Insulator and self-exposure defect detection method and device thereof

By constructing a defect detection model of filtering, rain removal and fog removal modules, the problem of low detection efficiency and accuracy of insulators under severe weather conditions is solved, and efficient insulator self-exposed defect detection under rain and fog and other conditions is achieved.

CN118967626APending Publication Date: 2024-11-15GUANGDONG ELECTRIC POWER SCI RES INST ENERGY TECH CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411058298.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-02
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The existing insulators and their self-exposed defect detection methods have low detection efficiency and accuracy under severe weather conditions, especially the low efficiency of manual inspection and the performance of the deep learning technology has significantly declined under fog, rain, snow and other conditions.

Method used

Build a defect detection model including filtering module, image rain removal module, image fog removal module and defect detection module. Separate rain fog images through physical models, merge them into clear images after rain removal and fog removal treatment to identify insulators and their self-exposed defects. .

Benefits of technology

It improves the efficiency and accuracy of insulators and their self-exposed defect detection in harsh weather conditions such as rain and fog, and is suitable for efficient detection in harsh environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118967626A_ABST
    Figure CN118967626A_ABST
Patent Text Reader

Abstract

The invention discloses an insulator and a self-exposure defect detection method and device thereof, and the method comprises the steps: constructing a defect detection model, and training the defect detection model through a training data set; and identifying the insulator in the rain and fog image and the specific position of the self-exposure defect of the insulator according to the trained defect detection model. The rain and fog image is processed through a filtering module, and a high-frequency image containing raindrops and a low-frequency image containing fog are separated out. Then, a rain removal module and a defogging module are constructed for the two images respectively to generate a clear rain-free and fog-free image, defect detection is carried out on the processed image, the insulator and the specific position of the self-exposure defect of the insulator are recognized, the method is suitable for defect detection under severe weather conditions such as rain and fog, and the detection accuracy is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of insulation self-exposure defect detection, and in particular to an insulator and a self-exposure defect detection method and device thereof. Background Art

[0002] In the power system, insulators play a vital role. They are a key component of high-voltage transmission lines, responsible for providing the necessary insulation support between the conductors and the brackets, preventing current leakage, and avoiding electrical short circuits between the conductors or between the conductors and the ground. The performance of insulators plays a decisive role in the safe and stable operation of high-voltage transmission lines. However, in various harsh environments, such as rain, fog, lightning, salt spray, ultraviolet radiation and other extreme weather conditions, insulators are very prone to self-exposure and other defects, which leads to a decrease in insulation performance, increases the risk of transmission line failure, and has a serious impact on the safe and stable operation of the power system. It may even cause safety accidents such as fire and explosion, causing huge economic losses. Therefore, it is particularly important to conduct regular inspection and maintenance of insulators and their self-exposure defects.

[0003] Existing detection methods for insulators and their self-exposure defects are generally carried out through manual inspections and detection based on deep learning technology. Manual inspections rely on manual field inspections, which is inefficient and easily affected by subjective factors of personnel, which may cause incomplete and inaccurate detection results. In addition, due to the limitations of the working environment, manual inspections may also face safety risks, such as natural disasters and threats from wild animals. On the other hand, deep learning-based detection technologies, especially convolutional neural networks (CNNs), have demonstrated excellent performance on standard datasets. However, when faced with adverse weather conditions such as fog, rain, and snow, these deep learning-based methods may encounter significant performance degradation due to reduced image quality. Summary of the invention

[0004] The present invention provides an insulator and self-exposure defect detection method and device, so as to solve the technical problem that the existing insulator and self-exposure defect detection has low detection efficiency and accuracy under adverse weather conditions such as fog, rain and snow.

[0005] In order to solve the above technical problems, an embodiment of the present invention provides an insulator and a method for detecting self-exposure defects thereof, comprising:

[0006] Constructing a defect detection model, and training the defect detection model according to a training data set;

[0007] Constructing a defect detection model, and training the defect detection model according to a training data set;

[0008] According to the trained defect detection model, the specific locations of insulators and their self-exposure defects in rain and fog images are identified. Specifically:

[0009] The defect detection model includes a filtering module, an image deraining module, an image defogging module and a defect detection module;

[0010] Constructing a physical model of a rain and fog image, and filtering the rain and fog image according to the filtering module to generate a high-frequency rain image and a low-frequency fog image;

[0011] Performing rain removal processing on the high-frequency rainy image according to the image rain removal module to obtain a high-frequency rain removal image;

[0012] Performing defogging processing on the low-frequency foggy image according to the image defogging module to obtain a low-frequency defogging image;

[0013] The high-frequency derained image and the low-frequency defogging image are fused into a clear image according to the physical model of the rain-fog image, and the clear image is detected according to the defect detection module to determine the specific positions of the insulator and its self-exposure defects.

