Method and Device for Generating Defect Images of Wind Turbine Blades and Defect Recognition Based on CycleGAN

Abstract

This invention proposes a CycleGAN-based method and device for generating and identifying defect images of fan blades. This method generates multiple defect image generation models based on CycleGAN for different types of fan blade defects. Using the CycleGAN model, normal fan blade images can be converted into different types of defect images, addressing the scarcity of fan blade defect images. Using the original defect image and a large number of generated different types of fan blade defect images, a YOLO-based fan blade image defect recognition model is trained to address the low accuracy of fan blade image defect recognition. This invention enables the conversion of a large number of normal fan blade images into a large number of blade defect images, thereby providing the blade image defect recognition model with sufficient defect image data and improving the recognition model's defect recognition performance.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a method, device, equipment and medium for generating defect images and identifying defects of wind turbine blades based on CycleGAN. Background Art

[0002] With the rapid development of the wind power industry, wind power is playing an increasingly important role in my country's energy mix. However, due to the extended operation of many turbines, which have already exceeded their warranty period, wind turbine maintenance is becoming a bottleneck hindering the rapid development of the wind power industry. According to statistics, wind turbine maintenance costs account for approximately one-fifth of the total cost of wind power generation. Wind turbines often operate in harsh and complex environments for extended periods of time. The blades, subject to constant and rapid shear loads due to variable wind conditions, are prone to surface defects such as pinholes, cracks, and peeling. Wind turbine blades are crucial components for converting wind energy into electricity and are also vulnerable to wear. Blade defects can severely impact the safe operation and power generation efficiency of wind turbines.

[0003] With the continuous development of smart devices and imaging technology, recent research has focused on blade defect recognition using drones. Drones equipped with high-definition cameras can perform automated inspections using trajectory planning, accurately and clearly capturing the surface of wind turbine blades and wirelessly transmitting the captured high-definition images or videos. Deep learning algorithms then intelligently identify blade images, automatically identifying and precisely locating blade defects. However, blade defect recognition methods are primarily implemented using deep learning algorithms, which typically require a large number of image samples for training. Since drone technology is only just beginning to be used in blade inspection, while there is ample data available on normal blade images, images of blade defects are extremely scarce. Numerous studies on wind turbine blade defect recognition have noted that the very limited number of defect samples in the training set results in insufficient generalization of the model, leading to missed and false positives. Summary of the Invention

[0004] The present invention provides a method and device for generating and identifying defect images of wind turbine blades based on CycleGAN, aiming to convert a large number of normal images of wind turbine blades into a large number of blade defect images, thereby providing sufficient defect image data for the blade image defect recognition model and improving the defect recognition effect of the recognition model.

[0005] To this end, the purpose of the present invention is to propose a method for generating defect images and identifying defects in wind turbine blades based on CycleGAN, comprising:

[0006] Construct a fan blade defect image generation and defect recognition network, which includes a fan blade defect image generation network and a fan blade defect image recognition network;

[0007] Obtain a wind blade image dataset, divide the wind blade into normal images and wind blade defect images, and mark the defect types and defect locations of the wind blade defect images; train a wind blade defect image generation network using the marked wind blade image dataset; and train a wind blade defect image recognition network based on the wind blade defect images generated by the trained wind blade defect image generation network;

[0008] The real-time captured fan blade defect image is input into the trained fan blade defect image recognition network, and the output result is the defect recognition result of the captured fan blade defect image.

[0009] The wind blade image dataset is divided into a normal wind blade image dataset and N types of wind blade defect image datasets; N is the defect type, including at least gel coat peeling, paint peeling, cracking, surface cracks, corrosion, oil stains, sand holes and ice.

[0010] Among them, the fan blade defect image generation network uses the CycleGAN model structure. The fan blade defect image generation network is designed with the same number of fan blade defect image generation subnetworks as the defect types. Each fan blade defect image generation subnetwork includes two generators G and F and two discriminators D. X 、D Y The input of each fan blade defect image generation subnetwork is a normal fan blade image X and a fan blade defect image Y with one defect type. The generator G generates a fan blade defect image X' with the same defect type as the fan blade defect image Y by inputting the normal fan blade image X; the generator F generates a normal fan blade image Y' by inputting the fan blade defect image Y; D X Used to distinguish the real normal image X of the wind blade and the generated normal image Y', D Y Used to truly distinguish the fan blade defect image Y from the generated fan blade defect image X'.

