Unmanned aerial vehicle-based live transmission line fitting damage detection method

Abstract

The present application relates to a power transmission line hardware damage live detection method based on a UAV, comprising the following steps: S1: obtaining power transmission line basic data, establishing a three-dimensional digital model of the line; S2: based on the three-dimensional digital model of the line, using an improved A-path search method, considering flight efficiency, energy consumption optimization and safety constraints, and generating an optimal inspection path according to the dynamic safety distance determined according to different voltage levels; S3: obtaining the estimated position of the power transmission line hardware, setting the hovering coordinates and detection parameters for each detection point; S4: the UAV inspects and obtains the original image data according to the optimal inspection path, the hovering coordinates and detection parameters set for each detection point; S5: preprocessing the collected original image data to obtain preprocessed image data; S6: inputting the preprocessed image data into an improved YOLOv8 detection network to obtain the hardware damage type and position. The present application effectively improves the live detection efficiency and reliability of hardware damage.

Description

Technical Field

[0001] This invention relates to the field of power transmission line inspection, and in particular to an intelligent method for detecting live damage to power transmission line fittings based on unmanned aerial vehicles (UAVs). Background Technology

[0002] With the continuous expansion of my country's power grid and the continuous improvement of voltage levels, the traditional method of inspecting transmission line fittings by manual inspection is no longer able to meet the requirements of modern power grids for safe and stable operation in terms of efficiency, safety and precision. On the one hand, the complex environment of working at high altitudes, in live conditions, and across valleys makes routine inspections pose high personal risks, and the inspection cycle is long, costly and has a high rate of missed inspections, making it difficult to detect early cracks, corrosion and loosening of fittings in a timely manner. On the other hand, existing infrared and ultraviolet detection methods mostly rely on power outages or close-range operations, and have limited ability to identify geometric damage to fittings. Summary of the Invention

[0003] To address the aforementioned problems, the present invention aims to provide an intelligent detection method for live-line damage of power transmission line fittings based on unmanned aerial vehicles (UAVs), which effectively improves the efficiency and reliability of live-line detection of fitting damage.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: A live-line intelligent detection method for power transmission line hardware damage based on unmanned aerial vehicles (UAVs) includes the following steps: S1: Obtain basic data of transmission lines, including the three-dimensional coordinates of towers, conductor suspension height, line direction and voltage level information, and establish a three-dimensional digital model of the line; S2: Based on the three-dimensional digital model of the line, and according to the dynamic safety distance determined by different voltage levels, an improved A... The path search method comprehensively considers flight efficiency, energy consumption optimization, and safety constraints to generate the optimal inspection path. S3: Obtain the estimated location of the transmission line fittings and set the hovering coordinates and detection parameters for each detection point; S4: The UAV performs inspections and acquires raw image data based on the optimal inspection path, hovering coordinates and detection parameters set for each detection point; S5: Preprocess the acquired raw image data to obtain preprocessed image data; S6: Input the preprocessed image data into the improved YOLOv8 detection network to obtain the type and location of hardware damage.

[0005] Furthermore, basic data of the transmission line is obtained, including the three-dimensional coordinates of the towers, conductor suspension height, line direction, and voltage level information, to establish a three-dimensional digital model of the line, as detailed below: The basic engineering data of the lines were exported from the GIS and SCADA systems of the power grid company, including tower number, tower type, coordinate information, conductor suspension point height, and line route information; for newly built lines not in the database or areas with low geographic coordinate accuracy, high-precision RTK mapping technology using unmanned aerial vehicles was used to perform supplementary surveys. Collect voltage level, electrical clearance parameters and standard design data for fitting arrangement to define line operation safety parameters, and automatically configure minimum safety distance and live-line work safety boundaries according to voltage level; A three-dimensional digital model is established by integrating BIM modeling and point cloud reconstruction technology. Towers, conductors, and metal structures are imported into the modeling engine in a parametric form. An interactive 3D model is generated through three-dimensional geometric reconstruction and spatial constraints. The suspension points of conductors are fitted with spline curves to fit the sag of the real line. The tower structure is automatically generated by matching standard component libraries. Finally, each object in the model is given spatial coordinates, physical attributes, and a unique number, realizing the spatial semantic integrity of the digital line.

[0006] Furthermore, based on the three-dimensional digital model of the line and the dynamic safety distance determined according to different voltage levels, an improved A... The path search method, taking into account flight efficiency, energy consumption optimization, and safety constraints, generates the optimal inspection path, as detailed below: First, in the three-dimensional digital line space model, the UAV inspection space state S(x,y,z) is defined, and the feasible space Ω is set. s No-fly zone Ω f : ; Where Li represents the i-th segment of the conductor or tower body; D((x,y,z),L i D represents the shortest Euclidean distance from the drone to the line or equipment. safe (V) is a safety distance function related to voltage level V; The dynamic safety distance is modeled using an empirical formula as follows: ; Where α and β are empirical coefficients; In the feasible space Ω s A 3D raster map is constructed using an improved A... The algorithm searches for the optimal inspection path.

