A fan defect detection method and system based on machine vision

Abstract

The application discloses a fan defect detection method and system based on machine vision, and belongs to the technical field of image data processing. The method comprises the following steps: acquiring images of historical shooting of fan blades, generating a training data set, establishing a deep learning network architecture based on a convolutional neural network, establishing a fan blade motion recognition model, performing calculation and processing by using a dark channel prior algorithm, acquiring a dynamically adjusted weight, training an AOD-Net model based on the dynamically adjusted weight, establishing a fan blade blur correction model, performing real-time shooting on the fan blades, performing motion recognition correction by using the fan blade motion recognition model, removing dust and fog by using the fan blade blur correction model, establishing a defect detection model, and performing defect detection by using the defect detection model. The method can accurately recognize high-speed moving fan blades, make the fan blade contour details clear, accurately analyze the fan blade state in time during defect detection, and improve the working performance of the fan blades.

Description

Technical Field

[0001] This application belongs to the field of image data processing technology, specifically, it relates to a method and system for detecting wind turbine defects based on machine vision. Background Technology

[0002] Wind power generation is a renewable energy technology that uses wind energy to convert into electrical energy. As a renewable, environmentally friendly, and economically efficient form of energy, it is of great significance to achieving sustainable development goals. The core component of a wind turbine, the wind turbine blade, is the rotor blade. Its performance directly affects the working efficiency and service life of the wind turbine. Therefore, it is necessary to promptly identify and address defects on the surface of the wind turbine blades in order to improve their performance.

[0003] Traditional methods for detecting blade damage and malfunctions involve observation with binoculars and manual inspection by rappelling. These methods are costly due to their high-altitude nature and the fact that they are manual inspections result in high labor intensity for workers, long inspection times, and low efficiency.

[0004] Due to the low efficiency and accuracy of manual inspection, a method based on machine vision theory and image processing technology was proposed to inspect wind turbine blades based on surface image data. However, some wind turbines are located in harsh environments, such as sandstorms, fog, and rain and snow. In addition, the wind turbine blades are in motion, which makes the blade details in the captured images unclear and reduces the accuracy of the inspection results. Summary of the Invention

[0005] To address the aforementioned problems and technical deficiencies, this application adopts the following technical solution: a machine vision-based method for detecting wind turbine defects, comprising the following steps:

[0006] Step 1: Acquire historical images of the wind turbine blades, preprocess the historical images, and generate a training dataset;

[0007] Step 2: Establish a deep learning network architecture based on convolutional neural networks, train it using the training dataset, and build a wind turbine blade motion recognition model;

[0008] Step 3: Use the dark channel prior algorithm to process the historical images in the training dataset to obtain dynamically adjusted weights, and train the AOD-Net model based on the dynamically adjusted weights to establish a wind turbine blade blur correction model.

[0009] Step 4: Take real-time photos of the wind turbine blades, use the wind turbine blade motion recognition model to perform motion recognition correction on the real-time captured wind turbine blade images, and use the wind turbine blade blur correction model to remove dust and fog.

[0010] Step 5: Establish a defect detection model and use the defect detection model to detect defects in the wind turbine blade images after motion recognition correction and dust removal.

[0011] Preferably, establishing the wind turbine blade motion recognition model includes:

[0012] The historical images in the training dataset are divided into continuous image segments of fixed length. The input data dimensions are adjusted according to the data dimensions to meet the requirements of 3D convolutional neural networks.

[0013] The first two C3 modules in the 3D convolutional neural network retain 2D convolution to capture dynamic features, while the 2D convolution of the last two C3 modules is replaced with 3D convolution to extract spatiotemporal features.

[0014] The dynamic features are labeled and filtered, and then input into the recurrent neural network;

[0015] In the fully connected layer of the recurrent neural network, dynamic features are mapped to the probability distribution of the captured images. The sequence length of the probability distribution is used for iterative iteration. By combining the cyclic structure and spatiotemporal features, a wind turbine blade motion recognition model is established.

[0016] Preferably, the calculation process of the dark channel prior algorithm includes:

[0017] By using guided filtering to accelerate the calculation of minimum value filtering in the dark channel of historically captured images, a dark channel transmittance map is obtained.

