Real-time identification and dynamic adjustment inspection method of photovoltaic unmanned aerial vehicle

Abstract

The invention belongs to the technical field of photovoltaic unmanned aerial vehicle inspection, and particularly relates to a photovoltaic unmanned aerial vehicle real-time identification and dynamic adjustment inspection method, which comprises the following steps: dividing a photovoltaic field station into a plurality of inspection unit areas as minimum inspection unit areas; an initial aerial photography point is set at the center position of each minimum inspection unit area, and all the initial aerial photography points are connected in series according to a preset inspection sequence to generate an unmanned aerial vehicle global inspection route; on the basis of the global inspection route of the unmanned aerial vehicle, after the unmanned aerial vehicle completes an aerial photography task of a certain minimum inspection unit, an inspection image of a current minimum inspection unit area is obtained, and photovoltaic module fault identification is carried out based on the inspection image of the current minimum inspection unit area; if the fault is not identified, continuing to inspect the next minimum inspection unit area according to the global inspection route of the unmanned aerial vehicle; and if the fault photovoltaic module is identified, generating a local route, and executing a fine shooting task of the photovoltaic module.

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic UAV inspection, and in particular relates to a real-time identification and dynamic adjustment inspection method for a photovoltaic UAV. Background Art

[0002] As photovoltaic power generation expands and large-scale photovoltaic stations become increasingly common, the importance of daily operation and maintenance becomes increasingly prominent. As core equipment, the operating status of photovoltaic modules directly impacts the power generation efficiency and economic benefits of power plants. However, photovoltaic power plants are often built in outdoor environments such as deserts, mountains, or plateaus. They are constantly exposed to factors such as wind and sand, obstruction, hot spots, module aging, and wiring anomalies, making them prone to failure and performance degradation. Therefore, regular inspections and the timely detection and resolution of faults are key to ensuring the safe and stable operation of power plants.

[0003] Traditional manual inspections rely on operations and maintenance personnel carrying equipment to conduct ground inspections, collecting data using infrared thermometers and thermal imagers. This method is labor-intensive, requires long inspection cycles, and relies heavily on personnel experience. Guaranteeing accuracy and consistency is difficult, leading to significant efficiency bottlenecks and labor cost pressures for large power plants.

[0004] To address the limitations of traditional methods, drone-based intelligent inspection solutions are emerging as a key component of intelligent photovoltaic power plant operation and maintenance. Equipped with dual-spectrum infrared and visible light cameras, combined with artificial intelligence recognition algorithms, these solutions can automatically identify and locate fault characteristics such as abnormal temperature rise, obstruction, and damage in photovoltaic modules, significantly improving inspection efficiency and recognition accuracy.

[0005] Currently, mainstream PV drone inspection systems use a pre-set route and waypoint approach to perform tasks. This involves pre-planning flight paths and fixed shooting points, guiding the drone to each waypoint in sequence and completing fixed-point photography. Inspection images are then collected and transmitted back to a backend platform for unified identification and fault analysis. Although this approach is simple to implement and widely used, the following technical bottlenecks have gradually emerged as PV power plants expand in size and operating environments become more complex: (1) Inspection image data is seriously redundant: full coverage and indiscriminate shooting methods are generally adopted, without differentiated processing based on the actual status of components or regional risk levels. This leads to the generation of a large number of redundant inspection images, which occupies storage and transmission resources, increases the back-end computing pressure, and affects diagnostic efficiency and platform performance.

[0006] (2) Lack of flexibility in mission execution strategies: Most inspection missions are executed along fixed routes, requiring equal photography and complete identification of all components. This makes it difficult to focus on key areas, resulting in wasted flight and computing resources. Limited by the drone's flight time, it is difficult to support efficient coverage of larger-scale power plants. Summary of the Invention

[0007] In order to overcome the problems in the prior art, the present invention proposes a real-time identification and dynamic adjustment inspection method for photovoltaic UAVs.

