High-voltage transmission line insulator defect detection method based on LightWeight-YOLOv8n

By embedding the LightWeight-YOLOv8n model on the drone, combined with the MoblieNetV3 and SimAM attention mechanism, the problem of real-time evaluation of the insulator discharge severity of high-voltage transmission line is solved, and fast and accurate defect detection is achieved, which improves patrol efficiency and power grid Security.

CN119027367APending Publication Date: 2024-11-26CHINA THREE GORGES UNIV
View PDF 0 Cites 5 Cited by

Patent Information

Application Number
CN202410915314.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-09
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The prior art is difficult to accurately evaluate the severity of insulator discharge in high-voltage transmission line in real time on site, resulting in insulated inspection efficiency and the inability to detect potential failure risks in time.

Method used

Using the insulator defect detection method of high-voltage transmission line based on LightWeight-YOLOv8n, the LightWeight-YOLOv8n model is embedded into the drone, and data is collected using high-definition cameras, combined with the MoblieNetV3 backbone network, SimAM attention mechanism and WIoU v3 border loss function, image recognition and Defect detection.

Benefits of technology

It realizes the rapid and accurate identification of insulator defects in high-voltage transmission line on-site, improves detection accuracy and speed, reduces model calculation and occupancy, meets the requirements of transmission line patrol, and ensures the safe and stable operation of the power grid.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027367A_ABST
    Figure CN119027367A_ABST
Patent Text Reader

Abstract

The invention discloses a high-voltage transmission line insulator defect detection method based on LightWeight-YOLOv8n. The high-voltage transmission line insulator defect detection method based on LightWeight-YOLOv8n comprises the following steps: S1, collecting and sorting a collected high-voltage transmission insulator data set by using a high-definition camera of an unmanned aerial vehicle; s2, performing image preprocessing operation on the arranged data set, and expanding the data set; s3, calling Label, labeling the images one by one, labeling the images, and dividing a final data set according to a certain proportion to obtain a high-voltage transmission line insulator data set; s4, configuring an algorithm basic environment, and setting five types of online image enhancement items before starting training so as to ensure the diversity and generalization ability of a data set; s5, the MoblieNetV3 backbone network, the SimAM attention mechanism, the WIoU v3 frame loss function and the SlideLoss classification loss function are combined into the YOLOv8n base line model, and a Light Weight-YOLOv8n model is reconstructed, and the MoblieNetV3 backbone network, the SimAM attention mechanism, the WIoU v3 frame loss function and the SlideLoss classification loss function are combined into the YOLOv8n base line model; the technical problem to be solved by the invention is to provide the LightWeight-YOLOv8n-based high-voltage transmission line insulator defect detection method, and the LightWeight-YOLOv8n-based high-voltage transmission line insulator defect detection method is embedded into an inspection unmanned aerial vehicle so as to improve the detection precision and speed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of power transmission line inspection, and in particular to a method for detecting defects in insulators of high-voltage transmission lines based on LightWeight-YOLOv8n. Background Art

[0002] Insulators are an important part of transmission lines. During long-term operation, they may age and be damaged due to factors such as the environment and power load, resulting in a decrease in insulation performance, which in turn causes discharge problems. Insulator discharge is one of the important causes of transmission line failures and accidents. Therefore, inspectors need to assess the severity of insulator discharge, discover potential fault risks early, and take maintenance measures early to avoid expanding faults and causing fluctuations in the power system. However, the current assessment of the severity of insulator discharge by inspectors mainly relies on video analysis after the inspection or on-site judgment based on experience. These methods are inefficient and may not be able to accurately assess the degree of discharge in real time. Therefore, there is an urgent need for an algorithm that can perform accurate assessments in real time at the edge to assist inspectors in assessing the severity of discharge on site, thereby improving inspection efficiency.

