Distribution network equipment defect identification method, system, equipment and medium

Abstract

The invention discloses a distribution network equipment defect identification method and system, equipment and a medium, and relates to the technical field of image processing and machine vision, and the method comprises the steps: obtaining an original image of distribution network equipment of which the defect is to be identified; global brightness adjustment processing is carried out on the original image to obtain a corrected image, and the global brightness adjustment processing comprises brightness normalization and nonlinear brightness mapping; performing multi-scale decomposition on the corrected image, performing local contrast enhancement on different scales, and fusing the enhanced images of the scales by adopting a weighted fusion strategy to obtain a final fused image; and performing preprocessing, symmetric feature extraction, feature enhancement based on a self-attention mechanism and classification and positioning of defect types on the fused final image through a pre-trained bionic vision neural network to obtain the defect types and position information. According to the invention, the problem that the defects of the distribution network equipment cannot be accurately identified in real time in a complex environment in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the fields of image processing and machine vision technology, and in particular to a method, system, device and medium for identifying defects in power distribution network equipment. Background Technology

[0002] Power distribution network equipment failures typically stem from abnormal equipment operation, especially latent faults such as cable faults, switch faults, and insulation aging. Failure to detect and repair these faults promptly will severely impact the security and reliability of power supply. In complex environments, traditional manual inspections suffer from high workload, low efficiency, high cost, and personnel safety hazards, limiting the timeliness of fault diagnosis. Automated fault diagnosis systems are widely used in power distribution networks to improve the accuracy and efficiency of fault detection.

[0003] Despite this, existing power distribution network fault diagnosis technologies still face challenges. Traditional image processing methods, such as edge detection-based defect identification, can identify some faults, but their detection performance is poor in low light, noise interference, and when equipment surface features are not obvious. Meanwhile, while neural network-based diagnostic methods improve accuracy, they still require significant computational resources, have long training times, and are highly dependent on data quality and sample size, limiting their application in real-time. Specifically, the invention patent CN119738655A discloses a power distribution network fault selection method based on image transformation and target detection networks. This method processes power distribution network images through a target detection network and combines it with image transformation technology for fault diagnosis, which can address fault diagnosis in complex environments to some extent. However, this method has shortcomings in processing equipment surface details, lacking fine-grained optimization of local image features, and is prone to missing detections, especially in the identification of minute defects. Furthermore, the combination of image transformation and target detection networks pays less attention to equipment surface details, potentially leading to the overlooking of some important fault details. For example, invention patent CN119963498A discloses a method for identifying defects in distribution network equipment based on image processing technology. This method uses image acquisition and processing techniques to detect faults in distribution network equipment and combines multi-sensor fusion to improve diagnostic accuracy. Although this technology can effectively improve the accuracy of fault identification, it may lack targeted adjustments for identifying different surface areas or different types of defects on the equipment, leading to a decrease in diagnostic accuracy when the surface features of some complex equipment are not obvious. Especially in high-noise environments or when there are fine cracks on the equipment surface, the lack of specific model optimization may result in inaccurate diagnostic results.

[0004] Therefore, there is an urgent need to design a method that can identify defects in distribution network equipment in real time and accurately in complex environments. Summary of the Invention

[0005] This invention provides a method, system, device, and medium for identifying defects in distribution network equipment, which solves the problem that existing technologies cannot identify defects in distribution network equipment in real time and accurately under complex environments.

[0006] In view of this, the first aspect of the present invention provides a method for identifying defects in distribution network equipment, the method comprising:

[0007] Obtain the original images of the distribution network equipment with defects to be identified;

[0008] The original image is subjected to global brightness adjustment processing to obtain a corrected image. The global brightness adjustment processing includes brightness normalization and non-linear brightness mapping.

[0009] The corrected image is decomposed into multiple scales, and local contrast enhancement is performed at different scales. A weighted fusion strategy is then used to fuse the enhanced images at each scale to obtain the final fused image.

[0010] The fused final image is preprocessed sequentially using a pre-trained bionic visual neural network, which includes symmetry feature extraction, feature enhancement based on a self-attention mechanism, and defect type classification and localization to obtain defect type and location information.

[0011] Optionally, the brightness normalization includes:

[0012] The brightness value of each pixel in the original image is converted from RGB space to brightness value space, wherein the brightness value of each pixel is calculated by weighting the intensity values ​​of its red channel, green channel and blue channel.

