Fault detection method for wind turbine blade based on terahertz imaging technology

Abstract

The application discloses a wind turbine blade fault detection method based on a terahertz imaging technology, belongs to the technical field of nondestructive testing and wind power equipment monitoring, and performs terahertz scanning imaging on a wind turbine blade, acquires and pre-processes a terahertz image, and obtains a blade defect image; the blade defect image is sequentially subjected to compressed sensing sampling, linear mapping reconstruction and multi-branch optimization reconstruction through an image reconstruction model, and a reconstructed image is generated; the reconstructed image is subjected to multi-level feature extraction of fusion convolution and a Transformer through a fault detection model, and a detection result is output through weighted classification of a multi-classifier, so that the performance of terahertz imaging in wind turbine blade defect detection is improved.

Description

Technical Field

[0001] This invention belongs to the field of non-destructive testing and wind power equipment monitoring technology, specifically relating to a method for detecting wind turbine blade faults based on terahertz imaging technology. Background Technology

[0002] As a core piece of equipment in the renewable energy sector, the long-term stable operation of wind turbines is crucial for ensuring energy supply. The blades, as key components of wind turbines, are constantly exposed to a complex and ever-changing natural environment, making them susceptible to wind and rain erosion, fatigue loads, lightning strikes, and other factors. This can lead to defects such as internal or surface cracks, delamination, debonding, and corrosion. If these defects are not detected and addressed promptly, they will severely impact the structural integrity of the blades, and may even cause serious accidents such as blade breakage, resulting in significant economic losses and safety risks.

[0003] Currently, fault detection in wind turbine blades mainly relies on technologies such as manual inspection, UAV visual inspection, ultrasonic testing, and infrared thermal imaging. Manual inspection is inefficient, highly subjective, and struggles to detect hidden internal defects; while UAV visual inspection can cover a wide area, it is primarily limited to surface defect identification and has limited capability for detecting internal damage; ultrasonic testing requires coupling agents and has poor adaptability to complex curved surfaces; and infrared thermal imaging is greatly affected by ambient temperature, with limited detection depth and resolution. Therefore, developing an efficient, accurate, non-contact, and non-destructive testing technology suitable for detecting internal defects in wind turbine blades has become an urgent need in the field of wind power equipment health monitoring.

[0004] Terahertz waves (THz), located between microwaves and infrared, possess unique advantages such as strong penetration, sensitivity to non-metallic materials, low photon energy, and high safety, making them ideal for detecting internal defects in composite materials (such as fiberglass and carbon fiber commonly used in wind turbine blades). Terahertz imaging technology can visualize the internal structure of blades and identify defects such as delamination, porosity, inclusions, and cracks. However, existing terahertz imaging methods still have shortcomings in image resolution, signal-to-noise ratio, and reconstruction efficiency. Especially in blade inspection under complex operating conditions, image quality is easily affected by noise and scattering, leading to a decrease in defect identification accuracy. Therefore, how to combine advanced image reconstruction and deep learning technologies to improve the performance of terahertz imaging in wind turbine blade defect detection is a pressing technical problem to be solved in this field. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing terahertz imaging methods in terms of image resolution, signal-to-noise ratio, and reconstruction efficiency, and to propose a wind turbine blade fault detection method based on terahertz imaging technology.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for detecting faults in wind turbine blades based on terahertz imaging technology, comprising the following steps: Terahertz scanning imaging was performed on the wind turbine blades to acquire and preprocess the terahertz images, thereby obtaining images of blade defects. The image reconstruction model is used to sequentially perform compressed sensing sampling, linear mapping reconstruction, and multi-branch optimization reconstruction on the blade defect image to generate a reconstructed image. The fault detection model performs multi-level feature extraction by fusing convolution and Transformer on the reconstructed image, and outputs the detection results by weighting and classifying them through a multi-classifier.

[0007] Furthermore, the terahertz image is acquired and preprocessed, including: Obtain time-series terahertz images; Feature point matching and affine transformation registration of time series terahertz images; Background subtraction and noise filtering are performed on the registered image to obtain the blade defect image.

