Fan tower corrosion defect identification method and device based on image processing

Abstract

The invention provides a fan tower corrosion defect identification method and device based on image processing, and relates to the technical field of fan tower defect identification, and the method comprises the following steps: a shadow processing step: converting an input fan tower image with a shadow into a non-shadow image by using a GAN (Generative Adversarial Network); performing edge feature extraction on the shadow-free image output in the shadow processing step to obtain an edge feature map; inputting the edge feature map into a convolutional neural network (CNN) for feature extraction and classification, and outputting an identification result of the corrosion defect of the fan tower; the generative adversarial network (GAN) comprises a generator, which is used for executing shadow detection and shadow removal and comprises a shadow detection module used for outputting a shadow probability confidence map and an encoder based on negative residual learning; and the discriminator is used for evaluating the authenticity of the shadow-free image output by the generator by adopting a spectrum normalization method. According to the invention, the identification accuracy of the corrosion defect of the fan tower can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind turbine tower defect recognition, in particular to a wind turbine tower corrosion defect recognition method and device based on image processing. BACKGROUND

[0002] Since the wind turbine tower is approximately cylindrical, the wall-climbing robot will inevitably face problems such as the sun side, the shade side, self-shielding, and wind turbine blade shielding during the wind turbine tower operation process. The shooting image cannot be perpendicular to the direct angle of light at all times, and the image with poor light conditions and the fault being shielded by the shadow is often collected. The conventional visual algorithm cannot distinguish the shadow edge and the fault edge, resulting in a decrease in recognition accuracy.

[0003] Under the complex and harsh light conditions in which the wind turbine tower is located, the conventional visual detection algorithm (such as image segmentation depending on a fixed threshold, simple edge detection, or template matching) faces great difficulties: it is difficult to effectively distinguish which edges in the image are caused by the true physical outline of an object (such as a crack or a protrusion) and which edges are only caused by the change in light (such as the boundary of a shadow). The boundary line of the shadow may be visually very similar to the true fault edge (such as a crack). Uneven light and shadows will significantly change the key visual features such as texture, color, and contrast of the target area, making it difficult for the algorithm to stably and accurately extract the inherent features of the fault. SUMMARY

[0004] The purpose of the embodiments of the present application is to provide a wind turbine tower corrosion defect recognition method and device based on image processing, which can improve the accuracy of wind turbine tower corrosion defect recognition and solve the above-mentioned problems existing in the prior art.

[0005] In a first aspect, a wind turbine tower corrosion defect recognition method based on image processing is provided, which can include: A shadow processing step: using a generative adversarial network (GAN) to convert the input wind turbine tower image with shadows into an image without shadows; Edge feature extraction is performed on the shadow-free image output by the shadow processing step to obtain an edge feature map; The edge feature map is input into a convolutional neural network (CNN) for feature extraction and classification, and the recognition result of the corrosion defect of the wind turbine tower is output; The generative adversarial network (GAN) includes: A generator for performing shadow detection and shadow removal, including a shadow detection module for outputting a shadow probability confidence map and an encoder based on negative residual learning, which generates a negative residual by fusing the original image and the probability confidence map, and applies the negative residual to the original image to output a shadow-free image; A discriminator, which adopts a spectral normalization method, is configured to evaluate the authenticity of the shadow-free image output by the generator.

[0006] In an optional implementation, the shadow detection module includes 10 convolutional layers, 1 LSTM layer, and 1 convolutional layer outputting the probability confidence map, and batch normalization and LeakyReLU activation functions are applied for feature extraction.

[0007] In an optional implementation, the encoder uses convolutional layers, batch normalization layers, and LeakyReLU activation functions to extract features from the image fused with shadow information, and uses a batch-normalized deconvolutional layer and a ReLU activation function to generate the negative residual error.

[0008] In an optional implementation, the network structure of the discriminator includes five convolutional layers, batch normalization layers, ReLU activation functions, and a fully connected layer, and the fully connected layer outputs a probability value of the input image being a real image.

