the cosines of the angle between the vector and the normal vector of each point in its neighborhood; The normal consistency threshold is used to determine... Is it high enough to indicate the point? And the neighborhood belongs to a smooth, continuous surface (such as the surface of a wire). Point The linear exponent, through point Principal component analysis was performed on the neighborhood of the sample. The linear exponential threshold is indicated. It should be noted that the curvature consistency threshold is determined based on the curvature statistical analysis of typical conductors and backgrounds, and its value range is typically 0.05 to 0.15. The linear exponential threshold is a key parameter defined based on the physical properties of the conductors and the statistical characteristics of the point cloud, and its value range is typically 0.6 to 0.9.

[0046] It should be noted that the scale-based geometric feature matching algorithm is used to identify and extract traverses from a single frame point cloud. Traverses exhibit different geometric features at different observation distances and angles. This algorithm analyzes the local geometric properties of points (such as curvature and normal vector direction) at different spatial scales to find point sets that conform to the cylindrical or linear features of the traverse, thereby accurately screening out candidate point clouds belonging to the traverse and laying the foundation for subsequent processing.

[0047] During multi-frame data registration, a high-precision spatiotemporal alignment is obtained through an iterative nearest-point algorithm. Finally, an optimized reconstruction algorithm is used to generate an initial 3D trajectory of the conductor that includes the precise spatial position, direction, and sag characteristics of the conductor, while outputting quality assessment parameters.

[0048] objective function ; in, Represent the optimal rotation matrix and optimal translation vector that minimize the objective function value; This represents the rotational transformation of a point cloud in three-dimensional space. Represents the three-dimensional coordinate vector of the i-th point in the source point cloud (such as the point cloud of the current frame) to be registered; This represents the translation transformation of a point cloud in three-dimensional space. Indicates the relationship between the target point cloud (reference frame) and the source point. The three-dimensional coordinate vectors of the points matched by the nearest neighbor search are used; N is the number of matched point pairs.

[0049] The optimized reconstruction algorithm expression is: ; in, This represents the weighted sum of squares with reprojection errors; This indicates that it contains the coordinates of all three-dimensional points in space to be optimized and the pose of the millimeter-wave radar sensor. This represents the confidence coefficient assigned based on the quality of feature point matching (typically ranging from 0 to 1). Indicates the first Rotation and translation of a millimeter-wave radar sensor relative to the world coordinate system; Indicates the first in the scene The three-dimensional position of a point in the world coordinate system; Indicates the first Observed in the frame radar point cloud, compared with global points Matched 3D point coordinates.

[0050] It should be noted that the optimization reconstruction algorithm is responsible for generating a smooth, accurate and complete 3D model of the conductor based on the registered multi-frame point cloud data. This algorithm optimizes and smooths the spatial position, orientation and sag of the conductor by establishing a mathematical model (such as spline curve fitting and least squares optimization). By minimizing the overall fitting error between the reconstructed trajectory and all observation data points, and incorporating physical constraints, it finally outputs a 3D spatial trajectory that accurately reflects the true shape of the conductor, and calculates parameters such as fitting residuals to evaluate the reconstruction quality.

[0051] S1.4 Based on the initial traverse 3D trajectory, the original traverse direction vector is calculated by the spatial coordinate difference between the first and last points, and the traverse axial unit vector is generated by normalizing the modulus. Specifically, in this embodiment, the spatial coordinate data of the first and last points are first extracted from the reconstructed three-dimensional trajectory of the conductor. The original conductor direction vector is obtained by subtracting the three-dimensional coordinates. This vector reflects the spatial direction and extension trend of the conductor. Then, the modulus of the original conductor direction vector is calculated to obtain the geometric length of the vector. The three spatial coordinate components (e.g., X component, Y component, Z component) of the original conductor direction vector are converted into unit vector components by vector normalization operation, and finally the conductor axial unit vector is generated.

[0052] S1.5. Based on the unit vector of the conductor axis and the constant gravitational acceleration vector, a unit vector of the horizontal and vertical flight direction is generated through vector cross product operation; Specifically, in this embodiment, a normalized unit vector representing the conductor's axial direction (e.g., with a standard vector length of 1) is first obtained. This unit vector accurately represents the conductor's extension direction in three-dimensional space. Simultaneously, a calibrated constant gravitational acceleration vector is read. This vector points vertically to the Earth's center and has a magnitude equal to the standard gravitational acceleration value. The cross product of the conductor's axial unit vector and the gravitational acceleration vector is performed using the right-hand rule to generate an orthogonal vector perpendicular to both the conductor's axial direction and the gravitational direction, forming a horizontal and vertical reference plane. Subsequently, the orthogonal vector is normalized so that its magnitude is strictly equal to 1, thereby generating a standard horizontal and vertical flight direction unit vector.

[0053] It should be noted that the constant gravitational acceleration vector is obtained by using the static calibration data of the high-precision inertial measurement unit, through the six-position calibration method and polynomial regression fitting.

[0054] S1.6. Based on the unit vector of horizontal and vertical flight direction, a basic waypoint sequence is generated by the waypoint linear offset algorithm, and then the coordinates of each basic waypoint are calculated by the safe distance radial offset algorithm to generate flight trajectory parameters perpendicular to the axis of the guide. Specifically, in this embodiment, the initial heading of the UAV is first determined based on the axis of the guide wire, according to the unit vector of the horizontal and vertical flight direction. The unit vector of the horizontal and vertical flight direction is strictly orthogonal to the axis of the guide wire. A series of basic waypoints are generated along the extension direction of the guide wire at fixed intervals (e.g., every 5 meters) using a waypoint linear offset algorithm. The spatial coordinates of each waypoint are obtained by vector linear superposition calculation to ensure that the waypoint spacing error is controlled within the centimeter-level accuracy range (e.g., ±0.05 meters). A safe distance radial offset algorithm is adopted, with a safe distance (e.g., 20 meters in the horizontal direction) preset based on historical accident statistics and electrical safety specifications as the radius, and each basic waypoint is offset along the horizontal and vertical directions. The coordinates of the offset three-dimensional waypoints are calculated by coordinate transformation formula, and finally a flight trajectory that is strictly perpendicular to the direction of the guide wire and maintains a constant safe distance is generated.