[0008] Furthermore, compressed sensing sampling includes: The blade defect image is divided into non-overlapping image blocks; Vectorize each image block; Compressed sensing data is generated by sampling vectorized image patches using a convolutional sampling matrix.

[0009] Furthermore, linear mapping reconstruction includes: Spatial integration of compressed sensing data is achieved by using a convolutional sampling transpose matrix; Perform a pixel shuffle operation on the spatially integrated data to generate a mapped reconstructed image.

[0010] Furthermore, multi-branch optimization and refactoring includes: A two-branch multi-scale feature extraction and fusion process is performed on the mapped and reconstructed image to obtain the fused extracted features; The fused extracted features are subjected to joint denoising processing to obtain joint denoised features; The joint denoising features are subjected to bi-branch optimization to obtain the first optimized feature and the second optimized feature. The first and second optimized features are fused by optimizing the mask to generate a reconstructed image.

[0011] Furthermore, multi-level feature extraction combining convolution and Transformer is implemented, including: The feature image extracted by convolution is divided into non-overlapping patch blocks; Linear projection is performed on each patch block to generate a patch vector; Multi-level feature extraction and fusion of patch vectors through convolution and Transformer fusion are performed to generate a combined feature image.

[0012] Furthermore, the detection results, after being weighted and classified by a multi-classifier, include: The features of the combined feature image are classified using multiple classifiers. The outputs of each classifier are weighted and fused. Output the five-level health status test results for wind turbine blades.

[0013] Secondly, the present invention provides a wind turbine blade fault detection system based on terahertz imaging technology, comprising: The defect image acquisition module is used to perform terahertz scanning imaging on wind turbine blades, acquire and preprocess terahertz images to obtain blade defect images. The reconstructed image generation module is used to sequentially perform compressed sensing sampling, linear mapping reconstruction, and multi-branch optimization reconstruction on the blade defect image through the image reconstruction model to generate a reconstructed image. The output detection result module is used to perform multi-level feature extraction of the reconstructed image by fusing convolution and Transformer through the fault detection model, and output the detection result after weighting and classification by a multi-classifier.

[0014] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a method for detecting wind turbine blade faults based on terahertz imaging technology.

[0015] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements a method for detecting wind turbine blade faults based on terahertz imaging technology.

[0016] Compared with the prior art, the present invention has the following beneficial technical effects: This invention proposes a wind turbine blade fault detection method based on terahertz imaging technology. By constructing an image reconstruction model that includes defect image sampling, linear mapping reconstruction, and optimized reconstruction, and combining compressed sensing and deep convolutional networks, the method effectively improves the image quality and resolution of terahertz imaging, clearly revealing minute internal defects of the blade and significantly improving the accuracy and reliability of defect identification. The method employs advanced network structures such as hybrid attention mechanisms, residual connections, and self-attention mechanisms, enabling the model to adaptively extract defect features, suppress noise interference, and enhance the ability to identify complex defect patterns. Optimized masking and multi-branch fusion mechanisms further improve the detail preservation and overall consistency of the reconstructed image. The fault detection model integrates convolutional networks and Transformer structures to achieve multi-level feature extraction and fusion from local features to global semantics. Combined with a multi-classifier weighting mechanism, it achieves five-level fine classification of blade health status, providing a scientific basis for maintenance decisions. Terahertz imaging requires no coupling agent, and non-contact scanning adapts to various blade surface structures. The detection process does not damage the blade, and the operation is safe and convenient, suitable for rapid on-site detection and long-term health monitoring. The method proposed in this invention does not depend on specific blade materials or defect types. Through data-driven learning, it has good generalization ability and can be extended to the field of non-destructive testing of other composite material structures, with broad application prospects. Attached Figure Description

[0017] The accompanying drawings described herein are for illustrative purposes only and are not intended to limit the scope of the invention in any way. Furthermore, the shapes and proportions of the components in the drawings are merely schematic to aid in understanding the invention and do not specifically limit the shapes and proportions of the components. In the drawings: Figure 1 This is a flowchart of a wind turbine blade fault detection method based on terahertz imaging technology.

