A kind of power line insulator defect detection device based on binocular unmanned aircraft marshalling

Abstract

This invention relates to the field of power line inspection equipment technology, and discloses a power line insulator defect detection device based on binocular UAV ensemble, comprising: a UAV ensemble execution module, a binocular vision inspection module, an ensemble collaborative control module, a data processing and defect identification module, a wireless communication module, and a ground monitoring terminal; the UAV ensemble execution module adopts a master-slave ensemble architecture, consisting of one master-controlled UAV and several slave-controlled inspection UAVs forming an inspection ensemble. By employing binocular synchronous imaging technology, it simultaneously acquires two-dimensional appearance and three-dimensional depth information of the insulator, completely solving the problems of single-eye detection lacking depth and prone to false detection and missed detection; moreover, the master-slave ensemble architecture enables multi-machine synchronous collaborative operation, with a wide coverage area, high inspection efficiency, and the ability to complete long-distance line inspections in one go, possessing fault redundancy capability, and significantly improving operational stability and continuity.

Description

Technical Field

[0001] This invention belongs to the technical field of power line inspection equipment, specifically relating to a power line insulator defect detection device based on binocular UAV ensemble. Background Technology

[0002] Insulators are critical insulation and support components of overhead transmission lines. Operating in complex outdoor environments for extended periods, they are prone to defects such as insulator breakage, cracks, contamination, missing sheds, corrosion of fittings, and insulator string tilting. If these defects are not detected and addressed promptly, they can easily trigger line flashovers, tripping, and even widespread power outages, seriously threatening the safe and stable operation of the power grid.

[0003] Currently, defect detection of power line insulators mainly relies on drone inspections, and this mainstream technology has significant shortcomings:

[0004] 1. Conventional monocular UAV inspection can only acquire two-dimensional planar images, lacking three-dimensional depth information and spatial positioning capabilities, which affects the quantitative calculation of defects; it is greatly affected by lighting, shooting angle, and background interference, resulting in a high false detection rate and missed detection rate for minor defects; and the inspection range is limited, making it difficult to meet the needs of long-distance and cross-regional line inspection.

[0005] 2. Existing multi-drone inspections are mostly conducted by individual drones in separate segments, lacking unified formation control, attitude coordination, data linkage, and a mechanism for precise defect localization based on captured images, feeding the defect results back to the host, and then having the host control and coordinate the slave drones to capture images of defective insulators from multiple angles. This easily leads to repeated inspections or blind spots, resulting in low inspection efficiency.

[0006] 3. Although binocular vision has three-dimensional detection capabilities, it is not deeply integrated with the drone grouping mode. Furthermore, general binocular devices have not been optimized for specific parameters in scenarios such as insulator shape, high altitude and long distance, and flight shaking. As a result, image distortion correction is poor, stereo matching accuracy is low, and the three-dimensional reconstruction effect is not good. It cannot meet the requirements for accurate identification of multiple types of defects in power line insulators and high-precision location and quantitative calculation of defects.

[0007] After comparison, the closest existing technologies to this invention are: the monocular vision autonomous inspection scheme in publication number CN112164015A, representing the traditional single-machine monocular inspection mode; and the UAV swarm inspection system in publication number CN120237561A, representing the conventional multi-machine inspection scheme. Neither of these schemes achieves deep integration of binocular vision and UAV swarm collaboration, and therefore cannot simultaneously ensure detection accuracy, inspection efficiency, and defect location reliability. Therefore, developing an insulator defect detection device that integrates UAV swarm collaborative control and binocular vision recognition has become a key technical problem urgently needing to be solved in the field of intelligent power operation and maintenance. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a power line insulator defect detection device based on binocular UAV ensembles. This device addresses the insufficient fusion between UAV ensembles and general binocular vision systems by constructing a deeply integrated collaborative perception and operation mechanism. This effectively overcomes the technical bottleneck of existing solutions in balancing detection types, detection accuracy, inspection efficiency, and defect location reliability, thereby achieving synergistic optimization of multi-objective performance.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a power line insulator defect detection device based on binocular UAV grouping, comprising: a UAV grouping execution module, a binocular vision detection module, a grouping coordination control module, a data processing and defect identification module, a wireless communication module, and a ground monitoring terminal; The drone formation execution module adopts a master-slave formation architecture, consisting of one master drone and several slave detection drones forming an inspection formation; The binocular vision inspection module is installed on each slave inspection drone and is used to collect left and right view images of the insulator and obtain three-dimensional depth information. The grouping and coordination control module is built into the main control UAV and is used to coordinate formation path planning, attitude coordination, task allocation and fault redundancy control. The data processing and defect recognition module is used to preprocess, stereo matching, 3D reconstruction and intelligent defect recognition of binocular images; The wireless communication module is used to realize data transmission and command interaction between the UAV group and the ground monitoring terminal.

[0010] Preferably, in the UAV grouping execution module, the number of slave detection UAVs is 2-4, and a lightweight multi-rotor structure is adopted; Each controlled inspection drone is equipped with a three-axis stabilization gimbal, which can rotate 360° horizontally and ±90° vertically to ensure that the binocular camera is always pointed at the insulator string and to eliminate the impact of flight shaking on image quality. Both the master-controlled UAV and the slave-controlled detection UAV use differential GPS + inertial navigation combined positioning, achieving a positioning accuracy of millimeters.