[0014] The present invention processes rain and fog images through a filtering module to separate high-frequency images containing raindrops and low-frequency images containing fog. Subsequently, rain removal and fog removal modules are constructed for the two images to generate clear rain-free and fog-free images, and then defect detection is performed on the processed images to identify the specific locations of insulators and their self-exposure defects. The method is suitable for defect detection under adverse weather conditions such as rain and fog, and improves the efficiency and accuracy of detection.

[0015] Furthermore, the physical model of the rain and fog image is constructed, and the rain and fog image is filtered according to the filtering module to generate a high-frequency rain image and a low-frequency fog image, specifically:

[0016] Constructing a physical model of a rain and fog image according to an atmospheric scattering model, wherein the physical model includes a sum of a high-frequency rain image and a low-frequency fog image;

[0017] The rain and fog image is filtered according to the guided filter to generate a high-frequency rainy image and a low-frequency foggy image.

[0018] Furthermore, the construction of the defect detection model specifically includes:

[0019] Constructing an image deraining module, wherein the image deraining module includes two convolution blocks, two deconvolution blocks and three dense blocks;

[0020] The convolution block includes a convolution layer, a batch normalization layer and an activation function;

[0021] The deconvolution block includes an involution feature extraction neural network operator, a batch normalization layer and an activation function;

[0022] The dense block includes several connected convolution blocks.

[0023] Furthermore, the step of performing rain removal processing on the high-frequency rainy image according to the image rain removal module to obtain the high-frequency rain removed image is specifically as follows:

[0024] Extracting a first feature of the high-frequency rainy image according to a first convolution block, and generating three first feature maps according to the three dense blocks and the first feature;

[0025] Connect and merge the three first feature maps to generate a second feature map;

[0026] Performing an inner convolution operation on the second feature map according to the two deconvolution blocks to obtain a third feature map;

[0027] The third feature map is input into the second convolution block to generate a high-frequency derained image.

[0028] Furthermore, the construction of the defect detection model specifically includes:

[0029] Constructing the image defogging module, wherein the image defogging module includes three convolution blocks and four deconvolution blocks;

[0030] The convolution block includes a convolution layer, a batch normalization layer and an activation function;

[0031] The deconvolution block includes an involution feature extraction neural network operator, a batch normalization layer and an activation function.

[0032] Furthermore, the construction of the defect detection model specifically includes:

[0033] Constructing a defect detection module, wherein the defect detection module includes a backbone network, an attention module, and a detection head;

[0034] The attention module is connected to the backbone network, and the attention module includes a channel attention module and a spatial attention module; the detection head includes a Wise IoU loss function.

[0035] Furthermore, before constructing the defect detection model, the method further includes:

[0036] Collect image datasets of insulators and their self-exposure defects, and annotate them to generate training datasets, specifically:

[0037] Collect rain and fog images containing insulators and their self-exposure defects under rain and fog weather conditions, and perform preprocessing operations on the rain and fog images, wherein the preprocessing operations include data enhancement and data expansion to generate a rain and fog image set;

[0038] According to the image annotation tool, rectangular frames are drawn on insulators and their self-exposure defects in the rain and fog image to mark them, a training data set is generated, and the training data set is divided into a training set, a validation set and a test set according to a preset ratio.