[0018] Figure 2 This is a simplified structural diagram of a wind turbine blade fault detection system based on terahertz imaging technology.

[0019] Figure 3 This is an electronic device diagram for a wind turbine blade fault detection method based on terahertz imaging technology.

[0020] Figure 4 This is a schematic diagram of the wind turbine blade fault detection method based on terahertz imaging technology in an embodiment of the present invention. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] Example 1 See Figure 1 A method for detecting wind turbine blade faults based on terahertz imaging technology includes the following steps: Terahertz scanning imaging is performed on wind turbine blades to acquire and preprocess terahertz images, resulting in blade defect images. The blade defect images are then subjected to compressed sensing sampling, linear mapping reconstruction, and multi-branch optimization reconstruction sequentially through an image reconstruction model to generate reconstructed images. The reconstructed images are then subjected to multi-level feature extraction using fusion convolution and Transformer through a fault detection model, and the detection results are output after weighted classification by a multi-classifier.

[0023] Specifically, the method includes the following steps: S1, scanning and imaging the wind turbine blade using a terahertz time-domain system to obtain a blade defect image; S2, constructing an image reconstruction model, and reconstructing the blade defect image using the image reconstruction model to obtain a reconstructed image; S3, constructing a fault detection model, and detecting wind turbine blade faults based on the fault detection model and the reconstructed image to obtain detection results. This invention provides an efficient, accurate, and intelligent method for detecting wind turbine blade faults, overcoming the shortcomings of existing detection technologies in internal defect identification, image quality, and automation, and providing strong technical support for the preventive maintenance and safe operation of wind power equipment.

[0024] See Figure 4 The specific steps are as follows: S1. Use a terahertz time-domain system to scan and image the wind turbine blades to obtain images of blade defects.

[0025] During wind turbine operation, a terahertz imaging device installed on the nacelle or tower scans the same blade region periodically (e.g., every 24 hours) to acquire a time-series terahertz image sequence. Feature point matching and affine transformation are used to register the sequence images, eliminating image shifts caused by blade vibration or attitude changes. Subsequently, background subtraction and noise filtering are performed to highlight the temporal changes in defect areas, resulting in blade defect images.

[0026] S2. Construct an image reconstruction model and use the image reconstruction model to reconstruct the blade defect image to obtain a reconstructed image.

[0027] In this embodiment, the image reconstruction model includes: a defect image sampling module, a linear mapping reconstruction module, and an optimization reconstruction module; wherein, the defect image sampling module is used to divide the blade defect image into blocks, and to sample and learn the image blocks obtained after block division to obtain compressed sensing data; the linear mapping reconstruction module is used to obtain a mapped reconstruction image based on the compressed sensing data; and the optimization reconstruction module is used to perform quality reconstruction on the mapped reconstruction image to obtain a reconstructed image.

[0028] The defect image sampling module includes: a block partitioning submodule and a block sampling submodule; wherein, the block partitioning submodule will sample images of size [size missing]. H × W × l Blade defect images Divide into non-overlapping sizes S × S × l The image blocks; where, H , W , l These represent the height, width, and number of channels of the blade defect image, respectively; the block sampling submodule vectorizes the divided image blocks to obtain a size of... N ×1 vector block ,in, N = S × S × l And the vectorized image patch (i.e., vector patch) is processed by the convolution sampling matrix. Sampling is performed to obtain compressed sensing data.

[0029] In this embodiment, the compressed sensing data includes: ; In the formula, y This represents compressed sensing data; C S * indicates a sampling operation; * indicates a convolution operation; W Ω This represents the convolution sampling matrix. In this embodiment, the sampling matrix is ​​used... Reorganized into M The size is S × S Step size is S The convolution kernel is used to obtain the convolution sampling matrix; where, M = τ × N , τ This indicates the sampling rate.

[0030] The linear mapping reconstruction module obtains the mapped and reconstructed image using the convolutional sampling matrix and compressed sensing data.