[0013] Optionally, the nonlinear brightness mapping includes:

[0014] A nonlinear brightness mapping is performed on the normalized brightness values ​​based on the cumulative distribution function of image brightness. The mapping function used for the nonlinear brightness mapping is as follows:

[0015] ;

[0016] In the formula, pixels in an image pixel brightness value, This is the brightness value after global brightness adjustment. These are parameters that are adaptively adjusted based on the image brightness distribution, and they have the following segmented settings:

[0017] ;

[0018] In the formula, The cumulative distribution function representing the image brightness.

[0019] Optionally, the step of performing multi-scale decomposition on the corrected image and performing local contrast enhancement at different scales includes:

[0020] A guided filter is used to perform multi-scale decomposition on the corrected image to obtain base images with different smoothness levels;

[0021] For each scale of the base image, local contrast enhancement is performed using the formula for local contrast enhancement operation to obtain the enhanced image at each scale;

[0022] The formula for local contrast enhancement is as follows:

[0023] ;

[0024] in, For pixels in the image The original brightness value, It is in the Image pixel values ​​after smoothing at each scale In the first Enhanced images at various scales.

[0025] Optionally, the expression for fusing the enhanced images at each scale using a weighted fusion strategy is:

[0026] ;

[0027] In the formula, This is the final image after multi-scale fusion. In the first Enhanced images at various scales For the first Weights of images at each scale The total number of scales.

[0028] Optionally, the symmetry feature extraction includes: calculating the gradients of the preprocessed image in the x and y directions using the Sobel operator, expressed as:

[0029] ;

[0030] In the formula, and These represent the images in and Gradient in the direction.

[0031] Optionally, the feature enhancement based on the self-attention mechanism includes:

[0032] By calculating the relationship between the query vector, key vector, and value vector of the image after symmetry feature extraction, the pre-trained bionic visual neural network adaptively enhances its focus on defect regions. The expression is as follows:

[0033] ;

[0034] In the formula, It is a query vector. It is a key vector. It is a value vector. It is the dimension of the key vector.

[0035] A second aspect of the present invention provides a distribution network equipment defect identification system, the system comprising:

[0036] The acquisition unit is used to acquire the original image of the distribution network equipment with defects to be identified;

[0037] A global brightness adjustment unit is used to perform global brightness adjustment processing on the original image to obtain a corrected image. The global brightness adjustment processing includes: brightness normalization and non-linear brightness mapping.

[0038] The local contrast enhancement unit is used to decompose the corrected image into multiple scales, perform local contrast enhancement at different scales, and use a weighted fusion strategy to fuse the enhanced images at each scale to obtain the final fused image.

[0039] The recognition unit is used to perform preprocessing, symmetry feature extraction, feature enhancement based on self-attention mechanism, and defect type classification and localization on the fused final image through a pre-trained bionic visual neural network to obtain defect type and location information.

[0040] A third aspect of the present invention provides a distribution network equipment defect identification device, the device comprising a processor and a memory:

[0041] The memory is used to store program code and transmit the program code to the processor;

[0042] The processor is used to execute the steps of the distribution network equipment defect identification method as described in the first aspect above, according to the instructions in the program code.

[0043] A fourth aspect of the present invention provides a computer-readable storage medium for storing program code for executing the distribution network equipment defect identification method described in the first aspect above.

[0044] As can be seen from the above technical solutions, the present invention has the following advantages:

[0045] This invention provides a method for identifying defects in distribution network equipment. By combining adaptive brightness correction, multi-scale contrast enhancement, and a biomimetic visual neural network, it significantly improves the accuracy and robustness of defect identification. Adaptive brightness correction effectively solves the problems of low light and noise interference, ensuring accurate representation of equipment surface details. Multi-scale contrast enhancement further enhances the features of minute defects in the image, while the biomimetic visual neural network accurately extracts symmetry and detail features from the image, enhancing the accuracy of defect identification. Compared to existing technologies, this invention not only provides more stable and efficient detection capabilities in complex environments but also reduces reliance on computing resources by optimizing the network structure, improving real-time performance and system adaptability, making it suitable for various complex monitoring environments. This solves the problem that existing technologies cannot identify distribution network equipment defects in real-time and accurately in complex environments. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 A flowchart illustrating a method for identifying defects in distribution network equipment provided in an embodiment of the present invention;

[0048] Figure 2 This is a schematic diagram of the structure of the biomimetic visual neural network provided in an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram of a power distribution network equipment defect identification system provided in an embodiment of the present invention. Detailed Implementation

[0050] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0051] Please see Figure 1 The present invention provides a method for identifying defects in distribution network equipment, comprising:

[0052] Step 101: Obtain the original image of the distribution network equipment with defects to be identified.

