A high-precision photovoltaic module defect positioning system and method based on a compound wing unmanned aerial vehicle

Abstract

The present application belongs to the technical field of intelligent operation and maintenance of photovoltaic power stations, and particularly relates to a high-precision photovoltaic module defect positioning method based on a composite-wing unmanned aerial vehicle, comprising the following steps: acquiring image data streams collected by the unmanned aerial vehicle in real time during inspection; using a defect recognition model deployed on the aerial vehicle to perform online processing and target recognition on the image data streams to detect suspected defect targets on the photovoltaic module and output pixel-level bounding box information of the suspected defect targets; when a defect target is recognized, the unmanned aerial vehicle is controlled to enter an automatic hovering state, and high-definition zoom image collection is performed on the defect target; the present application can realize accurate mapping from aerial images to ground module assets by fusing centimeter-level RTK positioning, visual geometric solving and high-precision GIS map matching technology, achieve a positioning effect of "what you see is what you get", and improve the decision-making efficiency of power station operation and maintenance through intelligent management of defect data.

Description

Technical Field

[0001] This invention belongs to the field of intelligent operation and maintenance technology of photovoltaic power plants, and specifically relates to a high-precision photovoltaic module defect location system and method based on a compound-wing UAV. Background Technology

[0002] With the continuous expansion of photovoltaic power plant construction and the increasingly complex terrain, efficient and accurate operation and maintenance inspections have become a key link in ensuring the safe and stable operation of power plants and improving power generation efficiency. Drones, with their advantages of flexibility, maneuverability, and high operational efficiency, have become an important tool for photovoltaic power plant inspections, widely used for identifying defects such as hot spots, microcracks, and damage based on infrared thermal imaging or visible light imaging.

[0003] Problems with existing technology: However, current drone-based inspection solutions still suffer from a significant bottleneck: insufficient defect location accuracy. The GNSS modules on typical consumer drones usually only offer meter-level positioning accuracy, and their inspection reports often only provide approximate location information, including a blurry global image and an error range of a few hundred meters. Maintenance personnel must rely on experience to repeatedly compare and troubleshoot among thousands of similar-looking photovoltaic modules to ultimately locate the defective module—a process that is inefficient and prone to errors.

[0004] Although RTK technology can provide centimeter-level positioning capabilities, most existing solutions simply equate the RTK position of the drone itself with the defect location, ignoring the spatial offset between the camera optical axis and the drone's center of mass, the specific pixel position of the defect in the image, and the geometric errors introduced by factors such as the undulation of the terrain. As a result, the final positioning results still cannot meet the actual requirements of component-level precision repair.

[0005] In addition, existing technical solutions generally lack the ability to deeply integrate and correlate inspection data with the digital model of power plant assets. Defect information often exists in the form of isolated data points, which cannot be effectively correlated and traced with historical inspection records, making it difficult to support advanced operation and maintenance needs such as defect evolution trend analysis, predictive maintenance, and full life cycle management of power plant assets. Summary of the Invention

[0006] The purpose of this invention is to provide a high-precision photovoltaic module defect location system and method based on a compound-wing UAV. It can achieve accurate mapping from aerial images to ground module assets by integrating centimeter-level RTK positioning, visual geometric calculation and high-precision GIS map matching technology, achieving a "what you see is what you get" positioning effect. Furthermore, it can improve the decision-making efficiency of power plant operation and maintenance through intelligent management of defect data.