[0039] In a second aspect, the present invention provides an insulator and a self-exposure defect detection device thereof, comprising: a model building module and a detection module;

[0040] The model building module is used to build a defect detection model and train the defect detection model according to the rain and fog image set;

[0041] The detection module is used to identify the specific locations of insulators and their self-exposure defects in the rain and fog image according to the trained defect detection model. Specifically:

[0042] The defect detection model includes a filtering module, an image deraining module, an image defogging module and a defect detection module;

[0043] Constructing a physical model of a rain and fog image, and filtering the rain and fog image according to the filtering module to generate a high-frequency rain image and a low-frequency fog image;

[0044] Performing rain removal processing on the high-frequency rainy image according to the image rain removal module to obtain a high-frequency rain removal image;

[0045] Performing defogging processing on the low-frequency foggy image according to the image defogging module to obtain a low-frequency defogging image;

[0046] The high-frequency derained image and the low-frequency defogging image are fused into a clear image according to the physical model of the rain-fog image, and the clear image is detected according to the defect detection module to determine the specific positions of the insulator and its self-exposure defects.

[0047] In a third aspect, the present invention provides a computer device, comprising: a processor, a communication interface and a memory, wherein the processor, the communication interface and the memory are interconnected, wherein the memory stores an executable program code, and the processor is used to call the executable program code to execute the insulator and self-exposure defect detection method.

[0048] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the insulator and self-exposure defect detection method thereof are implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A schematic structural diagram of an insulator and a self-exposure defect detection method thereof provided by an embodiment of the present invention;

[0050] Figure 2 A structural diagram of an image rain removal module of an insulator and a self-exposure defect detection method thereof provided in an embodiment of the present invention;

[0051] Figure 3 A schematic diagram of a ConvB module structure of an insulator and a self-exposure defect detection method thereof provided in an embodiment of the present invention;

[0052] Figure 4 A schematic diagram of an InvoB module structure of an insulator and a self-exposure defect detection method thereof provided in an embodiment of the present invention;

[0053] Figure 5 A schematic diagram of a DenseBlock module structure of an insulator and a self-exposure defect detection method thereof provided in an embodiment of the present invention;

[0054] Figure 6 A schematic diagram of an Invo operator structure of an insulator and a self-exposure defect detection method thereof provided in an embodiment of the present invention;

[0055] Figure 7 A schematic structural diagram of an image defogging module of an insulator and a self-exposure defect detection method thereof provided by an embodiment of the present invention;

[0056] Figure 8 A schematic structural diagram of a defect detection module of an insulator and a self-exposure defect detection method thereof provided by an embodiment of the present invention;

[0057] Fig. 9 Another structural schematic diagram of an insulator and a self-exposure defect detection method thereof provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] Embodiment 1

[0060] Please refer to Figure 1 , Figure 1 A schematic diagram of a structure of an insulator and a self-exposure defect detection method thereof provided by an embodiment of the present invention, comprising step 101 and step 102, which are specifically as follows;

[0061] Step 101: construct a defect detection model, and train the defect detection model according to a training data set;

[0062] Step 102: Identify the specific locations of the insulators and their self-exposure defects in the rain and fog image according to the trained defect detection model. Specifically:

[0063] The defect detection model includes a filtering module, an image deraining module, an image defogging module and a defect detection module;

[0064] Constructing a physical model of a rain and fog image, and filtering the rain and fog image according to the filtering module to generate a high-frequency rain image and a low-frequency fog image;

[0065] Performing rain removal processing on the high-frequency rainy image according to the image rain removal module to obtain a high-frequency rain removal image;

[0066] Performing defogging processing on the low-frequency foggy image according to the image defogging module to obtain a low-frequency defogging image;

[0067] The high-frequency derained image and the low-frequency defogging image are fused into a clear image according to the physical model of the rain-fog image, and the clear image is detected according to the defect detection module to determine the specific positions of the insulator and its self-exposure defects.

[0068] In this embodiment, before constructing the defect detection model, the method further includes:

[0069] Collect image datasets of insulators and their self-exposure defects, and annotate them to generate training datasets, specifically:

[0070] Collect rain and fog images containing insulators and their self-exposure defects under rain and fog weather conditions, and perform preprocessing operations on the rain and fog images, wherein the preprocessing operations include data enhancement and data expansion to generate a rain and fog image set;

[0071] According to the image annotation tool, rectangular frames are drawn on insulators and their self-exposure defects in the rain and fog image to mark them, a training data set is generated, and the training data set is divided into a training set, a validation set and a test set according to a preset ratio.

