A transmission line strain clamp crimping defect detection method based on improved YOLOv5 algorithm

Abstract

The application discloses a kind of transmission line strain clamp crimping defect detection methods based on improved YOLOv5 algorithm, first infrared imaging is carried out to transmission line strain clamp, the trichromatic value in strain clamp infrared image is used, it is judged whether the strain clamp is abnormal;X-DR imaging is carried out to the strain clamp of abnormal operation, and the X-DR image dataset of strain clamp is constructed;The improved model of YOLOv5 is constructed by CSPDarknet-C2f, BiFPN-M, SPPCSPC-M module, Mish activation function and YoloHead, and further judge the defect type of strain clamp in combination with X-DR image;According to the harm that defect can bring, defect is divided into three categories, such as emergency defect, serious defect and general defect, so as to give corresponding countermeasures;The application not only can intelligently detect the defect of strain clamp, but also greatly reduces the detection cost.

Description

Technical Field

[0001] This invention belongs to the field of power transmission line technology, specifically relating to a method for detecting crimping defects in tension clamps of power transmission lines based on an improved YOLOv5 algorithm. Background Technology

[0002] Tension clamps are a crucial component of power transmission lines, and their operational status directly impacts the stability of the transmission line's operation. Therefore, it is essential to inspect the operational quality of tension clamps. While external defects can be detected through observation, internal defects cannot be observed. With the rapid development of science and technology, X-ray non-destructive testing (XDR) technology has been increasingly applied in many engineering projects. X-rays can penetrate the internal components and display internal information as X-ray images. This method can detect defects without damaging the components. However, this technology is costly; X-ray imaging of a single tension clamp costs approximately 2000 RMB. A tension tower typically has dozens of tension clamps, and defective clamps are relatively few. Performing X-ray non-destructive testing on all tension clamps on a tension tower would be extremely costly. Summary of the Invention

[0003] To address the shortcomings of existing methods for detecting defects in tension clamps, this invention aims to provide a method for detecting crimping defects in transmission line tension clamps based on an improved YOLOv5 algorithm. By combining infrared and X-DR images, the method first uses infrared images to initially identify defective tension clamps, and then uses X-DR detection technology to further determine the defect type, avoiding blind detection. An improved approach to the YOLOv5 algorithm is provided, and the optimal network dimension is derived experimentally. This network can effectively detect the defect types of tension clamps. Based on the potential hazards of each defect type, the defects are further classified into three levels, providing a technical reference for maintenance personnel to detect crimping defects in tension clamps.

[0004] To achieve this objective, the present invention is implemented through the following scheme: a method for detecting crimping defects in transmission line tension clamps based on an improved YOLOv5 algorithm, comprising the following steps:

[0005] S1. Perform infrared imaging on the tension clamp of the transmission line, and use the three primary color values ​​in the infrared image of the tension clamp to determine whether the working status of the tension clamp is abnormal.

[0006] S2. Perform X-DR imaging on the abnormally functioning tension clamps and construct an X-DR image dataset of tension clamps.

[0007] S3. An improved YOLOv5 model was constructed using CSPDarknet-C2f, BiFPN-M, SPPCSPC-M modules, the Mish activation function, and YOLOHead. The network dimension was adjusted experimentally to obtain the optimal network dimension. The defect type of the tension clamp was further determined by combining X-DR images. The improved model was used to detect the crimping defects of the tension clamp.

[0008] S4. Based on the potential harm caused by the defect, classify the defect into three categories: urgent defect, critical defect, and general defect, and provide corresponding countermeasures.

[0009] Furthermore, in step S1, when performing infrared imaging on the tension clamp, the imaging should be focused on the tension clamp area as much as possible to avoid the influence of other instruments on the imaging; when obtaining the approximate range of the three primary color values ​​in the infrared image of the defective tension clamp, no less than 20 sets of defective tension clamps should be subjected to infrared imaging, and the values ​​of R, G, and B in the infrared images of the 20 sets of defective tension clamps should be obtained respectively; the range of the primary color R in these 20 sets of data [R min ~R max ], the size range of the primary color G [G min ~G max and the size range of the primary color B [B min ~B max The range of the three primary colors is defined as the defect interval U. If the values ​​of the three primary colors in the infrared image of the tension clamp to be tested belong to the defect interval U, the working state of the tension clamp is determined to be abnormal; otherwise, the working state is normal. In addition, when obtaining the defect interval U, the defective tension clamp removed from the tension tower can be used as the target for infrared imaging, which not only saves the imaging difficulty but also obtains the defect interval U more accurately.