[0055] It should be noted that the basic waypoint sequence (including the coordinates of each point) refers to the initial set of spatial coordinates generated at fixed intervals along the direction of the traverse. The three-dimensional spatial coordinates of the basic waypoints are calculated using a linear interpolation algorithm and an equidistant sampling algorithm, and are used to define the basic inspection path of the UAV. The flight trajectory parameters refer to the final flight control parameters after correction by the normal offset algorithm, ensuring that the UAV maintains a constant safe distance from the traverse and meets the inspection stability requirements.

[0056] S2. Based on flight trajectory parameters, synchronously collect spatiotemporally synchronized multi-source sensing data to generate a multi-sensor observation dataset; like Figure 2 As shown, in this preferred embodiment, S2.1, based on the flight trajectory parameters, a timestamp synchronization instruction set is generated through the sensor's built-in time counter and fiber optic transmission delay compensation; Specifically, in this embodiment, based on the flight trajectory parameters (including spatial coordinates, timestamps, and attitude angles), the FPGA hardware circuit converts them into a hardware trigger instruction set. The high-precision time counter built into the sensor records the acquisition time of millimeter-wave radar point cloud data, visible light uncompressed RAW images, and positioning and attitude data at nanosecond-level resolution. At the same time, a two-way time synchronization protocol is used to measure and compensate for fiber optic transmission delay in real time. The flight trajectory parameters (including spatial coordinates, timestamps, and attitude angles) are spatiotemporally aligned with the local timestamp after delay compensation to generate a timestamp synchronization instruction set with spatiotemporal calibration information.

[0057] S2.2 Acquire millimeter-wave radar point cloud data, uncompressed visible light RAW images, and positioning and attitude data to generate raw sensor datasets; Specifically, in this embodiment, upon receiving a timestamp synchronization command, the millimeter-wave radar acquires 3D point cloud data at a specific operating frequency and scanning rate. Each scanning cycle generates a dataset containing a large number of spatial points. The synchronously triggered global shutter camera outputs uncompressed visible light RAW image data with high pixel resolution and bit depth, and the exposure time is strictly controlled within a precise time window. The positioning and attitude acquisition method records spatial position and attitude data at a high sampling frequency to ensure that the positioning accuracy and attitude measurement error meet the requirements. All acquired data is embedded with timestamp information generated by a high-precision clock source before transmission. Subsequently, the multi-source data is transmitted to the central processing unit (CPU) through a high-speed fiber optic network. The spatiotemporal alignment processor completes time synchronization and coordinate transformation in a very short time, and finally generates the original sensor dataset in the standard coordinate system.

[0058] It should be noted that the integrated navigation unit refers to a multi-source fusion positioning system that integrates a global satellite navigation system, an inertial measurement unit, and a visual sensor. It is used to provide high-precision three-dimensional spatial positioning and flight attitude data for UAV power line inspection operations. The precise time window refers to the exposure time interval controlled by high-precision clock synchronization and strictly matched to the sensor triggering sequence. Its duration and start time are precisely set by timestamp synchronization commands, with errors controlled within the nanosecond range.

[0059] S2.3 Based on the original sensor dataset, the millimeter-wave radar point cloud data is converted to the local geographic rectangular coordinate system through step-by-step trigonometric function rotation and translation calculations, and combined with the positioning and attitude data and translation vectors, to generate the complete three-dimensional trajectory of the conductor. Specifically, in this embodiment, spatial coordinate information is extracted from the original point cloud data, and high-precision positioning and attitude data are acquired simultaneously. The millimeter-wave radar point cloud is transformed from the local coordinate system to the carrier coordinate system by constructing a three-dimensional rotation matrix. The three-dimensional rotation matrix is ​​obtained by the sine and cosine functions of the attitude angles (heading angle, pitch angle, roll angle). Then, the translation vector provided by the positioning and attitude data is superimposed to transform the millimeter-wave radar point cloud data to the geodetic coordinate system. Finally, the geodetic coordinate system is further transformed into a local rectangular coordinate system through map projection transformation. In the entire coordinate transformation process, an optimized numerical calculation method is used to ensure the calculation accuracy. Real-time error verification and compensation are performed on the transformation results at each step. The generated complete three-dimensional trajectory of the conductor fully preserves the spatial orientation angle, sag characteristics and geometric shape of the conductor.

[0060] It should be noted that the complete 3D trajectory of the conductor generated in this step is fundamentally different from the trajectory in step S1.3: the trajectory in S1.3 is a relative trajectory reconstructed from a single frame of millimeter-wave radar point cloud in the sensor coordinate system, mainly used for flight path planning; while this step uses coordinate transformation to fuse multiple frames of radar point cloud with high-precision positioning and attitude data, unifying them into a local geographic rectangular coordinate system with a geodetic reference, generating a high-precision absolute trajectory, providing an accurate spatial reference for subsequent spatiotemporal synchronization of multi-source data.

[0061] It should be noted that map projection refers to the method of mapping three-dimensional spatial data in a geodetic coordinate system to a two-dimensional plane coordinate system through mathematical transformation; Optimized numerical calculation methods refer to constructing rotation matrices using algorithms such as quaternion methods or Rodriguez rotation formulas to avoid gimbal lock problems that may occur when directly using Euler angles, and to reduce the cumulative error of trigonometric function calculations, thereby ensuring numerical stability and calculation accuracy.

[0062] S2.4 Based on the complete three-dimensional trajectory of the conductor, the data frames (millimeter-wave radar point cloud, visible light uncompressed RAW image and positioning and attitude data) of the same moment are matched by the time window alignment algorithm, and the spatiotemporal synchronous multi-source sub-item data are encapsulated into a structured set to generate a multi-sensor observation dataset.

[0063] Specifically, in this embodiment, a time alignment window is used as the reference for the complete three-dimensional trajectory of the conductor. Within this time window, millimeter-wave radar point cloud data, uncompressed visible light RAW images, and positioning and attitude determination data are matched using high-precision timestamps. An improved ICP algorithm is used to perform spatiotemporal alignment of the multi-source data, and cross-modal data association is established through feature points. The successfully matched millimeter-wave radar point cloud, uncompressed visible light RAW images, and positioning and attitude determination data are encapsulated into structured data entities, and finally multi-sensor observation data is generated.

