Model training method, blade photographing method and related device

Abstract

The application provides a model training method, a blade shooting method and related devices, and relates to the technical field of fan inspection. The model training method constructs a first loss function L1, a second loss function L2 and a third loss function L3, finally obtains a target blade detection model which can effectively filter out the background and other object interference in the fan blade image and can frame the smallest directed area of the blade in the fan blade image, on the other hand, the fan blade shooting method detects the object entering through laser ranging, triggers the blade target detection, adjusts the camera shooting parameters according to the blade position information obtained by the blade target detection, and shoots the blade image. The shot blade image has the advantages of being complete, good and high in clarity.

Description

Technical Field

[0001] This application relates to the technical field of wind turbine inspection, and in particular to a model training method, a blade imaging method, and related devices. Background Technology

[0002] Wind power, as an important new energy source, is being widely developed and utilized. Due to the wide geographical distribution and large installed capacity of wind turbines in my country, some turbines frequently experience blade failures such as paint peeling, scratches, edge cracking, and fatigue fractures during operation due to harsh environmental conditions. Therefore, image acquisition of wind turbine blades can enable reasonable analysis of their health status, thereby effectively preventing wind turbine failures.

[0003] Currently, manual inspection methods exist, such as using handheld binoculars or human double-crossers. However, due to the large size and high rotation speed of wind turbine blades, these manual methods suffer from low efficiency and inaccurate detection. In addition, traditional image recognition or target detection technologies are used to acquire images of wind turbine blades. However, the resulting images often contain incomplete, unclear, and complex backgrounds of the wind turbine blades. Summary of the Invention

[0004] The purpose of this application is to solve the technical problems mentioned in the background section and to provide a model training method, a leaf imaging method, and related devices.

[0005] In a first aspect, this application provides a method for training a blade detection model, comprising: acquiring a training sample set, the training sample set consisting of wind turbine blade images with different angles and rotation directions; constructing a first loss function L1, a second loss function L2, and a third loss function L3, wherein the first loss function L1 calculates the loss degree between the true value and the predicted value of the blade in the training sample set, the second loss function L2 calculates the loss degree between the angles of the tilt directions of the blades in the training sample set, and the third loss function L3 calculates the loss degree between the predicted geometric center and the actual geometric center in the training sample set; obtaining a total loss function L = L1 + L2 + L3 based on the first loss function L1, the second loss function L2, and the third loss function L3; training a blade detection model based on the training sample set, performing backpropagation iterations until the total loss function L reaches its minimum value, and generating a target blade detection model.

[0006] Optionally, the step of training the blade detection model based on the training sample set and performing backpropagation iterations until the total loss function L reaches its minimum value to generate the target blade detection model specifically includes: inputting the wind turbine blade image after grayscale preprocessing into the feature extraction network of the blade detection model to extract feature maps; setting a fixed number of regions of interest for each pixel position of the feature maps and sending the regions of interest into the region generation network for binary classification and coordinate regression; performing backpropagation iterations on the total loss function L until the total loss function L reaches its minimum value, thereby obtaining the optimal network model parameters and the target blade detection model.

[0007] Optionally, the leaf detection model is a convolutional neural network model, which includes: a feature extraction network, a pooling layer, and a region generation network.

[0008] Optionally, after obtaining the training sample set, which consists of wind turbine blade images with different angles and rotation directions, the method further includes: performing image processing on the wind turbine blade images, including: adding random noise, mirroring, rotating clockwise, and grayscale processing.

[0009] Secondly, this application provides a method for capturing images of wind turbine blades, which can capture clear and complete images of rotating wind turbine blades, including: sending a laser to a target wind turbine; receiving a laser ranging signal, and when the laser ranging signal is greater than a set threshold, inputting the acquired image into a target blade detection model for blade detection to determine the blade position information, wherein the target blade detection model is trained using the aforementioned blade detection model training method; and adjusting camera shooting parameters according to the blade position information to capture the blade image.

[0010] Optionally, it also includes: after taking a leaf image, incrementing the value of the leaf image counter by one, and ending the leaf image taking when the value of the leaf image counter reaches a set value.