[0007] The specific technical solution adopted by this invention is as follows: A high-precision photovoltaic module defect location method based on a compound-wing UAV includes the following steps: Acquire real-time image data streams collected by drones during inspections; Using an airborne defect identification model and a deep learning-based defect target detection algorithm, geometric correction, size and pixel normalization preprocessing are performed on real-time acquired video stream images; Based on the preprocessed image tensor, forward inference is performed using a lightweight deep learning model to extract multi-scale hierarchical features of the image and simultaneously predict defect categories and bounding boxes. The coordinate regression parameters are mapped to the image pixel coordinate system through coordinate transformation operations to obtain the absolute coordinates of the bounding box; a predefined confidence threshold is applied to filter the decoded initial detection boxes; The non-maximum suppression algorithm is executed independently for each defect category to eliminate spatially redundant prediction boxes and obtain accurate pixel-level defect target information, so as to detect suspected defect targets on photovoltaic modules and output their pixel-level bounding box information. When a defective target is detected, the drone is controlled to enter an automatic hovering state to acquire absolute geodetic coordinates from the RTK module, high-definition sequential image frames from the image sensor of the airborne optoelectronic pod, and angular velocity and linear acceleration data from the inertial measurement unit in real time. The obtained RTK absolute positioning coordinates, visual odometry relative pose observations calculated based on continuous image frame sequences, and inertial measurement data are input into state estimators such as extended Kalman filters or factor graph optimizers. Tightly coupled fusion calculations are performed under a unified spatiotemporal reference to output high-frequency, high-precision six-degree-of-freedom pose estimation results for the UAV in real time. The obtained real-time pose estimation value is compared with the hovering target setpoint to generate position error and attitude error signals; through the proportional-integral-derivative cascade control loop of the flight controller, the electronic speed controller and motor system of the UAV are driven to generate compensation control quantities to counteract position drift and attitude disturbances. The system evaluates the quality indicators of each sensor data in real time and dynamically adjusts its weight coefficient in state estimation based on the confidence index of each sensor. When the performance of a single sensor degrades or fails, the system automatically switches to the degradation fusion mode to maintain the continuous stability of the hovering state. At the same time, it controls the gimbal to adjust the zoom camera and perform high-definition zoom image acquisition on the defective target. In addition, multi-view geometry optimization was performed on multiple frames of images captured during hovering to acquire multiple high-definition zoom images of the same defective target captured by the UAV during stable hovering. Based on the principle of multi-view stereo vision, the bundle adjustment method is used to jointly optimize and adjust the coordinates of the defect points and the camera extrinsic parameters of each frame image, thereby eliminating random errors and system residuals and outputting the optimized high-precision geodetic coordinates of the defect points. During the image acquisition process, a multi-source spatiotemporal dataset is acquired synchronously based on a unified time reference. The multi-source spatiotemporal dataset includes UAV RTK positioning data, UAV pose data, gimbal attitude data, and defect target coordinates. Using camera intrinsic parameters obtained in advance through offline calibration, including focal length, principal point coordinates and lens distortion coefficient, the pixel coordinates of the defective target in the original image are geometrically corrected. Based on the attitude angles output in real time by the UAV flight control system and the pitch and roll angles fed back by the gimbal stabilization platform, a multi-level coordinate transformation matrix chain is constructed, which transforms from the camera coordinate system to the UAV body coordinate system, then to the local geographic coordinate system, and finally to the geodetic coordinate system. Based on the transformation matrix chain and the camera imaging geometry model, the spatial resection algorithm is used to solve the mapping relationship from pixel coordinates to geodetic coordinates of the defect target, and a mapping model is constructed. Based on the UAV RTK positioning data, the three-dimensional geodetic coordinates corresponding to the defective target are calculated using the mapping model. Using the three-dimensional geodetic coordinates of the defective target, spatial matching is performed with a pre-stored GIS digital map of the photovoltaic power station. This GIS digital map is a digital twin model of the power station and is configured as follows: Each photovoltaic module is defined as a geographic polygon feature with precise geometric boundaries. Each polygon feature is associated with a unique asset identifier, which corresponds to the actual physical number of the photovoltaic module in the power plant. By performing a point-area inclusion query using a spatial database engine, the geodetic coordinates of the defect point are automatically matched with the digital twin model, and the unique asset number of the photovoltaic module to which it is located is output, thereby obtaining the module array location number of the photovoltaic module where the defect target is located.

[0008] Based on the component array location number of the defective target, a structured report of the defective target is generated; at the same time, the relevant records in the historical defect database are retrieved, and the historical defect records of the defective target are compared and analyzed and the defect evolution trend is evaluated to generate a corresponding intelligent operation and maintenance work order.