[0031] Specifically, the sampling matrix Reorganized into N 1×1× M The convolutional sampling transpose matrix performs preliminary spatial integration of compressed sensing data, and then recombines it through the PixelShuffle operation to obtain the mapped reconstructed image.

[0032] In this embodiment, the image mapping and reconstruction includes: ; In the formula, Y This represents a mapped and reconstructed image; This represents the transpose of the convolution sampling matrix. PixelShuffle() represents the PixelShuffle operation.

[0033] The optimization and reconstruction module includes: a feature extraction submodule, a joint denoising submodule, and an optimization submodule.

[0034] In this embodiment, the feature extraction submodule includes a first feature extraction branch and a second feature extraction branch. The first feature extraction branch includes two 3×3 convolutional kernels and a hybrid attention module (CBMA). The second feature extraction branch includes two 7×7 convolutional kernels and a hybrid attention module (CBMA). After the mapped and reconstructed image is input into the first and second feature extraction branches respectively, first extracted features and second extracted features are obtained. The first and second extracted features are then processed by a concat operation to obtain fused features. The fused features are then processed by a 3×3 convolutional kernel to obtain fused extracted features.

[0035] The joint denoising submodule consists of two parts: the first denoising part obtains a first denoised feature map by performing gradient descent operation on the fused extracted features; the second denoising part processes the first denoised feature map using residual blocks to obtain a joint denoised feature map.

[0036] In this embodiment, the process of obtaining the first denoised feature map from the first denoised part includes: ; In the formula, This represents the first denoised feature map; This indicates feature extraction through fusion; Indicates the dynamic step size.

[0037] The structure of the second denoising part is as follows: one convolutional layer, two residual blocks, and one convolutional layer. Specifically, after the first denoising feature map is input into the second denoising part, it passes through one convolutional layer and two residual blocks in sequence to obtain a residual denoising feature map. Then, the corresponding elements of the residual denoising feature map and the first denoising feature map are added together, and the resulting phase feature map is input into the last convolutional layer to obtain the joint denoising feature.

[0038] In this embodiment, the process of the residual block processing the input includes: ; In the formula, f This represents the input to the residual block; W C1 , W C2 Both are two-layer convolutional kernels; ReLU() represents a non-linear activation function.

[0039] The optimization submodule includes a first optimization branch and a second optimization branch. The first optimization branch adopts an encoder-decoder structure. The encoder extracts key information from the joint denoising features through multi-layer downsampling, and the decoder performs upsampling using transposed convolution to restore spatial resolution. Finally, the first optimization branch features are obtained through channel attention and spatial attention mechanisms. The second optimization branch performs layer normalization on the input joint denoising features, and then passes them through a self-attention block to obtain attention features. The corresponding elements of the joint denoising features and attention features are added together to obtain optimized additive features. The optimized additive features are then subjected to layer normalization and input into a multilayer perceptron to obtain perceptual features. The perceptual features are added to the optimized additive features element-wise, and then passed through two Transform Blocks to obtain the second optimization branch features. The first and second optimization branch features are processed using an optimization mask, and then added element-wise with the joint denoising features to obtain the reconstructed image.

[0040] The process of obtaining the reconstructed image includes: ; In the formula, F 1() represents the feature of the first optimization branch; F 2() represents the characteristic of the second optimization branch; F l represents the joint denoising feature; gate represents the optimized mask.

[0041] The mask optimization process includes: ; In the formula, Sigmiod ( ) represents a non-linear activation function used for gate weights; Wy1 This represents the convolution kernel.

[0042] The loss function of the image reconstruction model includes: ; In the formula, λ Indicates the weighting coefficient; M Indicates the amount of compressed sensing data; E Represents the identity matrix.

[0043] S3. Construct a fault detection model, and detect wind turbine blade faults based on the fault detection model and the reconstructed image to obtain the detection results.