[0011] Thirdly, this application provides a blade detection model training device, comprising the following modules: a sample acquisition module configured to acquire a training sample set, the training sample set consisting of labeled wind turbine blade images with different angles and rotation directions; a loss function module configured to construct a first loss function L1, a second loss function L2, and a third loss function L3, wherein the first loss function L1 calculates the loss degree between the true value and the predicted value of the blade in the training sample set, the second loss function L2 calculates the loss degree between the angles of the blade tilt directions in the training sample set, and the third loss function L3 calculates the loss degree between the predicted geometric center and the actual geometric center in the training sample set; a total loss function module configured to obtain a total loss function L = L1 + L2 + L3 based on the first loss function L1, the second loss function L2, and the third loss function L3; and a training module configured to train a blade detection model based on the training sample set, performing backpropagation iterations until the total loss function L reaches its minimum value, thereby generating a target blade detection model.

[0012] Fourthly, this application provides a wind turbine blade imaging device, comprising the following modules: a laser transmitting module configured to transmit a laser to a target wind turbine; a blade detection module configured to receive a laser ranging signal, wherein when the laser ranging signal is greater than a set threshold, the acquired image is input into a target blade detection model for blade detection to determine the blade position information, wherein the target blade detection model is trained using the aforementioned blade detection model training method; and an image capturing module configured to adjust camera shooting parameters according to the blade position information to capture a blade image.

[0013] Fifthly, this application also provides an electronic device, the electronic device comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to execute the blade detection model training method or the wind turbine blade imaging method described above.

[0014] Sixthly, this application also provides a computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the above-described blade detection model training method or the above-described wind turbine blade imaging method.

[0015] The beneficial effects of this application are:

[0016] This application proposes a blade detection model training method. By constructing a first loss function L1, the loss between the ground truth value and the predicted value of the blade in the first region of the training sample set is calculated. By constructing a second loss function L2, the loss between the ground truth value and the predicted value of the blade in the second region of the training sample set is calculated. Finally, a target blade detection model is obtained that can effectively filter out background and other object interference in the wind turbine blade image and can select the smallest directed region where the blade is located in the wind turbine blade image.

[0017] On the other hand, this application proposes a wind turbine blade imaging method that combines laser ranging with target detection and tracking triggering. By detecting the entry of an object through laser ranging, the blade target detection is triggered. Then, the camera imaging parameters are adjusted according to the blade position information obtained from the blade target detection to capture the blade image. This wind turbine blade method has the advantages of capturing complete and clear blade images, requiring fewer target detections, and having high detection efficiency. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly described below.

[0019] Figure 1 A flowchart illustrating a blade detection model training method provided in an embodiment of this application;

[0020] Figure 2 A flowchart illustrating a wind turbine blade imaging method according to an embodiment of this application;

[0021] Figure 3 This is a schematic diagram of the blade detection process in one embodiment of this application;

[0022] Figure 4 This is a block diagram of a blade detection model training device provided in an embodiment of this application;

[0023] Figure 5 This is a block diagram of a wind turbine blade imaging device provided in one embodiment of this application;

[0024] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0025] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of this application, but not all embodiments.

[0027] Please refer to Figure 1 The present application provides a flowchart of a blade detection model training method, which includes steps S11 to S14.

[0028] Step S11: Obtain a training sample set, which consists of wind turbine blade images with different angles and rotation directions. The methods for obtaining these images include, but are not limited to, drone aerial photography, manual photography, and camera capture.

[0029] In this step, a large number of wind turbine blade images with different angles and rotation directions can be manually labeled in advance, mainly labeling the pixels on the blades in the wind turbine blade images as positive samples.

[0030] Correspondingly, pixels located in the wind turbine blade image but not on the blade, such as pixels in the background or on other objects, are used as negative samples. These negative samples are used as training sample sets for training the blade detection model by partially or completely annotating the acquired wind turbine blade images.

