An unmanned aerial vehicle-based fan visual inspection system and device and readable storage medium
By using drones equipped with Yolov8m and Yolov8s-OBB models for wind turbine inspection, the inefficiency and safety risks of traditional inspection methods have been solved. This has enabled efficient and accurate wind turbine defect detection and tower sweeping early warning, thereby improving the reliability and service life of wind turbines.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-27
- Publication Date
- 2026-08-04
AI Technical Summary
Traditional wind turbine inspection methods are inefficient and pose safety risks. They cannot be fully inspected under adverse weather conditions, making it difficult to detect equipment defects in a timely manner, which may lead to equipment failure and safety accidents.
A wind turbine vision inspection system based on drones is adopted. The YOLOv8m model is used to detect surface defects, and the YOLOv8s-OBB model is used to identify the blade and tower positions. The dual-model parallel architecture is combined to perform tower scanning inspection, realizing automated data collection and early warning.
It improves detection efficiency and coverage, ensures detection accuracy and robustness, enables early warning functions, reduces maintenance costs, and enhances the reliability and service life of wind turbines.
Smart Images
Figure CN121191023B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wind turbine blade inspection technology, and in particular to a wind turbine visual inspection system, device and readable storage medium based on unmanned aerial vehicles (UAVs). Background Technology
[0002] Against the backdrop of global energy transition, wind energy, as a clean and renewable energy source, is playing an increasingly important role. Wind power generation, with its mature technology, relatively low cost, and environmental friendliness, has become a key component of the energy structure of many countries and regions. In recent years, my country's wind power industry has experienced rapid development, with the scale and number of wind farms continuing to grow.
[0003] However, wind turbines are typically installed in remote mountainous areas, offshore locations, and other complex environments, enduring harsh weather conditions such as strong winds, torrential rain, sandstorms, and low temperatures. These factors can cause varying degrees of wear, corrosion, and cracks in the turbine's components. Failure to detect and address these defects in a timely manner can lead to equipment malfunctions, reduced power generation efficiency, and even serious safety accidents, resulting in significant economic losses. Therefore, regular, comprehensive, and meticulous inspections of wind turbines are crucial for ensuring their safe and stable operation.
[0004] Traditional wind turbine inspection methods primarily rely on manual climbing and the use of suspended platforms, which are not only inefficient but also pose significant safety risks. Inspectors expend considerable time and energy, and in complex environments, it's difficult to conduct comprehensive and accurate inspections of all parts of the turbine. Furthermore, manual inspections are limited by weather and lighting conditions, making them unsuitable for operation in adverse conditions. The emergence of drone inspection technology has brought a revolutionary change to wind turbine inspection. Drones offer advantages such as flexibility, efficiency, and safety, enabling them to quickly reach the location of wind turbines and conduct comprehensive imaging and monitoring. Equipped with high-definition cameras and infrared thermal imagers, drones can acquire detailed information about the surface and internal structure of the wind turbine. Therefore, how to use drones for wind turbine blade inspection has become a pressing issue. Summary of the Invention
[0005] To address the aforementioned shortcomings, the present invention aims to provide a wind turbine visual inspection system, device, and readable storage medium based on unmanned aerial vehicles (UAVs).
[0006] To achieve this objective, the present invention adopts the following technical solution: a wind turbine visual inspection method based on unmanned aerial vehicles (UAVs), comprising the following steps:
[0007] Step S1: Use a drone to move around the wind turbine along a preset path and record video of the wind turbine;
[0008] Step S2: Extract frames from the wind turbine video recording to obtain multiple detection images;
[0009] Step S3: Input the detected image into the trained Yolov8m model to detect surface defects of the wind turbine;
[0010] Step S4: Input the detected image into the trained Yolov8s-OBB model to identify the blades and tower. Based on the positional relationship between the blades and tower identified by the Yolov8s-OBB model, detect the blades and tower.