[0010] Furthermore, in step S2, X-DR imaging is performed on the abnormally functioning tension clamps. When constructing the tension clamp X-DR image dataset, the defective tension clamps removed from the tension tower can be used as targets to investigate the influence of their model, size, and X-ray imaging parameters on the imaging quality. The tension clamp X-ray image dataset is constructed by using an X-ray source, imaging plate, and computer system to build a digital X-ray imaging system. This system is used to perform X-ray digital imaging on the removed defective tension clamps and collect imaging data.

[0011] Furthermore, in step S3, an improved YOLOv5 model is constructed using CSPDarknet-C2f, BiFPN-M, SPPCSPC-M modules, the Mish activation function, and YOLOHead. The optimal network dimension is determined through experimental adjustments. The type of defect in the tension clamp is further determined by combining X-DR images, specifically including:

[0012] S3.1 Basic modules required to build the network: The Bottleneck module consists of two 3×3 convolutions and one residual edge; the C2f structure consists of 1×1 convolutions, n Bottleneck modules and n+1 residual edges, where the number of n determines the depth of the network; the CBM structure consists of ordinary convolution Conv, normalized BN and activation function Mish.

[0013] S3.2. Set the input image size to a multiple of 32. The image is then processed through Focus and CBM structures to perform preliminary feature extraction. The extracted feature layers are then processed through four C2f structures to complete the construction of the backbone network CSPDarknet-C2f. The activation function in the feature fusion network BiFPN is replaced with Mish to construct BiFPN-M. The activation function in the CSP pooling network is replaced with Mish to construct SPPCSPC-M. SPPCSPC-M modules are added to the backbone network CSPDarknet-C2f and the feature fusion network BiFPN-M. Based on the four feature layers output by the feature fusion network BiFPN-M, a corresponding YoloHead is added. At this point, the construction of the improved YOLOv5 model is complete, and the network depth is set to 1.33 and the network width is set to 1.25.

[0014] S3.3 When further determining the defect type of the tension clamp by combining X-DR images, the defect types are subdivided into five categories: under-pressure groove, leakage groove, steel anchor burrs, steel anchor bending, and loose strands of aluminum strand.

[0015] Furthermore, when constructing the improved YOLOv5 model using CSPDarknet-C2f, BiFPN-M, SPPCSPC-M modules, the Mish activation function, and YOLOHead in step S3, the input feature layer sizes must be consistent. If the feature layer sizes are different, they can be adjusted using 3×3 convolutions, and then the number of channels can be adjusted using 3×3 convolutions.

[0016] Furthermore, in step S3, the optimal network dimension is obtained. This optimal network dimension is selected based on the four network dimension sizes set according to the present invention, using the mAP value and FPS value from the experimental results.

[0017] Further, in step S3, a backbone network CSPDarknet-C2f is constructed. The input image size is set to a multiple of 32. The image first passes through a Focus structure, which halves the size of the feature layer and quadruples the dimension. Then it passes through a CBM structure, which uses a 1×1 convolution size to adjust the number of channels in the feature layer. It passes through four C2f structures at a time. Each time it passes through a structure, the feature layer size halves the size and the dimension doubles the dimension.

[0018] Furthermore, in step S4, defects are classified into three categories: emergency defects, serious defects, and general defects. Specifically, the following are classified as emergency defects: groove underpressure and steel anchor bending are classified as serious defects; groove underpressure is classified as serious defects; steel anchor flash and aluminum strand loosening are classified as general defects.

[0019] Compared with the prior art, the beneficial effects of the present invention include:

[0020] This invention provides a method for detecting crimping defects in tension clamps of transmission lines based on an improved YOLOv5 algorithm. It combines infrared and X-DR images. Infrared images are used to initially identify defective tension clamps, and then X-DR detection technology is used to further determine the defect type, avoiding blind detection and significantly reducing detection costs. An improved approach to the YOLOv5 algorithm is provided, and the optimal network dimension is derived experimentally. This network can effectively detect the defect type of tension clamps. Based on the potential hazards of each defect type, the defects are further classified into three levels, providing a technical reference for maintenance personnel to detect crimping defects in tension clamps. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention;

[0022] Figure 2 This is a structural diagram of the improved YOLOv5 algorithm used in this invention. Detailed Implementation

[0023] The flowchart of the implementation of this invention is as follows: Figure 1 As shown below, the embodiments will be described in detail. A method for detecting crimping defects in transmission line tension clamps based on an improved YOLOv5 algorithm includes the following steps:

[0024] S1. Perform infrared imaging on the tension clamps of the transmission line. Use the three primary color values ​​in the infrared image of the tension clamp to determine whether the working status of the tension clamp is abnormal. When performing infrared imaging on the tension clamp, the image should be focused on the tension clamp area as much as possible to avoid the influence of other instruments on the imaging. When obtaining the approximate range of the three primary color values ​​in the infrared image of the defective tension clamp, use the defective tension clamp removed from the tension tower as the target for infrared imaging to construct an infrared image dataset. Take no less than 20 sets of defective tension clamps for infrared imaging and obtain the R, G, and B values ​​in the infrared images of the 20 sets of defective tension clamps respectively. Record the range of the primary color R in these 20 sets of data [R min ~R max ], the size range of the primary color G [G min ~G max and the size range of the primary color B [B min ~B max The range of the three primary colors is defined as the defect interval U. If the values ​​of the three primary colors in the infrared image of the tension clamp to be tested belong to the defect interval U, the working state of the tension clamp is determined to be abnormal; otherwise, the working state is normal.