[0031] In this embodiment, only the pixels located on the blades in the wind turbine blade image are labeled, reducing the workload and offering the advantages of convenience and speed. In other embodiments, pixels in the background and other objects in the wind turbine blade image may also be labeled, or pixels in the background, other objects, and on the blades may be labeled simultaneously; all of these fall within the scope of protection of this application.

[0032] In one embodiment, the wind turbine blade images in the training sample set can also be processed by, for example, adding random noise, mirroring, rotating clockwise, or grayscale, so as to reduce the amount of computation and speed up the training process in the subsequent training process.

[0033] Step S12: Construct a first loss function L1, a second loss function L2, and a third loss function L3. The first loss function L1 calculates the loss between the true value and the predicted value of the blade in the training sample set. The second loss function L2 calculates the loss between the angles of the blade tilt direction in the training sample set. The third loss function L3 calculates the loss between the predicted geometric center and the actual geometric center in the training sample set.

[0034] In this step, the loss functions in the blade detection model are constructed. The first loss function L1 represents the blade classification loss, the second loss function L2 represents the blade orientation loss, and the third loss function L3 represents the blade localization loss.

[0035] Specifically, the formula for calculating the first loss function can be expressed as:

[0036]

[0037] Where, N cls D represents the total number of points in the image of the wind turbine blades. img Image representing wind turbine blades. F represents the true value of the blade in the wind turbine blade image. cls This represents the calculation of the loss between the set of points predicted as blades and the actual set of blade points in a wind turbine blade image.

[0038] Accordingly, the formula for calculating the second loss function can be expressed as:

[0039]

[0040] Where, N ori B represents the total number of points located on the blades in the wind turbine blade image. ori This represents the predicted minimum directed bounding box value for the leaf. F represents the truth value of the minimum directed frame of the blade. ori This represents the loss between the predicted point set and the true point set of the blades in the minimum directed frame.

[0041] Accordingly, the formula for calculating the third loss function L3 is:

[0042] L3=σ(C gt -C pre )

[0043] Among them, C gt C represents the true geometric center of the smallest directed frame of the blade. pre σ represents the geometric center of the minimum directed frame prediction for the blade, and σ is the loss weight.

[0044] Considering that during the training process, the blade detection model is easily affected by different types of objects and cluttered backgrounds in the wind turbine blade image, causing the minimum directed box to move outside the ground truth bounding box, this application introduces an effective spatial constraint to avoid this situation.

[0045] In addition, the blade detection model generates the minimum bounding rectangle of the blade according to each loss function in sequence. The specific details of the minimum directed box of the blade will be explained in detail in the section on wind turbine blade imaging methods below.

[0046] Step S13: Based on the first loss function L1, the second loss function L2, and the third loss function L3, obtain the total loss function L = L1 + L2 + L3.

[0047] In this embodiment, the total loss function L is obtained by adding the first loss function L1, the second loss function L2, and the third loss function L3 together.

[0048] In other embodiments, hyperparameters corresponding to the first loss function L1, the second loss function L2, and the third loss function L3 can be set, and the influence ratio of the three loss functions on the result can be adjusted by adjusting the size of the corresponding hyperparameters.

[0049] Step S14: Based on the training sample set, train the blade detection model and perform backpropagation iterations until the total loss function L reaches its minimum value, and generate the target blade detection model.

[0050] In this embodiment, a convolutional neural network model is used as the leaf detection model. The leaf detection model includes a structure of a feature extraction network, a pooling layer, and a region generation network.

[0051] In other embodiments, models employing other structures such as GAN networks (Generative Adversarial Networks) are also within the scope of protection of this application.

[0052] When training the blade detection model, the wind turbine blade image after grayscale preprocessing can be input into the feature extraction network of the blade detection model to extract the feature map. After the pooling layer, a fixed number of regions of interest are set for each pixel position of the feature map, and the regions of interest are sent to the region generation network for binary classification and coordinate regression calculation. Then, the total loss function L is backpropagated iteratively until the total loss function L reaches the minimum value, thereby obtaining the optimal network model parameters and the target blade detection model.