[0064] like Figure 3 As shown, S3 obtains image edge points and radar peak points through feature extraction and recognition, and dynamically calculates Bayesian confidence weights in combination with atmospheric extinction coefficients to generate an initial point cloud of the conductor with confidence labels. Preferably, in this embodiment, step S3 is as follows: S3.1 Based on the multi-sensor observation dataset, the visible light uncompressed RAW image data is processed by the edge detection algorithm to generate the visible light image edge point set, and the millimeter-wave radar point cloud data is processed by the constant false alarm rate detection algorithm to generate the millimeter-wave radar reflection peak point set. Specifically, in this embodiment, the uncompressed visible light RAW image data acquired by the visible light camera is subjected to de-mosaicing, white balance correction, and nonlinear noise reduction processing. Then, an optimized edge detection algorithm is used (this algorithm accurately extracts the contours of the conductor and its fittings from the preprocessed uncompressed visible light RAW image. It not only needs to detect the edges but also needs to generate a point set containing accurate pixel coordinates and gradient direction information, which provides a crucial data foundation for subsequent possible fusion with millimeter-wave point clouds, 3D reconstruction, or accurate measurement) to extract the contour features of the conductor and its auxiliary fittings, generating a visible light image edge point set containing accurate pixel coordinates and gradient direction information. At the same time, constant false alarm rate detection is performed on the 3D point cloud data acquired by millimeter-wave radar point cloud data. The effective reflection points on the surface of the conductor are identified by an adaptive threshold segmentation algorithm. After filtering out environmental noise and clutter interference, a high-confidence millimeter-wave radar reflection peak point set is generated.

[0065] It should be noted that the adaptive threshold segmentation algorithm is a method that dynamically defines a threshold based on the local statistical characteristics of millimeter-wave radar point cloud data. It adaptively determines the segmentation threshold by analyzing the attributes of local regions of the data, rather than using a globally fixed value. This algorithm typically calculates a local threshold for each point or local region in the point cloud to adapt to changes in environmental noise and reflection intensity in different areas. It is particularly suitable for scenarios where the distribution of reflection points on the conductor surface is uneven or where clutter interference exists. The adaptive threshold value is usually defined by sensor performance, environmental constraints, and industry standards.

[0066] It should be noted that constant false alarm rate (CFAR) detection typically employs the cell average (CA-CFAR) algorithm. CFAR detection parameters include the reference window length (typically 16-24 cells), the guard window length (typically 2-8 cells), and the false alarm probability (typically set to 10). -3 ~10 -6 ).

[0067] It should be noted that demosaicing refers to the process where, due to the Bayer filter layout in front of the camera sensor, each pixel can only capture information from one of the three primary colors: red, green, and blue. RAW image data is essentially a single-channel mosaic array. The demosaicing process uses complex interpolation algorithms to estimate and reconstruct the values ​​of the other two missing color channels of each pixel based on the color information of the neighboring pixels, thereby converting the original Bayer array data into a full-color image with complete RGB three-channel information.

[0068] White balance correction refers to the color cast that occurs when a camera sensor captures an image under different light sources. It aims to eliminate color distortion caused by differences in the color temperature of the light source, ensuring that white objects in the image are rendered as true white under any lighting conditions. Typically, based on a preset scene light source color temperature, the gain values ​​of the R, G, and B channels are adjusted non-linearly to correct the overall color balance of the image, ensuring the consistency and accuracy of the color features extracted subsequently from wires and hardware.

[0069] Nonlinear noise reduction refers to the various noises introduced by the sensor during exposure and signal reading, which are more pronounced in low-light environments. To effectively suppress noise while preserving the edge and texture details of the image to the maximum extent, a nonlinear advanced noise reduction algorithm is required. This algorithm can adaptively weight the noise based on the spatial distance between pixels and color / grayscale similarity, smoothing noise in uniform areas while protecting edge areas with drastic gradient changes to prevent them from becoming blurred due to smoothing. This step is crucial, as it provides a higher signal-to-noise ratio and clearer details as an input source for subsequent high-precision edge detection.

[0070] S3.2 Based on the visible light image edge point set, millimeter-wave radar reflection peak point set, positioning and attitude data and translation vector, the spatiotemporal matching point pair set is generated by matching data points at the same time through spatiotemporal alignment conditions. Specifically, in this embodiment, high reflectivity feature points are first extracted from the visible light image edge point set and the millimeter-wave radar reflection peak point set. At the same time, synchronously acquired positioning and attitude data are obtained. The data from different sensors are unified to the carrier reference frame through coordinate transformation. The optimal spatial transformation relationship is calculated according to the feature matching algorithm (using the ICP algorithm based on nearest neighbor search or its variant, combined with robust estimation methods such as RANSAC, and iteratively solving the optimal rotation matrix and translation vector by minimizing the Euclidean distance error or feature descriptor difference between matched point pairs). The data sources are aligned within a strictly synchronized time range. Successfully matched point pairs must simultaneously meet the geometric distance constraint and feature consistency requirements. Finally, a set of matched point pairs containing complete spatiotemporal information is generated.

[0071] It should be noted that the spatiotemporal alignment condition refers to establishing a strict correspondence between multi-source sensor data in time and space dimensions by unifying the time reference and spatial coordinate system, so as to ensure that data from different modalities can be accurately correlated.

[0072] S3.3 Based on the spatiotemporal matching point pair set, the weight value of each data point pair is calculated by the Bayesian confidence weight formula, and dynamically updated by combining the atmospheric extinction coefficient to generate a spatiotemporal matching point pair set with Bayesian confidence weight. Specifically, in this embodiment, the geometric consistency parameters (including spatial position deviation, feature similarity index, and time synchronization accuracy data) of each point pair are first extracted from the established spatiotemporal matching point pair set. A weighted evaluation method is then constructed using a Bayesian probabilistic framework, including analyzing the sensor observation accuracy characteristics (quantifying the pixel positioning error distribution of visible light cameras and the distance measurement error characteristics of millimeter-wave radar, constructing an observation likelihood function P(O|W) of a Gaussian-Weber mixture distribution) and examining the impact of environmental interference (establishing a conditional probability distribution P(E|W) based on parameters such as atmospheric extinction coefficient and light intensity). Then, these two... The element and prior weight distribution P(W) are fused using Bayes' theorem to construct a joint probability distribution, in which the observation error of the visible light camera is modeled as pixel-level Gaussian noise, and the observation error of the millimeter-wave radar is modeled as a Weber distribution including multipath effects; the environmental interference model quantifies the impact of atmospheric attenuation on optical signal transmission using the Beer-Lambert law and introduces temperature and humidity compensation factors; the initial confidence of each point pair is calculated using the Markov chain Monte Carlo method to solve the posterior distribution, and the optical observation weights are dynamically attenuated by combining the real-time monitored atmospheric extinction coefficient, generating a set of spatiotemporally matched point pairs with confidence weights.