[0011] Preferably, the binocular vision detection module adopts a binocular synchronous high-definition imaging unit, which includes two sets of industrial-grade high-definition CMOS cameras with identical parameters, a synchronous trigger controller, and an adaptive fill light unit. Two sets of cameras are horizontally parallel and are strictly calibrated for baseline distance and optical axis parallelism to eliminate imaging distortion errors; The camera has a resolution of at least 4K and a frame rate of at least 25 FPS. The synchronous trigger controller ensures that dual-channel images are exposed and acquired simultaneously, ensuring the consistency of imaging time for the left and right views, and providing a foundation for subsequent stereo matching and depth calculation.

[0012] Preferably, the grouping and coordination control module integrates a formation path planning algorithm, an attitude coordination algorithm, an obstacle avoidance control algorithm, and a task allocation unit; It can automatically generate the optimal inspection route based on GIS data of power transmission lines, dynamically adjust the formation and flight parameters, and allocate inspection sections to each slave aircraft. It has a fault redundancy and replacement function. When a single slave drone fails, the master drone can be reassigned tasks and the remaining slave drones can automatically take over.

[0013] Preferably, the data processing and defect identification module is equipped with an edge computing chip and executes the following sequentially: The left and right views acquired by binoculars are preprocessed with distortion correction, grayscale conversion, noise reduction and enhancement. A stereo matching algorithm based on feature matching quickly matches corresponding pixels in the left and right views and calculates the disparity map. Reconstruct a 3D point cloud model of the insulator to obtain the 3D contour and spatial coordinate information of the insulator surface; Using RGB images captured by a camera as input, an improved small target recognition deep learning model is used to accurately identify the types and locations of defects such as insulator damage, cracks, dirt, missing sheds, metal corrosion, and insulator string tilt. Furthermore, the two-dimensional images and three-dimensional point cloud data of the insulator are fused and analyzed to achieve quantitative calculation of the damaged area, crack length, and width of the insulator. The accuracy rate of defect type detection is ≥95%, and the accuracy of detecting minute defects reaches 0.5mm.

[0014] Preferably, the wireless communication module adopts 5G and WiFi dual-link communication to realize high-speed and stable data transmission between the UAV group and the ground monitoring terminal, and to transmit data such as binocular imaging video, defect detection results, UAV flight status, and position coordinates in real time. The ground monitoring terminal has functions such as real-time data display, storage, manual defect review, report generation, and remote control.

[0015] A method for detecting defects in power line insulators based on binocular UAV ensembles includes the following steps: S1. Establish a master-slave drone formation, consisting of 1 master drone and 2-4 slave detection drones, to complete pre-flight self-check and positioning calibration; S2, the grouping and collaborative control module plans the optimal inspection path based on GIS data, and the main control UAV coordinates the formation attitude, speed and task allocation; S3. Each slave-controlled inspection drone drives the binocular vision inspection module through a three-axis stabilized gimbal to simultaneously acquire left and right view images of the insulator. S4, the data processing and defect recognition module, performs preprocessing, stereo matching, and 3D point cloud reconstruction on binocular images, and integrates a deep learning model to complete defect recognition and localization. S5. The wireless communication module transmits defect information, flight status and imaging data back to the ground monitoring terminal in real time. S6. The ground monitoring terminal completes data display, storage, verification and report output, realizing full controllability of the inspection process.

[0016] Preferably, the binocular vision detection module in S3 operates at an inspection distance of 1–5m, simultaneously acquiring two-dimensional appearance features and three-dimensional depth information of the insulator to achieve millimeter-level spatial positioning of defects.

[0017] Preferably, the defect identification types in S4 include: insulator damage, cracks, dirt, missing sheds, metal corrosion, insulator string tilting, and other defects.

[0018] Preferably, when any slave-controlled inspection drone malfunctions, the master-controlled drone automatically activates the redundancy strategy, re-divides the inspection section, and the remaining slave drones take over to perform the inspection, ensuring continuous and stable operation.

[0019] Compared with the prior art, the present invention provides a power line insulator defect detection device based on binocular UAV formation, which has the following advantages: 1. Significantly improved binocular inspection accuracy: Adopting binocular synchronous imaging technology, it simultaneously acquires two-dimensional appearance and three-dimensional depth information of insulators, completely solving the problems of monocular inspection lacking depth and being prone to false detection and missed detection. It has high accuracy in identifying minute defects, and the spatial positioning of defects reaches the millimeter level, which is suitable for the high-precision inspection requirements of power systems.

[0020] 2. High efficiency of drone group collaboration: The master-slave group architecture enables multiple drones to work synchronously and collaboratively, with a wide coverage area and high inspection efficiency. It can complete long-distance line inspections in one go, avoid the short endurance of single drones, and has fault redundancy capability, significantly improving the stability and continuity of operations.