[0025] S2. Perform X-DR imaging on the abnormally operating tension clamps to construct a tension clamp X-DR image dataset. When constructing the tension clamp X-DR image dataset, defective tension clamps removed from the tension tower can be used as targets to investigate the influence of their model, size, and X-ray imaging parameters on the imaging quality. Construct a tension clamp X-ray image dataset by using an X-ray source, imaging plate, and computer system to build a digital X-ray imaging system. Use this system to perform X-ray digital imaging on the removed defective tension clamps and collect imaging data.

[0026] S3, such as Figure 2 As shown, an improved YOLOv5 model was constructed using CSPDarknet-C2f, BiFPN-M, SPPCSPC-M modules, the Mish activation function, and YOLOHead. The network dimension was adjusted experimentally to obtain the optimal network dimension. The defect type of the tension clamp was further determined by combining X-DR images. The improved model was used to detect the crimping defects of the tension clamp.

[0027] S3.1. Construct the original YOLOv5 model using CSPDarknet, SPP, PANet, and YOLOHead. Set the input image size to 640×640×3. After passing through a Focus structure, the feature layer size becomes 320×320×12. Then, after convolution, normalization, and SiLU activation, the image size becomes 320×320×64. At this point, after passing through four large residual modules CSP_Layer consisting of CSP structures and convolution, normalization, and SiLU activation, the size of the image feature layer is further compressed. The dimensions are further increased by shrinking the original size; and an SPP structure is added to the last large residual block to complete the construction of the backbone network CSPDarknet; three feature layers F1, F2 and F3 with sizes of 80×80×256, 40×40×512 and 20×20×1024 are taken from the backbone network, and PANet modules are constructed using bottom-up and top-down FPN and PAN structures; finally, three YoloHead modules with the same size as F1, F2 and F3 are used to construct the YOLOv5 decoding network; at this point, the basic network construction of YOLOv5 is completed;

[0028] S3.2. CSPDarknet, VGG, Resnet and Efficientnet were used as the backbone networks of YOLOv5 respectively to test the detection performance of different backbone networks. By comparing the mAP values ​​of different backbone networks, CSPDarknet was finally selected as the backbone network of YOLOv5.

[0029] S3.3. Based on the selection of CSPDarknet as the backbone network of YOLOv5, FPN, PAN, FPN+PAN and BiFPN are used as feature fusion networks of YOLOv5 respectively. By comparing the mAP values ​​of different feature fusion networks, BiFPN is finally selected as the feature fusion network of YOLOv5.

[0030] S3.4. Based on the modules selected in S3.2 and S3.3, the SPPCSPC module is added to the backbone network CSPDarknet and the feature fusion network BiFPN to further improve the model.

[0031] S3.5. Based on S3.4, the large residual module CSP_Layer in the backbone network is replaced with the C2f module to construct the backbone network CSPDarknet-C2f. The mAP values ​​before and after the replacement are compared through experiments, and finally CSPDarknet-C2f is selected as the model backbone.

[0032] S3.6. Based on S3.5, SiLU, Mish, ReLU, and LeakyReLU are used as activation functions for YOLOv5 respectively. By comparing the mAP value and detection speed FPS of different activation functions, Mish is finally selected as the activation function for YOLOv5. The Mish activation function is also used as the activation function of the feature fusion network BiFPN to construct the BiFPN-M structure. Based on the number of feature layers output by BiFPN-M, a YoloHead is added to construct the network. At this point, all improvements to the YOLOv5 network are completed.

[0033] S3.7 Since the network dimension has a significant impact on the detection results, this invention tested four different network dimensions based on S3.6, and finally selected a network depth of 1.33 and a network width of 1.25 as the final dimension of the improved YOLOv5 network. Statistical analysis was performed on the X-ray images of tension clamps, and the crimping defects were subdivided into five categories: under-crimping groove, groove leakage, steel anchor burrs, steel anchor bending, and loose strands of aluminum strands. The LabelImg tool was used for annotation, and the annotated dataset images were input into the improved YOLOv5 network for model training and testing to obtain the model training weights.