[0007] Further, improved A The algorithm introduces an energy consumption term and an attitude penalty term. The improved cost function is as follows: ; Where g(n) is the cumulative flight cost from the starting point to the current node; h(n) is the Euclidean distance heuristic function from the target point to the current node; E c (n) is the energy consumption function per unit distance; θ p (n) represents the attitude change penalty term, indicating the flight control adjustment cost caused by the nodal turn angle or pitch angle; w1, w2, w3, and w4 are the weight coefficients of the four terms, respectively. ; ; Where ρ is the air density; C d denoted as drag coefficient; A is the projected area of ​​the UAV; v is the instantaneous velocity of the UAV; φ is the flight tilt angle; P0 is the hovering base power; Δψ(n) represents the change in heading angle between adjacent nodes; γ is the attitude penalty factor. The overall objective function is optimized as follows: ; Where path is the complete flight path from the starting point to the ending point, passing through all detection points; J is the total path cost; n is a single node on the path; and f(n) is the comprehensive cost function of a single node. Under the following constraints: (x n ,y n ,z n )∈Ω s That is, safe flight space; v min ≤v(t)≤v max ; |a(t)|≤a max ; Among them, (x n ,y n ,z n Let be the three-dimensional spatial coordinates of path node n; v(t) be the instantaneous flight velocity of the UAV at time t; v min Minimum flight speed limit; v max The maximum flight speed limit is defined by a(t), where a(t) is the acceleration vector of the UAV at time t, and |a(t)| is the magnitude of the acceleration vector. max The maximum permissible acceleration; During algorithm execution, the safety boundary D is dynamically adjusted based on real-time wind speed, electric field strength, and UAV attitude. safe ,Right now: ; Among them, v w For real-time wind speed, δ w E is the wind speed safety correction factor. fieldThis is a measured value of the electric field strength; The safe flight distance is updated in real time; δ E The weighting of the electric field influence; D safe (V) represents the baseline safety distance; Meanwhile, a detection visibility constraint C is introduced into the UAV vision modeling. v Node planning is valid only when the target detection point is within the camera's field of view: ; Where, d view r is the camera's line-of-sight vector. target is the target pointing vector; FOV is the field of view angle.

[0008] Furthermore, the attitude control of the UAV inspection adopts quaternion representation to avoid the gimbal lock-up problem. By combining PID controller and feedforward compensation, the UAV can achieve stable flight in complex electromagnetic environment, monitor the distance to the power transmission line in real time, and automatically start the avoidance program when the safe distance is detected to be close to the warning threshold, and ensure the safe distance by adjusting the flight altitude or lateral displacement.

[0009] Furthermore, quaternion representation is used to avoid gimbal lock-up. Stable flight of the UAV in complex electromagnetic environments is achieved through a combination of PID controller and feedforward compensation, as detailed below: Using quaternions q = [q0, q1, q2, q3] T Indicates the attitude of the drone, where: q = q0 + q1a + q2b + q3c; Where a represents a 90° rotation about the x-axis; b represents a 90° rotation about the y-axis; and c represents a 90° rotation about the z-axis. The conversion relationship between quaternions and Euler angles: ; in, φ, θ, ψ These are roll angle, pitch angle, and yaw angle, respectively. Quaternion differential equation: ; Where ω=[0,ω x ,ω y ,ω z ] T The angular velocity quaternion, This represents quaternion multiplication. Expanded format: ; The PID controller adopts a dual-loop control structure, including an outer-loop attitude controller and an inner-loop angular velocity controller; The outer loop attitude controller: ; in, ω ref The reference angular velocity vector; This is the attitude scaling gain matrix; This is the attitude integral gain matrix; ω is the attitude differential gain matrix; ff It is the feedforward angular velocity; Where the attitude error e att Using quaternion errors: ; in, For the sign function; q e For error quaternions; For reference to the conjugate of quaternions; The actual measured attitude quaternion; Inner ring angular velocity controller: ; Where τ is the control torque vector; This is the angular velocity proportional gain matrix; This is the integral gain matrix of angular velocity; The differential gain matrix of angular velocity; τ represents the angular velocity error. ff This is the feedforward torque; Feedforward compensation includes wind feedforward compensation and electromagnetic interference compensation, where wind feedforward compensation is based on real-time wind speed v. wind Calculate wind disturbance moment : ; Where ρ is the air density; C d R is the drag coefficient; A is the frontal area; r cp This is the distance vector from the center of pressure to the center of mass; For the drone's velocity vector; Electromagnetic interference compensation: Establishing an error compensation model for the electromagnetic field on the IMU sensor: ; Among them, K em Here is the electromagnetic field compensation coefficient matrix; ω corrected ω is the corrected angular velocity. measured Raw angular velocity measured by the IMU; E field The electric field intensity vector; The control parameters are dynamically adjusted based on the electromagnetic environment strength. ; in, This is the adjusted proportional gain; α is the reference proportional gain; E is the proportional gain adjustment coefficient; norm This represents the normalized electric field strength. This is the adjusted differential gain; β is the base differential gain; β is the differential gain adjustment coefficient.