[0018] The preset pixel ratio is used to select the brightest pixel in the dark channel transmittance map and the selected pixel is used as the highest intensity pixel in the corresponding original image.

[0019] Transmittance is calculated using an atmospheric scattering model combined with the transmittance calculation results to generate a preliminary deblurred image.

[0020] The sharpness of the initial deblurred image is calculated using the Laplacian variance algorithm, the noise level of the initial deblurred image is calculated using block region variance analysis, and the contrast of the initial deblurred image is calculated using the image grayscale histogram.

[0021] The sharpness, noise level, and contrast are normalized according to preset weights to generate dynamically adjusted weights.

[0022] Furthermore, the training of the AOD-Net model based on dynamically adjusted weights includes:

[0023] A three-layer fully connected network is constructed. In the input layer, the statistical characteristics of the quality index and the dark channel transmittance map are calculated.

[0024] Activate the linear rectified function in the hidden layer to calculate the weight dimension;

[0025] In the output layer, the weight offsets of each convolutional layer of the AOD-Net model are calculated based on statistical features and weight dimensions.

[0026] The backbone parameters of the AOD-Net model are fixed, and the weighted loss is calculated by combining the dehazing loss and quality index. The weight offset is corrected based on the weighted loss to obtain the dynamically adjusted weights after training.

[0027] Furthermore, the establishment of the fuzzy correction model for the wind turbine blades includes:

[0028] The initially deblurred image is concatenated with the original blurred image along the channel dimension, and the concatenated data along the channel dimension is received at the input layer of the AOD-Net model.

[0029] The last three convolutional layers input dynamically adjusted weights, and the data concatenated along the channel dimension is modified based on the dynamically adjusted weights to output the modified deblurred image.

[0030] The deblurred image is compared with the corresponding clear image. Based on the comparison results, the deblurred image is judged. If it is not a deblurred image that meets the requirements, retraining is performed until it meets the requirements.

[0031] Obtain the dynamic adjustment weights corresponding to all deblurred images that meet the requirements, mark the dynamic adjustment weights, and establish a wind turbine blade fuzz correction model based on the marked dynamic adjustment weights.

[0032] Furthermore, the motion recognition correction of the real-time captured wind turbine blade images includes:

[0033] Edge detection and contour detection algorithms are used to determine multiple corner points of a static device object, and the target shape and size are determined based on the corner points.

[0034] The corner points of the device object in the image frames of the captured video are determined, and the determined corner points are calculated with the target shape and size to obtain the perspective transformation matrix of the device object.

[0035] Apply the perspective transformation matrix to the target device object in the image frame to correct the deformation of the device object to its original shape and size.

[0036] Furthermore, the process of establishing the defect detection model includes:

[0037] Acquire historical defect images of wind turbine blades, extract features from the historical defect images, obtain defect features, and generate a set of wind turbine blade image defects;

[0038] The defect set of wind turbine blade images is dimensionality reduced to generate a subset of image defects. The subset of image defects is then trained and calculated to generate multiple defect sub-classifiers. The defect sub-classifiers are then weighted and fused to generate a final defect classifier.

[0039] A defect classifier is used to detect defects in real-time captured images of wind turbine blades. The classifier determines whether the images of wind turbine blades contain defect features. If defect features are found, the defects are classified to determine the type of defect in the wind turbine blades.

[0040] A machine vision-based wind turbine defect detection system includes:

[0041] The data storage module is used to acquire historical images of wind turbine blades, preprocess the historical images, and generate a training dataset.

[0042] The motion recognition module is used to build a deep learning network architecture based on a convolutional neural network, train it using a training dataset, build a wind turbine blade motion recognition model, and use the wind turbine blade motion recognition model to perform motion recognition correction on real-time captured wind turbine blade images.

[0043] The blur correction module is used to process historical images in the training dataset using the dark channel prior algorithm to obtain dynamically adjusted weights. Based on the dynamically adjusted weights, the AOD-Net model is trained to establish a wind turbine blade blur correction model. The wind turbine blade blur correction model is then used to remove dust and fog from real-time captured wind turbine blade images.

[0044] The defect detection module is used to capture real-time images of wind turbine blades, establish a defect detection model, and use the defect detection model to detect defects in the wind turbine blade images after motion recognition correction and dust removal.