[0044] The fault detection model includes convolutional blocks, embedding blocks, feature extraction blocks, and weighting blocks. The convolutional blocks use a two-dimensional convolutional network to extract time-frequency features from the input reconstructed image, resulting in a convolutionally extracted feature image. The embedding block divides the convolutionally extracted feature image into non-overlapping patch blocks, learns the global spatial features of the patch blocks using matrix transpose and linear projection, and flattens the patch blocks into patch vectors. The feature extraction block combines the patch vectors to obtain a combined feature image. The weighting block weights the features of the combined feature image to obtain weighted features. Finally, a fully connected layer is used for classification to obtain the fault detection result.

[0045] In this embodiment, the process of the feature extraction block processing the patch vector includes: ; ; In the formula, Represents a combined feature image; w 1. w 2. w 3. w 4 represents the weights of the first, second, third, and fourth fully connected layers, respectively; b 1. b 2. b 3. b 4 represents the bias of the first fully connected layer, the second fully connected layer, the third fully connected layer, and the fourth fully connected layer, respectively; LN ( ) indicates layer normalization; X * indicates a patch vector; This represents the output of the second fully connected layer; This represents a non-linear activation function.

[0046] The weighted block assigns weights to the features of the combined feature image using multiple classifiers. The final decision includes: ; In the formula, N Indicates the number of classifiers; w i Indicates the first i The weights of each classifier; x i Indicates the first i The output of each classifier.

[0047] Then, the fully connected layer outputs the detection results: ; In the formula, f ( ) represents the activation function; k q Represents the weight matrix; b q This represents the bias vector.

[0048] In this embodiment, the fault detection results are divided into five levels, with different levels representing the health status of the wind turbine blades. The first level is healthy; the second level is minor cracks; the third level is cracks; the fourth level is spalling; and the fifth level is breakage.

[0049] This invention constructs an image reconstruction model that includes defect image sampling, linear mapping reconstruction, and optimized reconstruction modules. By combining compressed sensing and deep convolutional networks, it effectively improves the image quality and resolution of terahertz imaging, clearly presents minute defects inside blades, and significantly improves the accuracy and reliability of defect identification. This invention has high precision and high resolution detection capabilities. By constructing an image reconstruction model that includes defect image sampling, linear mapping reconstruction and optimized reconstruction modules, and combining compressed sensing and deep convolutional networks, it effectively improves the image quality and resolution of terahertz imaging, can clearly present tiny defects inside the blade, and significantly improves the accuracy and reliability of defect identification.

[0050] This invention enables adaptive and intelligent processing, employing advanced network structures such as hybrid attention mechanisms, residual connections, and self-attention mechanisms. This allows the model to adaptively extract defect features, suppress noise interference, and enhance the ability to recognize complex defect patterns. Optimized masking and multi-branch fusion mechanisms further improve the preservation of detail and overall consistency of the reconstructed image.

[0051] This invention employs a multi-level fault classification and assessment model. The fault detection model integrates convolutional networks and Transformer structures to achieve multi-level feature extraction and fusion from local features to global semantics. Combined with a multi-classifier weighting mechanism, it achieves a five-level refined classification of blade health status, providing a scientific basis for maintenance decisions.

[0052] This invention is non-contact, highly efficient, and safe. Terahertz imaging requires no coupling agent, and the non-contact scanning adapts to various blade surface structures. The detection process does not damage the blade, and the operation is safe and convenient. It is suitable for rapid on-site detection and long-term health monitoring.

[0053] This invention has strong generalization and scalability. The method proposed in this invention does not depend on specific blade materials or defect types. Through data-driven learning, it has good generalization ability and can be extended to the field of non-destructive testing of other composite material structures, and has broad application prospects.

[0054] Example 2 See Figure 2 A wind turbine blade fault detection system based on terahertz imaging technology includes: The defect image acquisition module is used to perform terahertz scanning imaging on wind turbine blades, acquire and preprocess terahertz images to obtain blade defect images. The reconstructed image generation module is used to sequentially perform compressed sensing sampling, linear mapping reconstruction, and multi-branch optimization reconstruction on the blade defect image through the image reconstruction model to generate a reconstructed image. The output detection result module is used to perform multi-level feature extraction of the reconstructed image by fusing convolution and Transformer through the fault detection model, and output the detection result after weighting and classification by a multi-classifier.