[0009] In an optional implementation, the edge feature map is obtained by performing edge feature extraction on the shadow-free image output by the shadow processing step, including: An edge enhancement operator is independently applied to the red, green, and blue color component channels of the shadow-free image, and the gradient amplitude of each pixel point is calculated through convolution operation to obtain a gradient amplitude map. The gradient amplitude map is subjected to non-maximum suppression and double-threshold processing to output a binary edge feature map.

[0010] In an optional implementation, the edge feature map is subjected to normalization preprocessing before being input into the convolutional neural network CNN.

[0011] In an optional implementation, the convolutional neural network CNN includes cross-layer connections, specifically, the output feature map of the first convolutional layer Conv1 is directly connected to the input of the third convolutional layer Conv3, and / or the output feature map of the third convolutional layer Conv3 is directly connected to the input of the fifth convolutional layer Conv5.

[0012] In a second aspect, a fan tower corrosion defect recognition device based on image processing is provided, which can include: A shadow processing module configured to convert an input shadowed image into a shadow-free image by using a generative adversarial network GAN. An edge feature extraction module configured to perform edge feature extraction on the shadow-free image to obtain an edge feature map. A recognition module configured to input the edge feature map into a convolutional neural network CNN for feature extraction and classification, and output a recognition result.

[0013] In a third aspect, an electronic device is provided, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus. The memory is configured to store a computer program. The processor is configured to execute the program stored in the memory to implement the method steps of any one of the first aspect.

[0014] In a fourth aspect, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to implement the method steps of any one of the first aspect.

[0015] The present application is aimed at images with poor lighting conditions and faults covered by shadows. By introducing a shadow detection and elimination network in the edge recognition algorithm, the influence of shadow areas on recognition is weakened, and a residual neural network is used for corrosion defect recognition, thereby improving the accuracy of fan tower corrosion defect recognition. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments of the present application. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0017] Figure 1 An architecture diagram of a fan tower corrosion defect recognition system based on image processing provided by the embodiments of the present application; Figure 2 A flowchart of a fan tower corrosion defect recognition method based on image processing provided by the embodiments of the present application; Figure 3 A structure diagram of a convolutional neural network (CNN) provided by the embodiments of the present application; Figure 4 A structure diagram of a fan tower corrosion defect recognition device based on image processing provided by the embodiments of the present application; Figure 5 A structure diagram of an electronic device provided by the embodiments of the present application. DETAILED DESCRIPTION

[0018] 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 only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. Unless otherwise defined, the technical or scientific terms used in this application should have the ordinary meaning understood by those skilled in the art. The words "first," "second," and similar terms used in this application do not indicate any order, quantity, or importance, but are only used to distinguish different components. The words "comprising" or "including," etc., mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, but do not exclude other elements or objects. The words "connected," "coupled," or "connected," etc., are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up," "down," "left," "right," etc., are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0019] The image processing-based wind turbine tower corrosion defect identification method provided in this application embodiment can be applied to... Figure 1 In the system architecture shown, such as Figure 1 As shown, the system may include: a server in the enterprise's backend and terminals for enterprise employees. The server can be a physical server, a server cluster consisting of multiple physical servers, or a distributed system. It can also be a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. The terminal may be a user equipment (UE) such as a mobile phone, smartphone, laptop, digital radio receiver, personal digital assistant (PDA), tablet computer (PAD), handheld device, in-vehicle device, wearable device, computing device, or other processing device connected to a wireless modem, mobile station (MS), or mobile terminal. The terminal and server can be directly or indirectly connected via wired or wireless communication methods; this application does not limit the connection.

[0020] The preferred embodiments of the present application are described below in conjunction with the accompanying drawings of the specification, it should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application, and the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0021] Figure 2 A flowchart of a fan tower corrosion defect recognition method based on image processing provided by an embodiment of the present application is shown in FIG. 1. As shown in FIG. 1, the method can include the following steps. Figure 2 Step S210, shadow processing step: using a generative adversarial network (GAN) to convert the input shadowed image of the fan tower into a non-shadowed image.