[0011] Preferably, in step S1, the preset route of the drone is set as follows:
[0012] With the tower as the central axis, the drone spirals and flies from top to bottom around the central axis, with the drone's flight starting point above the top of the tower and the drone's flight ending point below the lowest horizontal height of the blade.
[0013] Preferably, before performing step S3, the CIoU loss function in the YOLOv8 model needs to be replaced with the WIoU bounding box loss function.
[0014] Preferably, the WIoU bounding box loss function is as follows:
[0015] L WIoUv3 =r·L WIoUv1 ;
[0016]
[0017]
[0018] Where β represents the outlier of the anchor frame, used to distinguish samples of different qualities; β and a are adjustment coefficients that control the focusing coefficient. L' represents the adjusted loss value, while L' represents the loss value after adjustment. IoU It is L IoU The moving average,
[0019] L WIoUv1 =R WIoU L IoU ;
[0020]
[0021] L IoU =1-IoU(B,B gt );
[0022] R WIoU L is a nonlinear transformation factor based on the distance between the center of the predicted box and the center of the ground truth box. IoUFor the classic IoU loss, IoU(B,B) gt ) represents the predicted bounding box B and the ground truth bounding box B. gt The intersection and union ratio, the superscript * indicates that W g and H g Separating from the computational graph, (x, y) represents the coordinates of the prediction box, (x... gt y gt ) represents the coordinates of the actual bounding box.
[0023] Preferably, the training process of the Yolov8m model is as follows:
[0024] An image dataset is constructed, which includes a first type of image data and a second type of image data. The first type of image data consists of publicly available images of wind turbines under different types and operating conditions, while the second type of image data consists of wind turbine images taken by drones from different angles.
[0025] The first and second type of image data were labeled using the semi-automatic annotation software isat-sam. The annotation yielded a segmentation dataset. The script was used to export the segmentation dataset as 0-degree minimum bounding rectangles and minimum bounding rectangles with angles, resulting in an object detection dataset and a rotated object detection dataset.
[0026] The object detection dataset and the rotated object detection dataset are divided into training set, validation set and test set in a ratio of 7:2:1;
[0027] Transfer learning was performed using the YOLOv8m network and the YOLOv8m.pt pre-trained model. After multiple rounds of training, the object detection model gradually converged to the optimal level.
[0028] Preferably, step S4 is as follows:
[0029] Step S41: Identify the leaves in the detected image and obtain the first included angle between the leaves;
[0030] Step S42: Select the detection image whose first included angle meets the angle threshold as the first image, and construct the first image set based on the first image;
[0031] Step S43: Obtain the second angle between the blade and the tower in the first image; count all the second angles in the first image set, calculate the average value of the second angles, and determine whether there is a risk of blade sweeping the tower based on the average value of the second angles.
[0032] Preferably, step S41 is as follows:
[0033] The YOLOV8s-OBB model is used to detect and output the first detection box of the leaf in the detection image;
[0034] Select the two largest detection frames as the first frame body;
[0035] Obtain the first extension line of the center line of the width of the first frame, and take the included angle formed by the two first extension lines as the first included angle.
[0036] Preferably, the step of obtaining the second included angle in step S43 is as follows:
[0037] The YOLOV8s-OBB model is used to detect and output the detection box of the tower in the detection image, which is used as the second detection box;
[0038] Select the largest first detection frame as the second frame;
[0039] Obtain the second extension lines of the center lines of the second frame and the second detection frame respectively, and use the included angle formed by the two second extension lines as the second included angle.
[0040] A wind turbine visual inspection device based on unmanned aerial vehicles (UAVs) includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the aforementioned wind turbine visual inspection method based on UAVs when the computer program is executed.
[0041] A readable storage medium storing a computer program that, when executed by a processor, implements a drone-based visual inspection method for wind turbines.
[0042] One of the above technical solutions has the following advantages or beneficial effects: 1. Using a dual-model parallel architecture, Yolov8m focuses on high-precision surface defect recognition, while Yolov8s-OBB is specifically optimized for rotating target detection, forming complementary advantages and improving detection efficiency and coverage.