[0072] In this embodiment, image data of insulators and their self-exposure defects taken under rainy and foggy weather conditions are captured using a high-definition camera or collected from open source data sets on the Internet, and a preprocessing operation is performed on the rainy and foggy images, and the preprocessing operation includes data enhancement and data expansion.

[0073] In this embodiment, a data enhancement operation is performed on the captured or collected image data, and the data enhancement operation includes scaling, cropping, rotating, color jittering, contrast adjustment, and noise elimination.

[0074] In this embodiment, the original images containing insulators and their self-exposure defects and the data-enhanced images are integrated into a sufficient and diverse data set, and data expansion is completed to generate an image data set.

[0075] In this embodiment, a professional image annotation tool (such as Label img software) is used to draw rectangular boxes on the insulators and their self-exposure defects in the images of the image dataset to annotate them, construct an image dataset with precise annotation information, and randomly allocate the image dataset into a training set, a validation set, and a test set according to a predetermined ratio. The preset ratio is set to 6:1:1 by default.

[0076] In this embodiment, the defect detection model includes a filtering module, an image deraining module, an image defogging module and a defect detection module.

[0077] In this embodiment, a physical model of rain and fog images is constructed, so that the original image is processed according to the filtering module to obtain a high-frequency rainy image and a low-frequency foggy image.

[0078] In this embodiment, the physical model of the rain and fog image is constructed, and the rain and fog image is filtered according to the filtering module to generate a high-frequency rain image and a low-frequency fog image, specifically:

[0079] Constructing a physical model of a rain and fog image according to an atmospheric scattering model, wherein the physical model includes a sum of a high-frequency rain image and a low-frequency fog image;

[0080] The rain and fog image is filtered according to the guided filter to generate a high-frequency rainy image and a low-frequency foggy image.

[0081] In this embodiment, the rain and fog image is modeled as the sum of a high-frequency rain image and a low-frequency fog image according to the atmospheric scattering model. Specifically:

[0082] O = O H +O L (1)

[0083] Among them, O represents the observed image with rain and fog, O H =(I H +S)·T represents high-frequency rainy images, O L =I L T+A(1-T) represents a low-frequency foggy image, S represents rain streaks, T represents the transmittance of the fog layer between the shooting system and the scene, T represents the global atmospheric light value, A(1-T) represents the atmospheric light value intensity, I H Represents the high-frequency details of the real scene image, I L Represents the low-frequency background of the real scene image. Therefore, the rain and fog image can be modeled as I = IH +I L ;

[0084] In this embodiment, the rain and fog image is filtered by using a guided filter according to the filtering module to obtain a high-frequency rain image and a low-frequency fog image.

[0085] In this embodiment, the construction of the defect detection model specifically includes:

[0086] Constructing an image deraining module, wherein the image deraining module includes two convolution blocks, two deconvolution blocks and three dense blocks;

[0087] The convolution block includes a convolution layer, a batch normalization layer and an activation function;

[0088] The deconvolution block includes an involution feature extraction neural network operator, a batch normalization layer and an activation function;

[0089] The dense block includes several connected convolution blocks.

[0090] Please refer to Figure 2 , Figure 2 A structural diagram of an image rain removal module for an insulator and a self-exposure defect detection method thereof provided in an embodiment of the present invention.

[0091] In this embodiment, the image deraining module includes two convolution blocks (ConvB modules), two inverse convolution blocks (InvoB modules) and three dense blocks (DenseB lock modules), which are used to process and improve raindrop interference in the image.

[0092] In this embodiment, the image rain removal module performs rain removal processing on the high-frequency rainy image to obtain the high-frequency rain-removed image, specifically:

[0093] Extracting a first feature of the high-frequency rainy image according to a first convolution block, and generating three first feature maps according to the three dense blocks and the first feature;

[0094] Connect and merge the three first feature maps to generate a second feature map;

[0095] Performing an inner convolution operation on the second feature map according to the two deconvolution blocks to obtain a third feature map;

[0096] The third feature map is input into the second convolution block to generate a high-frequency derained image.

[0097] In this embodiment, please refer to Figure 3 , Figure 4 and Figure 5 , Figure 3A schematic diagram of a ConvB module structure of an insulator and a self-exposure defect detection method thereof provided in an embodiment of the present invention, Figure 4 A schematic diagram of an InvoB module structure of an insulator and a self-exposure defect detection method thereof provided in an embodiment of the present invention, Figure 5 A schematic diagram of a DenseBlock module structure of an insulator and a self-exposure defect detection method thereof provided in an embodiment of the present invention.