[0022] The generative adversarial network (GAN) includes: a generator for performing shadow detection and shadow removal; wherein the generator includes a shadow detection module for outputting a shadow probability confidence map and an encoder based on negative residual learning; The shadow detection module includes 10 convolutional layers (64 channels), an LSTM layer, and a convolutional layer for outputting a probability confidence map, and applies batch normalization and LeakyReLU activation functions for feature extraction; the value range of the shadow probability confidence map is 0-1; the value of each pixel point in the shadow probability confidence map quantifies the possibility of belonging to the shadow area, achieving accurate positioning of the shadow area in the shadowed image of the fan tower; The encoder is designed using the negative residual learning idea, generates a negative residual by fusing the original image and the probability confidence map, and applies the negative residual to the original image to output a non-shadowed image; the encoder uses convolutional layers, batch normalization layers, and LeakyReLU activation functions to extract features from the image fused with shadow information, and uses batch normalization deconvolution layers and ReLU activation functions to generate negative residuals. The encoder is also used to remove the shadow information in the shadowed image of the fan tower based on the negative residual, effectively suppressing or eliminating the shadow in the shadowed image, and obtaining a non-shadowed image. The encoder uses Conv+BN+LReLU to extract features from the image, and then uses batch processing deconvolution layers and ReLU activation functions to generate an image with specific distribution characteristics, i.e. a non-shadowed image; a discriminator including five convolutional layers (Conv+BN+Relu function), a batch normalization layer, a ReLU activation function, and a fully connected layer; a spectral normalization method is used to evaluate the authenticity of the non-shadowed image output by the generator. The fully connected layer in the discriminator outputs the probability value of the non-shadowed image being a real image, and distinguishes between pseudo images and real images by calculating the cross-entropy loss of pseudo images and real images.

[0023] ​In an embodiment of the present application, a training method of a generative adversarial network (GAN) comprises: Obtaining a plurality of original images of fan towers and a plurality of corrosion images of fan towers; wherein the original images of fan towers are uncorroded images; Based on an atmospheric scattering model, simulating fog effects (adjusting image contrast and saturation), generating rain lines (simulating raindrop falling trajectories) based on Poisson distribution, and simulating dust adhesion by random texture superposition, random injection of rain, fog, dust and other environmental disturbances into each original image of a fan tower to obtain a plurality of first images; Performing style transfer on each corrosion image of a fan tower by StyleGAN to generate different morphological defects such as pitting, uniform corrosion, and crevice corrosion, and combining random rotation (0-360°), scaling (0.8-1.2 times), and shear (5%-10% area) to obtain a plurality of second images; Based on the plurality of original images of fan towers, the plurality of corrosion images of fan towers, the plurality of first images, and the plurality of second images, a training data set is constructed; Using the training data set to train the generative adversarial network (GAN) to obtain a trained generative adversarial network (GAN).

[0024] In another embodiment of the present application, the generative adversarial network (GAN) uses a multi-task loss function = α × GAN adversarial loss (maintaining the authenticity of the unshadowed image) + β × generator pixel-level MSE loss (ensuring the accuracy of shadow removal) + γ × subsequent CNN classification cross-entropy loss (adapting to the defect recognition requirements), wherein α, β, and γ are adaptive weights (satisfying α + β + γ = 1 and being dynamically adjusted through training), the CNN classification loss is back-propagated to the generator to realize collaborative optimization of shadow removal and defect recognition.

[0025] Step S220, edge feature extraction is performed on the unshadowed image output by the shadow processing step to obtain an edge feature map.

[0026] In a specific implementation, edge feature extraction is performed on the unshadowed image output by the shadow processing step to obtain an edge feature map, comprising: An edge enhancement operator is independently applied to the red, green, and blue color component channels of the unshadowed image, and the gradient amplitude of each pixel point in the unshadowed image is calculated by convolution operation to obtain a gradient amplitude; wherein the gradient amplitude is used to represent the change in intensity (gray value) of each pixel point and its neighborhood pixel points, i.e., the strength of the edge intensity at each pixel point, the more dramatic the change, the larger the gradient amplitude. The edge enhancement operator quantifies the intensity gradient of the local region through convolution operation; The gradient amplitude graph is subjected to non-maximum suppression and double threshold processing, and a binary edge feature graph is output; wherein, the Canny edge detection algorithm is used to detect the edge of the gradient amplitude graph, the edge width of the gradient amplitude graph is thinned to a single pixel through non-maximum suppression, and the weak edge is connected and the noise false edge is suppressed by using the double threshold, and finally a clear and continuous binary edge feature graph is output.