[0010] Furthermore, the warning threshold settings are as follows: For power transmission lines, model them as spatial line segments, and find the shortest distance from the drone to the transmission line: ; Where P1 and P2 are the coordinates of the endpoints of the conductor; d min P represents the shortest distance from the drone to the wire. drone The current coordinates of the drone; Analytical solution: ; Among them, t The optimal projection parameters; Constrain to the interval [0,1]: ; Among them, t clamp These are the parameter values ​​after constraints. Shortest distance: ; Establish a multi-level early warning mechanism: dmin<1.5×D safe (V) indicates a Level 1 warning; dmin < 1.2 × D safe (V) indicates a Level II warning; dmin < 1.0 × D safe (V) indicates a Level 3 warning; Among them, D safe (V) represents the safe distance for voltage levels.

[0011] Furthermore, safe distances can be ensured by adjusting flight altitude or lateral displacement, as follows: An artificial potential field is established, with the transmission line acting as a repulsive potential field: ; Avoidance force: ; Where d is the current distance; d0 is the potential field's influence range; d vec U is the vector pointing to the nearest obstacle; rep(p) represents the repulsive potential energy function; P is the current position vector of the UAV. η d is the potential field strength coefficient; d is the actual distance from the UAV to the power transmission line; d0 is the potential field influence range; F rep It is the repulsive force vector; The negative gradient of the potential energy function; d vec This represents the distance vector from the obstacle to the drone; When horizontal avoidance is limited, vertical avoidance is used: ; Among them, z avoid The target avoidance height is the height the drone should reach after a vertical avoidance maneuver; z current This is the current altitude of the drone; z safe The safe flight altitude; sign is the avoidance direction determination function; Δz min Minimum vertical avoidance distance; Lateral avoidance strategy, calculate the optimal lateral avoidance direction: ; Where n is a unit vector perpendicular to the ground, v line d is the line direction vector; lateral The unit vector for lateral avoidance direction; Generate a smooth avoidance trajectory using Bézier curves: ; ; ; Where P0 is the current position; P3 is the target position; P(t) is the position of any point on the trajectory; L avoid The characteristic length of the avoidance trajectory; d avoid d is the unit vector of the avoidance direction at the starting point; target The target direction unit vector at the endpoint.

[0012] Furthermore, the acquired raw image data undergoes preprocessing as follows: First, the BRISQUE algorithm is used to calculate the naturalness score of the image; only images with scores higher than a preset value are included in the subsequent processing. An adaptive gamma correction algorithm is used to adjust the brightness and contrast of the image, ensuring a relatively uniform image effect under different lighting conditions. An edge-preserving bilateral filter is employed to remove sensor noise while retaining edge details, which is crucial for subsequent defect boundary identification. A contrast-limited adaptive histogram equalization algorithm is used to adjust the contrast of local areas of the image, highlighting the texture features and potential defects on the hardware surface. Finally, all images are adjusted to a standard size, and a mapping relationship between the original coordinates and standardized coordinates is established to ensure that the detection results can be accurately mapped back to real-world coordinates.

[0013] Furthermore, the improved YOLOv8 detection network is based on the standard YOLOv8 and improved. The backbone network adopts the improved CSPDarknet53 and incorporates an attention mechanism to enhance the feature extraction capability of small target hardware. The neck network uses an enhanced FPN and PAN structure and adds a low-level feature fusion path to improve the detection performance of hardware targets of different scales. The detection head part designs a multi-scale anchor strategy for fine-grained features of hardware defects.

[0014] The present invention has the following beneficial effects: 1. This invention uses an improved A The algorithm's intelligent path planning breaks through the limitations of traditional fixed paths. It can automatically generate the optimal inspection trajectory based on the dynamic safety distance requirements of different voltage levels, taking into account multiple objective factors such as flight efficiency, energy consumption optimization, and safety constraints. This ensures both the comprehensiveness of the inspection and maximizes operational efficiency. 2. This invention avoids the gimbal lock-up problem through quaternion attitude control, and achieves stable flight in complex electromagnetic environments by combining PID control and feedforward compensation. At the same time, it establishes a complete safety monitoring and automatic avoidance mechanism to ensure the safe operation of UAVs in electrified environments. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0016] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments: refer to Figure 1 In this embodiment, a live-line intelligent detection method for damage to power transmission line fittings based on unmanned aerial vehicles (UAVs) is provided, including the following steps: S1: Obtain basic data of transmission lines, including the three-dimensional coordinates of towers, conductor suspension height, line direction and voltage level information, and establish a three-dimensional digital model of the line; S2: Based on the three-dimensional digital model of the line, the dynamic safety distance is determined according to different voltage levels: a minimum safety distance of 1.5 meters for 110kV lines, 3.0 meters for 220kV lines, and 5.0 meters for 500kV lines, using an improved A... The path search method comprehensively considers flight efficiency, energy consumption optimization, and safety constraints to generate the optimal inspection path. S3: Obtain the estimated location of the transmission line fittings and set the hovering coordinates and detection parameters for each detection point; S4: The UAV performs inspections and acquires raw image data based on the optimal inspection path, hovering coordinates and detection parameters set for each detection point; S5: Preprocess the acquired raw image data to obtain preprocessed image data; S6: Input the preprocessed image data into the improved YOLOv8 detection network to obtain the type and location of hardware damage.