[0021] 3. High level of intelligence: It integrates edge computing and deep learning models to realize real-time automatic detection, location and classification of defects, eliminating the need for manual frame-by-frame image analysis, reducing manual workload, improving the level of intelligent inspection, and adapting to all-weather operation in complex field environments.

[0022] 4. High adaptability and reliability: The three-axis stabilization gimbal and adaptive lighting design eliminate flight jitter and light interference. Formation obstacle avoidance and redundant control functions ensure the safety of high-altitude operations. The overall device structure is lightweight and adaptable to various power transmission line insulator testing scenarios. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the power line insulator defect detection device based on binocular UAV formation according to the present invention; Figure 2 This is a schematic diagram illustrating the steps of the power line insulator defect detection device based on binocular UAV formation according to the present invention; Figure 3 This is a schematic diagram of the structure of the UAV grouping execution module of the present invention; Figure 4 This is a schematic diagram of the structure of the main control drone of the present invention; Figure 5 This is a schematic diagram of the structure of the controlled detection drone of the present invention.

[0024] In the diagram: 100, UAV grouping execution module; 110, master control UAV; 120, slave control detection UAV; 200, binocular vision detection module; 300, grouping and collaborative control module; 400, data processing and defect identification module; 500, wireless communication module; 600, ground monitoring terminal. Detailed Implementation

[0025] 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.

[0026] Example 1: See attached document Figure 1 , Figure 3 , Figure 4 and Figure 5 A power line insulator defect detection device based on binocular UAV grouping includes: UAV grouping execution module 100, binocular vision detection module 200, grouping coordination control module 300, data processing and defect identification module 400, wireless communication module 500, and ground monitoring terminal 600.

[0027] The above six modules constitute a complete technical system of master-slave collaborative inspection + binocular 3D imaging + edge intelligent recognition. Each module forms a closed-loop workflow through hardware electrical connection, wireless communication and software data interaction, realizing fully automated operation from UAV formation control, insulator 3D imaging, defect intelligent detection to data feedback monitoring. Standardized communication protocols and data interfaces are used between modules to ensure system compatibility, scalability and operational stability.

[0028] The UAV grouping execution module 100 adopts a master-slave grouping architecture, consisting of one master-controlled UAV 110 and several slave-controlled detection UAVs 120 forming an inspection formation, which is suitable for the inspection needs of power transmission lines of different lengths.

[0029] The master-slave architecture uses the master-controlled UAV 110 as the core control node of the formation, which undertakes global path planning, task scheduling, status monitoring and fault handling functions. The slave-controlled detection UAV 120 is a distributed execution node, which is dedicated to completing insulator image acquisition and local detection tasks. This architecture can flexibly adjust the number of slaves according to the transmission line mileage, tower density and terrain complexity, so as to realize long-distance, large-area and multi-section synchronous inspection, and solve the problems of limited coverage and low operation efficiency of single-machine inspection.

[0030] A binocular vision inspection module 200 is installed on each slave-controlled inspection drone 120 to acquire left and right view images of the insulator and obtain three-dimensional depth information.

[0031] The binocular vision inspection module 200 is based on the principle of binocular stereo vision. It acquires the two-dimensional texture features and three-dimensional spatial coordinates of the insulator simultaneously through two sets of cameras with the same parameters, upgrading planar image detection to three-dimensional stereo detection. It can accurately extract the geometric features and spatial location information of tiny defects on the surface of the insulator. It overcomes the technical defects of monocular vision, such as lack of depth, ambiguous positioning, and susceptibility to light interference, from the principle level, and improves the reliability of defect detection.

[0032] The grouping and coordination control module 300 is built into the main control UAV 110 and is used to coordinate formation path planning, attitude coordination, task allocation and fault redundancy control.

[0033] The 300 group collaborative control module integrates multi-machine collaborative control algorithms and real-time status feedback mechanisms. It collects operational parameters such as position, attitude, power, and gimbal angle of each slave UAV in real time through a high-speed data link. Combined with the geographical information of power transmission lines and obstacle distribution data, it dynamically issues control commands to achieve formation maintenance, attitude synchronization, balanced task allocation, and automatic fault redundancy, ensuring that multi-machine operations are conflict-free, blind-spot-free, and uninterrupted.

[0034] The data processing and defect recognition module 400 is used for preprocessing, stereo matching, three-dimensional reconstruction and intelligent defect recognition of binocular images.

[0035] The data processing and defect identification module 400 uses an edge computing chip as its hardware core and is equipped with a binocular stereo matching algorithm to achieve local real-time processing of image data. This eliminates the need to transmit massive amounts of raw images back to the ground, reducing wireless transmission bandwidth pressure and transmission latency. Through the fusion analysis of two-dimensional images and three-dimensional point cloud data, it comprehensively improves the accuracy of defect type identification and positioning precision.