[0027] In an embodiment of the present application, the edge feature graph is subjected to normalization preprocessing before being input into the CNN; wherein, the normalization preprocessing includes scaling to [0, 1] or standardization, so as to accelerate the network convergence and improve the training stability.

[0028] In another embodiment of the present application, the method further comprises: The shadow probability confidence graph is converted into an attention weight graph through a 1x1 convolution layer; the binary edge feature graph is weighted by using the attention weight graph, and a weighted edge feature graph is obtained; the weighted edge feature graph is input into the convolutional neural network CNN for feature extraction and classification, and the recognition result of the corrosion defect of the fan tower drum is output; wherein, the weight of the area with high shadow probability, i.e. the possible area of shadow residue, can be 0.3-0.5, and the weight of the area with low shadow probability is set to 0.8-1.0, so as to reduce the interference of the shadow residue on the edge feature.

[0029] Step S230, input the edge feature graph into the convolutional neural network CNN for feature extraction and classification, and output the recognition result of the corrosion defect of the fan tower drum.

[0030] Wherein, as Figure 3As shown, the convolutional neural network comprises: a first convolutional layer Conv1, a second convolutional layer Conv2, a third convolutional layer Conv3, a fourth convolutional layer Conv4, and a fifth convolutional layer Conv5; the first convolutional layer Conv1, the second convolutional layer Conv2, the third convolutional layer Conv3, the fourth convolutional layer Conv4, and the fifth convolutional layer Conv5 perform sliding convolution operations on the edge feature map using learnable convolution kernels (Filters) respectively, automatically extracting spatial features (such as edge direction, shape contour, local structure, etc.) of different levels and degrees of abstraction in the edge feature map; the first convolutional layer Conv1, the second convolutional layer Conv2, the third convolutional layer Conv3, the fourth convolutional layer Conv4, and the fifth convolutional layer Conv5 are all applied with ReLU activation functions, introducing nonlinear transformations, enhancing the expression ability of the model, and helping to alleviate the gradient vanishing problem; a maximum pooling layer is arranged between the first convolutional layer Conv1, the second convolutional layer Conv2, the third convolutional layer Conv3, the fourth convolutional layer Conv4, and the fifth convolutional layer Conv5. The maximum pooling layer selects the maximum value in a local region (such as a 2x2 window) as the output of the region. After multiple convolution and pooling operations, a three-dimensional feature tensor with low spatial resolution but high channel dimension is obtained. The maximum pooling layer is used to flatten the three-dimensional feature tensor into a one-dimensional feature vector; the convolutional neural network further comprises a first fully connected layer and a second fully connected layer. The first fully connected layer is used to receive the one-dimensional feature vector output by the maximum pooling layer, compress the high-dimensional features into lower-dimensional discriminative features through linear transformation, introduce nonlinearity with the ReLU activation function, filter out features more critical to corrosion defect recognition and quantification, and obtain a target feature vector; the second fully connected layer is used to map the target feature vector to a classification space of whether or not to corrode through linear transformation, and then convert the output into a probability value through the Softmax activation function to obtain the recognition result of the corrosion defect of the fan tower drum.

[0031] In an embodiment of the present application, the convolutional neural network CNN adopts cross-layer connection; specifically: the output feature map of the first convolutional layer Conv1 is directly connected to the input of the third convolutional layer Conv3, and / or the output feature map of the third convolutional layer Conv3 is directly connected to the input of the fifth convolutional layer Conv5, to allow the gradient to be returned more directly, helping to learn more complex feature representations.

[0032] In another embodiment of the present application, a Grad-CAM module is further connected after the fifth convolutional layer Conv5, used to calculate the gradient weight of the classification result on the output feature map of the fifth convolutional layer Conv5, and generate a corrosion region heat map; the corrosion region heat map is threshold segmented by an Otsu algorithm to obtain a binary defect region mask; a Canny edge detection and contour approximation algorithm is used to extract the contour of the defect region mask, and an external rectangular frame of the defect region is output to determine the defect position.