[0045] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the content of the machine vision-based wind turbine defect detection method described above.

[0046] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the content of the machine vision-based wind turbine defect detection method as described above.

[0047] Compared to existing technologies, the beneficial effects of this application are as follows:

[0048] This application establishes a wind turbine blade motion recognition model to identify the motion of wind turbine blades in operation. Even wind turbine blades moving at high speeds can be accurately identified. A wind turbine blade fuzzy correction model is established to remove dust and fog from external factors such as sand, fog, rain, and snow, thereby improving image quality and making the outline details of the wind turbine blades clear. When performing defect detection, the state of the wind turbine blades can be analyzed in a timely and accurate manner, thereby improving the working performance of the wind turbine blades. Attached Figure Description

[0049] In the attached diagram:

[0050] Figure 1 This is a schematic diagram of the method steps in an embodiment of this application;

[0051] Figure 2 This is a schematic diagram of the system structure according to an embodiment of this application;

[0052] Figure 3 This is a schematic diagram of the device structure according to an embodiment of this application. Detailed Implementation

[0053] 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. Generally, the components of the embodiments of this application described and shown in the accompanying drawings can be arranged and designed in various different configurations.

[0054] Example 1

[0055] like Figure 1 As shown, a machine vision-based method for wind turbine defect detection includes the following steps:

[0056] Acquire historical images of wind turbine blades, preprocess the historical images, and generate a training dataset;

[0057] A deep learning network architecture was established based on convolutional neural networks, and a training dataset was used to train the model to recognize the motion of wind turbine blades.

[0058] The establishment of a wind turbine blade motion recognition model includes:

[0059] The historical images in the training dataset are divided into continuous image segments of fixed length. The input data dimensions are adjusted according to the data dimensions to meet the requirements of 3D convolutional neural networks.

[0060] The first two C3 modules in the 3D convolutional neural network retain 2D convolution to capture dynamic features, while the 2D convolution of the last two C3 modules is replaced with 3D convolution to extract spatiotemporal features.

[0061] The dynamic features are labeled and filtered, and then input into the recurrent neural network;

[0062] In the fully connected layer of the recurrent neural network, dynamic features are mapped to the probability distribution of the captured images. The sequence length of the probability distribution is used for iterative iteration. By combining the cyclic structure and spatiotemporal features, a wind turbine blade motion recognition model is established.

[0063] The dark channel prior algorithm is used to process historical images in the training dataset to obtain dynamically adjusted weights. The AOD-Net model is then trained based on the dynamically adjusted weights to establish a wind turbine blade blur correction model.

[0064] The dark channel prior algorithm performs the following calculations:

[0065] By using guided filtering to accelerate the calculation of minimum value filtering in the dark channel of historically captured images, a dark channel transmittance map is obtained.

[0066] The preset pixel ratio is used to select the brightest pixel in the dark channel transmittance map and the selected pixel is used as the highest intensity pixel in the corresponding original image.

[0067] Transmittance is calculated using an atmospheric scattering model combined with the transmittance calculation results to generate a preliminary deblurred image.

[0068] The sharpness of the initial deblurred image is calculated using the Laplacian variance algorithm, the noise level of the initial deblurred image is calculated using block region variance analysis, and the contrast of the initial deblurred image is calculated using the image grayscale histogram.

[0069] The sharpness, noise level, and contrast are normalized according to preset weights to generate dynamically adjusted weights.

[0070] Training the AOD-Net model based on dynamically adjusted weights includes:

[0071] A three-layer fully connected network is constructed. In the input layer, the statistical characteristics of the quality index and the dark channel transmittance map are calculated.

[0072] Activate the linear rectified function in the hidden layer to calculate the weight dimension;

[0073] In the output layer, the weight offsets of each convolutional layer of the AOD-Net model are calculated based on statistical features and weight dimensions.

[0074] The backbone parameters of the AOD-Net model are fixed, and the weighted loss is calculated by combining the dehazing loss and quality index. The weight offset is corrected based on the weighted loss to obtain the dynamically adjusted weights after training.

[0075] The establishment of a fuzzy correction model for wind turbine blades includes:

[0076] The initially deblurred image is concatenated with the original blurred image along the channel dimension, and the concatenated data along the channel dimension is received at the input layer of the AOD-Net model.