[0036] The wireless communication module 500 is used to realize data transmission and command interaction between the UAV ensemble and the ground monitoring terminal 600. It adopts dual-link communication of 5G and WiFi. The 5G link meets the requirements of long-distance, high-speed, and high-bandwidth data transmission, while the WiFi link is used for short-range low-latency control and video transmission. The dual links automatically switch and serve as backups for each other, solving the problem of unstable communication in the field. Control commands and status data are transmitted through the MAVLink 2.0 protocol, and high-definition video is transmitted in real time through the RTSP / RTP protocol. When the main link signal strength is lower than -95 dBm, the packet loss rate exceeds 15%, the round-trip latency exceeds 1000 ms, or it is completely disconnected, the main and backup links are switched within 200ms. Critical data is sent simultaneously through both links to ensure zero loss. End-to-end latency indicators: control commands ≤50ms (5G), defect detection results ≤100ms, high-definition video ≤150ms (5G), meeting the real-time control requirements of inspection operations in complex electromagnetic environments in the field and ensuring stable and reliable air-to-ground communication.

[0037] In this embodiment, the number of slave detection drones 120 in the drone grouping execution module 100 is 2-4, and they adopt a lightweight multi-rotor structure.

[0038] The 120 slave-controlled inspection drone uses a lightweight multi-rotor platform, balancing flight maneuverability, payload capacity, and endurance. A configuration of 2–4 slave drones can cover the inspection needs of conventional power transmission lines from 10km to 30km. The multi-rotor structure enables hovering, fixed-point shooting, and small-range maneuvering adjustments, making it suitable for high-altitude fixed-point inspection of insulator strings. The fuselage adopts lightweight composite materials and a compact structural design, reducing flight energy consumption and extending the duration of a single operation.

[0039] Each of the 120 slave-controlled inspection drones is equipped with a three-axis stabilization gimbal. The gimbal can rotate 360° horizontally and ±90° vertically to ensure that the binocular camera is always pointed at the insulator string, eliminating the impact of flight shaking on image quality.

[0040] The three-axis stabilized gimbal, based on the principle of inertial stabilization control, compensates in real time for imaging deviations caused by drone flight vibrations, attitude shifts, and airflow disturbances. Its 360° horizontal omnidirectional rotation and ±90° vertical pitch adjustment can cover the entire insulator string with no blind spots, ensuring that the optical axis of the binocular camera is always perpendicular to the insulator detection surface, and acquiring high-quality images without distortion, blur, or obstruction, providing a data foundation for subsequent 3D reconstruction and defect identification.

[0041] Both the master control drone 110 and the slave control inspection drone 120 use differential GPS + inertial navigation combined positioning, with positioning accuracy reaching the millimeter level. The master control drone 110 issues formation maintenance commands in real time, and each slave drone maintains a fixed safe distance and cooperative attitude, flying in parallel along the power transmission line. This enables simultaneous inspection of multiple sections of a single line, greatly improving inspection coverage efficiency and avoiding inspection interruption caused by insufficient single-drone endurance.

[0042] Differential GPS eliminates the ionospheric and tropospheric errors of conventional GPS, while inertial navigation compensates for positioning blind spots caused by satellite signal blockage. The fusion of the two achieves millimeter-level high-precision positioning, providing a precise position reference for master-slave formation maintenance, attitude coordination, and task partitioning. The parallel flight mode can simultaneously cover multiple sections of insulators on the left, middle, and right sides of the line. The efficiency of multi-machine collaborative operation is 3-5 times that of single-machine operation. The distributed layout of slave machines avoids the short endurance of single machines and enables continuous inspection of long-distance lines.

[0043] In this embodiment, the binocular vision detection module 200 employs a binocular synchronous high-definition imaging unit, comprising two sets of industrial-grade high-definition CMOS cameras with identical parameters, a synchronous trigger controller, and an adaptive illumination unit. The two sets of industrial-grade high-definition CMOS cameras possess high resolution, high frame rate, and low noise characteristics, enabling clear capture of minute defect textures on the insulator surface. The synchronous trigger controller ensures that both cameras are exposed and imaged simultaneously, eliminating stereo matching errors caused by time deviations. The adaptive illumination unit automatically adjusts brightness and angle according to ambient light intensity, solving the problem of poor imaging quality in strong light, backlight, and low light environments. The cameras are horizontally parallel, with a binocular baseline distance of 50 mm and a depth measurement range of 1 m to 5 m. After rigorous calibration using the Zhang Zhengyou calibration method and the Bouguet stereo correction algorithm, by acquiring 20-30 sets of multi-view checkerboard image pairs, the intrinsic parameter matrices, distortion coefficients, and relative poses (rotation matrix R, translation vector T) of the left and right cameras are solved. The original images are then corrected using radial and tangential distortion correction formulas to achieve horizontal baseline alignment. By rigorously calibrating the baseline distance and optical axis parallelism, imaging distortion errors are eliminated, providing a foundation for subsequent binocular stereo matching and high-precision 3D reconstruction.

[0044] The horizontal parallel layout provides a standard geometric structure for binocular stereo matching. The baseline distance can be optimized based on the inspection distance of 1–5m to ensure the accuracy of depth calculation. Optical calibration and algorithm correction eliminate radial distortion, tangential distortion and optical axis non-parallelism errors of the lens, ensuring strict correspondence of pixels in the left and right views, and improving the accuracy of 3D point cloud reconstruction and defect spatial positioning.