[0098] In this embodiment, the ConvB module consists of a convolutional layer, a batch normalization (BN) layer, and a ReLU activation function.

[0099] In this embodiment, the InvoB module consists of an involution feature extraction neural network operator (Invo operator), a BN layer and a ReLU activation function; the Invo operator overcomes the limitations of traditional convolution and has the characteristics of being lighter and more efficient. It can improve the computational efficiency while ensuring the accuracy of the model, thus achieving a dual improvement in performance and efficiency. Figure 6 shown.

[0100] In this embodiment, the involution feature extraction neural network operator (Invo) is introduced to replace the traditional convolution operation. It is not only used in the rain removal and fog removal modules, but also integrated into the YOLOv8 backbone network. This operator overcomes the limitations of traditional convolution and has the characteristics of being lighter and more efficient. It can improve the computational efficiency while ensuring the accuracy of the model, thus achieving a double improvement in performance and efficiency.

[0101] In this embodiment, the DenseB lock module is composed of four ConvB modules, and the outputs of these ConvB modules are directly connected to form a dense connection structure, so that the feature map can transmit information at different spatial positions, thereby enhancing the model's understanding and utilization of image features.

[0102] In this embodiment, the high-frequency rainy image is convolved through a ConvB module to extract preliminary features. Then, the output results of the ConvB module are respectively input into three DenseB l ock modules, which further process the features and generate three processed feature maps. Next, the output feature maps of the three DenseB l ock modules are concat-connected and merged to form a feature map containing rich contextual information, which provides a rich information basis for subsequent processing. This merged feature map is then sent to the InvoB module for an inner convolution operation, aiming to restore more details of the image and improve the image quality. Then, the feature map processed by the InvoB module enters another InvoB module again for a second inner convolution operation to further optimize the image quality and clarity. Finally, after completing the processing of the InvoB module, the feature map passes through a ConvB module to obtain the final high-frequency rain-removed image, that is, a clear image output with raindrop interference eliminated.

[0103] In this embodiment, the rain removal module performs rain removal processing on the high-frequency rainy image to obtain a high-frequency rain removal image.

[0104] In this embodiment, the construction of the defect detection model specifically includes:

[0105] Constructing the image defogging module, wherein the image defogging module includes three convolution blocks and four deconvolution blocks;

[0106] The convolution block includes a convolution layer, a batch normalization layer and an activation function;

[0107] The deconvolution block includes an involution feature extraction neural network operator, a batch normalization layer and an activation function.

[0108] Please refer to Figure 7 , Figure 7 A structural schematic diagram of an image defogging module for an insulator and a self-exposure defect detection method thereof provided in an embodiment of the present invention.

[0109] In this embodiment, the defogging network module is composed of three ConvB modules and four InvoB modules, which are used to process and improve the cloud and fog interference in the image;

[0110] In this embodiment, the ConvB module consists of a convolutional layer, a batch normalization (BN) layer, and a ReLU activation function.

[0111] In this embodiment, the InvoB module consists of an involution feature extraction neural network operator (Invo operator), a BN layer and a ReLU activation function; the Invo operator overcomes the limitations of traditional convolution and has the characteristics of being lighter and more efficient. It can improve the computational efficiency while ensuring the accuracy of the model, thus achieving a dual improvement in performance and efficiency. Figure 6 shown.

[0112] In this embodiment, the involution feature extraction neural network operator (Invo) is introduced to replace the traditional convolution operation. It is not only used in the rain removal and fog removal modules, but also integrated into the YOLOv8 backbone network. This operator overcomes the limitations of traditional convolution and has the characteristics of being lighter and more efficient. It can improve the computational efficiency while ensuring the accuracy of the model, thus achieving a double improvement in performance and efficiency.

[0113] In this embodiment, the image defogging module performs defogging processing on the low-frequency foggy image to output a low-frequency defogged image.