[0009] A high-precision photovoltaic module defect location system based on a compound-wing UAV includes: The composite wing unmanned aerial vehicle platform is equipped with an RTK positioning module, a high-definition optoelectronic pod, an onboard computing unit, and a flight control system; An image defect recognition module, deployed in the airborne computing unit, is used to process image data streams in real time and identify defects in photovoltaic modules. The high-precision positioning and calculation module is used to construct a coordinate mapping model based on synchronously acquired multi-source spatiotemporal data and calculate the three-dimensional geodetic coordinates of the defective target. The GIS matching asset management module is used to spatially match defect coordinates with pre-stored digital twin models of power plants to determine the defect component number; The data processing report generation module is used to correlate historical data, perform trend analysis, and generate structured operation and maintenance work orders.

[0010] An electronic device includes: a memory storing a computer program thereon; and a processor communicatively connected to the memory for executing the computer program to implement the method.

[0011] A computer-readable storage medium having a computer program stored thereon that, when executed by an electronic device, implements the method.

[0012] The technical effects achieved by this invention are as follows: This invention constructs a rigorous multi-level coordinate transformation model encompassing "image pixel coordinates - camera coordinates - UAV body coordinates - geographic coordinates - geodetic coordinates," and simultaneously integrates high-precision spatiotemporal data. This enables centimeter-level accuracy in calculating defect locations from two-dimensional images to three-dimensional geographic space, overcoming the systematic errors of directly using the UAV's position as the defect location. It improves defect location accuracy from meters to centimeters compared to traditional methods. Maintenance personnel can directly locate the specific defective component based on the component number in the report, effectively saving time spent on on-site searching and comparison, and improving maintenance efficiency.

[0013] This invention achieves fully automated processing across the entire chain, from automatic defect identification, automatic hovering triggering, automatic data acquisition and synchronization, automatic coordinate calculation, automatic component number matching, and automatic report generation. It reduces manual intervention, lowers the probability of human error, and makes the inspection and maintenance workflow of large-scale photovoltaic power plants more efficient and standardized.

[0014] This invention strongly correlates defect information with specific asset components in a high-precision GIS digital twin map, transforming one-time inspection data into traceable and analyzable digital assets. Combined with the trend analysis function of historical databases, the system can extract valuable operation and maintenance knowledge from massive amounts of data, promoting the transformation of operation and maintenance mode from "passive response" to "proactive early warning" and "predictive maintenance," which helps to extend the overall lifespan of power plant assets and optimize power generation efficiency. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a system block diagram of the present invention; Figure 3 This is a schematic diagram of the accurate acquisition of defect images in the human-machine hovering mode of the present invention; Figure 4 This is a schematic diagram of the operation and maintenance report generated by the present invention, which includes precise positioning information and a comparison with historical trends. Figure 5 This is a block diagram of the electronic device of the present invention. Detailed Implementation

[0016] To make the objectives and advantages of this invention clearer, the invention will be specifically described below with reference to embodiments. It should be understood that the following text is merely used to describe one or more specific embodiments of the invention and does not strictly limit the scope of protection specifically claimed by the invention.