[0045] The camera has a resolution of at least 4K and a frame rate of ≥25FPS. A synchronous trigger controller ensures simultaneous exposure and acquisition of dual-channel images, guaranteeing consistent imaging time for both left and right views. This provides a foundation for subsequent stereo matching and depth calculation, and it is suitable for inspection distances of 1-5m. Under normal lighting conditions and a 1m shooting distance, the detection error for small-area defects is less than 5%.

[0046] The 4K resolution can identify defects as small as 0.5mm, and the ≥25FPS frame rate ensures no ghosting or blurring during high-speed flight; the synchronous triggering accuracy reaches the microsecond level, completely avoiding parallax calculation errors caused by the time difference between the two images; the 1–5m inspection distance balances safe flight distance and imaging clarity, and complies with the safety specifications for high-altitude inspection of power lines.

[0047] The binocular vision inspection module 200 acquires images of the left and right views of the insulator in real time, and simultaneously obtains two-dimensional appearance features and three-dimensional depth information of the insulator. Compared with monocular imaging, it can accurately identify defects such as micro-cracks, damage, and skirt deformation, and at the same time obtain three-dimensional parameters such as the spatial location of defects, the insulator string tilt, and the spacing, completely solving the problems of monocular detection lacking depth information and unclear defect location.

[0048] Two-dimensional appearance features are used for defect type classification, while three-dimensional depth information is used for quantitative defect detection and spatial positioning. The fusion of the two enables integrated qualitative and quantitative detection. It can output quantitative parameters such as three-dimensional coordinates, size, depth, and tilt angle of defects, providing accurate data support for defect level determination and maintenance decisions, and solving the technical shortcomings of monocular vision, which can only perform qualitative analysis but cannot accurately quantify.

[0049] In this embodiment, the grouping and coordination control module 300 integrates the formation path planning algorithm, attitude coordination algorithm, obstacle avoidance control algorithm and task allocation unit to realize intelligent collaborative operation of UAV grouping.

[0050] Formation path planning algorithm based on transmission line GIS data and A The intelligent optimization algorithm generates safe, shortest, and collision-free inspection paths; the attitude coordination algorithm uses master-slave distributed PID control to ensure that the heading, altitude, and speed of multiple aircraft are synchronized; the obstacle avoidance control algorithm perceives obstacles such as poles, wires, and trees in real time and automatically adjusts the flight trajectory; the task allocation unit is based on a load balancing algorithm to achieve fair allocation of inspection sections and improve overall operation efficiency.

[0051] It can automatically generate the optimal inspection route based on the GIS data of the power transmission line, dynamically adjust the formation and flight parameters, allocate inspection sections to each slave aircraft, and avoid obstacles such as poles, conductors, and trees.

[0052] GIS data provides basic information such as line coordinates, tower locations, terrain, and no-fly zones. The optimal path balances inspection efficiency and flight safety. Dynamic formation adjustment can switch between parallel, serial, and circling formation modes according to line width and insulator distribution density to ensure full coverage without blind spots. The obstacle avoidance logic adopts a combination of advance prediction and real-time avoidance strategies to ensure the safety of high-altitude operations in formation.

[0053] The system receives real-time data on the position, attitude, and flight status of each slave drone, dynamically adjusts the flight speed, altitude, and heading of each drone to ensure stable formation, and simultaneously targets different sections of insulator strings with multiple binocular cameras to achieve multi-angle, full-coverage imaging.

[0054] The real-time data interaction frequency is ≥50Hz, ensuring that control commands are issued without delay; multi-machine multi-angle synchronous imaging can acquire all-round images of the insulator string from the front, side and top, avoiding missed detections caused by single-angle obstruction, forming a multi-view cross-verification mechanism, further reducing the false detection rate and missed detection rate of defects.

[0055] The task allocation unit rationally allocates each slave inspection section according to the line section length and insulator distribution density, avoiding duplicate inspections and blind spots, and achieving optimal resource allocation.

[0056] The task allocation unit adopts a refined allocation strategy based on sections, towers, and insulator strings. Long sections and high-density areas are assigned to multiple slave machines for collaborative testing, while short sections and low-density areas are completed independently by a single slave machine. Global scheduling eliminates duplicate operations and testing gaps, maximizing the use of UAV endurance and testing resources.

[0057] Equipped with fault redundancy and backup function, when a single slave drone fails, the master drone 110 can reassign tasks and enable the remaining slave drones to automatically take over, ensuring continuous inspection operations and greatly improving the reliability of the equipment.

[0058] The fault redundancy mechanism monitors slave device battery power, communication status, hardware faults, flight anomalies, and other information in real time. Once a fault is detected, the master control drone responds in 110 milliseconds, quickly re-divides the inspection section, and schedules adjacent slave devices to cover the faulty slave device's task area, ensuring that the inspection process is uninterrupted and data is not lost, thus meeting the high reliability requirements of power grid operation and maintenance.

[0059] In this embodiment, the data processing and defect identification module 400 is equipped with an edge computing chip and executes the following sequentially: The left and right views acquired by binoculars undergo distortion correction, grayscale conversion, and noise reduction enhancement preprocessing to improve image quality.