[0043] 2. Blade sweep detection based on location relationships enables early warning, prevents major accidents and reduces maintenance costs; in addition, the entire process can be quickly deployed in wind farms, supports large-scale continuous monitoring, optimizes resource utilization and enhances the reliability and service life of wind turbines. Attached Figure Description
[0044] Figure 1 This is a flowchart of one embodiment of the present invention.
[0045] Figure 2 A schematic diagram of the drone's flight path.
[0046] Figure 3 This is a schematic diagram illustrating the calculation of the blade tower angle according to an embodiment of the present invention. Detailed Implementation
[0047] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.
[0048] In the description of embodiments of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of embodiments of the present invention, "a plurality of" means two or more, unless otherwise explicitly specified.
[0049] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.
[0050] like Figures 1-3 As shown, a method for visual inspection of wind turbines based on unmanned aerial vehicles (UAVs) includes the following steps:
[0051] Step S1: Use a drone to move around the wind turbine along a preset path and record video of the wind turbine;
[0052] Step S2: Extract frames from the wind turbine video recording to obtain multiple detection images;
[0053] Step S3: Input the detected image into the trained Yolov8m model to detect surface defects of the wind turbine;
[0054] Step S4: Input the detected image into the trained Yolov8s-OBB model to identify the blades and tower. Based on the positional relationship between the blades and tower identified by the Yolov8s-OBB model, detect the blades and tower.
[0055] This technical solution first uses a drone to move around the wind turbine along a preset path and record video to achieve automated data acquisition. Then, the video is processed by frame extraction to extract multiple detection images as the basis for analysis. Next, these images are simultaneously input into a trained YOLOv8m model to detect surface defects (such as cracks or corrosion) of the wind turbine, and input into a trained YOLOv8s-OBB model. The position and outline of the blades and tower are accurately identified through directional bounding box technology. Finally, based on the output positions of the blades and tower, tower sweep detection is performed, automatically analyzing the distance and angle between the blades and tower, and detecting the risk of blade sweep (i.e., the possibility of blade collision with the tower) in real time, thereby completing the entire wind turbine health monitoring process.
[0056] The advantages of this technical solution lie in its use of a dual-model parallel architecture. The YOLO model 8m focuses on high-precision surface defect identification, while the YOLO 8s-OBB is specifically optimized for rotating target detection, forming complementary advantages and improving detection efficiency and coverage. The combined application of the YOLO models (YoLO 8m for high-precision defect identification and YOLO 8s-OBB for orientation and positioning) ensures the accuracy and robustness of the detection, effectively identifying subtle surface problems and dynamic positional anomalies. Simultaneously, blade scanning detection based on positional relationships enables early warning functions, preventing major accidents and reducing maintenance costs. Furthermore, the entire process can be rapidly deployed in wind farms, supporting large-scale continuous monitoring, optimizing resource utilization, and enhancing the reliability and lifespan of wind turbines.
[0057] Preferably, in step S1, the preset route of the drone is set as follows:
[0058] With the tower as the central axis, the drone spirals and flies from top to bottom around the central axis, with the drone's flight starting point above the top of the tower and the drone's flight ending point below the lowest horizontal height of the blade.
[0059] Flight path selection is a crucial aspect of inspection work. During a shutdown inspection, if the aircraft flies close to the blades, such as... Figure 2 While the left side allows for close-up observation of the blades, the risk of a collision is extremely high. Even with blade locking measures, yaw can still occur, and downtime operations incur significant costs. Non-stop inspections utilize a cylindrical flight path, such as... Figure 2 In traditional methods, drones need to perform circular motions around wind turbines, resulting in long travel distances and low inspection efficiency. The hemispherical flight path proposed in this invention is the optimal choice, such as... Figure 1 On the right, regardless of whether the fan is stopped or running, there is no need to consider the fan's attitude, making it highly versatile and effectively improving inspection efficiency.
[0060] Preferably, before performing step S3, the CIoU loss function in the YOLOv8 model needs to be replaced with the WIoU bounding box loss function.