[0017] like Figure 1 As shown, a high-precision photovoltaic module defect location method based on a compound-wing UAV includes the following steps: The composite-wing UAV is equipped with a high-definition optoelectronic pod, including visible light and infrared cameras, and automatically cruises and scans the photovoltaic power station according to a pre-planned route, collecting image data streams in real time. Using an airborne defect identification model and a deep learning-based defect target detection algorithm, geometric correction, size and pixel normalization preprocessing are performed on real-time acquired video stream images; Based on the preprocessed image tensor, forward inference is performed using a lightweight deep learning model to extract multi-scale hierarchical features of the image and simultaneously predict defect categories and bounding boxes. The coordinate regression parameters are mapped to the image pixel coordinate system through coordinate transformation operations to obtain the absolute coordinates of the bounding box; a predefined confidence threshold is applied to filter the decoded initial detection boxes; The non-maximum suppression algorithm is executed independently for each defect category to eliminate spatially redundant prediction boxes and obtain accurate pixel-level defect target information, so as to detect suspected defect targets on photovoltaic modules and output their pixel-level bounding box information. like Figure 3 As shown, when a defective target is detected, the UAV is controlled to enter an automatic hovering state, hovering stably roughly directly above the defective target, and acquiring in real time the absolute geodetic coordinates from the RTK module, high-definition sequential image frames from the image sensor of the airborne optoelectronic pod, and angular velocity and linear acceleration data from the inertial measurement unit. The obtained RTK absolute positioning coordinates, visual odometry relative pose observations calculated based on continuous image frame sequences, and inertial measurement data are input into state estimators such as extended Kalman filters or factor graph optimizers. Tightly coupled fusion calculations are performed under a unified spatiotemporal reference to output high-frequency, high-precision six-degree-of-freedom pose estimation results for the UAV in real time, including position and attitude angles. The obtained real-time pose estimation value is compared with the hovering target setpoint to generate position error and attitude error signals; through the proportional-integral-derivative cascade control loop of the flight controller, the electronic speed controller and motor system of the UAV are driven to generate compensation control quantities to counteract position drift and attitude disturbances. The system evaluates the quality indicators of each sensor data in real time and dynamically adjusts its weight coefficient in state estimation based on the confidence index of each sensor. When the performance of a single sensor degrades or fails, the system automatically switches to the degradation fusion mode to maintain the continuous stability of the hovering state. At the same time, it controls the gimbal to adjust the zoom camera and perform high-definition zoom image acquisition on the defective target. In addition, multi-view geometry optimization was performed on multiple frames of images captured during hovering to acquire multiple high-definition zoom images of the same defective target captured by the UAV during stable hovering. Based on the principle of multi-view stereo vision, the bundle adjustment method is used to jointly optimize and adjust the coordinates of the defect points and the camera extrinsic parameters of each frame image, thereby eliminating random errors and system residuals and outputting the optimized high-precision geodetic coordinates of the defect points. During the image acquisition process, a multi-source spatiotemporal dataset is acquired synchronously based on a unified time reference. The multi-source spatiotemporal dataset includes UAV RTK positioning data, UAV pose data, gimbal attitude data, and defect target coordinates. The gimbal attitude data consists of the pitch and roll angles fed back in real time by the optoelectronic pod stabilization platform. Using camera intrinsic parameters obtained in advance through offline calibration, including focal length, principal point coordinates and lens distortion coefficient, the pixel coordinates of the defective target in the original image are geometrically corrected. Based on the UAV attitude angle and gimbal attitude angle output in real time by the UAV flight control system, as well as the pre-calibrated camera installation parameters, including the stick arm offset vector and rotation matrix of the camera relative to the UAV body, a multi-level coordinate transformation matrix chain is constructed from the camera coordinate system to the UAV body coordinate system, then transformed to the local geographic coordinate system, and finally mapped to the geodetic coordinate system. The corresponding transformation matrix T0 can be expressed as: Among them, the T1 matrix is ​​determined by the UAV RTK position, the T2 matrix is ​​determined by the UAV attitude angle, and the T3 matrix is ​​determined by the gimbal angle and camera installation parameters. Based on the transformation matrix chain and the camera imaging geometry model, the spatial resection algorithm is used to solve the mapping relationship from pixel coordinates to geodetic coordinates of the defect target, and a mapping model is constructed. Based on the UAV RTK positioning data, the three-dimensional geodetic coordinates corresponding to the defective target are calculated using the mapping model. Using the three-dimensional geodetic coordinates of the defective target, spatial matching is performed with a pre-stored GIS digital map of the photovoltaic power station. This GIS digital map is a digital twin model of the power station and is configured as follows: Each photovoltaic module is defined as a geographic polygon feature with precise geometric boundaries. Each polygon feature is associated with a unique asset identifier, which corresponds to the actual physical number of the photovoltaic module in the power plant. By performing a point-area inclusion query using a spatial database engine, the geodetic coordinates of the defect point are automatically matched with the digital twin model, and the unique asset number of the photovoltaic module to which it is located is output, thereby obtaining the module array location number of the photovoltaic module where the defect target is located.