[0053] In other embodiments, the total loss function L can be iteratively updated using the Adam algorithm until it reaches its minimum value. The Adam algorithm is an adaptive momentum stochastic gradient descent algorithm. Furthermore, iterative updates of the total loss function L using other existing algorithms are also within the scope of this invention.

[0054] This application provides a blade detection model training method. By inputting a large number of wind turbine blade images into the blade detection model for training, the detection accuracy of the blade detection model is improved. It can effectively filter out the interference of background and other objects in the wind turbine blade images, identify and select the smallest directed region where the blade is located in the wind turbine blade image, and finally obtain the target blade detection model.

[0055] Based on the same inventive concept, this application also provides a method for photographing wind turbine blades, which is used to detect wind turbine blades according to a target blade detection model in order to capture clear and complete images of rotating wind turbine blades.

[0056] Please refer to Figure 2 The above is a flowchart of a wind turbine blade photography method provided in an embodiment of this application. The wind turbine blade photography method includes steps S21 to S23.

[0057] Step S21: Send a laser to the target wind turbine. In this step, laser ranging is used to detect the rotation status of the blades of the target wind turbine.

[0058] In this embodiment, the distance between the shooting device and the wind turbine blades is obtained by laser ranging, which can be achieved using a drone equipped with a gimbal shooting system. The gimbal shooting system hardware may include: a motorized gimbal, a laser rangefinder, a telephoto camera, a computer, and a power supply.

[0059] When performing laser ranging on a target wind turbine, first control the drone to fly to a predetermined altitude and hover below the blades. Tilt the gimbal downwards to aim at the wind turbine tower, activate the laser ranging function, and the laser is emitted from the drone's laser to the target wind turbine, which is the wind turbine whose blades need to be photographed.

[0060] Step S22: Receive the laser ranging signal. When the laser ranging signal is greater than a set threshold, input the acquired image into the target blade detection model for blade detection to determine the blade position information.

[0061] In this step, as the wind turbine blades rotate, the laser emitted by the laser is interrupted, causing a change in the value of the laser ranging signal. When the distance change is greater than or equal to a set threshold, it means that the blade has been detected to have rotated into the target range of the UAV. At this time, the image acquisition device set on the UAV acquires images in real time and inputs the acquired images into the target blade detection model for target detection to determine the blade position information. The target blade detection model is trained using the blade detection model training method described above.

[0062] During the real-time target detection process of inputting the acquired image into the target leaf detection model, the target leaf detection model performs leaf detection on the input image. When a clear and complete leaf image is detected, the outer region of the leaf is generated, and the position information of the leaf can be determined based on the outer region of the leaf.

[0063] Please refer to Figure 3 This is a schematic diagram of the blade detection process in one embodiment of this application.

[0064] The target leaf detection model predicts and generates the minimum bounding rectangle region of the leaf based on the input image. This minimum bounding rectangle region initially separates the leaf from the background, narrowing the search range. Then, based on the minimum bounding rectangle region, the target leaf detection model further predicts the positional relationship of the leaf, generates the minimum directed region of the leaf, and outputs the minimum directed bounding box of the leaf that encloses the minimum directed region of the leaf.

[0065] In this embodiment, the generated minimum oriented region of the blade limits the space where the blade is located to the greatest extent, eliminating interference from the background and other objects. Compared with existing methods, it can obtain blade position information with higher accuracy and lower error rate.

[0066] In other embodiments, the blade detection model can be updated in real time, thereby continuously adjusting the blade detection model based on the collected wind turbine blade images during the target detection process to adapt to wind turbine blade detection in different environments. Compared with traditional methods, this method has the advantage of being able to be extended to environments with different appearance distributions.

[0067] Step S23: Adjust the camera shooting parameters according to the blade position information to capture images of the blade. In this step, the gimbal can be controlled according to the blade position information to achieve automatic shooting with changing camera angles.