[0008] The technical solution of the present invention to solve the above technical problems is as follows: The present invention provides a real-time identification and dynamic adjustment inspection method for photovoltaic UAVs, which is characterized by comprising the following steps: The photovoltaic station is divided into multiple inspection unit areas, which are used as the minimum inspection unit area. An initial aerial photography point is set at the center of each minimum inspection unit area. All initial aerial photography points are connected in series according to the preset inspection sequence to generate the drone's global inspection route. Based on the drone's global inspection route, when the drone completes the aerial photography task of a certain minimum inspection unit, it obtains the inspection image of the current minimum inspection unit area, and identifies photovoltaic module faults based on the inspection image of the current minimum inspection unit area; if the faulty photovoltaic module is not identified, the drone will continue to inspect the next minimum inspection unit area according to the drone's global inspection route; if the faulty photovoltaic module is identified, a local route will be generated to perform the detailed photography task of the photovoltaic module.

[0009] Furthermore, the photovoltaic station is divided into multiple inspection unit areas, as the smallest inspection unit area, including: Obtain orthophotos of the photovoltaic station and use the pre-trained YOLO V10 inspection image recognition algorithm to detect photovoltaic modules in the orthophotos of the photovoltaic station. Extract the spatial coordinates of all photovoltaic modules and form a photovoltaic module point set. The DBSCAN clustering algorithm is used to cluster the photovoltaic module point set. The DBSCAN clustering algorithm uses the spatial distance between photovoltaic modules as the clustering criterion. By setting the neighborhood radius and the minimum number of photovoltaic modules, multiple photovoltaic modules that meet the clustering criterion are divided into a cluster area. After clustering is completed, the shape of each cluster area is optimized: when the number of PV modules in a cluster area exceeds the preset maximum PV module number threshold, it is further subdivided into multiple sub-cluster areas to ensure that the shape of the cluster area is close to the visual range of a single drone inspection image.

[0010] Furthermore, the clustering criteria include a preset neighborhood radius and a minimum number of photovoltaic components, MinPts. If there are at least MinPts photovoltaic components within the neighborhood radius of a certain photovoltaic component, then this photovoltaic component is the core point, and the photovoltaic components in its neighborhood belong to the same cluster, thereby dividing multiple components into a cluster unit.

[0011] Furthermore, photovoltaic module fault identification is performed based on the inspection image of the current minimum inspection unit area, including: Based on the inspection image of the current minimum inspection unit area, a pre-trained fault diagnosis model is used to identify photovoltaic module faults in the inspection image and obtain the photovoltaic module fault area.

[0012] Furthermore, if a faulty PV module is identified, a local route is generated to perform detailed photography of the PV module, including: While using a pre-trained fault diagnosis model to identify photovoltaic module faults in the inspection image, a pre-trained photovoltaic module recognition model is called to detect and locate all photovoltaic modules in the inspection image. The position area of each photovoltaic module in the inspection image coordinate system, namely the center position and contour boundary, is extracted to generate a photovoltaic module set. Based on the spatial overlap between the fault area of the photovoltaic module and the location of the photovoltaic module in the inspection image coordinate system, the identified fault area is attributed to the corresponding photovoltaic module, realizing a one-to-one mapping between the fault and the specific photovoltaic module. In combination with the longitude and latitude calibration parameters in the inspection image metadata, the affine transformation method is used to construct the mapping relationship between the inspection image pixel coordinates and the geographic coordinates. Through the mapping relationship between the inspection image pixel coordinates and geographic coordinates, the location of the faulty PV module in the inspection image is converted into the actual spatial coordinates in the PV station to form a coordinate set of the faulty PV module; For the acquired set of faulty PV panel locations, a nearest neighbor search algorithm is used to generate a local shortest path route.

[0013] Furthermore, it also includes: real-time monitoring of the remaining power of the drone during flight, and when it is determined that the current power is insufficient to support the completion of the next area inspection task, triggering the return command and charging; when charging is completed, it returns to the last interruption point and continues to perform subsequent inspection tasks.

[0014] Furthermore, it also includes: all inspection images are synchronously uploaded to the background system after shooting, and the inspection image analysis is automatically completed based on the pre-trained fault recognition model, suspected fault information is extracted, faulty components are located, and geographic location mapping is performed in combination with the inspection image metadata.