[0018] like Figure 4 As shown, based on the component array location number of the defect target, a structured report is automatically generated, including information such as defect image, precise coordinates, component number, defect type, and confidence level. At the same time, the system uses the component number as an index to automatically query the historical defect database. If a historical record exists, a comparative analysis is performed to calculate the defect area change rate, abnormal temperature difference growth trend, etc. Based on the analysis results, auxiliary decision-making suggestions are generated, and suggestions such as "continue to observe," "prioritize repair," or "may spread" are given. Finally, an operation and maintenance work order containing precise location information and intelligent analysis conclusions is generated to guide on-site maintenance work.

[0019] like Figure 2 As shown, a high-precision photovoltaic module defect location system based on a compound-wing UAV includes: The composite wing unmanned aerial vehicle platform is equipped with an RTK positioning module, a high-definition optoelectronic pod, an onboard computing unit, and a flight control system; An image defect recognition module, deployed in the airborne computing unit, is used to process image data streams in real time and identify defects in photovoltaic modules. The high-precision positioning and calculation module is used to construct a coordinate mapping model based on synchronously acquired multi-source spatiotemporal data and calculate the three-dimensional geodetic coordinates of the defective target. The GIS matching asset management module is used to spatially match defect coordinates with pre-stored digital twin models of power plants to determine the defect component number; The data processing report generation module is used to correlate historical data, perform trend analysis, and generate structured operation and maintenance work orders.

[0020] like Figure 5 As shown, an electronic device includes: a memory storing a computer program thereon; and a processor communicatively connected to the memory for executing the computer program to implement the method described thereon.

[0021] A computer-readable storage medium having a computer program stored thereon that, when executed by an electronic device, implements the method.

[0022] Example 1: Routine Inspection and Precise Hot Spot Location of Large-Scale Flat-Ground Photovoltaic Power Station Scene Overview: A large centralized photovoltaic power station is located in a plain area and has more than 100,000 standard photovoltaic modules, which are arranged in a neat manner. The operation and maintenance team uses the method of this patent to carry out quarterly routine inspections using a composite wing UAV equipped with a high-precision RTK module and a visible light / infrared dual-light pod. The aim is to quickly detect and locate hot spot defects caused by bypass diode failure or microcracks in the cells. Specific implementation steps: The drone flies at a fixed altitude of about 20 meters above the surface of the component, following a pre-planned route. The optoelectronic pod continuously scans the component array below, generating real-time visible light and infrared image data streams. The airborne defect identification module, based on the lightweight YOLOv5 model, processes the infrared image stream in real time. When it flies over the 5th subarray, the model identifies a local abnormal high temperature area in the infrared image of one of the components with a confidence level of 95%, which is determined to be a "hot spot" defect. The module then outputs the pixel-level bounding box coordinates (u,v,w,h) of the hot spot in the image. After receiving the defect identification signal, the flight control system immediately controls the UAV to enter an automatic hovering mode based on RTK and vision fusion over the target; the UAV's position is stabilized with centimeter-level accuracy. The gimbal automatically adjusts to center the zoom camera on the defect area, enabling high-definition zoom shooting and capturing detailed visible light and infrared close-up images. Simultaneous acquisition of multi-source spatiotemporal data: At the moment of camera exposure, the system synchronously records the data based on a unified timestamp. UAV RTK positioning data: (Longitude: 118.123456°, Latitude: 32.654321°, Altitude: 50.12m); UAV attitude data: Pitch angle -85°, camera approximately vertically downward, roll angle 0.5°, yaw angle 120°; Gimbal attitude data: To compensate for minor body sway, the gimbal pitch angle is -85.2° and the roll angle is 0.3°; Defect target coordinates: The center pixel coordinates of the hot spot after distortion correction (1024,768). The high-precision positioning and calculation module is started. First, it uses the pre-calibrated camera intrinsic parameters, including focal length f, principal point (cx, cy), and distortion coefficients k1, k2, p1, p2, to perform geometric correction on the pixel coordinates. Next, based on the UAV pose, gimbal attitude, and the fixed installation relationship between the camera and the UAV IMU, a mapping model is constructed from the camera coordinate system → UAV body coordinate system → northeast-sky coordinate system → WGS-84 geodetic coordinate system. Using the precise location provided by the UAV RTK as the control point, and employing a spatial resection algorithm, the three-dimensional geodetic coordinates corresponding to the center point of the hot spot were calculated as follows: (longitude: 118.123502°, latitude: 32.654298°, altitude: 25.34m). The GIS matching asset management module performs spatial overlay analysis with the calculated geodetic coordinates and the pre-generated high-precision digital twin model of the power station; through the "point-area containment" query, the system immediately determines that the coordinate point falls within the polygonal face of the photovoltaic module numbered PV-Array-05-R12-C08; The data processing report generation module automatically generates a structured inspection report, which includes: close-up image of the defect, defect type, confidence level, precise location coordinates, component number (PV-Array-05-R12-C08), and discovery time. The system also queries the historical database and finds that the component has no hot spot records in the past two inspections. Therefore, it is marked as a "new defect" and automatically generates an intelligent operation and maintenance work order to send the component to the precise location for replacement or repair.