[0114] In this embodiment, after obtaining the high-frequency derained image and the low-frequency defogging image, the high-frequency derained image and the low-frequency defogging image are fused into a clear image according to the physical model of the rain and fog image. Specifically:

[0115] I=I H +I L (2)

[0116] Among them, I represents the clear image after removing rain and fog, I H represents the high-frequency derained image, I L Represents a low-frequency dehazed image.

[0117] In this embodiment, the construction of the defect detection model specifically includes:

[0118] Constructing a defect detection module, wherein the defect detection module includes a backbone network, an attention module, and a detection head;

[0119] The attention module is connected to the backbone network, and the attention module includes a channel attention module and a spatial attention module; the detection head includes a Wise IoU loss function.

[0120] In this embodiment, please refer to Figure 8 , Figure 8 A structural schematic diagram of a defect detection module of an insulator and a self-exposure defect detection method thereof provided in an embodiment of the present invention.

[0121] In this embodiment, the backbone network of the defect detection module is a YOLOv8 network model. In the backbone network of the original YOLOv8 network model, the original ConvB module is replaced by an InvoB module, and a CBAM (Convolutional Block Attention Module) module is added before the backbone network SPPF module to obtain an improved backbone network; the DFL (Distance-aware Focus Loss) loss function and the CIOU (Complete Interception over Union) loss function in the Bbox Loss module in the detection head of the original YOLOv8 network model are replaced by the Wiseloud (Weighed Interception over Union) loss function to obtain an improved detection head.

[0122] In this embodiment, the CBAM module is an attention module, which is composed of a channel attention module and a spatial attention module. The CBAM module is introduced to extract the key features of the feature map and suppress noise, thereby improving the accuracy of the model.

[0123] In this embodiment, in the channel attention module, global maximum pooling and global average pooling are first performed on the front input feature map of size H×W×C in the height and width dimensions, where H represents the height, W represents the width, and C represents the number of channels, to obtain two 1×1×C feature maps; then the two feature maps are respectively input into a two-layer neural network, and the output feature maps are element-wise added, and then a channel attention weight vector is generated through a sigmoid activation function; finally, the channel attention weight vector is element-wise multiplied with the original input feature map to obtain the output feature map.

[0124] In this embodiment, in the spatial attention module, the output feature map of the channel attention module is subjected to global maximum pooling and global average pooling to obtain two H×W×1 feature maps, and the two feature maps are spliced ​​in the channel dimension; then, the spliced ​​feature map is subjected to convolution operation and dimensionality reduction processing, and then the spatial attention weight matrix is ​​generated through the si gmo id activation function; finally, the spatial attention weight matrix and the output feature map of the channel attention module are multiplied at the element level to obtain the final output feature map of the CBAM module.

[0125] In this embodiment, the detection head includes a Wise IoU loss function.

[0126] In this embodiment, the Wi se I oU loss function is specifically:

[0127]

[0128] Among them, Loss IoU Represents the Wise IoU loss function value, W i , H i Respectively represent the width and height of the overlapping part, w gt 、h gt They represent the width and height of the target box, respectively, and w and h represent the width and height of the anchor box, respectively.

[0129] In this embodiment, the Wise IoU loss function is used to replace the DFL loss function and the CIOU loss function in the Bbox Loss module in the YOLOv8 detection head. The Wise IoU loss function not only takes into account the differences in relative position and size between objects, but also includes an intelligent weight adjustment mechanism that can adaptively adjust the weight coefficients, thereby improving the flexibility and robustness of object detection.

[0130] In this embodiment, the defect detection model is trained by using a training data set. Specifically, the training data set is input into the defect detection model for iterative training, and in each iteration, a validation set is used to calculate the value of the loss function, and the ADAM algorithm is used to update the weights of the network by back propagation. The weight file with the smallest loss function value is saved as the best weight parameter obtained from training, and these parameters are used to generate a defect detection model.

[0131] Please refer to Fig. 9 , Fig. 9 Another structural schematic diagram of an insulator and a self-exposure defect detection method thereof provided in an embodiment of the present invention.