[0077] The last three convolutional layers input dynamically adjusted weights, and the data concatenated along the channel dimension is modified based on the dynamically adjusted weights to output the modified deblurred image.

[0078] The deblurred image is compared with the corresponding clear image. Based on the comparison results, the deblurred image is judged. If it is not a deblurred image that meets the requirements, retraining is performed until it meets the requirements.

[0079] Obtain the dynamic adjustment weights corresponding to all deblurred images that meet the requirements, mark the dynamic adjustment weights, and establish a wind turbine blade fuzz correction model based on the marked dynamic adjustment weights.

[0080] Real-time images of wind turbine blades are captured, and a wind turbine blade motion recognition model is used to perform motion recognition correction on the captured images. A wind turbine blade blur correction model is used to remove dust and fog.

[0081] Motion recognition and correction of real-time captured images of wind turbine blades includes:

[0082] Edge detection and contour detection algorithms are used to determine multiple corner points of a static device object, and the target shape and size are determined based on the corner points.

[0083] The corner points of the device object in the image frames of the captured video are determined, and the determined corner points are calculated with the target shape and size to obtain the perspective transformation matrix of the device object.

[0084] Apply the perspective transformation matrix to the target device object in the image frame to correct the deformation of the device object to its original shape and size.

[0085] A defect detection model was established, and the model was used to detect defects in the wind turbine blade images after motion recognition correction and dust removal.

[0086] The process of establishing a defect detection model includes:

[0087] Acquire historical defect images of wind turbine blades, extract features from the historical defect images, obtain defect features, and generate a set of wind turbine blade image defects;

[0088] The defect set of wind turbine blade images is dimensionality reduced to generate a subset of image defects. The subset of image defects is then trained and calculated to generate multiple defect sub-classifiers. The defect sub-classifiers are then weighted and fused to generate a final defect classifier.

[0089] A defect classifier is used to detect defects in real-time captured images of wind turbine blades. The classifier determines whether the images of wind turbine blades contain defect features. If defect features are found, the defects are classified to determine the type of defect in the wind turbine blades.

[0090] Example 2

[0091] like Figure 2 As shown, a machine vision-based wind turbine defect detection system includes:

[0092] The data storage module is used to acquire historical images of wind turbine blades, preprocess the historical images, and generate a training dataset.

[0093] The motion recognition module is used to build a deep learning network architecture based on a convolutional neural network, train it using a training dataset, build a wind turbine blade motion recognition model, and use the wind turbine blade motion recognition model to perform motion recognition correction on real-time captured wind turbine blade images.

[0094] The blur correction module is used to process historical images in the training dataset using the dark channel prior algorithm to obtain dynamically adjusted weights. Based on the dynamically adjusted weights, the AOD-Net model is trained to establish a wind turbine blade blur correction model. The wind turbine blade blur correction model is then used to remove dust and fog from real-time captured wind turbine blade images.

[0095] The defect detection module is used to capture real-time images of wind turbine blades, establish a defect detection model, and use the defect detection model to detect defects in the wind turbine blade images after motion recognition correction and dust removal.

[0096] Example 3

[0097] like Figure 3 As shown, from a hardware perspective, this application provides an embodiment of an electronic device containing all or part of a machine vision-based wind turbine defect detection method. The electronic device includes a service processor and a distributed memory. The service processor is connected to the memory. The distributed memory stores a service self-management program configured to store machine-readable instructions. The service processor executes the service self-management program. When the instructions are executed by the processor, they implement the machine vision-based wind turbine defect detection method as described above.

[0098] From a hardware perspective, in order to effectively improve the flexibility, versatility, and efficiency of data acquisition, this application provides an embodiment of an electronic device comprising all or part of a machine vision-based wind turbine defect detection method. The electronic device specifically includes the following components:

[0099] The system includes a processor, memory, a communications interface, and a bus; wherein the processor, memory, and communications interface communicate with each other via the bus; the communications interface is used to realize information transmission between the core business system of the machine vision-based wind turbine defect detection method, user terminals, and related databases and other related devices; the logic controller can be a desktop computer, tablet computer, or mobile terminal, etc., but this embodiment is not limited to these.