[0053] It should be noted that original images of the distribution network equipment with defects to be identified can be obtained in various ways for subsequent defect identification. For example, high-definition industrial cameras deployed along the inspection route of the distribution network equipment can capture images in real time, which will not be elaborated upon here.

[0054] Step 102: Perform global brightness adjustment processing on the original image to obtain the corrected image. Global brightness adjustment processing includes brightness normalization and nonlinear brightness mapping.

[0055] It should be noted that in traditional image processing, uneven image brightness and low-light areas often result in indistinct surface features of the device, thus affecting the accurate identification of defects. The adaptive brightness correction technology of this invention improves the overall visual effect of the image through a two-step process: global brightness adjustment in step 102 and local contrast enhancement in step 103, ensuring that the details of image features are effectively restored.

[0056] In one embodiment, luminance normalization includes converting the luminance value of each pixel in the original image from RGB space to a luminance value. Space, where the brightness value of each pixel It is calculated by weighting the intensity values ​​of its red, green, and blue channels.

[0057] Nonlinear brightness mapping includes: performing nonlinear brightness mapping on normalized brightness values ​​based on the cumulative distribution function of image brightness.

[0058] Regarding global brightness adjustment, it should be noted that this global brightness adjustment is the first step in the adaptive brightness correction process. It aims to address the loss of detail in low-light areas by adjusting the overall brightness distribution of the image, while ensuring that high-brightness areas remain sharp. This process uses a non-linear mapping function to adjust the brightness distribution of the image, thereby enhancing low-brightness areas without affecting the bright parts of the image. Specific steps include brightness normalization and non-linear brightness mapping.

[0059] First, to ensure consistent brightness adjustment, the input image is normalized. This is done by converting the brightness value of each pixel from the RGB space to a normalized value. This provides space, thus simplifying subsequent brightness adjustments. The brightness value for each pixel... It is calculated by weighting the intensity values ​​of its red, green, and blue channels, as shown in the following formula:

[0060]

[0061] in, These are the pixels in the image. The intensity values ​​of the red, green, and blue channels for the location.

[0062] Secondly, a nonlinear brightness mapping function was employed to enhance low-brightness areas in the image. This function uses an adaptive mapping coefficient z based on the distribution of brightness values ​​in the image, resulting in stronger enhancement in low-brightness areas while maintaining minimal change in high-brightness areas, thus avoiding artifacts or loss of detail caused by excessive brightness adjustment.

[0063] The expression for this mapping function is:

[0064]

[0065] in, pixels in an image The original brightness value, This is the brightness value after global brightness adjustment. These are parameters that are adaptively adjusted based on the image brightness distribution, and they have the following segmented settings:

[0066]

[0067] in, The cumulative distribution function (CDF) represents the image brightness. This mapping function ensures that low-brightness areas are effectively enhanced, while high-brightness areas remain unchanged or undergo only minor changes.

[0068] Furthermore, in practical applications, since the brightness distribution of an image often exhibits strong local differences, it is necessary to dynamically adjust the mapping parameters. By analyzing the overall brightness distribution of the image, a frequency analysis standard was defined, and parameters were optimized based on the proportion of low-brightness areas in the image. The value of . Specifically, when the proportion of low-brightness areas in an image exceeds a certain threshold, It is set to a lower value, thus producing a stronger brightness enhancement effect in the low-brightness range. Conversely, when the overall image brightness is high, The value is close to 1, thus keeping the highlighted parts of the image unaffected.

[0069] Understandably, step 102 effectively improves the detail representation of low-brightness areas in the image through adaptive brightness correction, while avoiding overexposure or distortion in high-brightness areas, which is used for the subsequent extraction of defect features of distribution network equipment.

[0070] Step 103: Decompose the corrected image into multiple scales, perform local contrast enhancement at different scales, and use a weighted fusion strategy to fuse the enhanced images at each scale to obtain the final fused image.

[0071] In one embodiment, step 103, which involves multi-scale decomposition of the corrected image and local contrast enhancement at different scales, includes:

[0072] A guided filter is used to decompose the corrected image into multiple scales to obtain base images at different smoothness levels. For each scale of the base image, local contrast enhancement is performed using the formula for local contrast enhancement operation to obtain the enhanced image at each scale.