[0060] The preprocessing process employs hardware-accelerated algorithms. Distortion correction eliminates lens optical errors, grayscale reduction lowers data dimensionality and improves processing speed, and Gaussian denoising and histogram enhancement suppress environmental noise and highlight defect edge features. The signal-to-noise ratio and contrast of the preprocessed image are significantly improved, providing high-quality input for subsequent stereo matching and defect recognition.

[0061] A stereo matching algorithm based on feature matching can quickly match corresponding pixels in the left and right views and calculate the disparity map.

[0062] The feature matching algorithm selects scale-invariant features such as SIFT and SURF to achieve stable matching in complex backgrounds. The disparity map calculation transforms the differences in two-dimensional images into depth information, with disparity accuracy reaching the sub-pixel level, providing accurate data for three-dimensional coordinate calculation. The matching speed meets the requirements of real-time detection.

[0063] Reconstruct a three-dimensional point cloud model of the insulator to obtain the three-dimensional contour and spatial coordinate information of the insulator surface.

[0064] Three-dimensional point cloud reconstruction is based on the principle of triangulation. It transforms parallax data into three-dimensional coordinates on the surface of the insulator, generating a high-density, high-precision point cloud model. This model can intuitively present three-dimensional features such as the shape of the insulator, the depth of damage, the deformation of the skirt, and the tilt of the string, thus realizing the visualization and quantitative detection of defects.

[0065] Using RGB images captured by a camera as input, an improved small target recognition deep learning model is used to accurately identify the types and locations of defects such as insulator damage, cracks, dirt, missing skirts, metal corrosion, and insulator string tilt. Furthermore, the two-dimensional images and three-dimensional point cloud data of the insulator are fused and analyzed to achieve quantitative calculation of the damaged area, crack length, and width of the insulator.

[0066] Defect classification and recognition of two-dimensional images of insulators employs a deep learning model combining slice segmentation ensemble and super-resolution reconstruction. The slice segmentation ensemble module adaptively slices the acquired original high-resolution image, focusing the detection process on potential target regions and improving defect detectability and localization accuracy without increasing input resolution. This mainly includes adaptive image slicing, slice-aware annotation projection, slice validity verification, and the selection and ensemble of high-quality slices. The image super-resolution reconstruction module, based on image slicing, uses the Long-Range Spatial-Texture Enhanced ESRGAN super-resolution model as its backbone, and performs targeted optimizations in four aspects: input feature rearrangement, generator channel modeling, discriminant strategy design, and perceptual constraint methods, improving the structural consistency and discriminativeness of the reconstruction results in the defect region. Finally, to fully exploit the effective feature information after spatial reconstruction and texture enhancement, a lightweight target detection model, STG-YOLO11, integrating the Swin Transformer and GSConv, is adopted. By introducing the global modeling capability of the Transformer and the local discriminative advantages of the efficient convolutional structure, the robustness of insulator defect detection is improved while ensuring computational efficiency.

[0067] The defect type accuracy is ≥95%, the detection accuracy of minute defects reaches 0.5mm, and it supports real-time inference at the edge. It eliminates the need to transmit massive amounts of image data back to the ground, reducing the pressure on wireless communication and enabling real-time defect detection and early warning, thus meeting the high-precision requirements of power line inspection.

[0068] With a real-time inference latency of ≤100ms at the edge, it can perform on-flight inspection and reporting. The 0.5mm detection accuracy can capture early micro-cracks, minor damage and other hidden defects, preventing defects from expanding and causing safety accidents. Local processing greatly reduces the amount of data transmission, reduces communication bandwidth usage and traffic costs, and is suitable for outdoor scenarios without broadband coverage.

[0069] In this embodiment, the wireless communication module 500 adopts dual-link communication of 5G and WiFi to realize high-speed and stable data transmission between the UAV group and the ground monitoring terminal, and to transmit data such as binocular imaging video, defect detection results, UAV flight status, and position coordinates in real time.

[0070] 5G links meet the requirements for long-distance, high-speed, and high-bandwidth data transmission, while WiFi links are used for short-range, low-latency control and video backhaul. The dual links automatically switch and back each other up, solving the problem of unstable communication in the field. The backhauled data includes video streams, text data, location information, fault alarms, etc., meeting the needs of real-time ground monitoring and decision-making.

[0071] The ground monitoring terminal has functions such as real-time data display, storage, manual review of defects, report generation, and remote control. Staff can view the inspection screen and defect information in real time, manually review suspected defects, and remotely adjust the flight path and binocular camera parameters to achieve full control of the inspection operation.

[0072] The ground monitoring terminal adopts a visual interactive interface to display information such as formation position, inspection trajectory, defect list, and 3D point cloud model in real time. Data is stored locally to support historical query and traceability. The manual review function is used to correct algorithm misjudgments, and the report generation complies with power operation and maintenance management standards. Remote control enables path modification, pan-tilt adjustment, and detection parameter adjustment, forming an integrated air-ground closed-loop management system.

[0073] See attached document Figure 2 A method for detecting defects in power line insulators based on binocular UAV ensembles includes the following steps: S1. Establish a master-slave drone formation, consisting of 1 master drone 110 and 2-4 slave detection drones 120, to complete pre-flight self-check and positioning calibration.