[0100] In this embodiment, the logic controller can be implemented with reference to the embodiment of the machine vision-based wind turbine defect detection method, the content of which is incorporated herein, and repeated parts will not be described again.

[0101] It is understood that the user terminal may include smartphones, tablet electronic devices, network set-top boxes, portable computers, desktop computers, personal digital assistants (PDAs), in-vehicle devices, smart wearable devices, etc., wherein the smart wearable devices may include smart glasses, smartwatches, smart bracelets, etc.

[0102] In practical applications, parts of the machine vision-based wind turbine defect detection method can be executed on the electronic device side as described above, or all operations can be completed in the client device. The choice depends on the processing power of the client device and the limitations of the user's usage scenario, and this application does not impose any limitations on this. If all operations are completed in the client device, the client device may further include a processor.

[0103] The aforementioned client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission with the server. The server may include a server on the task scheduling center side, and in other implementation scenarios, it may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster composed of multiple servers, or a server structure of a distributed device.

[0104] Example 4

[0105] The embodiments of this application also provide a computer-readable storage medium capable of implementing the machine vision-based wind turbine defect detection method with the execution subject as a server or client in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all the contents of the machine vision-based wind turbine defect detection method with the execution subject as a server or client in the above embodiments.

[0106] The embodiments of this application may be provided as methods, apparatus, or computer program products. Therefore, this application may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0107] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0108] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0109] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0110] The embodiments described above are merely preferred embodiments of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications, improvements, and substitutions without departing from the concept of this application, and these all fall within the protection scope of this application.

Claims

1. A machine vision-based method for detecting defects in wind turbines, characterized in that, Includes the following steps: Step 1: Acquire historical images of the wind turbine blades, preprocess the historical images, and generate a training dataset; Step 2: Establish a deep learning network architecture based on convolutional neural networks, train it using the training dataset, and build a wind turbine blade motion recognition model; Step 3: Use the dark channel prior algorithm to process the historical images in the training dataset to obtain dynamically adjusted weights, and train the AOD-Net model based on the dynamically adjusted weights to establish a wind turbine blade blur correction model. The establishment of a fuzzy correction model for wind turbine blades includes: The initially deblurred image is concatenated with the original blurred image along the channel dimension, and the concatenated data along the channel dimension is received at the input layer of the AOD-Net model. The last three convolutional layers input dynamically adjusted weights, and the data concatenated along the channel dimension is modified based on the dynamically adjusted weights to output the modified deblurred image. The deblurred image is compared with the corresponding clear image. Based on the comparison results, the deblurred image is judged. If it is not a deblurred image that meets the requirements, retraining is performed until it meets the requirements. Obtain the dynamic adjustment weights corresponding to all deblurred images that meet the requirements, mark the dynamic adjustment weights, and establish a wind turbine blade blur correction model based on the marked dynamic adjustment weights. Step 4: Take real-time photos of the wind turbine blades, use the wind turbine blade motion recognition model to perform motion recognition correction on the real-time captured wind turbine blade images, and use the wind turbine blade blur correction model to remove dust and fog. Step 5: Establish a defect detection model and use the defect detection model to detect defects in the wind turbine blade images after motion recognition correction and dust removal.

2. The machine vision-based wind turbine defect detection method according to claim 1, characterized in that, The establishment of the wind turbine blade motion recognition model includes: The historical images in the training dataset are divided into continuous image segments of fixed length. The input data dimensions are adjusted according to the data dimensions to meet the requirements of 3D convolutional neural networks. The first two C3 modules in the 3D convolutional neural network retain 2D convolution to capture dynamic features, while the 2D convolution of the last two C3 modules is replaced with 3D convolution to extract spatiotemporal features. The dynamic features are labeled and filtered, and then input into the recurrent neural network; In the fully connected layer of the recurrent neural network, dynamic features are mapped to the probability distribution of the captured images. The sequence length of the probability distribution is used for iterative iteration. By combining the cyclic structure and spatiotemporal features, a wind turbine blade motion recognition model is established.