[0132] In this embodiment, an image data set including insulators and their self-exposure defects taken under rainy and foggy weather conditions is collected; a marking tool is used to perform precise marking work on the insulators and their self-exposure defects in the image, and a data set for training a neural network model is established; a physical model of rainy and foggy images is constructed, and the original images are processed using filters to obtain high-frequency rainy images and low-frequency foggy images; an image deraining module is constructed, and the high-frequency rainy images are input into this module for processing to obtain high-frequency derained images; an image defogging module is constructed, and the low-frequency foggy images are input into this module for processing to obtain low-frequency defogging images; a defect detection module is constructed, and the high-frequency derained images and the low-frequency defogging images are synthesized to obtain a clear image in combination with the physical model of rainy and foggy images, and then input into the defect detection module for processing; the network model including deraining, defogging and defect detection modules is trained using the marked insulator defect image data set; the rainy and foggy weather images to be detected are input into the trained network model to automatically identify and mark the specific locations of the insulators and their self-exposure defects in the image.

[0133] The present invention processes rain and fog images through a filtering module to separate high-frequency images containing raindrops and low-frequency images containing fog. Subsequently, rain removal and fog removal modules are constructed for the two images to generate clear rain-free and fog-free images, and then defect detection is performed on the processed images to identify the specific locations of insulators and their self-exposure defects. The method is suitable for defect detection under adverse weather conditions such as rain and fog, and improves the efficiency and accuracy of detection.

[0134] The embodiment of the present invention also provides an insulator and a self-exposure defect detection device thereof, comprising: a model building module and a detection module;

[0135] The model building module is used to build a defect detection model and train the defect detection model according to the rain and fog image set;

[0136] The detection module is used to identify the specific locations of insulators and their self-exposure defects in the rain and fog image according to the trained defect detection model. Specifically:

[0137] The model building module is used to build a defect detection model and train the defect detection model according to the rain and fog image set;

[0138] The detection module is used to identify the specific locations of insulators and their self-exposure defects in the rain and fog image according to the trained defect detection model. Specifically:

[0139] The defect detection model includes a filtering module, an image deraining module, an image defogging module and a defect detection module;

[0140] Constructing a physical model of a rain and fog image, and filtering the rain and fog image according to the filtering module to generate a high-frequency rain image and a low-frequency fog image;

[0141] Performing rain removal processing on the high-frequency rainy image according to the image rain removal module to obtain a high-frequency rain removal image;

[0142] Performing defogging processing on the low-frequency foggy image according to the image defogging module to obtain a low-frequency defogging image;

[0143] The high-frequency derained image and the low-frequency defogging image are fused into a clear image according to the physical model of the rain-fog image, and the clear image is detected according to the defect detection module to determine the specific positions of the insulator and its self-exposure defects.

[0144] An embodiment of the present invention further provides a computer device, comprising: a processor, a communication interface and a memory, wherein the processor, the communication interface and the memory are interconnected, wherein the memory stores an executable program code, and the processor is used to call the executable program code to execute the insulator and self-exposure defect detection method.

[0145] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the insulator and the self-exposure defect detection method thereof are implemented.

[0146] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. An insulator and a method for detecting self-exposure defects thereof, characterized in that: include: Constructing a defect detection model, and training the defect detection model according to a training data set; According to the trained defect detection model, the specific locations of insulators and their self-exposure defects in the rain and fog image are identified. Specifically: The defect detection model includes a filtering module, an image deraining module, an image defogging module and a defect detection module; Constructing a physical model of a rain and fog image, and filtering the rain and fog image according to the filtering module to generate a high-frequency rain image and a low-frequency fog image; Performing rain removal processing on the high-frequency rainy image according to the image rain removal module to obtain a high-frequency rain removal image; Performing defogging processing on the low-frequency foggy image according to the image defogging module to obtain a low-frequency defogging image; The high-frequency derained image and the low-frequency defogging image are fused into a clear image according to the physical model of the rain-fog image, and the clear image is detected according to the defect detection module to determine the specific positions of the insulator and its self-exposure defects.

2. The insulator and the self-exposure defect detection method thereof according to claim 1, characterized in that: The physical model of the rain and fog image is constructed, and the rain and fog image is filtered according to the filtering module to generate a high-frequency rain image and a low-frequency fog image, specifically: Constructing a physical model of a rain and fog image according to an atmospheric scattering model, wherein the physical model includes a sum of a high-frequency rain image and a low-frequency fog image; The rain and fog image is filtered according to the guided filter to generate a high-frequency rainy image and a low-frequency foggy image.