[0034] S3.8 For tension clamps that show abnormalities in infrared imaging, X-DR images are used for further detection; after obtaining the X-DR image of the abnormal tension clamp, the trained model weights are input into the improved YOLOv5 model to detect the specific type of crimping defect.

[0035] S4. Based on the potential harm caused by the defects, the defects are classified into three categories: emergency defects, serious defects, and general defects. The groove leakage and steel anchor bending are classified as emergency defects, the groove underpressure is classified as serious defects, and the steel anchor burrs and aluminum strand loosening are classified as general defects. Corresponding countermeasures are given.

[0036] The above description merely illustrates preferred embodiments of the present invention, and while the description is relatively specific and detailed, it should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications, improvements, and substitutions without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A method for detecting crimping defects in tension clamps of transmission lines based on an improved YOLOv5 algorithm, characterized in that, Includes the following steps: S1. Perform infrared imaging on the tension clamp of the transmission line, and use the three primary color values ​​in the infrared image of the tension clamp to determine whether the working status of the tension clamp is abnormal. S2. Perform X-DR imaging on the abnormally functioning tension clamps and construct an X-DR image dataset of tension clamps. S3. An improved YOLOv5 model was constructed using CSPDarknet-C2f, BiFPN-M, SPPCSPC-M modules, the Mish activation function, and YoloHead. The optimal network dimension was determined through experiments, including: setting the input image size to a multiple of 32; the image was then processed sequentially through Focus and CBM structures to initially extract image features; the initially extracted feature layers were then processed sequentially through four C2f structures to construct the backbone network CSPDarknet-C2f; and the feature fusion network was then... The activation function in BiFPN is replaced with Mish to construct BiFPN-M; the activation function in the CSP pooling network is replaced with Mish to construct SPPCSPC-M; an SPPCSPC-M module is added between the backbone network CSPDarknet-C2f and the feature fusion network BiFPN-M; based on the four feature layers output by the feature fusion network BiFPN-M, four YOLOHeads are set accordingly. At this point, the construction of the improved YOLOv5 model is completed, and the network depth is set to 1.33 and the network width is set to 1.

25. An improved model was used to detect crimping defects in tension clamps. Based on the potential hazards of the defects, they were classified into three categories: emergency defects, serious defects, and general defects, and corresponding countermeasures were given. S4. Further determine the type of defect in the tension clamp by combining the X-DR image.

2. The method for detecting crimping defects in transmission line tension clamps based on the improved YOLOv5 algorithm according to claim 1, characterized in that, Step S1, which involves using the three primary color values ​​from the infrared image of the tension clamp to determine whether the working status of the tension clamp is abnormal, specifically includes: S1.1 Obtain the approximate range of the three primary color values ​​in the infrared images of the defective tension clamps: Take no less than 20 sets of defective tension clamps for infrared imaging, and obtain the values ​​of R, G, and B in the infrared images of the 20 sets of defective tension clamps respectively; record the range of the primary color R in these 20 sets of data [R min ~R max ], the size range of the primary color G [G min ~G max and the size range of the primary color B [B min ~B max ]; S1.

2. The range of the three primary colors is defined as the defect interval U. If the values ​​of the three primary colors in the infrared image of the tension clamp to be tested belong to the defect interval U, the working state of the tension clamp is determined to be abnormal; otherwise, the working state is normal.

3. The method for detecting crimping defects in transmission line tension clamps based on the improved YOLOv5 algorithm according to claim 1, characterized in that, In step S3, an improved YOLOv5 model is constructed using CSPDarknet-C2f, BiFPN-M, SPPCSPC-M modules, the Mish activation function, and YOLOHead. The network dimension is adjusted experimentally to obtain the optimal network dimension. The defect type of the tension clamp is further determined by combining X-DR images, specifically including: S3.1, Basic modules required to build the network: The Bottleneck module consists of two 3×3 convolutions and a residual edge; ... n Each Bottleneck module and n The residual edge of +1 constructs the C2f structure, where n The number of elements determines the depth of the network. The CBM structure is constructed from ordinary convolutional Conv, normalized BN, and activation function Mish. S3.2 Detect tension clamp crimping defects using an improved model. Based on the potential hazards of the defects, classify them into three categories: emergency defects, serious defects, and general defects, and provide corresponding countermeasures. For example, classify groove under-crimping and steel anchor bending as emergency defects, groove under-crimping as serious defects, and steel anchor flash and aluminum strand loosening as general defects.

4. The method for detecting crimping defects in transmission line tension clamps based on the improved YOLOv5 algorithm according to claim 3, characterized in that, In step S4, the X-DR image is used to further determine the type of defect in the tension clamp. Specifically, the X-DR image is used to further determine the type of defect in the tension clamp, and the defect type is subdivided into five categories: under-pressure groove, groove leakage, steel anchor burrs, steel anchor bending, and loose strands of aluminum strand.

Images