[0015] Compared with the prior art, the present invention has the following technical effects: (1) This invention dynamically determines whether to enter a localized, detailed inspection process based on the intelligent recognition results of inspection images, breaking through the limitations of traditional static preset path models and achieving greater real-time performance and flexibility. The system can instantly adjust flight plans based on the on-site environment and fault distribution, adapting to the complex and diverse inspection needs of photovoltaic stations, significantly improving the system's scenario adaptability and operational efficiency.

[0016] (2) Compared with the traditional coverage acquisition method, the present invention focuses on the identified faulty components for detailed photography, and only collects high-value inspection image data with diagnostic significance, thereby reducing the amount of redundant inspection images from the source, reducing the inspection image storage and processing costs, and significantly improving data processing efficiency and system response speed.

[0017] (3) This invention adopts a task organization method of "regular area shooting + local fine shooting", which enables the UAV to focus on key areas while maintaining the coverage breadth, achieving a balance between accuracy and breadth. By reducing ineffective flights and repeated shooting, it effectively saves power and flight resources, indirectly improving the UAV's inspection endurance and enhancing the coverage range of a single flight. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 This is a schematic diagram of the area division of the inspection unit of the present invention; Figure 2 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0020] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose of the invention, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementation methods, structures, features, and effects of the technical solutions proposed by the present invention. Specific features, structures, or characteristics in one or more embodiments may be combined in any suitable form. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention belongs.

[0021] In one embodiment of the present invention, referring to Figure 1-Figure 2 , provides a real-time identification and dynamic adjustment inspection method for photovoltaic UAVs, including the following steps: The photovoltaic station is divided into multiple inspection unit areas, which are used as the minimum inspection unit area. An initial aerial photography point is set at the center of each minimum inspection unit area. All initial aerial photography points are connected in series according to the preset inspection sequence to generate the drone's global inspection route. Based on the drone's global inspection route, when the drone completes the aerial photography task of a certain minimum inspection unit, it obtains the inspection image of the current minimum inspection unit area, and identifies photovoltaic module faults based on the inspection image of the current minimum inspection unit area; if no photovoltaic module fault is identified, the drone continues to inspect the next minimum inspection unit area according to the global inspection route. If no fault is identified, a local route is generated to perform the detailed photography task of the photovoltaic module.

[0022] The following is a detailed explanation of each of the above steps: Step 100: Divide the photovoltaic station into multiple inspection unit areas as the minimum inspection unit area; set an initial aerial photography point at the center of each minimum inspection unit area, and connect all the initial aerial photography points in series according to the inspection order to generate a global inspection route for the drone.

[0023] Reference Figure 1 Obtain orthophotos of the PV station. Based on these orthophotos, a regional division strategy is employed to implement grid management of PV panels within the station. Based on the spatial distribution of PV panels, the PV station is divided into multiple standard inspection units. Each unit typically covers approximately 4×4 PV panels, serving as the minimum inspection unit.

[0024] For each minimum inspection unit area, an initial aerial photography point is automatically set at its center. Combined with the set flight altitude, this ensures that a single inspection image fully covers the corresponding inspection unit area. All initial aerial photography points are linked together according to the inspection sequence to generate the drone's global inspection route. During the global inspection route generation process, the drone's endurance, flight path length, and inspection image coverage redundancy are comprehensively considered to rationally plan the flight sequence, improving mission execution efficiency and allowing for adjustment for subsequent dynamic fault data collection.

[0025] In order to achieve the automatic division of the above minimum inspection unit, the present invention adopts a density-based spatial clustering method to divide the photovoltaic station into multiple inspection unit areas as the minimum inspection unit area. The specific process is as follows: First, the pre-trained YOLO V10 inspection image recognition algorithm is used to detect photovoltaic module targets on the orthophoto map of the photovoltaic station, and the spatial position coordinates of all photovoltaic modules are extracted to form the photovoltaic module point set C = {c1, c2,…, c n}, where c n Indicates the spatial position coordinates of the nth photovoltaic module.