[0023] Example 2: Microcrack Detection and Coordinate Optimization of Photovoltaic Power Stations in Complex Mountainous Topography Scene Overview: A photovoltaic power station in a mountainous area has its modules laid out following the undulating terrain, with significant elevation differences and slopes. One of the key points of operation and maintenance is to detect potential "microcracks" in the solar cells that may occur during transportation or installation. This terrain places higher demands on positioning accuracy. Inspection and identification and stable hovering: The drone adaptively flew over complex terrain; the airborne defect identification model detected a thin line shadow on the surface of a component in a visible light image, which is suspected to be a "hidden crack"; The drone quickly entered an adaptive hovering state; due to the strong gusts of wind in the mountains, the flight control system relied on the extended Kalman filter to deeply fuse RTK, visual odometry and IMU data, and dynamically adjusted the weights of each sensor, successfully maintaining a stable centimeter-level hovering over the target point; During the approximately 10-second stable hovering period, the gimbal controlled the zoom camera to capture 15 high-resolution zoom images of the same hidden crack target from slightly different angles by making minor adjustments to the gimbal's yaw and pitch. The high-precision positioning and solving module not only performs coordinate solving for a single frame image, but also initiates a multi-view geometry optimization step; The system utilizes these 15 images, based on the principles of multi-view stereo vision and bundle adjustment, to jointly optimize the pixel coordinates of the hidden crack feature points in each image, the camera extrinsic parameters at 15 time points, calculated from the synchronous pose data, and the three-dimensional coordinates of the defect points to be determined. This process effectively eliminates random errors and possible system residuals in single-frame calculations, such as tiny calibration errors, and further improves the accuracy of the three-dimensional coordinates of the defect points from the centimeter level in a single frame to the sub-centimeter level. The optimized high-precision coordinates were sent to the GIS matching asset management module; the digital twin model of the power station accurately modeled the three-dimensional geometric information of the slope where each component was located; The system performs three-dimensional spatial relationship judgment and accurately matches the defect point to the PV-Hillside-03-R08-C15 component located on the 3rd slope, 8th row, and 15th column; the report will also provide a schematic diagram of the relative position of the defect on the component panel. The system's historical records revealed that the component had a minor "microcrack" six months ago; the length of the microcrack detected this time has increased by 30%. Based on this, the data processing report generation module performs trend analysis and assesses that the defect shows signs of expansion. In the generated maintenance work order, the priority is raised to "urgent" and it is recommended to "inspect immediately to prevent it from developing into fragmentation that could lead to a significant drop in power or hot spots".