[0068] In this embodiment, the camera is mounted on top of the gimbal and can rotate with the gimbal, adjusting shooting parameters such as camera position, focus distance, rotation angle, and rotation speed based on the blade position information. For example, the gimbal rotation angle can be calculated based on the angle information in the blade position information and the gimbal position information, and then the rotation angle of the UAV gimbal can be controlled according to the gimbal rotation angle information. As another example, the rotation speed of the UAV gimbal can be adjusted based on the speed information in the blade position information, thereby enabling the capture of complete, high-definition images of the wind turbine blades.

[0069] In one embodiment, the coordinates of multiple waypoints required for drone inspection can be preset and the inspection route of the drone can be planned. Then, the drone can be controlled to carry out inspections according to the inspection route and take pictures of the wind turbine blades during the inspection process.

[0070] In one embodiment, after a blade image is captured, the value in the blade image capture counter is incremented by one. When the value in the blade image capture counter reaches a set value, the capture of the blade image ends, thereby realizing the function of automatically collecting all the blades of a wind turbine.

[0071] In summary, the wind turbine blade imaging method provided in this application combines laser ranging with target detection and tracking triggering. When an object is detected entering the blade via laser ranging, target detection is triggered. The blade position information obtained from the target detection is used to adjust the camera's imaging parameters for capturing the blade image. The wind turbine blade imaging method of this application produces complete and high-resolution images. Compared to existing target detection methods that require extensive computation and are relatively time-consuming, it reduces the number of target detection operations, improves detection efficiency, and makes capturing wind turbine blades more timely and accurate.

[0072] Based on the same inventive concept, one embodiment of this application also provides a blade detection model training device.

[0073] Please refer to Figure 4 This is a block diagram of a blade detection model training device according to an embodiment of this application. The blade detection model training device includes the following modules:

[0074] The sample acquisition module 41 is used to acquire a training sample set, which consists of labeled images of wind turbine blades with different angles and different rotation directions.

[0075] The loss function module 42 is configured to construct a first loss function L1, a second loss function L2, and a third loss function L3. The first loss function L1 calculates the loss between the true value of the leaf and the predicted value of the leaf in the training sample set. The second loss function L2 calculates the loss between the angles of the tilt direction of the leaves in the training sample set. The third loss function L3 calculates the loss between the predicted geometric center and the actual geometric center in the training sample set.

[0076] The total loss function module 43 is configured to obtain the total loss function L = L1 + L2 + L3 based on the first loss function L1, the second loss function L2, and the second loss function L3.

[0077] Training module 44 is configured to train the blade detection model based on the training sample set and perform backpropagation iterations until the total loss function L reaches its minimum value, thereby generating the target blade detection model.

[0078] The specific implementation process of the functions and roles of each module in the above-mentioned device can be found in the implementation process of the corresponding steps in the above-mentioned blade detection model training method, and will not be repeated here.

[0079] Based on the same inventive concept, one embodiment of this application also provides a wind turbine blade imaging device.

[0080] Please refer to Figure 5 This is a block diagram of a wind turbine blade imaging device according to an embodiment of this application. The wind turbine blade imaging device includes the following modules:

[0081] Laser transmitting module 51 is configured to transmit a laser to the target wind turbine;

[0082] The blade detection module 52 is configured to receive a laser ranging signal. When the laser ranging signal is greater than a set threshold, the acquired image is input into the target blade detection model for blade detection to determine the blade position information. The target blade detection model is trained using the blade detection model training method described above.

[0083] The image capturing module 53 is configured to adjust the camera capturing parameters according to the leaf position information in order to capture leaf images.

[0084] This application also provides an electronic device, please refer to... Figure 6 Here is a block diagram of the electronic device, which includes at least one processor 61 and a memory 62. Figure 6 Taking a processor 61 as an example. The processor 61 and the memory 62 are connected via a bus 60. The memory 62 stores instructions that can be executed by the processor 61. The instructions are executed by the processor 61 so that the electronic device 6 can execute all or part of the processes in the blade detection model training method and the blade imaging method in the above embodiments.

[0085] The memory 62 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable red-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0086] It should be noted that the electronic device 6 is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0087] This application also provides a computer-readable storage medium storing a computer program that can be executed by a processor 61 to complete a blade detection model training method or a blade imaging method provided in this application.