[0017] In this embodiment, basic data of the transmission line are acquired, including the three-dimensional coordinates of the towers, conductor suspension height, line route, and voltage level information, and a three-dimensional digital model of the line is established, as follows: The basic engineering data of the lines were exported from the GIS and SCADA systems of the power grid company, including tower number, tower type, coordinate information (longitude, latitude, elevation), conductor suspension point height, and line route information; for newly built lines not in the database or areas with low geographic coordinate accuracy, high-precision RTK mapping technology using unmanned aerial vehicles was used to perform supplementary surveys. Collect voltage level, electrical clearance parameters and standard design data for fitting arrangement to define line operation safety parameters. Based on voltage level (110kV, 220kV, 500kV and above), automatically configure minimum safety distance and live working safety boundary, and use different electric field strength models for different voltage areas so as to superimpose electromagnetic environment influences in subsequent modeling. A three-dimensional digital model is established by integrating BIM modeling and point cloud reconstruction technology. Towers, conductors, and hardware objects are imported into the modeling engine in a parametric form. An interactive 3D model is generated through three-dimensional geometric reconstruction and spatial constraints. The conductor suspension points are fitted with spline curves to fit the sag of the real line. The tower structure is automatically generated by matching standard component libraries. Finally, each object in the model (such as tower body, conductor, insulator string, and hardware) is given spatial coordinates, physical attributes, and a unique number, realizing the spatial semantic integrity of the digital line.

[0018] In this embodiment, based on the three-dimensional digital model of the line and the dynamic safety distance determined according to different voltage levels, an improved A... The path search method, taking into account flight efficiency, energy consumption optimization, and safety constraints, generates the optimal inspection path, as detailed below: First, in the three-dimensional digital line space model, the UAV inspection space state S(x,y,z) is defined, and the feasible space Ω is set. s No-fly zone Ω f : ; Where Li represents the i-th segment of the conductor or tower body; D((x,y,z),L i D represents the shortest Euclidean distance from the drone to the line or equipment. safe (V) is a safety distance function related to voltage level V; The dynamic safety distance is modeled using an empirical formula as follows: ; Where α and β are empirical coefficients; In the feasible space Ω s A 3D raster map is constructed using an improved A... The algorithm searches for the optimal inspection path.

[0019] In this embodiment, the improved A The algorithm introduces an energy consumption term and an attitude penalty term. The improved cost function is as follows: ; Where g(n) is the cumulative flight cost from the starting point to the current node; h(n) is the Euclidean distance heuristic function from the target point to the current node; E c (n) is the energy consumption function per unit distance, representing the energy loss caused by load, speed, wind resistance, etc.; θ p (n) represents the attitude change penalty term, indicating the flight control adjustment cost caused by the nodal turn angle or pitch angle; w1, w2, w3, and w4 are the weight coefficients of the four terms, respectively. ; ; Where ρ is the air density; C d denoted as drag coefficient; A is the projected area of ​​the UAV; v is the instantaneous velocity of the UAV; φ is the flight tilt angle; P0 is the hovering base power; Δψ(n) represents the change in heading angle between adjacent nodes; γ is the attitude penalty factor. The overall objective function is optimized as follows: ; Where path is the complete flight path from the starting point to the ending point, passing through all detection points; J is the total path cost; n is a single node on the path; and f(n) is the comprehensive cost function of a single node. Under the following constraints: (x n ,y n ,z n )∈Ω s That is, safe flight space; v min ≤v(t)≤v max ; |a(t)|≤a max ; Among them, (x n ,y n ,z n Let be the three-dimensional spatial coordinates of path node n; v(t) be the instantaneous flight velocity of the UAV at time t; v min Minimum flight speed limit; v max The maximum flight speed limit is defined by a(t), where a(t) is the acceleration vector of the UAV at time t, and |a(t)| is the magnitude of the acceleration vector. max The maximum permissible acceleration; During algorithm execution, the safety boundary D is dynamically adjusted based on real-time wind speed, electric field strength, and UAV attitude. safe ,Right now: ; Among them, v w For real-time wind speed, δ w E is the wind speed safety correction factor. field This is a measured value of the electric field strength; The safe flight distance is updated in real time; δ E The weighting of the electric field influence; D safe (V) represents the baseline safety distance; Meanwhile, a detection visibility constraint C is introduced into the UAV vision modeling. v Node planning is valid only when the target detection point is within the camera's field of view: ; Where, d view r is the camera's line-of-sight vector. target is the target pointing vector; FOV is the field of view angle.

[0020] In this embodiment, the attitude control of the UAV inspection adopts quaternion representation to avoid the gimbal lock-up problem. By combining PID controller and feedforward compensation, the UAV can achieve stable flight in complex electromagnetic environment, monitor the distance to the power transmission line in real time, and automatically start the avoidance program when the safe distance is detected to be close to the warning threshold, and ensure the safe distance by adjusting the flight altitude or lateral displacement.