[0061] In the loss function of YOLOv8, L CIoU By incorporating information such as center point distance and aspect ratio, the difference between the predicted bounding box and the ground truth bounding box can be measured more accurately, thereby guiding the model to learn more accurate bounding box positions and improving the localization accuracy of object detection. However, in actual detection tasks, there are often problems such as unclear boundaries and inconsistent standards in manually labeled datasets, leading to... CIoU It is difficult for it to exert its proper capabilities, resulting in a decline in model performance.
[0062] In wind turbine defect detection, defects often have indistinct boundaries, leading to insufficient quality of the model's predicted bounding box regression. Furthermore, some low-quality samples in the dataset negatively impact the model's final accuracy and generalization ability. Therefore, we introduce L... WIoUv3 This loss method focuses more on samples of average quality, thereby improving the overall performance and generalization ability of the network model.
[0063] Preferably, the WIoU bounding box loss function is as follows:
[0064] L WIoUv3 =r·L WIoUv1 ;
[0065]
[0066]
[0067] Where β represents the outlier of the anchor frame, used to distinguish samples of different qualities; β and a are adjustment coefficients that control the focusing coefficient. L′ represents the adjusted loss value, primarily used to measure sample quality; while L′ IoU It is L IoU The moving average is used to dynamically represent the average quality of several samples; through the above design, L WIoUv3 It can dynamically and accurately allocate gradient gain based on the real-time quality of the samples, thereby optimizing the model's learning efficiency for both ordinary and high-quality samples.
[0068] L WIoUv1 =R WIoU L IoU ;
[0069]
[0070] L IoU =1-IoU(B,B gt );
[0071] R WIoU L is a nonlinear transformation factor based on the distance between the center of the predicted box and the center of the ground truth box. IoU For the classic IoU loss, IoU(B,B) gt ) represents the predicted bounding box B and the ground truth bounding box B. gt The intersection and union ratio, the superscript * indicates that W g and H g Separating from the computational graph, (x, y) represents the coordinates of the prediction box, (x... gt y gt ) represents the coordinates of the actual bounding box.
[0072] Preferably, the training process of the Yolov8m model is as follows:
[0073] An image dataset is constructed, which includes a first type of image data and a second type of image data. The first type of image data consists of publicly available images of wind turbines under different types and operating conditions, while the second type of image data consists of wind turbine images taken by drones from different angles.
[0074] The first and second types of image data were labeled using the semi-automatic labeling software isat-sam. The labeling yielded a segmentation dataset. The segmentation dataset was then exported as 0-degree minimum bounding rectangles and minimum bounding rectangles with angles, resulting in a target detection dataset (defects) and a rotated target detection dataset (blades and towers).
[0075] To conduct visual inspection of wind turbine defects, a dedicated image dataset was constructed. The first type of image data was downloaded from publicly available online datasets. These images came from a wide range of sources, covering different types of wind turbines under various operating conditions. After screening and organization, approximately 5,000 relevant images were collected. The second type of image data was collected through in-house flight. A drone equipped with a high-definition camera was used to capture images of the wind turbines from different angles and distances, resulting in approximately 5,000 images. Finally, after image enhancement, the dataset containing defect annotations comprised a total of 8,000 images.
[0076] The object detection dataset and the rotated object detection dataset are divided into training set, validation set and test set in a ratio of 7:2:1;
[0077] Transfer learning was performed using the YOLOv8m network and the YOLOv8m.pt pre-trained model. After multiple rounds of training, the object detection model gradually converged to the optimal level.
[0078] Parameters were continuously adjusted on the validation set to optimize model performance. Finally, the model was evaluated on the test set, yielding excellent training results. As shown in Table 1, the accuracy reached 95%, indicating a high proportion of correctly detected defects; the recall rate was 91%, meaning the model could detect most actual defects. For example, the model could accurately identify minor scratches and pinholes, providing reliable technical support for wind turbine defect detection.