[0033] In another embodiment of the present application, the convolutional neural network further comprises: a third fully connected layer and a fourth fully connected layer; the third fully connected layer is used to map the target feature vector to a regression space of a corrosion area ratio (1-dimensional output) through linear transformation, and then limit the output to the interval of 0-1 through a Sigmoid activation function, directly outputting the proportion of the number of corrosion region pixels to the total number of image pixels, to realize the quantification of the corrosion area; the fourth fully connected layer is used to map the target feature vector to a regression space of a corrosion depth level (1-dimensional output) through linear transformation, and then limit the output range through a ReLU6 activation function, and cooperate with an integer operation to obtain an integer depth level of 1-5 (1 level of mildness, 5 level of severity), to realize the level quantification of the corrosion severity.

[0034] Corresponding to the above method, the present application also provides an image processing-based fan tower corrosion defect identification device, as shown in Figure 4 The device comprises: A shadow processing module 410 configured to convert an input shadow image into a shadow-free image using a generative adversarial network (GAN); An edge feature extraction module 420 configured to perform edge feature extraction on the shadow-free image to obtain an edge feature map; An identification module 430 configured to input the edge feature map into a convolutional neural network (CNN) for feature extraction and classification, and output an identification result.

[0035] The functions of each functional unit of the image processing-based fan tower corrosion defect identification device provided in the above embodiments of the present application can be realized through the above method steps, therefore, the specific working process and beneficial effects of each unit in the image processing-based fan tower corrosion defect identification device provided in the embodiments of the present application will not be repeated here.

[0036] The present application also provides an electronic device, as shown in Figure 5 The device comprises a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540.

[0037] a memory 530 for storing a computer program; a processor 510 for executing the program stored in the memory 530 to implement the following steps: a shadow processing step of converting an inputted image of a fan tower with shadows into an image without shadows by using a generative adversarial network (GAN); an edge feature extraction step of extracting an edge feature map from the image without shadows outputted by the shadow processing step; a feature extraction and classification step of inputting the edge feature map into a convolutional neural network (CNN) to output a recognition result of the corrosion defect of the fan tower; The GAN includes: a generator for performing shadow detection and shadow removal, including a shadow detection module for outputting a shadow probability confidence map and an encoder based on negative residual learning, which generates a negative residual by fusing the original image and the probability confidence map, and applies the negative residual to the original image to output the image without shadows; a discriminator for evaluating the authenticity of the image without shadows outputted by the generator by using a spectral normalization method.

[0038] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or only one type of bus.

[0039] The communication interface is used for communication between the electronic device and other devices.

[0040] The memory can include a Random Access Memory (RAM) and can also include a Non-Volatile Memory (NVM), such as at least one disk memory. Optionally, the memory can also be at least one storage device located away from the aforementioned processor.

[0041] The processor described above can be a general processor, including a central processing unit (CPU), a network processor (NP), etc.; can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0042] The implementation manners and beneficial effects of the electronic device in the above embodiments can be achieved by referring to the implementation manners of each step in the above embodiments, and thus, the specific working process and beneficial effects of the electronic device provided by the embodiments of the present application are not repeated here. Figure 2 The implementation manners and beneficial effects of the electronic device in the above embodiments can be achieved by referring to the implementation manners of each step in the above embodiments, and thus, the specific working process and beneficial effects of the electronic device provided by the embodiments of the present application are not repeated here.

[0043] In another embodiment provided by the present application, a computer readable storage medium is provided, and the computer readable storage medium stores instructions, when the instructions are run on a computer, the computer executes the image processing based fan tower corrosion defect identification method in any of the above embodiments.

[0044] In another embodiment provided by the present application, a computer program product containing instructions is provided, when the instructions are run on a computer, the computer executes the image processing based fan tower corrosion defect identification method in any of the above embodiments.

[0045] Those skilled in the art should understand that the embodiments in the embodiments of the present application can be provided as a method, a system or a computer program product. Therefore, the embodiments in the present application can be in the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the embodiments in the present application can be in 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 codes.