[0074] The pre-flight self-check covers all aspects of the UAV's power system, gimbal system, binocular camera, communication module, battery level, positioning module, etc. Positioning calibration completes the initial alignment of differential GPS and inertial navigation to ensure accurate positioning after formation takeoff. Only after the self-check calibration is passed can the inspection mission be carried out to ensure operational safety.

[0075] S2, the grouping and collaborative control module 300 plans the optimal inspection path based on GIS data, and the main control UAV 110 coordinates the formation attitude, speed and task allocation.

[0076] The optimal path planning comprehensively considers factors such as route direction, obstacles, no-fly zones, and endurance to generate a safe and efficient flight path. The master control UAV 110 uses the global clock as a reference to uniformly synchronize the attitude, speed, and mission start point of each slave UAV, ensuring the orderly conduct of formation collaborative operations.

[0077] S3. Each slave-controlled detection drone 120 drives the binocular vision detection module 200 through a three-axis stabilized gimbal to simultaneously acquire left and right view images of the insulator.

[0078] The gimbal provides real-time stability compensation for flight jitter, and the binocular cameras capture images synchronously to ensure no time difference or distortion in the left and right views. During the acquisition process, the adaptive lighting unit dynamically adjusts the illumination to ensure consistent imaging quality under different environments, providing a reliable data source for 3D detection.

[0079] S4, the data processing and defect identification module 400, performs preprocessing, stereo matching, and 3D point cloud reconstruction on the binocular images, and integrates a deep learning model to complete defect identification and localization.

[0080] This step is the core detection process. It improves image quality through image preprocessing, obtains parallax data through stereo matching, generates a spatial model of the insulator through 3D reconstruction, and completes defect classification, quantification, and localization through fusion recognition. The entire process is completed in real time at the edge, realizing automatic defect detection.

[0081] S5, the wireless communication module 500 transmits defect information, flight status and imaging data back to the ground monitoring terminal in real time.

[0082] The transmitted data is compressed and encrypted to ensure transmission efficiency and data security. The ground terminal receives and parses the data in real time, and updates defect alarms and flight status synchronously, realizing full-process visual monitoring of the inspection process.

[0083] The S6 and ground monitoring terminal 600 complete data display, storage, verification and report output, realizing full controllability of the inspection process.

[0084] The data displays the inspection results intuitively, the stored data is used for later analysis and archiving, manual review improves the accuracy of the inspection, the report output meets the management requirements of the power industry, and the operation strategy can be adjusted in real time through remote control commands, forming a closed-loop management system of inspection-feedback-control.

[0085] In this embodiment, the binocular vision detection module 200 in S3 operates at an inspection distance of 1–5m, simultaneously acquiring the two-dimensional appearance features and three-dimensional depth information of the insulator, and realizing millimeter-level spatial positioning of defects.

[0086] A safe operating distance of 1–5m for power line inspection avoids the risk of drones coming into contact with power lines. Within this distance range, the binocular vision depth calculation accuracy is optimal, and it can stably output millimeter-level spatial positioning data to meet the needs of precise defect location and maintenance navigation.

[0087] In this embodiment, the defect identification types in S4 include: insulator damage, cracks, dirt, missing sheds, metal corrosion, and tilting of insulator strings.

[0088] The aforementioned defects are common and high-risk types of defects in overhead transmission line insulators, covering common faults in all categories of porcelain, glass, and composite insulators. The identification results can be directly used for defect classification, hidden danger investigation, and maintenance plan formulation, fully supporting the safe operation and maintenance of the power grid.

[0089] In this embodiment, when any slave detection drone 120 malfunctions, the master control drone 110 automatically activates the redundancy strategy, re-divides the inspection section, and the remaining slave drones automatically fill in to perform the detection, ensuring continuous and stable operation.

[0090] The redundancy strategy is a system-level reliability assurance mechanism. Fault types include battery depletion, communication interruption, hardware failure, and flight anomalies. The master control UAV 110 quickly dispatches the remaining slave UAVs to cover the fault area, ensuring that the inspection mission is not interrupted and data is not lost, thus meeting the needs of continuous inspection of long power grid lines.

[0091] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A power line insulator defect detection device based on binocular UAV ensemble, characterized in that, include: The UAV grouping execution module (100), binocular vision inspection module (200), grouping and collaborative control module (300), data processing and defect identification module (400), wireless communication module (500) and ground monitoring terminal (600). The UAV grouping execution module (100) adopts a master-slave grouping architecture, consisting of one master-controlled UAV (110) and several slave-controlled detection UAVs (120) forming an inspection formation; The binocular vision inspection module (200) is installed on each slave-controlled inspection drone (120) to collect left and right view images of the insulator and obtain three-dimensional depth information; The grouping coordination control module (300) is built into the main control UAV (110) and is used to coordinate formation path planning, attitude coordination, task allocation and fault redundancy control. The data processing and defect recognition module (400) is used to preprocess, stereo match, 3D reconstruction and intelligent defect recognition of binocular images; The wireless communication module (500) is used to realize data transmission and command interaction between the UAV group and the ground monitoring terminal (600).