[0011] Among them, each fan blade defect image generation subnetwork is used to generate a specified type of fan blade defect; the specified type of fan blade defect image and the normal image of the fan blade are input into the corresponding fan blade defect image generation subnetwork for training, and the Adam optimizer is used to update the fan blade defect image generation subnetwork model parameters.

[0012] The step of using the Adam optimizer to update the parameters of the wind turbine blade defect image generation sub-network model includes:

[0013] Set the first loss function, combine the input normal image X of the wind turbine blade and the defect image Y of the wind turbine blade with one of the defect types, and train the generator G and the generator F. The formula of the first loss function is expressed as:

[0014]

[0015] Among them, Loss GAN It means that the defect image of the fan blade generated by the generator G is as close as possible to the discriminator D Y Determine it as a real fan blade defect image; make the fan blade normal image generated by generator F as close as possible to the discriminator D X It is determined to be a normal image of a real wind turbine blade; it is expressed as:

[0016]

[0017] Loss Cycle It means that the generated defective image of the wind turbine blade is input into the generator F so that it can be restored to the original image as much as possible, and the generated normal image of the wind turbine blade is input into the generator G so that it can be restored to the original image as much as possible; it is expressed as:

[0018]

[0019] Loss Identity ′ means that the images generated by generators G and F are basically consistent with the input images in terms of content, and only the conversion between normal and defective images is achieved; it is expressed as:

[0020]

[0021] Set the second loss function and train the discriminator D by inputting the real normal image X of the wind blade and the generated normal image Y' of the wind blade X , whose purpose is to distinguish between real and generated normal images of wind turbine blades; expressed as:

[0022]

[0023] Set the third loss function and train the discriminator D by inputting the real fan blade defect image Y and the generated fan blade defect image X' Y , whose purpose is to distinguish between real and generated wind turbine blade defect images; expressed as:

[0024]

[0025] Through iterative training, when the number of iterations is reached, the iterative training is stopped and the generator G, F and discriminator D are output. x 、D y .

[0026] Among them, the normal image dataset of the wind blade is input into each trained wind blade defect image generation sub-network to generate a wind blade defect image dataset of the corresponding defect type as training data for the wind blade defect image recognition network.

[0027] Among them, the fan blade defect image recognition network is a YOLO network model, which is trained by the fan blade defect images in the fan blade image dataset and the fan blade defect images generated by the fan blade defect image generation network; after the training is completed, the fan blade defect image acquired in real time is input into the trained fan blade defect image recognition network, and the output result is the defect type recognition result of the fan blade defect image acquired in real time.

[0028] In addition, the present invention also aims to propose a wind turbine blade defect image generation and defect recognition device based on CycleGAN, comprising:

[0029] A model building module is used to build a fan blade defect image generation and defect recognition network, which includes a fan blade defect image generation network and a fan blade defect image recognition network;

[0030] The model training module is used to obtain a wind blade image dataset, divide the wind blade normal images into wind blade defect images, and mark the defect types and defect locations of the wind blade defect images; train the wind blade defect image generation network using the marked wind blade image dataset; and train the wind blade defect image recognition network based on the wind blade defect images generated by the trained wind blade defect image generation network;

[0031] The defect recognition module is used to input the real-time captured fan blade defect image into the trained fan blade defect image recognition network, and the output result is the defect recognition result of the captured fan blade defect image.

[0032] The present invention also aims to provide a computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method of any one of the aforementioned technical solutions is implemented.

[0033] Another object of the present invention is to provide a non-temporary computer-readable storage medium having a computer program stored thereon, which implements the method of the aforementioned technical solution when the computer program is executed by a processor.

[0034] Different from the existing technology, the CycleGAN-based fan blade defect image generation and defect recognition method provided by the present invention generates multiple defect image generation models based on CycleGAN for different types of fan blade defects. Using the CycleGAN model, normal fan blade images can be generated into different types of defect images, solving the problem of the scarcity of fan blade defect images. Using the original defect image and a large number of different types of generated fan blade defect images, a fan blade image defect recognition model is trained based on YOLO, solving the problem of low accuracy in fan blade image defect recognition. Through the present invention, a large number of normal fan blade images can be converted into a large number of blade defect images, thereby providing sufficient defect image data to the blade image defect recognition model and improving the defect recognition effect of the recognition model. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments with reference to the accompanying drawings, in which:

[0036] Figure 1 This is a flow chart of a method for generating defect images and identifying defects in fan blades based on CycleGAN provided by the present invention.