[0079] Table 1 Comparative Experiment
[0080]
[0081] Preferably, step S4 is as follows:
[0082] Step S41: Identify the leaves in the detected image and obtain the first included angle between the leaves;
[0083] Step S42: Select the detection image whose first included angle meets the angle threshold as the first image, and construct the first image set based on the first image;
[0084] Step S43: Obtain the second angle between the blade and the tower in the first image; count all the second angles in the first image set, calculate the average value of the second angles, and determine whether there is a risk of blade sweeping the tower based on the average value of the second angles.
[0085] During tower sweep inspection, a frontal / side view image is needed to effectively show the angular relationship between the blades and the tower, thus predicting the risk of tower sweep. Because the drone flies in a hemispherical configuration, the captured images are taken from different angles. At different angles, the shapes of the blades and the angles between them vary. Therefore, the first frontal / side view image can be selected based on the angles between the blades.
[0086] Then, the angle between the blade and the tower in the first image is obtained. When the angle is too small, there is a risk of tower sweeping. This further improves the stability of the detection. This application uses the average value of the second angle as the basis for judging the risk of tower sweeping. Because a single image may have errors, such as slight differences in the shooting angle or errors in the detection model, the average value can reduce random errors and improve the accuracy of the judgment. Secondly, based on the average value of multiple images, it can more comprehensively reflect the overall angle between the blade and the tower, avoiding the randomness of a single image.
[0087] Preferably, step S41 is as follows:
[0088] The YOLOV8s-OBB model is used to detect and output the first detection box of the leaf in the detection image;
[0089] Select the two largest detection frames as the first frame body;
[0090] Obtain the first extension line of the center line of the width of the first frame, and take the included angle formed by the two first extension lines as the first included angle.
[0091] The model has already been trained using a rotated object detection dataset, so the YOLOV8s-OBB model can accurately obtain the bounding box (first detection box) of the leaf detection.
[0092] like Figure 3 As shown, the two largest detection boxes are selected as the first frames, and the first included angle β is constructed by extending the first line from the midline of the width of the two first frames. When the detection image is captured from the front, the first included angle β will be 120°. As the drone moves, the first included angle β will also increase. When the drone moves to the side of the wind turbine, the angle of the first included angle β will reach its maximum value. Therefore, the corresponding first image can be selected by the first included angle β.
[0093] Preferably, the step of obtaining the second included angle in step S43 is as follows:
[0094] The YOLOV8s-OBB model is used to detect and output the detection box of the tower in the detection image, which is used as the second detection box;
[0095] Select the largest first detection frame as the second frame;
[0096] Obtain the second extension lines of the center lines of the second frame and the second detection frame respectively, and use the included angle formed by the two second extension lines as the second included angle.
[0097] The model has already been trained using a rotating object detection dataset, so the OLOV8s-OBB model can accurately obtain the bounding box (second detection box) of the tower. Because the largest blade has the clearest visible bounding box, and it is located at the bottom, the relationship between the blade and the tower can be clearly understood by using the second angle α formed by the second extension line of the second bounding box and the center line of the second detection box. When the second angle α is less than a certain angle, the blade may sweep the tower during rotation, causing a tower sweeping problem. Therefore, the second angle α can clearly and accurately reflect the risk of tower sweeping.
[0098] A wind turbine visual inspection device based on unmanned aerial vehicles (UAVs) includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the aforementioned wind turbine visual inspection method based on UAVs when the computer program is executed.
[0099] A readable storage medium storing a computer program that, when executed by a processor, implements a drone-based visual inspection method for wind turbines.
[0100] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0101] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.