[0073] It should be noted that after global brightness adjustment, the brightness uniformity of the image is improved, but local contrast may still be weak, especially in edge, detail, and defect areas. To enhance image detail, this invention employs a multi-scale decomposition method based on guided filtering. Guided filters have strong local preservation capabilities, enabling image smoothing at different scales while retaining edge details. Compared to traditional Gaussian filters, guided filters are more effective at reducing noise interference and preserving edge information, thus avoiding the loss of image details.

[0074] The multi-scale decomposition process first decomposes the image into multiple scales, each representing a different degree of smoothness. By applying local contrast enhancement techniques to the image at each scale, details and edge information can be enhanced at different levels.

[0075] The formula for local contrast enhancement is as follows:

[0076]

[0077] in, For pixels in the image The original brightness value, It is in the Image pixel values ​​after smoothing at each scale In the first Images enhanced at various scales.

[0078] It should be further explained that the key to multi-scale contrast enhancement is how to effectively fuse enhanced images from different scales together. In this invention, a weighted fusion strategy is adopted, determining the weights of images at different scales based on the importance of image contrast and detail information at each scale. The final fused image is a weighted average of the images at each scale, as shown in the following formula:

[0079]

[0080] in, This is the final image after multi-scale fusion. In the first Enhanced images at various scales For the first Weights of images at each scale The total number of scales. Through weighted averaging, detailed information from images at each scale is effectively fused, and the risk of over-enhancement is reduced while preserving local features.

[0081] Understandably, step 103 achieves precise enhancement of the details in the corrected image through a multi-scale decomposition and weighted fusion strategy. Multi-scale decomposition using a guided filter separates the base images at different smoothness levels, ensuring the integrity of key details such as edges and defect areas during the decomposition process. Furthermore, local contrast enhancement operations on the base images at each scale further highlight the image's detailed features at different levels, effectively compensating for insufficient local contrast after global brightness adjustment. Based on this, a weighted fusion strategy, based on the importance of contrast and detail information at each scale, integrates the enhanced images at different scales through weighted averaging, forming a high-quality final image that retains rich local features while avoiding the risk of over-enhancement. This provides a clearer and more detailed image data foundation for the accurate identification of defects in distribution network equipment. This step, by combining image processing algorithms with a reasonable fusion strategy, improves image quality, making subtle defects in distribution network equipment easier to detect and identify in the image.

[0082] Step 104: The fused final image is preprocessed sequentially using a pre-trained bionic visual neural network, including symmetry feature extraction, feature enhancement based on self-attention mechanism, and defect type classification and localization, to obtain defect type and location information.

[0083] It should be noted that the biomimetic visual neural network mimics the visual system of insects, particularly the characteristics of the LPLC2 neuron in the fruit fly, which is highly sensitive to symmetry and detail in images. This network design extracts features from images using a convolutional neural network (CNN) and combines this with a self-attention mechanism to enhance attention to image details, thereby achieving accurate defect identification.

[0084] Specifically, neural network structure design such as Figure 2 The module includes: an image processing module for preprocessing, a symmetry feature extraction module for symmetry feature extraction, a symmetry enhancement module for feature enhancement self-attention mechanism, and a defect recognition module for defect type classification and localization.

[0085] The descriptions of each module are as follows:

[0086] A. Image preprocessing module:

[0087] The image preprocessing module standardizes the input image, removing interference from environmental factors (such as uneven lighting and noise). By performing brightness adjustment, normalization, and noise suppression, it ensures that subsequent modules can process the image stably and efficiently.

[0088] The formulaic description of the image brightness normalization process is as follows:

[0089] ;

[0090] in, Is the image in position Pixel brightness value at that location, It is the average brightness of the image. It is the standard deviation of image brightness. It is the normalized image brightness value.

[0091] By normalizing the image, the brightness range is standardized to a standard range, thereby eliminating the impact of lighting differences on subsequent processing.

[0092] B. Symmetry Feature Extraction Module:

[0093] This module extracts symmetry features from input images, with a particular focus on edges and symmetry information. Bionic visual neurons (such as the LPLC2 neuron in fruit flies) are highly sensitive to radial symmetry; therefore, this module employs an edge detection method based on the Sobel operator to capture crucial symmetry information in the image.

[0094] The formula for calculating the gradient of an image using the Sobel operator is:

[0095] ;

[0096] in, and These represent the images in and Gradient in the direction.