[0024] Example 3: Routine Inspection and Monitoring of Historical Defect Evolution in Coastal Power Stations Scene Overview: A photovoltaic power station built on a coastal mudflat is located in a humid environment, which makes it prone to defects such as component aging and EVA delamination. The operation and maintenance company hopes to establish "health records" for key components through regular inspections to achieve predictive maintenance. During routine inspections, the drone detected a localized color change on the surface of a component in a certain area using visible light images. The defect identification model determined it to be an "EVA delamination / aging" defect and triggered the standard hovering and data acquisition process. The coordinates of the defect point were obtained by solving the standard coordinate mapping model and matched with the component number PV-Coastal-02-R20-C10 in the power plant GIS digital twin model. When generating this defect report, the data processing report generation module deeply integrates with the "historical defect database"; The system automatically retrieved all inspection records for component PV-Coastal-02-R20-C10 from the past two years: showing that the component first showed slight yellowing 12 months ago, and the yellowed area expanded to 8cm 6 months ago.2 ; The system compares and analyzes the defect area measured this time with historical data and assesses the defect evolution trend; through linear or more complex model fitting, the system calculates the monthly expansion rate of the aging area of ​​the component and predicts that, following this trend, it may affect the battery string connection line after 3 months, resulting in a significant decrease in power. Based on the above analysis, the system no longer generates traditional "reactive" maintenance work orders, but instead generates a predictive intelligent maintenance work order. The work order suggests: "The aging of component PV-Coastal-02-R20-C10 is accelerating, and it is expected to have a substantial impact on power generation performance in the next inspection cycle; it is recommended to include it in the next month's maintenance plan and carry out preventive replacement in advance." At the same time, the system may mark the batch to which the component belongs. Because the installation time and material batch are the same, it prompts the maintenance personnel to pay special attention to other components in the same batch, realizing the upgrade of the maintenance mode from "single point maintenance" to "batch early warning".

[0025] In summary, these three embodiments demonstrate the application of this patented method in three typical scenarios: efficient positioning of photovoltaic power plants on conventional flat land, high-precision optimized positioning in complex terrain, and predictive maintenance based on historical data. Together, they verify how this method effectively addresses the core pain points of traditional UAV inspections, such as ambiguous positioning, low efficiency, and data silos, through a closed-loop process of "real-time identification - precise positioning - asset association - intelligent analysis," significantly improving the intelligence level and economic benefits of photovoltaic power plant operation and maintenance. The above description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained in this invention are implemented according to conventional methods in the art unless otherwise specified or limited.

Claims

1. A high-precision photovoltaic module defect location method based on a compound-wing UAV, characterized in that, Includes the following steps: Acquire real-time image data streams collected by drones during inspections; Using an airborne defect identification model, the image data stream is processed online and targets are identified to detect suspected defect targets on photovoltaic modules and output their pixel-level bounding box information. When a defective target is detected, the drone is controlled to enter automatic hovering mode and perform high-definition zoom image acquisition on the defective target. During the image acquisition process, a multi-source spatiotemporal dataset is synchronously acquired based on a unified time reference. The multi-source spatiotemporal dataset includes UAV RTK positioning data, UAV pose data, gimbal attitude data, and pixel coordinates of the defect target in the image. The coordinates of the defective target are corrected based on the camera intrinsic parameters, and a mapping model from the image pixel coordinate system to the geodetic coordinate system is constructed using the UAV pose data and the gimbal attitude data. Based on the UAV RTK positioning data, the three-dimensional geodetic coordinates corresponding to the defective target are calculated using the mapping model. By using the three-dimensional geodetic coordinates of the defective target and spatially matching it with a pre-stored photovoltaic power station GIS digital map, the component array location number of the photovoltaic module where the defective target is located can be obtained.

2. The method according to claim 1, characterized in that: The defect identification model is based on a deep learning-based target detection algorithm. Perform geometric correction, size and pixel normalization preprocessing on the real-time acquired video stream images; Based on the preprocessed image tensor, forward inference is performed using a lightweight deep learning model to extract multi-scale hierarchical features of the image and simultaneously predict defect categories and bounding boxes. The coordinate regression parameters are mapped to the image pixel coordinate system through coordinate transformation operations to obtain the absolute coordinates of the bounding box; a predefined confidence threshold is applied to filter the decoded initial detection boxes; Furthermore, a non-maximum suppression algorithm is independently executed for each defect category to eliminate spatially redundant prediction boxes and obtain accurate pixel-level defect target information.