2. The power line insulator defect detection device based on binocular UAV formation according to claim 1, characterized in that, In the UAV grouping execution module (100), the number of slave detection UAVs (120) is 2-4, and they adopt a lightweight multi-rotor structure; Each controlled inspection drone (120) is equipped with a three-axis stabilization gimbal. The gimbal can rotate 360° horizontally and ±90° vertically to ensure that the binocular camera is always aligned with the insulator string and to eliminate the impact of flight jitter on image quality. Both the master-controlled UAV (110) and the slave-controlled detection UAV (120) adopt differential GPS + inertial navigation combination positioning, with positioning accuracy reaching the millimeter level.

3. The power line insulator defect detection device based on binocular UAV formation according to claim 1, characterized in that, The binocular vision detection module (200) adopts a binocular synchronous high-definition imaging unit, which includes two sets of industrial-grade high-definition CMOS cameras with the same parameters, a synchronous trigger controller and an adaptive fill light unit. Two sets of cameras are horizontally parallel and are strictly calibrated for baseline distance and optical axis parallelism to eliminate imaging distortion errors; The camera has a resolution of at least 4K and a frame rate of at least 25 FPS. The synchronous trigger controller ensures that dual-channel images are exposed and acquired simultaneously, ensuring the consistency of imaging time for the left and right views, and providing a foundation for subsequent stereo matching and depth calculation.

4. The power line insulator defect detection device based on binocular UAV formation according to claim 1, characterized in that, The grouping and coordination control module (300) integrates the grouping path planning algorithm, attitude coordination algorithm, obstacle avoidance control algorithm and task allocation unit; It can automatically generate the optimal inspection route based on GIS data of power transmission lines, dynamically adjust the formation and flight parameters, and allocate inspection sections to each slave aircraft. It has a fault redundancy and replacement function. When a single slave drone fails, the master drone (110) can redistribute tasks and automatically replace the remaining slave drones.

5. The power line insulator defect detection device based on binocular UAV formation according to claim 1, characterized in that, The data processing and defect identification module (400) is equipped with an edge computing chip and executes the following sequentially: The left and right views acquired by binoculars are preprocessed with distortion correction, grayscale conversion, noise reduction and enhancement. A stereo matching algorithm based on feature matching quickly matches corresponding pixels in the left and right views and calculates the disparity map. Reconstruct a 3D point cloud model of the insulator to obtain the 3D contour and spatial coordinate information of the insulator surface; Using RGB images captured by a camera as input, an improved small target recognition deep learning model is used to accurately identify the types and locations of defects such as insulator damage, cracks, dirt, missing sheds, metal corrosion, and insulator string tilt. Furthermore, the two-dimensional images and three-dimensional point cloud data of the insulator are fused and analyzed to achieve quantitative calculation of the damaged area, crack length, and width of the insulator. The accuracy rate of defect type identification is ≥95%, and the detection accuracy of minute defects reaches 0.5mm.

6. The power line insulator defect detection device based on binocular UAV formation according to claim 1, characterized in that, The wireless communication module (500) adopts 5G and WiFi dual-link communication to realize high-speed and stable data transmission between the UAV group and the ground monitoring terminal, and to transmit data such as binocular imaging video, defect detection results, UAV flight status, and position coordinates in real time. The ground monitoring terminal has functions such as real-time data display, storage, manual defect review, report generation, and remote control.

7. The method for detecting defects in power line insulators based on binocular UAV formations as described in any one of claims 1-6, characterized in that, Includes the following steps: S1. Establish a master-slave UAV formation, consisting of 1 master UAV (110) and 2–4 slave detection UAVs (120) to complete pre-flight self-check and positioning calibration; S2, the grouping and coordination control module (300) plans the optimal inspection path based on GIS data, and the main control UAV (110) coordinates the formation attitude, speed and task allocation; S3. Each slave-controlled detection drone (120) drives the binocular vision detection module (200) through a three-axis stabilized gimbal to simultaneously acquire left and right view images of the insulator; S4, Data Processing and Defect Recognition Module (400) preprocesses binocular images, performs stereo matching and 3D point cloud reconstruction, and integrates deep learning models to complete defect recognition and localization; S5, the wireless communication module (500) transmits defect information, flight status and imaging data back to the ground monitoring terminal in real time. S6, Ground monitoring terminal (600) completes data display, storage, verification and report output, realizing full control of the inspection process.

8. The method for detecting defects in power line insulators based on binocular UAV formations according to claim 7, characterized in that, The binocular vision detection module (200) in S3 operates at an inspection distance of 1–5m, simultaneously acquiring two-dimensional appearance features and three-dimensional depth information of the insulator, and realizing millimeter-level spatial positioning of defects.

9. The method for detecting defects in power line insulators based on binocular UAV formation according to claim 7, characterized in that, The defect identification types in S4 include: insulator damage, cracks, dirt, missing skirts, metal corrosion, and tilted insulator strings.

10. The method for detecting defects in power line insulators based on binocular UAV formation according to claim 7, characterized in that, When any slave detection drone (120) malfunctions, the master control drone (110) automatically activates the redundancy strategy, re-divides the inspection section, and the remaining slave drones take over to perform the detection, ensuring continuous and stable operation.

Images