[0021] In this embodiment, quaternion representation is used to avoid gimbal lock-up. A combination of PID controller and feedforward compensation is used to achieve stable flight of the UAV in complex electromagnetic environments, as detailed below: Using quaternions q = [q0, q1, q2, q3] T Indicates the attitude of the drone, where: q = q0 + q1a + q2b + q3c; Where a represents a 90° rotation about the x-axis; b represents a 90° rotation about the y-axis; and c represents a 90° rotation about the z-axis. The conversion relationship between quaternions and Euler angles: ; in, φ, θ, ψ These are roll angle, pitch angle, and yaw angle, respectively. Quaternion differential equation: ; Where ω=[0,ω x ,ω y ,ω z ] T The angular velocity quaternion, This represents quaternion multiplication. Expanded format: ; The PID controller adopts a dual-loop control structure, including an outer-loop attitude controller and an inner-loop angular velocity controller; The outer loop attitude controller: ; in, ω ref The reference angular velocity vector; This is the attitude scaling gain matrix; This is the attitude integral gain matrix; ω is the attitude differential gain matrix; ff It is the feedforward angular velocity; Where the attitude error e att Using quaternion errors: ; in, For the sign function; q e For error quaternions; For reference to the conjugate of quaternions; The actual measured attitude quaternion; Inner ring angular velocity controller: ; Where τ is the control torque vector; This is the angular velocity proportional gain matrix; This is the integral gain matrix of angular velocity; The differential gain matrix of angular velocity; τ represents the angular velocity error. ff This is the feedforward torque; Feedforward compensation includes wind feedforward compensation and electromagnetic interference compensation, where wind feedforward compensation is based on real-time wind speed v. wind Calculate wind disturbance moment : ; Where ρ is the air density; C d R is the drag coefficient; A is the frontal area; r cp This is the distance vector from the center of pressure to the center of mass; For the drone's velocity vector; Electromagnetic interference compensation: Establishing an error compensation model for the electromagnetic field on the IMU sensor: ; Among them, K em Here is the electromagnetic field compensation coefficient matrix; ω corrected ω is the corrected angular velocity. measured Raw angular velocity measured by the IMU; E field The electric field intensity vector; The control parameters are dynamically adjusted based on the electromagnetic environment strength. ; in, This is the adjusted proportional gain; The reference proportional gain (nominal value without electromagnetic interference); α is the proportional gain adjustment factor; E norm This represents the normalized electric field strength. This is the adjusted differential gain; β is the base differential gain; β is the differential gain adjustment coefficient.

[0022] In this embodiment, the warning threshold is set as follows: For power transmission lines, model them as spatial line segments, and find the shortest distance from the drone to the transmission line: ; Where P1 and P2 are the coordinates of the endpoints of the conductor; d min P represents the shortest distance from the drone to the wire. drone The current coordinates of the drone; Analytical solution: ; Among them, t The optimal projection parameters; Constrain to the interval [0,1]: ; Among them, t clamp These are the parameter values ​​after constraints. Shortest distance: ; Establish a multi-level early warning mechanism: dmin<1.5×D safe (V) indicates a Level 1 warning; dmin < 1.2 × D safe (V) indicates a Level II warning; dmin < 1.0 × D safe (V) indicates a Level 3 warning; Among them, D safe (V) represents the safe distance for voltage levels.

[0023] In this embodiment, a safe distance is ensured by adjusting the flight altitude or lateral displacement, as follows: An artificial potential field is established, with the transmission line acting as a repulsive potential field: ; Avoidance force: ; Where d is the current distance; d0 is the potential field's influence range; d vec U is the vector pointing to the nearest obstacle; rep (p) represents the repulsive potential energy function; P is the current position vector of the UAV. η d is the potential field strength coefficient; d is the actual distance from the UAV to the power transmission line; d0 is the potential field influence range; F rep It is the repulsive force vector; The negative gradient of the potential energy function; d vec This represents the distance vector from the obstacle to the drone; When horizontal avoidance is limited, vertical avoidance is used: ; Among them, z avoid The target avoidance height is the height the drone should reach after a vertical avoidance maneuver; z currentThis is the current altitude of the drone; z safe The safe flight altitude; sign is the avoidance direction determination function; Δz min Minimum vertical avoidance distance; Lateral avoidance strategy, calculate the optimal lateral avoidance direction: ; Where n is a unit vector perpendicular to the ground, v line d is the line direction vector; lateral The unit vector for lateral avoidance direction; Generate a smooth avoidance trajectory using Bézier curves: ; ; ; Where P0 is the current position; P3 is the target position; P(t) is the position of any point on the trajectory; L avoid The characteristic length of the avoidance trajectory; d avoid d is the unit vector of the avoidance direction at the starting point; target The target direction unit vector at the endpoint.

[0024] In this embodiment, the acquired raw image data is preprocessed as follows: First, the BRISQUE (Blind Image Spatial Quality Assessment) algorithm is used to calculate the naturalness score of the image. Only images with scores higher than a preset value are included in the subsequent processing flow. An adaptive gamma correction algorithm is used to adjust the brightness and contrast of the image to ensure a relatively uniform image effect under different lighting conditions. An edge-preserving bilateral filter is used to remove sensor noise while retaining the edge details of the image, which is crucial for subsequent defect boundary identification. A contrast-limited adaptive histogram equalization (CLAHE) algorithm is used to adjust the contrast of local areas of the image to highlight the texture features and potential defects of the hardware surface. Finally, all images are adjusted to a standard size, and a mapping relationship between the original coordinates and the standardized coordinates is established to ensure that the detection results can be accurately mapped back to real-world coordinates.