3. The machine vision-based wind turbine defect detection method according to claim 1, characterized in that, The calculation process of the dark channel prior algorithm includes: By using guided filtering to accelerate the calculation of minimum value filtering in the dark channel of historically captured images, a dark channel transmittance map is obtained. The preset pixel ratio is used to select the brightest pixel in the dark channel transmittance map and the selected pixel is used as the highest intensity pixel in the corresponding original image. Transmittance is calculated using an atmospheric scattering model combined with the transmittance calculation results to generate a preliminary deblurred image. The sharpness of the initial deblurred image is calculated using the Laplacian variance algorithm, the noise level of the initial deblurred image is calculated using block region variance analysis, and the contrast of the initial deblurred image is calculated using the image grayscale histogram. The sharpness, noise level, and contrast are normalized according to preset weights to generate dynamically adjusted weights.

4. The machine vision-based wind turbine defect detection method according to claim 3, characterized in that, The training of the AOD-Net model based on dynamically adjusted weights includes: A three-layer fully connected network is constructed. In the input layer, the statistical characteristics of the quality index and the dark channel transmittance map are calculated. Activate the linear rectified function in the hidden layer to calculate the weight dimension; In the output layer, the weight offsets of each convolutional layer of the AOD-Net model are calculated based on statistical features and weight dimensions. The backbone parameters of the AOD-Net model are fixed, and the weighted loss is calculated by combining the dehazing loss and quality index. The weight offset is corrected based on the weighted loss to obtain the dynamically adjusted weights after training.

5. The machine vision-based wind turbine defect detection method according to claim 1, characterized in that, The motion recognition and correction of the real-time captured wind turbine blade images includes: Edge detection and contour detection algorithms are used to determine multiple corner points of a static device object, and the target shape and size are determined based on the corner points. The corner points of the device object in the image frames of the captured video are determined, and the determined corner points are calculated with the target shape and size to obtain the perspective transformation matrix of the device object. Apply the perspective transformation matrix to the target device object in the image frame to correct the deformation of the device object to its original shape and size.

6. The machine vision-based wind turbine defect detection method according to claim 1, characterized in that, The process of establishing the defect detection model includes: Acquire historical defect images of wind turbine blades, extract features from the historical defect images, obtain defect features, and generate a set of wind turbine blade image defects; The defect set of wind turbine blade images is dimensionality reduced to generate a subset of image defects. The subset of image defects is then trained and calculated to generate multiple defect sub-classifiers. The defect sub-classifiers are then weighted and fused to generate a final defect classifier. A defect classifier is used to detect defects in real-time captured images of wind turbine blades. The classifier determines whether the images of wind turbine blades contain defect features. If defect features are found, the defects are classified to determine the type of defect in the wind turbine blades.

7. A machine vision-based wind turbine defect detection system, characterized in that, include: The data storage module is used to acquire historical images of wind turbine blades, preprocess the historical images, and generate a training dataset. The motion recognition module is used to build a deep learning network architecture based on a convolutional neural network, train it using a training dataset, build a wind turbine blade motion recognition model, and use the wind turbine blade motion recognition model to perform motion recognition correction on real-time captured wind turbine blade images. The blur correction module is used to process historical images in the training dataset using the dark channel prior algorithm to obtain dynamically adjusted weights. Based on the dynamically adjusted weights, the AOD-Net model is trained to establish a wind turbine blade blur correction model. The wind turbine blade blur correction model is then used to remove dust and fog from real-time captured wind turbine blade images. The establishment of a fuzzy correction model for wind turbine blades includes: The initially deblurred image is concatenated with the original blurred image along the channel dimension, and the concatenated data along the channel dimension is received at the input layer of the AOD-Net model. The last three convolutional layers input dynamically adjusted weights, and the data concatenated along the channel dimension is modified based on the dynamically adjusted weights to output the modified deblurred image. The deblurred image is compared with the corresponding clear image. Based on the comparison results, the deblurred image is judged. If it is not a deblurred image that meets the requirements, retraining is performed until it meets the requirements. Obtain the dynamic adjustment weights corresponding to all deblurred images that meet the requirements, mark the dynamic adjustment weights, and establish a wind turbine blade blur correction model based on the marked dynamic adjustment weights. The defect detection module is used to capture real-time images of wind turbine blades, establish a defect detection model, and use the defect detection model to detect defects in the wind turbine blade images after motion recognition correction and dust removal.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the content of the machine vision-based wind turbine defect detection method as described in claim 1.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the content of the machine vision-based wind turbine defect detection method as described in claim 1.

Images