[0073] The expression for the Bayesian confidence weight formula is as follows: ; in, Indicates the first The final confidence weight of each matching point pair typically ranges from 0 to 1. Denotes the parameters of a binary proposition, and denotes the first... Each pair of points matches correctly. The determination needs to combine geometric consistency and semantic features (such as wire diameter matching); Represents the conditional likelihood probability, indicating the probability of a correct match ( Under the condition of (establishment), sensor observation data The probability of credibility; Represents the prior probability, obtained based on historical data statistics. The baseline probability of a successful match for a point pair; This represents the normalization constant, used to ensure that all weight values ​​fall within a reasonable range. The calculation formula is the sum of probabilities for all possible cases. This represents the atmospheric attenuation compensation term, where The atmospheric extinction coefficient is measured in real time (typically ranging from 0.001 to 0.02 m). -1 ); This is the optical transmission path length, used to compensate for the impact of the atmospheric environment on optical observations.

[0074] S3.4 Based on the set of spatiotemporal matching point pairs with Bayesian confidence weights, generate an initial point cloud of the traverse with confidence labels through weighted fusion coordinate calculation and confidence label annotation rules; Specifically, in this embodiment, the spatial coordinates of each pair of matching points are weighted and fused. The coordinates of the millimeter-wave radar point cloud and the coordinates of the visible light image feature points are weighted and averaged according to their confidence weight values. High-confidence point pairs directly retain the original coordinate data, while medium-confidence point pairs are smoothed and optimized using a Kalman filter algorithm. At the same time, by analyzing historical calibration data and sensor performance indicators, the preset weight threshold range is set as follows: high confidence (e.g., ≥0.8), medium confidence (e.g., 0.5-0.8), and low confidence (e.g., <0.5). The weights are visualized in the point cloud using color coding (e.g., red-yellow-green gradient). The final generated initial point cloud of the conductor not only contains accurate three-dimensional spatial coordinate information but also integrates reflection intensity features and multi-level confidence labels. Its data quality and accuracy fully meet the requirements of subsequent processing and analysis, providing a reliable data foundation for conductor status assessment and safety analysis.

[0075] S4. Based on the initial point cloud of the traverse with credibility labels, generate the catenary equation parameters, and generate the complete three-dimensional trajectory of the traverse under the physical constraints of the catenary using the locust optimization algorithm. Preferably, in this embodiment, step S4 is as follows: S4.1 Based on the initial point cloud of the traverse with credibility labels, the sag coefficient is calculated by the ratio of the traverse span to the maximum allowable sag, and the reference elevation parameters are obtained by combining the offset of the lowest elevation point to generate the initial parameters of the catenary equation. Specifically, in this embodiment, spatial feature points of the conductor are first extracted from the initial point cloud of the conductor with confidence labels. A spatial geometric model of the conductor is established using a curve fitting algorithm. The ratio of the conductor span to the maximum allowable sag is calculated. Simultaneously, the spatial position of the lowest point of the conductor sag is accurately located. The vertical distance and horizontal offset between the lowest point of the conductor sag and the two suspension points are calculated. Based on the ratio of the conductor span to the maximum allowable sag, the vertical distance, and the horizontal offset, the sag coefficient is derived using a geometric analytical method combined with a least squares optimization algorithm. Then, combined with the conductor material properties, an initial parameter set for the catenary equation is constructed using a nonlinear parameter optimization algorithm. The initial parameters of the catenary equation are precisely calibrated using an optimization algorithm to ensure that the sag calculation accuracy meets engineering requirements. The entire calculation process adopts an efficient computing architecture to achieve rapid processing, providing an accurate mathematical model foundation for subsequent conductor mechanical analysis and safety assessment.

[0076] The equation for the catenary is: ; in, Represents elevation coordinates. This represents the sag coefficient (typically ranging from 500 to 3000 meters). This represents the hyperbolic cosine function (which describes the geometric shape of a conductor hanging naturally under the influence of gravity). This indicates the offset from the reference elevation. Represents the horizontal coordinate.

[0077] It should be noted that the offset of the lowest elevation point refers to the vertical distance between the lowest point of the guideline sag and the line connecting the two suspension points (i.e., the maximum sag value), which is a correction term for the reference elevation of the catenary equation.

[0078] S4.2 Based on the initial parameters of the catenary equation, generate feasible solutions for solution vectors that violate the physical constraint of sag coefficient and are greater than zero, and obtain the optimized catenary equation parameter set; Preferably, in this embodiment, S4.2 specifically includes: Based on the initial parameters of the catenary equation, individual solution vectors are generated by initializing the locust population. The solution vectors are then driven to move towards the current optimal solution by distance-weighted position updates. A geometric series penalty function is applied to solution vectors that violate the sag coefficient and are greater than zero physical constraints to force the generation of feasible solutions, thereby obtaining the optimized catenary equation parameter set.

[0079] Specifically, in this embodiment, when using the locust optimization algorithm for parameter optimization based on the initial parameters of the catenary equation, specific parameterization design of the key components of the locust optimization algorithm is required. The population size N is usually set to 50 to 100 individual solution vectors to ensure population diversity and balance computational efficiency. The number of iterations is set to 500 to 1000 generations to ensure that the algorithm fully converges to the optimal solution domain. The convergence criterion adopts a dual standard: optimization is terminated when the improvement of the objective function value of the global optimal solution is less than the threshold 1e-6 or the maximum number of iterations is reached. For solution vectors that violate the physical constraint that the sag coefficient is greater than zero, a geometric series penalty function is used, specifically in the form of: ; in, Represents the penalty function value; penalty base. Take 10; geometric ratio Set it to 1.5; represents the current iteration number; 𝑎 represents the constraint violation amount (i.e., a≤0 indicates a constraint violation); the geometric series penalty function dynamically strengthens the constraint penalty force through an iterative exponential growth mechanism, forcing infeasible solutions to quickly return to the feasible region during the optimization process; the locust optimization algorithm drives population movement through a distance-weighted position update mechanism, combined with an adaptive shrinkage factor c (linearly decreasing from 1.0 to 0.00001) to coordinate global exploration and local development; after multiple generations of iterative optimization, the final output is the optimized catenary equation parameter set.