[0025] In this embodiment, the improved YOLOv8 detection network is based on the standard YOLOv8 and improved. The backbone network adopts the improved CSPDarknet53 and incorporates attention mechanisms (CBAM and SE modules) to enhance the feature extraction capability of small target fittings. The neck network uses an enhanced FPN and PAN structure and adds a low-level feature fusion path to improve the detection performance of fitting targets of different scales. The detection head part designs a multi-scale anchor strategy for fine-grained features of fitting defects, including 9 anchor boxes of different scales to adapt to the size range from small vibration dampers to large insulator strings.

[0026] Preferably, in this embodiment, the improved YOLOv8 network adopts a multi-stage detection strategy: the first stage performs hardware target detection, identifying the type and location of all hardware in the image; the second stage performs fine-grained defect detection on the detected hardware regions, identifying specific damage types and locations; the third stage performs multi-scale feature fusion, combining detection results at different resolutions to improve the recognition accuracy of small defects. The TensorRT acceleration engine is used during network inference, controlling the inference time for a single image to within 50ms, ensuring real-time detection capabilities. The confidence threshold is set to 0.6, and the NMS (non-maximum suppression) threshold is set to 0.4, effectively suppressing duplicate detection boxes. For multi-angle images of the same hardware, the system adopts a weighted fusion strategy, assigning weights to each detection result based on factors such as image quality, shooting angle, and lighting conditions, ultimately outputting the detection result with the highest confidence. The network output includes bounding box coordinates, category labels, confidence scores, and newly added severity scores (levels 1-5) and urgency indicators (normal / concern / alarm / severe) for defect detection.

[0027] Finally, post-processing and quality control verification of the detection results are implemented. The post-processing module first performs coordinate transformation on the raw network output, converting relative coordinates into image pixel coordinates, and further converting them into physical locations in the real-world coordinate system based on image metadata information. The result filter performs logical checks based on prior knowledge of the transmission line: verifying whether the detected fitting type conforms to the standard configuration of the line at this voltage level, checking whether the defect location is within a reasonable spatial range, and eliminating obvious false detection results. The multi-target tracking algorithm establishes temporal correlations for the same fitting in continuous images, and achieves cross-frame target matching through the DeepSORT algorithm to ensure the temporal consistency of the detection results. The anomaly detection module uses the isolated forest algorithm to identify abnormal detection results, marking results with abnormally low confidence or abnormal locations, triggering a manual review process. The system also establishes a three-level quality control system for the detection results: automatic quality control (based on algorithm rules), semi-automatic quality control (AI-assisted manual control), and expert quality control (reviewed by senior engineers) to ensure the accuracy and reliability of the output results. All detection results are automatically saved to a structured database, supporting multi-dimensional queries and statistical analysis by time, location, defect type, and other dimensions.

[0028] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0029] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0030] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0031] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0032] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A live-line intelligent detection method for power transmission line hardware damage based on unmanned aerial vehicles (UAVs), characterized in that, Includes the following steps: S1: Obtain basic data of transmission lines, including the three-dimensional coordinates of towers, conductor suspension height, line direction and voltage level information, and establish a three-dimensional digital model of the line; S2: Based on the three-dimensional digital model of the line, and according to the dynamic safety distance determined by different voltage levels, an improved A... The path search method comprehensively considers flight efficiency, energy consumption optimization, and safety constraints to generate the optimal inspection path. S3: Obtain the estimated location of the transmission line fittings and set the hovering coordinates and detection parameters for each detection point; S4: The UAV performs inspections and acquires raw image data based on the optimal inspection path, hovering coordinates and detection parameters set for each detection point; S5: Preprocess the acquired raw image data to obtain preprocessed image data; S6: Input the preprocessed image data into the improved YOLOv8 detection network to obtain the type and location of hardware damage.

2. The method for intelligent detection of live-line damage to transmission line fittings based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The acquisition of basic data for transmission lines, including the three-dimensional coordinates of towers, conductor suspension height, line alignment, and voltage level information, and the establishment of a three-dimensional digital model of the line, is detailed below: The basic engineering data of the lines were exported from the GIS and SCADA systems of the power grid company, including tower number, tower type, coordinate information, conductor suspension point height, and line route information; for newly built lines not in the database or areas with low geographic coordinate accuracy, high-precision RTK mapping technology using unmanned aerial vehicles was used to perform supplementary surveys. Collect voltage level, electrical clearance parameters and standard design data for fitting arrangement to define line operation safety parameters, and automatically configure minimum safety distance and live-line work safety boundaries according to voltage level; A three-dimensional digital model is established by integrating BIM modeling and point cloud reconstruction technology. Towers, conductors, and metal structures are imported into the modeling engine in a parametric form. An interactive 3D model is generated through three-dimensional geometric reconstruction and spatial constraints. The suspension points of conductors are fitted with spline curves to fit the sag of the real line. The tower structure is automatically generated by matching standard component libraries. Finally, each object in the model is given spatial coordinates, physical attributes, and a unique number, realizing the spatial semantic integrity of the digital line.