[0097] C. Feature enhancement module for self-attention mechanism:

[0098] Self-attention is an important method in deep learning to improve the expressive power and flexibility of networks. In this module, self-attention can dynamically adjust the model's focus on key regions based on the image content. By calculating the relationship between the image's query vector, key vector, and value vector, the network can adaptively enhance its focus on defective regions.

[0099] ;

[0100] in, It is a query vector. It is a key vector. It is a value vector. This refers to the dimension of the key vector. Through this mechanism, the model can dynamically assign different weights to each image region, thereby enhancing the features of defect-related regions.

[0101] D. Defect Identification Module:

[0102] The defect identification module is the output layer of the neural network, responsible for classifying and locating defect types based on extracted features. This module maps extracted image features to defect categories through a fully connected layer, outputting the defect type and its location. The classification output is calculated using the Softmax activation function, as shown in the following formula:

[0103] ;

[0104] in, and These are the weights and biases obtained during network training. These are the image features output from the previous feature extraction module. It refers to the type or category of the output defect.

[0105] Furthermore, it should be noted that the neural network in step 104 employs a supervised learning method, trained using a dataset of distribution network equipment images containing labeled defects. Data augmentation techniques (such as rotation, translation, and scaling) are used to enhance the diversity of the training dataset. The optimization process uses the Adam optimizer, adjusting the network's weight parameters by minimizing the cross-entropy loss function.

[0106] Understandably, in step 104, the network eliminates environmental interference through an image preprocessing module; the symmetry feature extraction module captures edge and symmetry information using the Sobel operator, aligning with the bionic vision's sensitivity to detail; the self-attention mechanism feature enhancement module uses dynamic weight allocation to focus the network on potential defect areas, improving the specificity of feature expression; and finally, the defect recognition module classifies and locates defect types through fully connected layers and the Softmax activation function. Throughout the process, supervised learning combined with data augmentation techniques ensures the model's generalization ability, while the Adam optimizer and cross-entropy loss function efficiently optimize network parameters. Through the synergistic effect of these modules, the pre-trained bionic vision neural network can deeply mine defect features in images and accurately output defect type and location information, providing core technical support for the automated and intelligent identification of defects in power distribution network equipment.

[0107] This invention provides a defect identification method for distribution network equipment, employing adaptive brightness correction, multi-scale contrast enhancement, and biomimetic visual neural network technology to identify defects in distribution network equipment under complex environments characterized by low light, high noise, and blurred surface features. Adaptive brightness correction optimizes image detail through global brightness adjustment and local contrast enhancement, ensuring effective recovery of minute features in low-light areas while avoiding over-adjustment in high-brightness areas. Multi-scale contrast enhancement, combined with guided filtering, enhances the representation of minute defects (such as cracks and corrosion) on the equipment surface, effectively improving detail information at different scales through a weighted fusion strategy. The biomimetic visual neural network simulates the insect visual system, particularly in the extraction of symmetry and detail features, and accurately identifies defect areas using a self-attention mechanism. Through testing in various environments, the inventors have verified the robustness and efficiency of this method in complex environments, significantly improving the accuracy and real-time performance of minute defect identification.

[0108] The above is a method for identifying defects in distribution network equipment provided in the embodiments of the present invention. The following is a system for identifying defects in distribution network equipment provided in the embodiments of the present invention.

[0109] Please see Figure 3 The present invention provides a distribution network equipment defect identification system, comprising:

[0110] Acquisition unit 201 is used to acquire the original image of the distribution network equipment with defects to be identified;

[0111] The global brightness adjustment unit 202 is used to perform global brightness adjustment processing on the original image to obtain the corrected image. The global brightness adjustment processing includes: brightness normalization and nonlinear brightness mapping based on the cumulative distribution function of image brightness.

[0112] The local contrast enhancement unit 203 is used to decompose the corrected image into multiple scales, perform local contrast enhancement at different scales, and use a weighted fusion strategy to fuse the enhanced images at each scale to obtain the fused final image.

[0113] The recognition unit 204 is used to perform preprocessing, symmetry feature extraction, feature enhancement based on self-attention mechanism, and defect type classification and localization on the fused final image through a pre-trained bionic visual neural network to obtain defect type and location information.

[0114] Furthermore, this embodiment of the invention also provides a distribution network equipment defect identification device, the device including a processor and a memory:

[0115] The memory is used to store program code and transfer the program code to the processor;

[0116] The processor is used to execute the steps of the distribution network equipment defect identification method as described in the above method embodiments, according to the instructions in the program code.