[0080] It should be noted that the zero physical constraint is based on the conductor sag safety boundary condition calibrated according to power safety regulations and real-time meteorological data. It is used to ensure that the catenary parameters generated by the optimization algorithm always meet the requirements of conductor-to-ground safety distance and mechanical strength. For example, (for 500kV lines, the sag coefficient a∈[1200,1800] and the lowest point is ≥6.5m above the ground).

[0081] The geometric series penalty function refers to the exponentially increasing penalty term imposed on the solution vector that violates the catenary parameter constraints.

[0082] S4.3 Based on the optimized catenary equation parameter set, the two-dimensional trajectory is mapped to three-dimensional space through three-dimensional spatial rotation and translation transformation to generate the complete three-dimensional trajectory of the conductor; Specifically, in this embodiment, the UAV's waypoint sequence in a two-dimensional vertical plane is calculated based on the optimized catenary equation parameter set. Taking the midpoint of the guide span as the origin, the catenary equation is discretized and sampled at fixed intervals along the horizontal direction of the guide span to generate a waypoint sequence containing two-dimensional coordinates. Then, the two-dimensional waypoints are mapped to a three-dimensional geographic coordinate system through a three-dimensional spatial transformation (rotation matrix). Finally, the translation vector provided by the positioning and attitude data is combined with an affine transformation to convert it to a three-dimensional geographic coordinate system, thus generating a complete three-dimensional trajectory of the guide.

[0083] S5. Based on the complete three-dimensional trajectory of the traverse and the multi-sensor observation dataset, sub-meter level three-dimensional traverse spatial structure data are generated by minimizing the reprojection error using trajectory points as spatial control points. like Figure 4 As shown, in this preferred embodiment, the specific steps of S5 are as follows: S5.1 Based on the complete three-dimensional trajectory of the traverse, the key positions of the complete three-dimensional trajectory of the traverse are obtained by using the curvature extreme point extraction rule, and the three-dimensional coordinates of the key positions are encapsulated by the spatial control point generation algorithm to generate a set of spatial control points; Specifically, in this embodiment, curvature analysis is performed on the complete three-dimensional trajectory of the conductor. The location of feature points in the trajectory (including key locations such as the lowest point of sag and the suspension point of the insulator) is obtained through the sliding window curvature calculation method. Since the original curvature feature points may contain noise and repetition, a density-based spatial clustering algorithm (which can effectively automatically divide feature points that are close in spatial location into the same cluster, while filtering out noise points distributed in sparse areas) is used to optimize the extracted feature points by grouping. Then, the center coordinates of each feature point group are obtained by weighted average calculation, and finally, a set of spatial control points is generated.

[0084] It should be noted that the curvature extremum point extraction rule is a calculation method for identifying significant geometric features in the 3D trajectory of a traverse. By analyzing local extrema of the trajectory curvature (such as the lowest point of sag, suspension points, etc.), key points affecting the spatial morphology of the traverse are determined. The spatial control point set is a set of 3D coordinates characterizing the key geometric features of the traverse, used to constrain the accuracy of subsequent 3D reconstruction. Its generation requires the fusion of curvature extremum points and physical constraints (such as the catenary equation and safety distance).

[0085] S5.2 Based on the spatial control point set and multi-sensor observation dataset, the reprojection error is minimized by the hybrid gradient trust region optimization algorithm to generate the optimized spatial control point coordinates and camera attitude parameter set; Specifically, in this embodiment, each three-dimensional coordinate point in the spatial control point set is mapped onto the imaging plane of the visible light camera using a pinhole projection model combined with the sensor pose matrix and translation vector from the camera pose parameter set and the multi-sensor observation dataset. This obtains the theoretical projected pixel coordinates of the spatial control points on the image. Then, the Euclidean distance between the projected pixel coordinates and the actual extracted image edge point pixel coordinates on the corresponding uncompressed visible light RAW image in the multi-sensor observation dataset is calculated as the reprojection error of a single point pair. Next, the reprojection errors of all successfully matched point pairs are weighted and summed, with the weights taken from the point cloud generation of a single point pair. The Bayesian confidence weights obtained in stages are used to construct the global reprojection error function. Subsequently, a hybrid gradient trust region optimization algorithm is adopted, using the spatial control point coordinates and camera attitude parameter set as optimization variables. In each iteration, based on the current error function value, gradient information, and approximate Hessian matrix, a search direction and step size within the trust region radius are calculated. By iteratively updating the spatial control point coordinates and camera attitude parameter set, the global reprojection error function value gradually decreases. When the error decrease is less than the convergence threshold or the maximum number of iterations is reached after several consecutive iterations, the algorithm terminates, generating the optimized spatial control point coordinates and camera attitude parameters.

[0086] It should be noted that the convergence threshold is defined based on the balance between the accuracy requirements of the catenary parameter optimization and the computational efficiency, and its value is usually in the range of 1e-6 to 1e-8.

[0087] S5.3 Based on the optimized spatial control point coordinates and camera attitude parameter set, the catenary sag coefficient is forced to be satisfied through the penalty function mechanism, and combined with the fixed span constraint and elevation safety constraint, a physically constrained optimized spatial control point coordinate and camera attitude parameter set is generated. Specifically, in this embodiment, based on the optimized spatial control point coordinates and camera attitude parameter set, a multi-constraint joint optimization framework is used for processing. For the catenary sag coefficient constraint, an exponential penalty function is used to impose a geometrically increasing penalty on solutions that violate physical constraints. For the fixed span constraint, the Lagrange multiplier method is used to strictly limit the length error within the allowable range. For the elevation safety constraint, a piecewise linear penalty function is used to ensure the safety margin. All constraints are processed synchronously through an interior point optimization algorithm. In each iteration, the weight of the geometric series penalty function and the Lagrange multiplier are dynamically adjusted. After multiple iterations, the output physical constraint optimization results significantly improve the accuracy of the spatial control point coordinates and camera attitude parameters.