[0037] Figure 2 It is a logical schematic diagram of a fan blade defect image generation network in a fan blade defect image generation and defect recognition method based on CycleGAN provided by the present invention.

[0038] Figure 3 This is a structural schematic diagram of a fan blade defect image generation and defect recognition device based on CycleGAN provided by the present invention.

[0039] Figure 4 It is a structural schematic diagram of a non-transitory computer-readable storage medium provided by the present invention. DETAILED DESCRIPTION

[0040] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present invention, but are not to be construed as limiting the present invention.

[0041] A wind turbine blade image stitching method based on homography and reconstruction network according to an embodiment of the present invention will be described below with reference to the accompanying drawings.

[0042] Figure 1A flow chart of a method for generating defect images and identifying defects in wind turbine blades based on CycleGAN is provided in an embodiment of the present invention. The method comprises the following steps:

[0043] Step 101 : constructing a fan blade defect image generation and defect recognition network, which includes a fan blade defect image generation network and a fan blade defect image recognition network.

[0044] The fan blade defect image generation network uses the CycleGAN model structure. The fan blade defect image generation network is designed with the same number of fan blade defect image generation subnetworks as the number of defect types. Each fan blade defect image generation subnetwork includes two generators G and F and two discriminators D. X 、D Y The input of each fan blade defect image generation subnetwork is a normal fan blade image X and a fan blade defect image Y with one defect type. The generator G generates a fan blade defect image X' with the same defect type as the fan blade defect image Y by inputting the normal fan blade image X; the generator F generates a normal fan blade image Y' by inputting the fan blade defect image Y; D X Used to distinguish the real normal image X of the wind blade and the generated normal image Y', D Y Used to truly distinguish the fan blade defect image Y from the generated fan blade defect image X'.

[0045] In the present invention, the fan blade defect image recognition network is a YOLO network model.

[0046] Step 102: Acquire a wind blade image dataset, divide the wind blade into normal images and wind blade defect images, and mark the defect types and defect locations of the wind blade defect images; train a wind blade defect image generation network using the marked wind blade image dataset; and train a wind blade defect image recognition network based on the wind blade defect images generated by the trained wind blade defect image generation network.

[0047] The wind blade image dataset is divided into a normal wind blade image dataset and N types of wind blade defect image datasets; N is the defect type, which at least includes gel coat peeling, paint peeling, cracking, surface cracks, corrosion, oil stains, sand holes and ice.

[0048] Each fan blade defect image generation subnetwork is used to generate a specified type of fan blade defect; the specified type of fan blade defect image and the normal fan blade image are input into the corresponding fan blade defect image generation subnetwork for training, and the Adam optimizer is used to update the fan blade defect image generation subnetwork model parameters.

[0049] The step of using the Adam optimizer to update the parameters of the wind turbine blade defect image generation sub-network model includes:

[0050] Set the first loss function, combine the input normal image X of the wind turbine blade and the defect image Y of the wind turbine blade with one of the defect types, and train the generator G and the generator F. The formula of the first loss function is expressed as:

[0051]

[0052] Among them, Loss GAN It means that the defect image of the fan blade generated by the generator G is as close as possible to the discriminator D Y Determine it as a real fan blade defect image; make the fan blade normal image generated by generator F as close as possible to the discriminator D X It is determined to be a normal image of a real wind turbine blade; it is expressed as:

[0053]

[0054] Loss Cycle It means that the generated defective image of the wind turbine blade is input into the generator F so that it can be restored to the original image as much as possible, and the generated normal image of the wind turbine blade is input into the generator G so that it can be restored to the original image as much as possible; it is expressed as:

[0055]

[0056] Loss Identity ′ means that the images generated by generators G and F are basically consistent with the input images in terms of content, and only the conversion between normal and defective images is achieved; it is expressed as:

[0057]

[0058] Set the second loss function and train the discriminator D by inputting the real normal image X of the wind blade and the generated normal image Y' of the wind blade X , whose purpose is to distinguish between real and generated normal images of wind turbine blades; expressed as:

[0059]

[0060] Set the third loss function and train the discriminator D by inputting the real fan blade defect image Y and the generated fan blade defect image X' Y , whose purpose is to distinguish between real and generated wind turbine blade defect images; expressed as:

[0061]

[0062] Through iterative training, when the number of iterations is reached, the iterative training is stopped and the generator G, F and discriminator D are output. x 、D y .