[0117] Furthermore, this embodiment of the invention also provides a computer-readable storage medium for storing program code, which is used to execute the distribution network equipment defect identification method described in the above method embodiments.

[0118] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0119] In the embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0120] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0121] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0122] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0123] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying a defect of a network device, characterized in that, The method comprises the following steps: obtaining an original image of a distribution network equipment with defects to be identified; performing global brightness adjustment processing on the original image to obtain a corrected image, wherein the global brightness adjustment processing comprises brightness normalization and non-linear brightness mapping; performing multi-scale decomposition on the corrected image, performing local contrast enhancement on different scales respectively, and adopting a weighted fusion strategy to fuse the enhanced images of each scale to obtain a fused final image; performing preprocessing, symmetry feature extraction, feature enhancement based on a self-attention mechanism, and classification and positioning of defect types on the fused final image in sequence through a pre-trained biomimetic visual neural network to obtain defect type and position information.

2. The network device defect identification method of claim 1, wherein, The brightness normalization comprises: converting the brightness value of each pixel in the original image from the RGB space to the brightness value space, wherein the brightness value of each pixel is calculated by weighting the intensity values of the red channel, green channel and blue channel.

3. The network device defect identification method of claim 1, wherein, The non-linear brightness mapping comprises: performing non-linear brightness mapping on the brightness value after brightness normalization based on the cumulative distribution function of image brightness, wherein the mapping function adopted by the non-linear brightness mapping is: ； wherein is the original luminance value of a pixel in the image, is the luminance value after global luminance adjustment, is a parameter for adaptive adjustment according to the luminance distribution of the image, with the following piecewise settings: ； In the formula, represents the cumulative distribution function of the image luminance.

4. The network device defect identification method of claim 1, wherein, The multi-scale decomposition of the corrected image and the local contrast enhancement on different scales comprise: performing multi-scale decomposition on the corrected image using a guided filter to obtain base images under different smoothing degrees; for each scale of the base image, respectively using the formula of the local contrast enhancement operation to perform local contrast enhancement to obtain the enhanced image of each scale; wherein the formula of the local contrast enhancement operation is as follows: ； in, For pixels in the image pixel brightness value, It is in the Image pixel values ​​after smoothing at each scale In the first Enhanced images at various scales.

5. The network device defect identification method of claim 1, wherein, The expression for fusing the enhanced images of each scale using the weighted fusion strategy is: ； In the formula, This is the final image after multi-scale fusion. In the first Enhanced images at various scales For the first Weights of images at each scale The total number of scales.

6. The network device defect identification method of claim 1, wherein, The symmetry feature extraction comprises: ； wherein and denote the gradient of the image in and directions, respectively.

7. The network device defect identification method of claim 1, wherein, calculating the gradients of the preprocessed image in the x and y directions using the Sobel operator, and the expression is: The feature enhancement based on the self-attention mechanism comprises: ； wherein is a query vector, is a key vector, is a value vector, is a dimension of the key vector.

8. A network device defect identification system, comprising: by calculating the relationship between the query vector, key vector and value vector of the image after symmetry feature extraction, the pre-trained biomimetic visual neural network adaptively enhances the attention to the defect area, and the expression is: The method comprises the following steps: an acquisition unit is configured to obtain an original image of a distribution network equipment with defects to be identified; a global brightness adjustment unit is configured to perform global brightness adjustment processing on the original image to obtain a corrected image, wherein the global brightness adjustment processing comprises brightness normalization and non-linear brightness mapping; a local contrast enhancement unit is configured to perform multi-scale decomposition on the corrected image, perform local contrast enhancement on different scales respectively, and adopt a weighted fusion strategy to fuse the enhanced images of each scale to obtain a fused final image; 9. A network device defect identification device, comprising: an identification unit is configured to perform preprocessing, symmetry feature extraction, feature enhancement based on a self-attention mechanism, and classification and positioning of defect types on the fused final image in sequence through a pre-trained biomimetic visual neural network to obtain defect type and position information. The device comprises a processor and a memory: the memory is configured to store program code and transmit the program code to the processor; The processor is used to execute the distribution network equipment defect identification method according to any one of claims 1-7 according to the instructions in the program code.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program code for executing the distribution network equipment defect identification method according to any one of claims 1-7.

Images