3. The method for intelligent detection of live-line damage to transmission line fittings based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The three-dimensional digital model based on the line, according to the dynamic safety distance determined by different voltage levels, adopts an improved A... The path search method, taking into account flight efficiency, energy consumption optimization, and safety constraints, generates the optimal inspection path, as detailed below: First, in the three-dimensional digital line space model, the UAV inspection space state S(x,y,z) is defined, and the feasible space Ω is set. s No-fly zone Ω f : ； Where Li represents the i-th segment of the conductor or tower body; D((x,y,z),L i D represents the shortest Euclidean distance from the drone to the line or equipment. safe (V) is a safety distance function related to voltage level V; The dynamic safety distance is modeled using an empirical formula as follows: ； Where α and β are empirical coefficients; In the feasible space Ω s A 3D raster map is constructed using an improved A... The algorithm searches for the optimal inspection path.

4. The method for intelligent detection of live-line damage to transmission line fittings based on unmanned aerial vehicles (UAVs) according to claim 3, characterized in that, The improved A The algorithm introduces an energy consumption term and an attitude penalty term. The improved cost function is as follows: ； Where g(n) is the cumulative flight cost from the starting point to the current node; h(n) is the Euclidean distance heuristic function from the target point to the current node; E c (n) is the energy consumption function per unit distance; θ p (n) represents the attitude change penalty term, indicating the flight control adjustment cost caused by the nodal turn angle or pitch angle; w1, w2, w3, and w4 are the weight coefficients of the four terms, respectively. ； ； Where ρ is the air density; C d denoted as drag coefficient; A is the projected area of ​​the UAV; v is the instantaneous velocity of the UAV; φ is the flight tilt angle; P0 is the hovering base power; Δψ(n) represents the change in heading angle between adjacent nodes; γ is the attitude penalty factor. The overall objective function is optimized as follows: ； Where path is the complete flight path from the starting point to the ending point, passing through all detection points; J is the total path cost; n is a single node on the path; and f(n) is the comprehensive cost function of a single node. Under the following constraints: (x n ,y n ,z n )∈Ω s That is, safe flight space; v min ≤v(t)≤v max ; ∣a(t)∣≤a max ; Among them, (x n ,y n ,z n Let be the three-dimensional spatial coordinates of path node n; v(t) be the instantaneous flight velocity of the UAV at time t; v min Minimum flight speed limit; v max The maximum flight speed limit is defined by a(t), where a(t) is the acceleration vector of the UAV at time t, and |a(t)| is the magnitude of the acceleration vector. max The maximum permissible acceleration; During algorithm execution, the safety boundary D is dynamically adjusted based on real-time wind speed, electric field strength, and UAV attitude. safe ,Right now: ； Among them, v w For real-time wind speed, δ w E is the wind speed safety correction factor. field This is a measured value of the electric field strength; The safe flight distance is updated in real time; δ E The weighting of the electric field influence; D safe (V) represents the baseline safety distance; Meanwhile, a detection visibility constraint C is introduced into the UAV vision modeling. v Node planning is valid only when the target detection point is within the camera's field of view: ； Where, d view r is the camera's line-of-sight vector. target is the target pointing vector; FOV is the field of view angle.

5. The method for intelligent detection of live-line damage to transmission line fittings based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The attitude control of the UAV inspection uses quaternion representation to avoid gimbal lock-up. By combining PID controller and feedforward compensation, the UAV can achieve stable flight in complex electromagnetic environments. It monitors the distance to the power transmission line in real time. When the safe distance is detected to be close to the warning threshold, the avoidance program is automatically started, and the safe distance is ensured by adjusting the flight altitude or lateral displacement.