[0026] Pre-training the YOLO V10 inspection image recognition algorithm involves processing a dataset of PV panel inspection images using annotation tools, annotating the location and bounding box of each PV panel in each inspection image, and constructing a labeled dataset containing the panel's location information. This dataset is then used to train the YOLO V10 model. During training, data augmentation techniques such as inspection image rotation, scaling, and brightness adjustment are employed to improve the model's recognition accuracy for PV panels in various scenarios and environmental conditions. This pre-training process results in the YOLO V10 inspection image recognition algorithm.

[0027] In order to divide the photovoltaic modules into regions, the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clustering algorithm is used to cluster the photovoltaic module point set C. The DBSCAN clustering algorithm uses the spatial distance between photovoltaic modules as the clustering criterion and pre-sets the neighborhood radius. ε The area is divided by two parameters: the minimum number of photovoltaic modules MinPts. If the neighborhood radius of photovoltaic module M1 ε If there are at least MinPts PV modules (including itself) within a region, PV module M1 is considered the core point, and all PV modules in its neighborhood belong to the same cluster. This allows multiple modules with close proximity and relatively high density to be grouped together as a single cluster unit. To accommodate the common spacing of PV modules, the neighborhood radius ε is typically set to approximately 1 meter, and MinPts is set to approximately 8 meters. This allows densely spaced modules to be grouped into the same inspection area.

[0028] To match the aspect ratio of the inspection images taken by drones, after clustering is completed, the shape of each cluster result is optimized: when the number of PV modules in a cluster area exceeds the set maximum PV module number threshold (such as 16), it will be further subdivided into multiple sub-cluster areas. While ensuring that the aspect ratio is approximately 4:3, its boundaries are adjusted to ensure that the area shape is close to the visual range of a single drone inspection image, so that the inspection image coverage is more uniform and the recognition effect is better.

[0029] An initial aerial photography point is automatically set at the center of each resulting inspection area. All these aerial photography points are connected in spatial order to form the initial global inspection route. By default, an "S-shaped" path is used, connecting from left to right and from top to bottom, to balance flight distance and mission efficiency.

[0030] Step 200: Based on the drone's global inspection route, when the drone completes the aerial photography task of a certain minimum inspection unit, it obtains the inspection image of the current minimum inspection unit area, and performs photovoltaic module fault identification based on the inspection image of the current minimum inspection unit area; if no photovoltaic module fault is identified, the drone continues to inspect the next minimum inspection unit area according to the global inspection route. If no fault is identified, a local route is generated to perform the detailed photography task of the photovoltaic module.

[0031] As an example, this step 200 includes the following sub-steps: Step 210: Based on the drone's global inspection route, when the drone completes the aerial photography mission of a certain minimum inspection unit, the inspection image of the current minimum inspection unit area is obtained, and photovoltaic module fault identification is performed based on the inspection image of the current minimum inspection unit area.

[0032] Based on the inspection image of the current minimum inspection unit area, the photovoltaic module fault in the inspection image is identified, and the position of the photovoltaic module in the inspection image is detected at the same time, and the photovoltaic module fault area and the position area of the photovoltaic module in the inspection image coordinate system are obtained.

[0033] First, after the drone completes the aerial photography mission of a certain minimum inspection unit, it obtains the inspection images taken by the drone and performs two types of inspection image recognition tasks on the obtained inspection images: Use the pre-trained fault diagnosis model to identify possible photovoltaic fault types in the inspection image, such as hot spots, occlusion, damage, etc., and output the inspection image pixel coordinate set of the photovoltaic module fault area , where each Indicates the fault area, its location, and type.

[0034] Among them, the fault diagnosis model is built based on the YOLO V10 target detection algorithm and is trained using a dataset of fault inspection images of photovoltaic modules. The collected photovoltaic inspection images are manually annotated using annotation tools to mark typical fault types, including hot spots, occlusions, damage, etc., and the type of each fault area and the pixel position of the inspection image are clearly defined in the annotation. The model is trained based on the annotated dataset, and data enhancement (such as rotation, scaling, brightness changes, etc.) is used to improve the recognition robustness and accuracy of the model. After training, the model can identify and classify photovoltaic fault areas in the input inspection image, and output the corresponding inspection image pixel coordinates to obtain a set of fault component areas. .