3. The insulator and the self-exposure defect detection method thereof according to claim 1, characterized in that: The construction of the defect detection model specifically includes: Constructing an image deraining module, wherein the image deraining module includes two convolution blocks, two deconvolution blocks and three dense blocks; The convolution block includes a convolution layer, a batch normalization layer and an activation function; The deconvolution block includes an involution feature extraction neural network operator, a batch normalization layer and an activation function; The dense block includes several connected convolution blocks.

4. The insulator and the self-exposure defect detection method thereof according to claim 3, characterized in that: The step of performing rain removal processing on the high-frequency rainy image according to the image rain removal module to obtain the high-frequency rain removed image is specifically as follows: Extracting a first feature of the high-frequency rainy image according to a first convolution block, and generating three first feature maps according to the three dense blocks and the first feature; Connect and merge the three first feature maps to generate a second feature map; Performing an inner convolution operation on the second feature map according to the two deconvolution blocks to obtain a third feature map; The third feature map is input into the second convolution block to generate a high-frequency derained image.

5. The insulator and the self-exposure defect detection method thereof according to claim 1, characterized in that: The construction of the defect detection model specifically includes: Constructing the image defogging module, wherein the image defogging module includes three convolution blocks and four deconvolution blocks; The convolution block includes a convolution layer, a batch normalization layer and an activation function; The deconvolution block includes an involution feature extraction neural network operator, a batch normalization layer and an activation function.

6. The insulator and the self-exposure defect detection method thereof according to claim 1, characterized in that: The construction of the defect detection model specifically includes: Constructing a defect detection module, wherein the defect detection module includes a backbone network, an attention module, and a detection head; The attention module is connected to the backbone network, and the attention module includes a channel attention module and a spatial attention module; the detection head includes a Wise IoU loss function.

7. The insulator and the self-exposure defect detection method thereof according to any one of claims 1 to 6, characterized in that: Before building the defect detection model, the method further includes: Collect image datasets of insulators and their self-exposure defects, and annotate them to generate training datasets, specifically: Collect rain and fog images containing insulators and their self-exposure defects under rain and fog weather conditions, and perform preprocessing operations on the rain and fog images, wherein the preprocessing operations include data enhancement and data expansion to generate a rain and fog image set; According to the image annotation tool, rectangular frames are drawn on insulators and their self-exposure defects in the rain and fog image to mark them, a training data set is generated, and the training data set is divided into a training set, a validation set and a test set according to a preset ratio.

8. An insulator and a self-exposure defect detection device thereof, characterized in that: include: Model building module and detection module; The model building module is used to build a defect detection model and train the defect detection model according to the rain and fog image set; The detection module is used to identify the specific locations of insulators and their self-exposure defects in the rain and fog image according to the trained defect detection model. Specifically: The defect detection model includes a filtering module, an image deraining module, an image defogging module and a defect detection module; Constructing a physical model of a rain and fog image, and filtering the rain and fog image according to the filtering module to generate a high-frequency rain image and a low-frequency fog image; Performing rain removal processing on the high-frequency rainy image according to the image rain removal module to obtain a high-frequency rain removal image; Performing defogging processing on the low-frequency foggy image according to the image defogging module to obtain a low-frequency defogging image; The high-frequency derained image and the low-frequency defogging image are fused into a clear image according to the physical model of the rain-fog image, and the clear image is detected according to the defect detection module to determine the specific positions of the insulator and its self-exposure defects.

9. A computer device, characterized in that: include: A processor, a communication interface and a memory, wherein the processor, the communication interface and the memory are connected to each other, wherein the memory stores an executable program code, and the processor is used to call the executable program code to execute the insulator and self-exposure defect detection method thereof according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by a processor, the insulator and the self-exposure defect detection method thereof according to any one of claims 1 to 7 are implemented.

Citation Information

Patent Citations

  • The invention discloses an iInsulator self-explosion defect detection method based on an inspection image

    CN109615611A

  • Image rain removal method based on rain and fog separation processing and multi-scale network

    CN115034999A

  • Insulator defect detection method in foggy day scene based on improved YOLOv7 algorithm

    CN116843636A

  • Rainy day insulator defect detection method based on improved MS-PReNet and GAM-YOLOv7

    CN117422689A