6. The method for intelligent detection of live-line damage to transmission line fittings based on unmanned aerial vehicles (UAVs) according to claim 5, characterized in that, The method employs quaternion representation to avoid gimbal lock-up issues and combines PID controllers with feedforward compensation to achieve stable flight of the UAV in complex electromagnetic environments, as detailed below: Using quaternions q = [q0, q1, q2, q3] T Indicates the attitude of the drone, where: q = q0 + q1a + q2b + q3c; Where a represents a 90° rotation about the x-axis; b represents a 90° rotation about the y-axis; and c represents a 90° rotation about the z-axis. The conversion relationship between quaternions and Euler angles: ; in, φ, θ, ψ These are roll angle, pitch angle, and yaw angle, respectively. Quaternion differential equation: ; Where ω=[0,ω x ,ω y ,ω z ] T The angular velocity quaternion, This represents quaternion multiplication. Expanded format: ； The PID controller adopts a dual-loop control structure, including an outer-loop attitude controller and an inner-loop angular velocity controller; The outer loop attitude controller: ； in, ω ref The reference angular velocity vector; This is the attitude scaling gain matrix; This is the attitude integral gain matrix; ω is the attitude differential gain matrix; ff It is the feedforward angular velocity; Where the attitude error e att Using quaternion errors: ； in, For the sign function; q e For error quaternions; For reference to the conjugate of quaternions; The actual measured attitude quaternion; Inner ring angular velocity controller: ； Where τ is the control torque vector; This is the angular velocity proportional gain matrix; This is the integral gain matrix of angular velocity; The differential gain matrix of angular velocity; τ represents the angular velocity error. ff This is the feedforward torque; Feedforward compensation includes wind feedforward compensation and electromagnetic interference compensation, where wind feedforward compensation is based on real-time wind speed v. wind Calculate wind disturbance moment : ； Where ρ is the air density; C d R is the drag coefficient; A is the frontal area; r cp This is the distance vector from the center of pressure to the center of mass; For the drone's velocity vector; Electromagnetic interference compensation: Establishing an error compensation model for the electromagnetic field on the IMU sensor: ； Among them, K em Here is the electromagnetic field compensation coefficient matrix; ω corrected ω is the corrected angular velocity. measured Raw angular velocity measured by the IMU; E field The electric field intensity vector; The control parameters are dynamically adjusted based on the electromagnetic environment strength. ； in, This is the adjusted proportional gain; α is the reference proportional gain; E is the proportional gain adjustment coefficient; norm This represents the normalized electric field strength. This is the adjusted differential gain; β is the base differential gain; β is the differential gain adjustment coefficient.

7. The method for intelligent detection of live-line damage to transmission line fittings based on unmanned aerial vehicles (UAVs) according to claim 6, characterized in that, The warning threshold is set as follows: For power transmission lines, model them as spatial line segments, and find the shortest distance from the drone to the transmission line: ； Where P1 and P2 are the coordinates of the endpoints of the conductor; d min P represents the shortest distance from the drone to the wire. drone The coordinates of the drone's current position; Analytical solution: ； Among them, t The optimal projection parameters; Constrain to the interval [0,1]: ； Among them, t clamp These are the parameter values ​​after constraints; Shortest distance: ； Establish a multi-level early warning mechanism: dmin<1.5×D safe (V) indicates a Level 1 warning; dmin < 1.2 × D safe (V) indicates a Level II warning; dmin < 1.0 × D safe (V) indicates a Level 3 warning; Among them, D safe (V) represents the safe distance for voltage levels.

8. The method for intelligent detection of live-line damage to transmission line fittings based on unmanned aerial vehicles (UAVs) according to claim 7, characterized in that, The method of ensuring a safe distance by adjusting flight altitude or lateral displacement is as follows: An artificial potential field is established, with the transmission line acting as a repulsive potential field: ； Avoidance force: ； Where d is the current distance; d0 is the potential field's influence range; d vec U is the vector pointing to the nearest obstacle; rep (p) represents the repulsive potential energy function; P is the current position vector of the UAV. η denoted as the potential field strength coefficient; d represents the actual distance from the UAV to the power transmission line. d0 is the range of influence of the potential field; F rep It is the repulsive force vector; The negative gradient of the potential energy function; d vec This represents the distance vector from the obstacle to the drone. When horizontal avoidance is limited, vertical avoidance is used: ； Among them, z avoid The target avoidance height is the height the drone should reach after a vertical avoidance maneuver; z current This is the current altitude of the drone; z safe The safe flight altitude; sign is the avoidance direction determination function; Δz min Minimum vertical avoidance distance; Lateral avoidance strategy, calculate the optimal lateral avoidance direction: ； Where n is a unit vector perpendicular to the ground, v line d is the line direction vector; lateral The unit vector for lateral avoidance direction; Generate a smooth avoidance trajectory using Bézier curves: ； ； ； Where P0 is the current position; P3 is the target position; P(t) is the position of any point on the trajectory; L avoid The characteristic length of the avoidance trajectory; d avoid d is the unit vector representing the avoidance direction at the starting point; target The target direction unit vector at the endpoint.

9. The method for intelligent detection of live-line damage to transmission line fittings based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The preprocessing of the acquired raw image data is as follows: First, the BRISQUE algorithm is used to calculate the naturalness score of the image; only images with scores higher than a preset value are included in the subsequent processing. An adaptive gamma correction algorithm is used to adjust the brightness and contrast of the image to ensure a relatively uniform image effect under different lighting conditions. An edge-preserving bilateral filter is used to remove sensor noise while retaining edge details, which is crucial for subsequent defect boundary identification. A contrast-limited adaptive histogram equalization algorithm is used to adjust the contrast of local areas of the image, highlighting the texture features and potential defects on the hardware surface. Finally, all images are adjusted to a standard size, and a mapping relationship between the original coordinates and standardized coordinates is established to ensure that the detection results can be accurately mapped back to real-world coordinates.

10. The method for intelligent detection of live-line damage to transmission line fittings based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The improved YOLOv8 detection network is based on the standard YOLOv8. The backbone network adopts the improved CSPDarknet53 and incorporates an attention mechanism to enhance the feature extraction capability for small target hardware. The neck network uses an enhanced FPN and PAN structure and adds a low-level feature fusion path to improve the detection performance of hardware targets of different scales. The detection head part designs a multi-scale anchor strategy for fine-grained features of hardware defects.

Images