[0035] At the same time, the pre-trained photovoltaic module recognition model is called to detect and locate all photovoltaic modules in the inspection image, and the position area of each photovoltaic module in the inspection image coordinate system, that is, the center position and contour boundary, is extracted to generate a photovoltaic module set. , where each Represents a PV panel and its pixel position in the inspection image.

[0036] Step 220: Matching and mapping faulty PV panels to their geographic locations: Based on the spatial overlap between the PV module fault area and the PV module location area, the method calculates the intersection-over-union (IoU) of the bounding box of the PV module fault area with the bounding box of the PV module location area. Typically, the IoU threshold is set at 0.5. The identified PV module fault area is assigned to the corresponding PV module, achieving a one-to-one mapping between the fault and the specific PV module. Subsequently, an affine transformation method is used to construct a mapping between the inspection image pixel coordinates and the geographic coordinates, combining the latitude and longitude calibration parameters in the inspection image metadata.

[0037] The mathematical model for constructing the mapping relationship between inspection image pixel coordinates and geographic coordinates is as follows: ; In the above formula, (x, y): pixel coordinates of the center of the PV module fault area; : The corresponding latitude and longitude (longitude, latitude); W, H represent the width and height of the image respectively (unit: pixel); : Longitude and latitude of the upper left corner of the image (minimum longitude, maximum latitude); : The latitude and longitude of the lower right corner of the image (maximum longitude, minimum latitude).

[0038] Through the mapping relationship between the inspection image pixel coordinates and geographic coordinates, the location of the faulty photovoltaic module in the inspection image can be accurately converted into the actual spatial coordinates in the photovoltaic station to form the coordinate set of the faulty photovoltaic module. , where each Indicates the center position of a faulty PV panel in the geographic coordinate system.

[0039] Step 230: Local path planning and precise shooting control: Based on the acquired set of faulty PV panel locations P, a nearest neighbor search algorithm is used to generate a local shortest path. This local shortest path starts from the current location and sequentially visits all faulty PV panels, ensuring the shortest flight path and the fastest response time, thus meeting real-time requirements.

[0040] Based on the generated local shortest path, the drone is controlled to perform high-precision imaging at a lower altitude and optimized angle, capturing only the components identified as faulty. This improves inspection image resource utilization, saves flight time, and generates high-quality local inspection data. This local refined route is destroyed upon completion of the inspection of the current area, and the system resumes the global inspection process, continuing with regular imaging tasks for the next standard inspection area.

[0041] Throughout the entire process, after completing each minimum inspection unit, the drone dynamically decides whether to proceed to a localized inspection mission for the abnormal sub-area based on the inspection image analysis results. If no obvious faults are found in the inspection images of the current area, the drone skips the local route and continues to the next area according to the drone's global inspection route. If an abnormal component is detected, the drone immediately switches to the local route and performs detailed imaging of the PV module. This strategy effectively balances system operational efficiency and fault response accuracy, improving the practicality and flexibility of overall intelligent inspection.

[0042] Following the planned global inspection route, the drone flies to the initial aerial photography points in each standard inspection area, completing both inspection image acquisition tasks at standard altitudes and local route inspections. During flight, the drone monitors the remaining battery power in real time. If the current battery level is insufficient to complete the next inspection area, it automatically triggers a return command to ensure flight safety. Once charging is complete, the system automatically resumes at the last interruption point and continues with subsequent inspections, ensuring a complete and closed-loop mission.

[0043] All inspection images are uploaded to the backend system immediately after capture. Based on a pre-trained fault recognition model, the system automatically analyzes the images, extracts suspected fault information, locates faulty components, and geo-maps them using inspection image metadata. The recognition results, combined with information such as the corresponding component location and inspection time, are automatically entered into the inspection management system to support subsequent maintenance decisions and troubleshooting.

[0044] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A real-time identification and dynamic adjustment inspection method for photovoltaic UAVs, characterized in that: The following steps are involved: The photovoltaic station is divided into multiple inspection unit areas, which are used as the minimum inspection unit area. An initial aerial photography point is set at the center of each minimum inspection unit area. All initial aerial photography points are connected in series according to the preset inspection sequence to generate the drone's global inspection route. Based on the drone's global inspection route, when the drone completes the aerial photography task of a certain minimum inspection unit, it obtains the inspection image of the current minimum inspection unit area, and identifies photovoltaic module faults based on the inspection image of the current minimum inspection unit area; if the faulty photovoltaic module is not identified, the drone will continue to inspect the next minimum inspection unit area according to the drone's global inspection route; if the faulty photovoltaic module is identified, a local route will be generated to perform the detailed photography task of the photovoltaic module.