[0088] The apparatuses and methods disclosed in the several embodiments provided in this application can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatuses, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0089] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for photographing wind turbine blades, characterized in that, include: Send a laser beam to the target wind turbine; The system receives laser ranging signals. When the laser ranging signal is greater than a set threshold, the acquired image is input into the target leaf detection model for leaf detection to determine the leaf position information. The camera shooting parameters are adjusted according to the leaf position information in order to capture images of the leaf. The target leaf detection model was obtained by training the following leaf detection model training method: Obtain a training sample set, which consists of images of wind turbine blades with different angles and rotation directions; Construct a first loss function L1, a second loss function L2, and a third loss function L3. The first loss function L1 calculates the loss between the true value and the predicted value of the leaf in the training sample set. The second loss function L2 calculates the loss between the angles of the tilt direction of the leaf in the training sample set. The third loss function L3 calculates the loss between the predicted geometric center and the actual geometric center in the training sample set. Based on the first loss function L1, the second loss function L2, and the third loss function L3, the total loss function L = L1 + L2 + L3 is obtained; The blade detection model is trained based on the training sample set, and backpropagation iteration is performed until the total loss function L reaches its minimum value, thereby generating the target blade detection model.

2. The wind turbine blade imaging method as described in claim 1, characterized in that, Also includes: After each leaf image is captured, the value in the leaf image capture counter is incremented by one. When the value in the leaf image capture counter reaches a set value, the leaf image capture ends.

3. The wind turbine blade imaging method as described in claim 1, characterized in that, The step of training a leaf detection model based on the training sample set, performing backpropagation iterations until the total loss function L reaches its minimum value, and generating a target leaf detection model specifically includes: The pre-processed grayscale image of the wind turbine blade is input into the feature extraction network of the blade detection model to extract feature maps. A fixed number of regions of interest are defined for each pixel position in the feature map, and these regions of interest are fed into a region generation network for binary classification and coordinate regression. The total loss function L is backpropagated iteratively until the total loss function L reaches its minimum value, thereby obtaining the optimal network model parameters and the target blade detection model.

4. The wind turbine blade imaging method as described in claim 1, characterized in that, The leaf detection model is a convolutional neural network model, which includes: a feature extraction network, a pooling layer, and a region generation network.

5. The wind turbine blade imaging method as described in claim 1, characterized in that, After obtaining the training sample set, which consists of wind turbine blade images with different angles and rotation directions, the method further includes: image processing of the wind turbine blade images, including: adding random noise, mirror flipping, clockwise rotation, and grayscale processing.

6. A wind turbine blade imaging device, characterized in that, Includes the following modules: The laser transmitting module is configured to send a laser beam to the target wind turbine; The blade detection module is configured to receive laser ranging signals. When the laser ranging signal is greater than a set threshold, the acquired image is input into the target blade detection model for blade detection to determine the blade position information. The image capturing module is configured to adjust the camera capturing parameters according to the blade position information in order to capture images of the blade. The target leaf detection model was obtained by training the following leaf detection model training method: Obtain a training sample set, which consists of images of wind turbine blades with different angles and rotation directions; Construct a first loss function L1, a second loss function L2, and a third loss function L3. The first loss function L1 calculates the loss between the true value and the predicted value of the leaf in the training sample set. The second loss function L2 calculates the loss between the angles of the tilt direction of the leaf in the training sample set. The third loss function L3 calculates the loss between the predicted geometric center and the actual geometric center in the training sample set. Based on the first loss function L1, the second loss function L2, and the third loss function L3, the total loss function L = L1 + L2 + L3 is obtained; The blade detection model is trained based on the training sample set, and backpropagation iteration is performed until the total loss function L reaches its minimum value, thereby generating the target blade detection model.

7. An electronic device, characterized in that, The electronic device includes: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to perform the wind turbine blade imaging method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores one or more programs, which can be executed by one or more processors to implement the wind turbine blade imaging method as described in any one of claims 1 to 5.

Images