[0055] Example 3 See Figure 3 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 computer program, it implements a method for detecting wind turbine blade faults based on terahertz imaging technology.

[0056] Example 4 A computer-readable storage medium storing a computer program, which, when executed by a processor, implements a method for detecting wind turbine blade faults based on terahertz imaging technology.

[0057] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, read-only optical discs, optical storage, etc.) containing computer-usable program code.

[0058] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will 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 flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0059] 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.

[0060] 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.

[0061] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for detecting faults in wind turbine blades based on terahertz imaging technology, characterized in that, Includes the following steps: Terahertz scanning imaging was performed on the wind turbine blades to acquire and preprocess the terahertz images, thereby obtaining images of blade defects. The blade defect image is sequentially subjected to compressed sensing sampling, linear mapping reconstruction, and multi-branch optimization reconstruction using an image reconstruction model to generate a reconstructed image. The reconstructed image is subjected to multi-level feature extraction using fusion convolution and Transformer through a fault detection model, and the detection results are output after weighted classification by a multi-classifier.

2. The method for detecting wind turbine blade faults based on terahertz imaging technology according to claim 1, characterized in that, The acquisition and preprocessing of the terahertz image includes: Obtain time-series terahertz images; Feature point matching and affine transformation registration are performed on the time series terahertz image; The registered image is subjected to background subtraction and noise filtering to obtain the blade defect image.

3. The method for detecting wind turbine blade faults based on terahertz imaging technology according to claim 1, characterized in that, The compressed sensing sampling includes: The blade defect image is divided into non-overlapping image blocks; Each of the aforementioned image blocks is vectorized; Compressed sensing data is generated by sampling vectorized image patches using a convolutional sampling matrix.

4. The method for detecting wind turbine blade faults based on terahertz imaging technology according to claim 1, characterized in that, The linear mapping reconstruction includes: The compressed sensing data is spatially integrated using a convolutional sampling transpose matrix. Perform a pixel shuffle operation on the spatially integrated data to generate a mapped reconstructed image.

5. The method for detecting wind turbine blade faults based on terahertz imaging technology according to claim 1, characterized in that, The multi-branch optimization and refactoring includes: The mapped and reconstructed image is subjected to bi-branch multi-scale feature extraction and fusion to obtain fused extracted features; The fused and extracted features are subjected to joint denoising processing to obtain joint denoised features; The joint denoising features are subjected to bi-branch optimization processing to obtain the first optimized feature and the second optimized feature; The reconstructed image is generated by fusing the first and second optimized features using an optimized mask.

6. The method for detecting wind turbine blade faults based on terahertz imaging technology according to claim 1, characterized in that, The multi-level feature extraction that combines convolution and Transformer includes: The convolution-extracted feature image is divided into non-overlapping patch blocks; Linear projection is performed on each of the patch blocks to generate a patch vector; The patch vector is subjected to multi-level feature extraction and fusion through convolution and Transformer fusion to generate a combined feature image.

7. The method for detecting wind turbine blade faults based on terahertz imaging technology according to claim 1, characterized in that, The detection results output by the multi-classifier weighted classification include: The features of the combined feature image are classified using multiple classifiers. The outputs of each classifier are weighted and fused. Output the five-level health status test results for wind turbine blades.

8. A wind turbine blade fault detection system based on terahertz imaging technology, characterized in that, include: The defect image acquisition module is used to perform terahertz scanning imaging on wind turbine blades, acquire and preprocess terahertz images to obtain blade defect images. The reconstructed image generation module is used to sequentially perform compressed sensing sampling, linear mapping reconstruction, and multi-branch optimization reconstruction on the blade defect image through an image reconstruction model to generate a reconstructed image. The output detection result module is used to perform multi-level feature extraction of the reconstructed image by fusing convolution and Transformer through the fault detection model, and output the detection result after weighting and classification by a multi-classifier.

9. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the wind turbine blade fault detection method based on terahertz imaging technology as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the wind turbine blade fault detection method based on terahertz imaging technology as described in any one of claims 1-7.

Images