2. The method for real-time identification and dynamic adjustment inspection of a photovoltaic UAV according to claim 1 is characterized in that: The photovoltaic station is divided into multiple inspection unit areas, as the smallest inspection unit area, including: Obtain orthophotos of the photovoltaic station and use the pre-trained YOLO V10 inspection image recognition algorithm to detect photovoltaic modules in the orthophotos of the photovoltaic station. Extract the spatial coordinates of all photovoltaic modules and form a photovoltaic module point set. The DBSCAN clustering algorithm is used to cluster the photovoltaic module point set. The DBSCAN clustering algorithm uses the spatial distance between photovoltaic modules as the clustering criterion and divides multiple photovoltaic modules that meet the clustering criterion into a cluster area. After clustering is completed, shape optimization is performed on each cluster area: when the number of photovoltaic modules in a cluster area exceeds the preset maximum photovoltaic module number threshold, it is further subdivided into multiple sub-cluster areas.

3. The real-time identification and dynamic adjustment inspection method of a photovoltaic UAV according to claim 2 is characterized in that: The clustering criteria include a preset neighborhood radius and a minimum number of photovoltaic components, MinPts. If there are at least MinPts photovoltaic components within the neighborhood radius of a photovoltaic component, then this photovoltaic component is the core point, and the photovoltaic components in its neighborhood belong to the same cluster, thereby dividing multiple components into a cluster unit.

4. The method for real-time identification and dynamic adjustment inspection of a photovoltaic UAV according to claim 1 is characterized in that: PV module fault identification based on the inspection image of the current minimum inspection unit area, including: Based on the inspection image of the current minimum inspection unit area, a pre-trained fault diagnosis model is used to identify photovoltaic module faults in the inspection image and obtain the photovoltaic module fault area.

5. The method for real-time identification and dynamic adjustment inspection of a photovoltaic UAV according to claim 4 is characterized in that: If a faulty PV panel is identified, a local route is generated to perform detailed photography of the PV panel, including: While using a pre-trained fault diagnosis model to identify photovoltaic module faults in the inspection image, a pre-trained photovoltaic module recognition model is called to detect and locate all photovoltaic modules in the inspection image. The position area of each photovoltaic module in the inspection image coordinate system, namely the center position and contour boundary, is extracted to generate a photovoltaic module set. Based on the spatial overlap between the fault area of the photovoltaic module and the location of the photovoltaic module in the inspection image coordinate system, the identified fault area is attributed to the corresponding photovoltaic module, realizing a one-to-one mapping between the fault and the specific photovoltaic module. In combination with the longitude and latitude calibration parameters in the inspection image metadata, the affine transformation method is used to construct the mapping relationship between the inspection image pixel coordinates and the geographic coordinates. Through the mapping relationship between the inspection image pixel coordinates and geographic coordinates, the location of the faulty PV module in the inspection image is converted into the actual spatial coordinates in the PV station to form a coordinate set of the faulty PV module; For the acquired set of faulty PV panel locations, a nearest neighbor search algorithm is used to generate a local shortest path route.

6. The method for real-time identification and dynamic adjustment inspection of a photovoltaic UAV according to claim 1, characterized in that: Also includes: The remaining battery power of the drone is monitored in real time during flight. When it is determined that the current battery power is insufficient to complete the next area inspection mission, the return command is triggered and charging is started. When charging is completed, the system resumes at the last interruption point and continues with subsequent inspection tasks.

7. The method for real-time identification and dynamic adjustment inspection of a photovoltaic UAV according to claim 1, characterized in that: Also includes: All inspection images are uploaded to the backend system synchronously after shooting. Inspection image analysis is completed based on the pre-trained fault recognition model, suspected fault information is extracted, faulty components are located, and geographic location mapping is performed based on the inspection image metadata.

Images