[0088] It should be noted that the distance between the lowest point of the elevation safety constraint guideline and the ground in three-dimensional space must meet the mandatory requirements of electrical safety regulations (such as conductor voltage level, operating conditions and environmental factors).

[0089] It should be noted that interior point optimization algorithm is a numerical algorithm used to handle constrained optimization problems. Interior point optimization algorithm introduces the constraints into the objective function in the form of penalty terms or barrier functions, so that the search path always stays within the feasible region, thereby simultaneously processing all constraints (including the sag coefficient constraint imposed by the penalty function, the fixed span constraint processed by the Lagrange multiplier method, and the elevation safety constraint processed by the piecewise linear penalty function).

[0090] S5.4. Through the spatial structure accuracy verification protocol (including sag parameter calculation, span verification algorithm and safety margin assessment algorithm), the accuracy of the three-dimensional coordinates of the spatial control points is collaboratively verified to generate three-dimensional traverse spatial structure data with sub-meter accuracy. Specifically, in this embodiment, based on the optimized set of spatial control points, the sag parameters are calculated by inverting the catenary equation to verify the compliance of the vertical geometry of the conductor. Then, a span verification algorithm is used to ensure the accuracy of the horizontal span. At the same time, a safety margin assessment algorithm is used to verify the spatial safety. All verification results are cross-compared with multi-sensor observation data (including the spatial positioning accuracy of millimeter-wave radar point clouds and the geometric feature matching results of visible light images). The generated three-dimensional conductor spatial structure data meets the sub-meter accuracy requirements.

[0091] S6. Based on the three-dimensional traverse spatial structure data, calculate the elevation difference between two different traverses by locating the intersection point, generate the actual vertical distance at the intersection point, and compare it with the safety regulation threshold to generate a traverse intersection identification report.

[0092] Preferably, in this embodiment, step S6 is as follows: S6.1 Based on the three-dimensional traverse spatial structure data, the traverse trajectory point set is retrieved through the spatial intersection detection algorithm. Point pairs with horizontal projection distance not exceeding the safety regulation threshold and elevation difference greater than the safety margin threshold are selected to generate the intersection point set. Specifically, in this embodiment, an improved R-tree (multidimensional spatial index structure) spatial indexing algorithm is used to perform a fast neighborhood search on the set of traverse points. When spatially adjacent traverse pairs are detected, the crossing points are screened through dual constraints: the horizontal projection distance must be less than the safety regulation threshold, and the elevation difference must be greater than the safety margin threshold. The candidate point pairs that meet the conditions are geometrically refined, and the precise crossing position is calculated by B-spline (parametric curve fitting method) curve interpolation. The crossing type is then marked, and finally, the set of crossing points is generated. The improved R-tree (multidimensional spatial index structure) spatial indexing algorithm expression is as follows: ; in, Indicates the optimal child node; This represents all possible new data points under the current node. child nodes; Indicates the point Insert child node Then, the area that the minimum bounding rectangle needs to expand; Indicates the point Insert child node Then, the increase in the overlapping area between the node and its sibling nodes; Indicates the measurement of points Insert child node Then, the difference between the spatial distribution density of data points within a node and the density of its sibling nodes.

[0093] It should be noted that the safety regulation threshold and the safety margin threshold are defined based on the power industry engineering design specifications and electrical safety standards. The safety regulation threshold is usually in the range of 0.3 to 1.0 meters (increasing with the voltage level), and the safety margin threshold is usually in the range of 0.8 to 2.0 meters (considering icing and temperature conditions).

[0094] It should be noted that the improved R-tree (multidimensional spatial index structure) spatial indexing algorithm refers to a multidimensional spatial indexing method optimized for the characteristics of traverse spatial data. Based on the traditional R-tree structure, it introduces a dynamic node splitting strategy and an adaptive search mechanism based on spatial distribution density. It reduces the node overlap area by pre-clustering the traverse trajectory point set, and combines Z-order curves or Hilbert curves to spatially fill and sort the high-dimensional data to improve the locality of data within nodes. This enables efficient filtering of non-nearby points and accelerates the neighborhood search process in intersection detection. At the same time, it supports fast pruning under dual constraints, significantly improving the screening efficiency and computational performance of intersection point pairs.

[0095] S6.2. Based on the set of intersection points, perform dynamic compensation to generate the actual vertical distance after dynamic compensation; Preferably, in this embodiment, S6.2 specifically includes: Based on the intersection point set, the absolute elevation difference between two different conductors is obtained through the vertical distance formula, and dynamic compensation is performed by combining the conductor thermal expansion coefficient and the measured temperature difference on site to generate the actual vertical distance after dynamic compensation. Specifically, in this embodiment, the corresponding spatial coordinates of the two conductors are extracted from the crossover point set. The initial absolute elevation difference is obtained through the elevation difference calculation formula. Combining the thermal expansion characteristics of the conductor material and the real-time collected ambient temperature data, the temperature difference between the real-time measured ambient temperature and the reference temperature is calculated. Then, based on the temperature difference and the original length of the conductor, the change in conductor length is calculated using the thermal expansion method. Since the conductor is suspended between towers, the change in length will cause the change in sag, which in turn will cause the conductor elevation to shift. The change in length is converted into the change in elevation of the conductor at the crossover point through a catenary. Finally, the elevation changes of the two conductors are summed algebraically to obtain the elevation difference compensation value caused by the temperature difference. The initial absolute elevation difference is superimposed with the elevation difference compensation value to complete the dynamic compensation correction. The final generated actual vertical distance is used as the core parameter output for the crossover safety assessment, and the temperature compensation value and confidence level are marked simultaneously.

[0096] It should be noted that the coefficient of thermal expansion of a conductor refers to the rate of change of length of the conductor material due to a unit temperature rise. The coefficient of thermal expansion of a conductor is typically taken in the range of 1.5~13.5×10⁻⁶. -6 ℃ -1 (The specific value depends on the type of conductor material).

[0097] S6.3 Based on the actual vertical distance, the minimum allowable vertical distance is identified by indexing the safety threshold through the voltage level, and dynamically corrected according to the working environment correction factor; Specifically, in this embodiment, the corresponding minimum allowable vertical distance is extracted from the safety specification database according to the line voltage level. Then, the current operating condition environmental parameters (including wind speed, icing thickness and ambient temperature) are collected in real time. The correction coefficient model established by the multivariate regression model trained by historical operating condition data is used to quantify the real-time collected current operating condition environmental parameters to generate the minimum allowable vertical distance applicable to the current operating condition.