Claims
1. A method for visual inspection of a fan based on a UAV, characterized in that, Includes the following steps: Step S1: Use a drone to move around the wind turbine along a preset path and record video of the wind turbine; Step S2: Extract frames from the wind turbine video recording to obtain multiple detection images; Step S3: Input the detected image into the trained Yolov8m model to detect surface defects of the wind turbine; Step S4: Input the detected image into the trained Yolov8s-OBB model to identify the blades and tower. Based on the positional relationship between the blades and tower identified by the Yolov8s-OBB model, detect the blades and tower. The steps of step S4 are as follows: Step S41: Identify the leaves in the detected image and obtain the first included angle between the leaves; Step S42: Select the detection image whose first included angle meets the angle threshold as the first image, and construct the first image set based on the first image; Step S43: Obtain the second angle between the blade and the tower in the first image; Statistically analyze all the second included angles within the first image set, calculate the average value of the second included angles, and determine whether there is a risk of blade sweeping the tower based on the average value of the second included angles. 2.The UAV-based visual inspection method of wind turbines according to claim 1, wherein, In step S1, the preset route of the drone is set as follows: With the tower as the central axis, the drone spirals and flies from top to bottom around the central axis, with the drone's flight starting point above the top of the tower and the drone's flight ending point below the lowest horizontal height of the blade. 3.The UAV-based visual inspection method of wind turbines according to claim 1, wherein, Before performing step S3, the CIoU loss function in the YOLOv8 model needs to be replaced with the WIoU bounding box loss function.
4. The wind turbine visual inspection method based on unmanned aerial vehicles (UAVs) according to claim 3, characterized in that, The WIoU bounding box loss function is shown below: ； ； ； in, Represents the outlier degree of the anchor frame, used to distinguish samples of different quality; and To control the adjustment range of the focusing coefficient; This represents the adjusted loss value. yes The moving average, For hyperparameters; ； ； ； This is a nonlinear transformation factor based on the distance between the centers of the predicted bounding box and the ground truth bounding box. This is the classic IoU loss. This represents the predicted bounding box B and the ground truth bounding box B. The intersection and union ratio, with the superscript * indicating that the intersection and union ratio will be... and Separate from the computational graph. and These represent the predicted bounding box B and the ground truth bounding box, respectively. The width and height of the constructed minimum closed bounding box, where (x, y) represent the coordinates of the predicted box. Represents the coordinates of the actual bounding box.
5. The wind turbine visual inspection method based on unmanned aerial vehicles (UAVs) according to claim 3, characterized in that, The training process of the Yolov8m model is as follows: An image dataset is constructed, which includes a first type of image data and a second type of image data. The first type of image data consists of publicly available images of wind turbines under different types and operating conditions, while the second type of image data consists of wind turbine images taken by drones from different angles. The first and second type of image data were labeled using the semi-automatic annotation software isat-sam. The annotation yielded a segmentation dataset. The script was used to export the segmentation dataset as 0-degree minimum bounding rectangles and minimum bounding rectangles with angles, resulting in an object detection dataset and a rotated object detection dataset. The object detection dataset and the rotated object detection dataset are divided into training set, validation set and test set in a ratio of 7:2:1; Transfer learning was performed using the YOLOv8m network and the YOLOv8m.pt pre-trained model. After multiple rounds of training, the object detection model gradually converged to the optimal level.
6. The wind turbine visual inspection method based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The steps of step S41 are as follows: The YOLOV8s-OBB model is used to detect and output the first detection box of the leaf in the detection image; Select the two largest detection frames as the first frame body; Obtain the first extension line of the center line of the width of the first frame, and take the included angle formed by the two first extension lines as the first included angle.
7. The wind turbine visual inspection method based on unmanned aerial vehicles (UAVs) according to claim 1, characterized in that, The steps for obtaining the second included angle in step S43 are as follows: The YOLOV8s-OBB model is used to detect and output the detection box of the tower in the detection image, which is used as the second detection box; Select the largest first detection frame as the second frame; Obtain the second extension lines of the center lines of the second frame and the second detection frame respectively, and use the included angle formed by the two second extension lines as the second included angle.
8. A wind turbine visual inspection device based on unmanned aerial vehicles (UAVs), characterized in that, It includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement, when executing the computer program, a wind turbine visual inspection method based on a drone as described in any one of claims 1-7.
9. A readable storage medium, characterized in that, The readable storage medium stores a computer program, which, when executed by a processor, implements a wind turbine visual inspection method based on a drone as described in any one of claims 1-7.