[0063] The normal image dataset of the fan blade is input into each trained fan blade defect image generation sub-network to generate a fan blade defect image dataset of the corresponding defect type as the training data of the fan blade defect image recognition network.

[0064] The fan blade defect image recognition network is a YOLO network model, which is trained by using fan blade defect images in a fan blade image dataset and fan blade defect images generated by a fan blade defect image generation network.

[0065] Step 103: inputting the real-time captured wind blade defect image into the trained wind blade defect image recognition network, and outputting a defect recognition result for the captured wind blade defect image.

[0066] After the training is completed, the real-time acquired fan blade defect image is input into the trained fan blade defect image recognition network, and the output result is the defect type recognition result of the real-time acquired fan blade defect image.

[0067] like Figure 3 As shown, the present invention also aims to propose a fan blade defect image generation and defect recognition device based on CycleGAN, comprising:

[0068] A model building module 310 is used to build a fan blade defect image generation and defect recognition network, which includes a fan blade defect image generation network and a fan blade defect image recognition network;

[0069] The model training module 320 is configured to obtain a wind blade image dataset, classify the wind blade normal images and the wind blade defect images, and mark the defect types and locations of the wind blade defect images; train a wind blade defect image generation network using the marked wind blade image dataset; and train a wind blade defect image recognition network based on the wind blade defect images generated by the trained wind blade defect image generation network.

[0070] The defect recognition module 330 is used to input the real-time captured wind blade defect image into the trained wind blade defect image recognition network, and the output result is the defect recognition result of the captured wind blade defect image.

[0071] The implementation process of the above device is similar to or even the same as the implementation process of the method in the aforementioned embodiment, and will not be repeated here.

[0072] In order to implement the embodiment, the present invention also proposes an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute each step in the wind blade defect image generation and defect identification method of the aforementioned technical solution.

[0073] like Figure 4 As shown, the non-transitory computer-readable storage medium includes a memory 810 of instructions and an interface 830. The instructions can be executed by a processor 820 for generating and identifying defects based on a defect image of a wind turbine blade to complete the method. Alternatively, the storage medium can be a non-transitory computer-readable storage medium, for example, a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.

[0074] In order to implement the embodiment, the present invention also proposes a non-temporary computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the wind turbine blade defect image generation and defect identification method as in the embodiment of the present invention is implemented.

[0075] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0076] 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 the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0077] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0078] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and a portable compact disc read-only memory (CDROM). Furthermore, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing it in another suitable manner if necessary, and then storing it in a computer memory.

[0079] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0080] Those skilled in the art will understand that all or part of the steps in the method of the above embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0081] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing module, or each unit may exist physically separately, or two or more units may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or in the form of software functional modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may also be stored in a computer-readable storage medium.

[0082] The storage medium mentioned above may be a read-only memory, a magnetic disk, or an optical disk, etc. Although the embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and are not to be construed as limiting the present invention. Persons skilled in the art may make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for generating defect images and identifying defects in wind turbine blades based on CycleGAN, characterized in that: include: Constructing a fan blade defect image generation and defect recognition network, wherein the fan blade defect image generation and defect recognition network includes a fan blade defect image generation network and a fan blade defect image recognition network; Acquire a wind blade image dataset, divide the wind blade into normal images and wind blade defect images, and mark the defect types and defect locations of the wind blade defect images; Training the fan blade defect image generation network using the labeled fan blade image dataset; Training the fan blade defect image recognition network based on the fan blade defect image generated by the trained fan blade defect image generation network; Inputting the real-time captured fan blade defect image into the trained fan blade defect image recognition network, and outputting a defect recognition result of the captured fan blade defect image; The fan blade defect image generation network uses the CycleGAN model structure. The fan blade defect image generation network is designed with the same number of fan blade defect image generation subnetworks as the defect types. Each fan blade defect image generation subnetwork includes two generators G and F and two discriminators. 、 ; The input of each fan blade defect image generation subnetwork is a normal fan blade image X and a fan blade defect image Y with one defect type. The generator G generates a fan blade defect image X' with the same defect type as the fan blade defect image Y by inputting the normal fan blade image X; the generator F generates a normal fan blade image Y' by inputting the fan blade defect image Y; D X Used to distinguish the real normal image X of the wind blade and the generated normal image Y', D Y Used to truly distinguish the fan blade defect image Y from the generated fan blade defect image X'.