[0098] It should be noted that the operating environment correction factor refers to a dynamic adjustment parameter that considers the combined effects of environmental factors such as temperature, wind speed, and icing on the conductor safety distance, and its value typically ranges from 0.8 to 1.5. The line voltage level refers to the rated operating voltage of the transmission line design (e.g., 110kV / 220kV / 500kV), which is obtained by matching and querying the conductor type, insulator type, and tower structure parameters in the power industry standard database (e.g., DL / T741-2019).

[0099] It should be noted that the input to the multiple regression model is a vector of environmental parameters collected in real time under the current operating conditions, specifically including wind speed, icing thickness, and ambient temperature. The output is a dynamic correction coefficient for the minimum permissible vertical distance based on the baseline. The mapping relationship between the input and output is a multiple linear or nonlinear regression function trained using historical operating data. The multiple linear or nonlinear regression function quantifies the coupled influence of environmental parameters on the safe distance. That is, by weighted fitting of the statistical relationship between environmental parameters and historical correction coefficients, the model can accurately predict a correction coefficient based on the real-time input parameters. Multiplying this correction coefficient by the minimum permissible vertical distance based on the baseline yields the dynamic safety threshold applicable to the current operating conditions.

[0100] The dynamic safety threshold is dynamically defined by a regression model based on the line voltage level, conductor material characteristics and real-time environmental operating conditions. It is usually 0.8 to 1.5 times the benchmark safety distance, and the specific value is generated in real time by the product of the correction coefficient and the benchmark value.

[0101] S6.4 Based on the actual vertical distance and the minimum allowable vertical distance, generate a conductor crossing identification report through safety margin calculation and early warning classification rules.

[0102] Specifically, in this embodiment, the safety margin value between the actual vertical distance between conductors measured in real time and the dynamically adjusted minimum allowable vertical distance is calculated. The calculated safety margin value is automatically judged according to the warning classification rules specified in the power industry standard. When the safety margin value is negative, the highest level red warning is triggered, indicating that there is an emergency danger that needs to be dealt with immediately. When the safety margin value is in the lower positive range, an orange warning is triggered, indicating that there is a potential risk that needs to be paid attention to. When the safety margin value reaches the safety standard, it is marked as a green safe state. The three-dimensional spatial coordinates of the crossing point, real-time environmental operating parameters, and dynamic threshold calculation process are recorded in full. The specific type and angle of the conductor crossing are determined through spatial geometric analysis, and finally a conductor crossing identification report is generated.

[0103] This embodiment also provides an intelligent detection system for UAV conductor crossings that integrates real-time atmospheric compensation, including: The data acquisition module is used to measure the atmospheric extinction coefficient in real time and obtain flight trajectory parameters perpendicular to the axis of the conductor. The multi-source data synchronous acquisition module is used to synchronously acquire spatiotemporally synchronized multi-source sensing data based on flight trajectory parameters, and generate a multi-sensor observation dataset. The point cloud generation module is used to obtain image edge points and radar peak points through feature extraction and recognition, and dynamically calculate Bayesian confidence weights by combining atmospheric extinction coefficients to generate an initial point cloud of the conductor with confidence labels. The catenary parameter optimization module is used to generate catenary equation parameters based on the initial point cloud of the traverse with confidence labels, and generate the complete 3D trajectory of the traverse under the physical constraints of the catenary through the locust optimization algorithm. The 3D structure optimization module is used to generate 3D traverse spatial structure data with sub-meter accuracy by minimizing reprojection error based on the complete 3D trajectory of the traverse and multi-sensor observation dataset, using trajectory points as spatial control points. The safety assessment report module is used to calculate the elevation difference between two different conductors by locating the intersection point based on the three-dimensional conductor spatial structure data, generate the actual vertical distance at the intersection point, compare it with the safety regulation threshold, and generate a conductor intersection identification report.

[0104] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0105] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for intelligent detection of UAV power line crossings integrating real-time atmospheric compensation, characterized in that, Includes the following steps: S1. Measure the atmospheric extinction coefficient in real time and obtain the flight trajectory parameters perpendicular to the conductor axis; S2. Based on flight trajectory parameters, synchronously collect spatiotemporally synchronized multi-source sensing data to generate a multi-sensor observation dataset; S3. Obtain image edge points and radar peak points through feature extraction and recognition, and dynamically calculate Bayesian confidence weights by combining atmospheric extinction coefficients to generate initial point clouds of conductors with confidence labels. S4. Based on the initial point cloud of the traverse with credibility labels, generate the catenary equation parameters, and generate the complete three-dimensional trajectory of the traverse under the physical constraints of the catenary using the locust optimization algorithm. S5. Based on the complete three-dimensional trajectory of the traverse and the multi-sensor observation dataset, sub-meter level three-dimensional traverse spatial structure data are generated by using trajectory points as spatial control points. S6. Based on the three-dimensional traverse spatial structure data, the actual vertical distance at the intersection point is generated by the elevation difference between two different traverses, and compared with the safety regulation threshold to generate a traverse intersection identification report.

2. The intelligent detection method for UAV power line crossings integrating real-time atmospheric compensation as described in claim 1, characterized in that, The specific steps of S1 are as follows: S1.1 Real-time acquisition of atmospheric transmitted light intensity data, and calculation of atmospheric extinction coefficient using Beer-Lambert law; S1.2 Based on the extinction coefficient, temperature and humidity compensation is performed through a dynamic temperature and humidity compensation mechanism to generate an atmospheric extinction coefficient sequence; S1.3 Based on the atmospheric extinction coefficient sequence, millimeter-wave radar activation markers are generated through static activation rules, and the initial three-dimensional trajectory of the conductor is reconstructed through an improved density clustering algorithm using point cloud data collected by millimeter-wave radar. S1.4 Based on the initial traverse 3D trajectory, the original traverse direction vector is calculated by the spatial coordinate difference between the first and last points, and the traverse axial unit vector is generated by normalizing the modulus. S1.

5. Based on the unit vector of the conductor axis and the constant gravitational acceleration vector, a unit vector of the horizontal and vertical flight direction is generated through vector cross product operation; S1.