3. The method according to claim 1, characterized in that: When the drone enters automatic hovering mode, the drone achieves centimeter-level precision hovering based on RTK and vision fusion positioning, specifically as follows: It acquires absolute geodetic coordinates from the RTK module, high-definition sequential image frames from the image sensor of the airborne optoelectronic pod, and angular velocity and linear acceleration data from the inertial measurement unit in real time. The obtained RTK absolute positioning coordinates, visual odometry relative pose observations calculated based on continuous image frame sequences, and inertial measurement data are input into state estimators such as extended Kalman filters or factor graph optimizers. Tightly coupled fusion calculations are performed under a unified spatiotemporal reference to output high-frequency, high-precision six-degree-of-freedom pose estimation results for the UAV in real time. The obtained real-time pose estimation value is compared with the hovering target setpoint to generate position error and attitude error signals; through the proportional-integral-derivative cascade control loop of the flight controller, the electronic speed controller and motor system of the UAV are driven to generate compensation control quantities to counteract position drift and attitude disturbances. The system evaluates the quality indicators of each sensor's data in real time and dynamically adjusts its weight coefficient in state estimation based on the confidence index of each sensor. When the performance of a single sensor degrades or fails, the system automatically switches to a degraded fusion mode to maintain the continuous stability of the hovering state.

4. The method according to claim 1, characterized in that: The specific construction method of the mapping model is as follows: Using camera intrinsic parameters obtained in advance through offline calibration, including focal length, principal point coordinates and lens distortion coefficient, the pixel coordinates of the defective target in the original image are geometrically corrected. Based on the attitude angles output in real time by the UAV flight control system and the pitch and roll angles fed back by the gimbal stabilization platform, a multi-level coordinate transformation matrix chain is constructed, which transforms from the camera coordinate system to the UAV body coordinate system, then to the local geographic coordinate system, and finally to the geodetic coordinate system. Based on the transformation matrix chain and the camera imaging geometry model, the spatial resection algorithm is used to solve the mapping relationship between the defect target from pixel coordinates to geodetic coordinates.

5. The method according to claim 1, characterized in that: The GIS digital map is a digital twin model of the power station, configured as follows: Each photovoltaic module is defined as a geographic polygon feature with precise geometric boundaries. Each polygon feature is associated with a unique asset identifier, which corresponds to the actual physical number of the photovoltaic module in the power plant. By performing a point-surface inclusion query using a spatial database engine, the geodetic coordinates of the defective point are automatically matched with the digital twin model, and the unique asset number of the photovoltaic module to which it is located is output.

6. The method according to claim 1, characterized in that: It also includes a multi-view geometry optimization step: Collect multiple high-resolution zoom images of the same defective target taken by the drone during stable hovering; Based on the principle of multi-view stereo vision, the bundle adjustment method is used to jointly optimize and adjust the coordinates of the initially calculated defect points and the camera extrinsic parameters of each frame image, eliminating random errors and system residuals, and outputting the optimized high-precision geodetic coordinates of the defect points.

7. The method according to claim 1, characterized in that: It also includes intelligent report generation and trend analysis steps: Based on the component array location number of the defective target, a structured report of the defective target is generated; at the same time, the relevant records in the historical defect database are retrieved, and the historical defect records of the defective target are compared and analyzed and the defect evolution trend is evaluated to generate a corresponding intelligent operation and maintenance work order.

8. A high-precision photovoltaic module defect location system based on a compound-wing unmanned aerial vehicle, implementing the method described in any one of claims 1-7, characterized in that, include: The composite wing unmanned aerial vehicle platform is equipped with an RTK positioning module, a high-definition optoelectronic pod, an onboard computing unit, and a flight control system; An image defect recognition module, deployed in the airborne computing unit, is used to process image data streams in real time and identify defects in photovoltaic modules. The high-precision positioning and calculation module is used to construct a coordinate mapping model based on synchronously acquired multi-source spatiotemporal data and calculate the three-dimensional geodetic coordinates of the defective target. The GIS matching asset management module is used to spatially match defect coordinates with pre-stored digital twin models of power plants to determine the defect component number; The data processing report generation module is used to correlate historical data, perform trend analysis, and generate structured operation and maintenance work orders.

9. An electronic device, characterized in that, The electronic device includes: a memory storing a computer program thereon; and a processor communicatively connected to the memory for executing the computer program to implement the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by an electronic device, it implements the method of any one of claims 1 to 7.

Images