2. The method for generating defect images and identifying defective blades based on CycleGAN according to claim 1 is characterized in that: The wind blade image dataset is divided into a normal wind blade image dataset and N types of wind blade defect image datasets; N is the defect type, including at least gel coat peeling, paint peeling, cracking, surface cracks, corrosion, oil stains, sand holes and ice.

3. The method for generating defect images and identifying defective blades based on CycleGAN according to claim 1, characterized in that: Each wind blade defect image generation subnetwork is used to generate a specified type of wind blade defect; The specified type of fan blade defect images and normal fan blade images are input into the corresponding fan blade defect image generation subnetwork for training, and the Adam optimizer is used to update the fan blade defect image generation subnetwork model parameters.

4. The method for generating defect images and identifying defective blades based on CycleGAN according to claim 3 is characterized in that: The steps of using the Adam optimizer to update the parameters of the wind turbine blade defect image generation sub-network model include: Set the first loss function, combine the input normal image X of the wind turbine blade and the defect image Y of the wind turbine blade with one of the defect types, and train the generator G and the generator F. The formula of the first loss function is expressed as: （1） in, It means that the defect image of the fan blade generated by the generator G is as close as possible to the discriminator D Y Determine it as a real fan blade defect image; make the fan blade normal image generated by generator F as close as possible to the discriminator D X It is determined to be a normal image of a real wind turbine blade; it is expressed as: It means that the generated defective image of the wind turbine blade is input into the generator F so that it can be restored to the original image as much as possible, and the generated normal image of the wind turbine blade is input into the generator G so that it can be restored to the original image as much as possible; it is expressed as: It means that the images generated by generators G and F are basically consistent with the input images in terms of content, and only the conversion between normal and defective images is achieved; it is expressed as: Set the second loss function and train the discriminator D by inputting the real normal image X of the wind blade and the generated normal image Y' of the wind blade X , whose purpose is to distinguish between real and generated normal images of wind turbine blades; expressed as: （2） Set the third loss function and train the discriminator D by inputting the real fan blade defect image Y and the generated fan blade defect image X' Y , whose purpose is to distinguish between real and generated wind turbine blade defect images; expressed as: （3） Through iterative training, when the number of iterations is reached, the iterative training is stopped and the generator G, F and discriminator D are output. x 、D y .

5. The method for generating defect images and identifying defective blades based on CycleGAN according to claim 4 is characterized in that: The normal image dataset of the wind blade is input into each trained wind blade defect image generation sub-network to generate a wind blade defect image dataset of the corresponding defect type as training data for the wind blade defect image recognition network.

6. The method for generating defect images and identifying defective blades based on CycleGAN according to claim 5 is characterized in that: The fan blade defect image recognition network is a YOLO network model, which is trained by fan blade defect images in a fan blade image dataset and fan blade defect images generated by a fan blade defect image generation network; After the training is completed, the fan blade defect image acquired in real time is input into the trained fan blade defect image recognition network, and the output result is the defect type recognition result of the fan blade defect image acquired in real time.

7. A fan blade defect image generation and defect recognition device based on CycleGAN, characterized in that: include: A model building module is used to build a fan blade defect image generation and defect recognition network, wherein the fan blade defect image generation and defect recognition network includes a fan blade defect image generation network and a fan blade defect image recognition network; A model training module is used to obtain a wind blade image dataset, divide the wind blade normal images into wind blade defect images, and mark the defect type and defect location of the wind blade defect images; Training the fan blade defect image generation network using the labeled fan blade image dataset; Training the fan blade defect image recognition network based on the fan blade defect image generated by the trained fan blade defect image generation network; A defect recognition module is used to input the real-time captured fan blade defect image into the trained fan blade defect image recognition network, and output a defect recognition result of the captured fan blade defect image; The fan blade defect image generation network uses the CycleGAN model structure. The fan blade defect image generation network is designed with the same number of fan blade defect image generation subnetworks as the defect types. Each fan blade defect image generation subnetwork includes two generators G and F and two discriminators. 、 ; The input of each fan blade defect image generation subnetwork is a normal fan blade image X and a fan blade defect image Y with one defect type. The generator G generates a fan blade defect image X' with the same defect type as the fan blade defect image Y by inputting the normal fan blade image X; the generator F generates a normal fan blade image Y' by inputting the fan blade defect image Y; D X Used to distinguish the real normal image X of the wind blade and the generated normal image Y', D Y Used to truly distinguish the fan blade defect image Y from the generated fan blade defect image X'.

8. A computer device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images