6. Based on the unit vectors of horizontal and vertical flight directions, a basic waypoint sequence is generated through the waypoint linear offset algorithm, and then the coordinates of each basic waypoint are calculated through the safe distance radial offset algorithm to generate flight trajectory parameters perpendicular to the axis of the guide.

3. The intelligent detection method for UAV conductor crossings integrating real-time atmospheric compensation as described in claim 1, characterized in that, The specific steps of S2 are as follows: S2.1 Based on flight trajectory parameters, a timestamp synchronization instruction set is generated through the sensor's built-in time counter and fiber optic transmission delay compensation; S2.2 Acquire millimeter-wave radar point cloud data, uncompressed visible light RAW images, and positioning and attitude data to generate raw sensor datasets; S2.

3. By converting millimeter-wave radar point cloud data to a local geographic rectangular coordinate system and combining it with positioning and attitude data and translation vectors, a complete three-dimensional trajectory of the traverse is generated. S2.

4. Using simultaneous data frames, encapsulate spatiotemporally synchronized multi-source sub-item data into a structured set to generate a multi-sensor observation dataset.

4. The intelligent detection method for UAV conductor crossings integrating real-time atmospheric compensation as described in claim 1, characterized in that, The specific steps of S3 are as follows: S3.1 Based on the multi-sensor observation dataset, the millimeter-wave radar point cloud data is processed by the constant false alarm rate detection algorithm to generate a set of millimeter-wave radar reflection peak points; S3.

2. Match data points at the same time through spatiotemporal alignment conditions to generate a set of spatiotemporal matching point pairs; S3.

3. Generate a set of spatiotemporal matching point pairs with Bayesian confidence weights by dynamically updating the weight values ​​of each data point pair. S3.

4. Generate the initial point cloud of the traverse line with confidence labels using the confidence label annotation rules.

5. The intelligent detection method for UAV conductor crossings integrating real-time atmospheric compensation as described in claim 1, characterized in that, The specific steps of S4 are as follows: S4.

1. Obtain the reference elevation parameters by combining the ratio of the conductor span to the maximum allowable sag with the offset of the lowest elevation point, and generate the initial parameters of the catenary equation. S4.2 Based on the initial parameters of the catenary equation, generate feasible solutions for solution vectors that violate the physical constraint of sag coefficient and are greater than zero, and obtain the optimized catenary equation parameter set; S4.3 Based on the optimized catenary equation parameter set, the two-dimensional trajectory is mapped to three-dimensional space through three-dimensional spatial rotation and translation transformation to generate a complete three-dimensional trajectory of the conductor.

6. The intelligent detection method for UAV power line crossings integrating real-time atmospheric compensation as described in claim 5, characterized in that, Specifically, S4.2 is as follows: Based on the initial parameters of the catenary equation, individual solution vectors are generated by initializing the locust population. The solution vectors are then driven to move towards the current optimal solution by distance-weighted position updates. A geometric series penalty function is applied to solution vectors that violate the sag coefficient and are greater than zero physical constraints to force the generation of feasible solutions, thereby obtaining the optimized catenary equation parameter set.

7. The intelligent detection method for UAV conductor crossings integrating real-time atmospheric compensation as described in claim 1, characterized in that, The specific steps of S5 are as follows: S5.1 Obtain the key positions of the complete three-dimensional trajectory of the conductor, encapsulate the three-dimensional coordinates of the key positions, and generate a set of spatial control points; S5.

2. Generate the optimized set of spatial control point coordinates and camera attitude parameters using the hybrid gradient trust region optimization algorithm. S5.

3. Based on the optimized spatial control point coordinates and camera attitude parameter set, generate a physically constrained optimized spatial control point coordinate and camera attitude parameter set. S5.

4. Through the spatial structure accuracy verification protocol, perform collaborative verification of the accuracy of the three-dimensional coordinates of spatial control points to generate three-dimensional traverse spatial structure data with sub-meter accuracy.

8. The intelligent detection method for UAV power line crossings integrating real-time atmospheric compensation as described in claim 1, characterized in that, The specific steps of S6 are as follows: S6.1 Based on the three-dimensional traverse spatial structure data, select point pairs whose horizontal projection distance does not exceed the safety regulation threshold and whose elevation difference is greater than the safety margin threshold, and generate an intersection point set; S6.

2. Based on the set of intersection points, perform dynamic compensation to generate the actual vertical distance after dynamic compensation; S6.

3. Based on the actual vertical distance, identify the minimum allowable vertical distance and make dynamic corrections according to the working environment; S6.

4. Generate a conductor crossing identification report based on safety margin calculation and early warning classification rules.

9. The intelligent detection method for UAV power line crossings integrating real-time atmospheric compensation as described in claim 8, characterized in that, Specifically, S6.2 is as follows: Based on the intersection point set, the absolute elevation difference between two different conductors is obtained through the vertical distance formula. Dynamic compensation is then performed by combining the conductor's thermal expansion coefficient with the measured temperature difference on site to generate the actual vertical distance after dynamic compensation.

10. A UAV intelligent detection system for bridging overhead power lines with real-time atmospheric compensation, based on the UAV intelligent detection method for bridging overhead power lines with real-time atmospheric compensation as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to measure the atmospheric extinction coefficient in real time and obtain flight trajectory parameters perpendicular to the axis of the conductor. The multi-source data acquisition module is used to synchronously acquire spatiotemporally synchronized multi-source sensing data based on flight trajectory parameters, and generate a multi-sensor observation dataset. The point cloud generation module is used to obtain image edge points and radar peak points through feature extraction and recognition, and dynamically calculate Bayesian confidence weights by combining atmospheric extinction coefficients to generate an initial point cloud of the conductor with confidence labels. The catenary parameter optimization module is used to generate catenary equation parameters based on the initial point cloud of the traverse with confidence labels, and generate the complete 3D trajectory of the traverse under the physical constraints of the catenary through the locust optimization algorithm. The 3D structure optimization module is used to generate 3D traverse spatial structure data with sub-meter accuracy based on the complete 3D trajectory of the traverse and the multi-sensor observation dataset, by using trajectory points as spatial control points. The safety assessment report module is used to generate the actual vertical distance at the crossing point based on the elevation difference between two different conductors, using three-dimensional conductor spatial structure data, and compare it with the safety regulation threshold to generate a conductor crossing identification report.

Images