[0046] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 The flowchart and / or block diagram in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to the present application. In this regard, each block in the flowchart and / or block diagrams can represent a module, segment, or portion of code, which comprises one or more executable Figure 1 instructions for implementing the specified logical functions. It should also be noted that each block of the flowchart and / or block diagrams and that a flowchart and / or block diagram can represent a combination of

[0047] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the Figure 1 functionality specified in the flowchart block or blocks. The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1

[0048] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 functionality specified in the flowchart block or blocks. The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1

[0049] While the preferred embodiments in the present application have been described, additional variations and modifications can be applied to those embodiments without departing from the spirit and scope of the present application. Therefore, the appended claims are intended to encompass all such variations and modifications as falling within the scope of the present application. It is to be understood that the above-referenced patents, patent applications, and publications are incorporated by reference herein.

[0050] It will be apparent to those skilled in the art that various modifications and variations can be made to the embodiments of the present application without departing from the spirit or scope of the application. Thus, it is intended that the present application cover the modifications and variations of this application provided they come within the scope of the appended claims and their equivalents.​​

Claims

1. An image processing-based method for identifying corrosion defects of a fan tower, characterized in that, Includes the following steps: Shadow processing steps: Use a generative adversarial network (GAN) to convert the input image of the wind turbine tower with shadows into an image without shadows; Edge features are extracted from the shadowless image output by the shadow processing step to obtain an edge feature map; The edge feature map is input into a convolutional neural network (CNN) for feature extraction and classification, and the identification results of corrosion defects in the wind turbine tower are output. The Generative Adversarial Network (GAN) includes: A generator for performing shadow detection and shadow removal includes a shadow detection module for outputting a shadow probability confidence map and an encoder based on negative residual learning, which generates a negative residual by fusing the original image with the probability confidence map and applies the negative residual to the original image to output a shadow-free image. The discriminator, employing a spectral normalization method, is used to evaluate the realism of the shadowless image output by the generator.

2. The method according to claim 1, characterized in that, The shadow detection module includes 10 convolutional layers, 1 LSTM layer, and 1 convolutional layer that outputs the probability confidence map, and applies batch normalization and LeakyReLU activation function for feature extraction.

3. The method according to claim 1, characterized in that, The encoder extracts features from an image that incorporates shadow information using convolutional layers, batch normalization layers, and the Leaky ReLU activation function, and generates the negative residual using batch normalized deconvolutional layers and the ReLU activation function.

4. The method according to claim 1, characterized in that, The discriminator's network structure includes five convolutional layers, a batch normalization layer, a ReLU activation function, and a fully connected layer. The fully connected layer outputs the probability value that the input image is a real image.

5. The method according to claim 1, characterized in that, The step of extracting edge features from the shadowless image output by the shadow processing step to obtain an edge feature map includes: Edge enhancement operators are applied independently to the red, green, and blue color component channels of the image without shadows. The gradient magnitude of each pixel is calculated through convolution operations to obtain a gradient magnitude map. The gradient magnitude map is subjected to non-maximum suppression and double thresholding to output a binarized edge feature map.

6. The method according to claim 1, characterized in that, Before inputting the edge feature map into the convolutional neural network (CNN), the edge feature map is first subjected to normalization preprocessing.

7. The method according to claim 1, characterized in that, The convolutional neural network (CNN) includes cross-layer connections, specifically: the output feature map of the first convolutional layer Conv1 is directly connected to the input of the third convolutional layer Conv3, and / or the output feature map of the third convolutional layer Conv3 is directly connected to the input of the fifth convolutional layer Conv5.

8. A wind turbine tower corrosion defect identification device based on image processing, characterized in that, include: The shadow processing module is configured to use a generative adversarial network (GAN) to convert the input shadowed image into a shadowless image; The edge feature extraction module is configured to extract edge features from the shadowless image to obtain an edge feature map. The recognition module is configured to input the edge feature map into a convolutional neural network (CNN) for feature extraction and classification, and output the recognition result.

9. 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 computer program, it implements the method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-7.

Images