[0003] At present, artificial intelligence is developing rapidly, and the deep learning technology derived from it is also booming. As one of the key categories of deep learning, image recognition algorithms have widely replaced traditional manual inspection methods due to their superior performance. The advantages of deep learning in image recognition are mainly reflected in its ability to automatically learn and extract features, handle complex backgrounds and noise, and its strong generalization ability. Among them, the two-stage algorithms represented by Cascade R-CNN and Faster R-CNN first generate candidate regions, and then classify and regress the candidate regions. The other is represented by the YOLO family and SSD. This algorithm converts the image recognition problem into a regression problem. This method can directly detect targets on the image without generating candidate regions first. The SSD algorithm has a large memory footprint and its speed performance is not as good as YOLO, so the YOLO family algorithm is currently the main research object. The YOLO series of algorithms are developing rapidly. At present, the YOLOv8 target detection algorithm has the advantages of high recognition accuracy and faster speed compared with other algorithms. YOLOV8 is divided into five models: YOLOV8n / s / l / m / x, among which YOLOV8n is the lightest model. This patent mainly makes the model lighter and faster, so YOLOV8n is used as the baseline algorithm and improved on this basis. This improved algorithm has a very broad application prospect and practical significance in the rapid and real-time identification of insulators and their defects in high-voltage transmission lines. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a high-voltage transmission line insulator defect detection method based on LightWeight-YOLOv8n, which is embedded in a patrol drone to improve the detection accuracy and speed.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for detecting defects of insulators of high-voltage transmission lines based on LightWeight-YOLOv8n, comprising the following steps: S1. Use the high-definition camera of the drone to collect and organize the collected high-voltage transmission insulator data set; S2, perform image preprocessing operations on the sorted data set to expand the data set; S3, call Labelme to annotate and label the images one by one, and divide the final data set according to a certain ratio to obtain the high-voltage transmission line insulator data set; S4. Configure the algorithm basic environment and set up 5 types of online image enhancement projects before starting training to ensure the diversity and generalization ability of the data set; S5. Combine the MoblieNetV3 backbone network, SimAM attention mechanism, WIoU v3 border loss function, and SlideLoss classification loss function into the YOLOv8n baseline model and reconstruct it into the LightWeight-YOLOv8n model. S6. Put the obtained data set into the LightWeight-YOLOv8n model, combine the model to identify the acquired insulator image, and use precision (Precision) and mean average precision (mAP) as the detection accuracy evaluation index; the number of frames per second (FPS) is used as the detection speed evaluation index; the floating point operation amount (Gflops), parameter amount (parameter), and model size are used as the evaluation indexes of the lightweight degree of the detection model; S7. Use the trained model to accurately identify the image to be detected.

[0006] Preferably, in step S2, the image preprocessing operation includes histogram equalization, Gaussian filtering, median filtering, and canny operator preprocessing methods.

[0007] Preferably, the step S3 is specifically as follows: S3.1. Call Labelme to annotate the images one by one. The labels are divided into four categories: glass insulator, porcelain insulator, composite insulator, and defect. The final data set is divided into training set, validation set, and test set according to the ratio of 6:3:1. S3.2. Convert the Labelme-annotated xml file into a txt file through the pycharm platform, and divide it into training set, validation set, and test set in the ratio of 6:3:1 to obtain a complete data set of high-voltage transmission line insulators.

[0008] Preferably, the online image enhancement items in step S4 include: Mosaic data enhancement (stopped at the 290th round), Blur data enhancement (p=0.01, blur_limit=(3,7)), MedianBlur data enhancement (p=0.01, blur_limit=(3,7)), Togray data enhancement (p=0.01), CLAHE data enhancement (p=0.01, clip_limit=(1,4.0), tile_grid_size=(8,8)). The online enhancement function is used to improve the recognition accuracy of the image, make the typical features of each category more distinct, and improve the training quality.

[0009] Preferably, step S5 is specifically as follows: all modules in the YOLOv8n backbone network are replaced with the MobileNetV3 network, and the SimAM attention mechanism is inserted into the last layer of the backbone network to form a new MobileNetV3-SimAM backbone network module.

[0010] Preferably, step S5 comprises the following steps: S5.1. Replace the original backbone network with the MobileNetV3 network and insert the SimAM attention mechanism after the MobileNetV3 network to form a lightweight new backbone network. S5.2. Replace all C2f structures in the neck with lightweight CSPPC structures to improve the feature fusion capability of the model on the basis of lightweight; S5.3. Replace the CIoU border loss function with the WIoU v3 border loss function, and replace the VFLoss classification loss function with the SlideLoss classification loss function.

[0011] Preferably, the LightWeight-YOLOv8n model includes a Backbone module, a feature fusion Neck module, and a detection head Head module. The Backbone module is the part of the model used to extract image features, and includes a MobileNetV3 module, a SimAM attention mechanism module, a CSPPC lightweight module, and an SPPF module; SimAM is composed of a channel attention module (CBAM) and a spatial attention module (BAM); the feature fusion Neck is the part of the model used to connect the Backbone module and the detection head Head module, and includes a CBS module, an Upsample layer, and a CSPPC module; wherein the Upsample layer is used for upsampling; the CBS module is composed of standard convolution, normalization, and SiLU activation functions; the CSPPC module is composed of a Dual new lightweight convolution layer and a Partial new lightweight convolution layer; the detection head Head module is the part of the model used to predict the target bounding box and category probability, and includes a WIoU v3 border loss function and a SlideLoss classification loss function.

[0012] Preferably, step S6 is specifically as follows: placing the obtained data set into the LightWeight-YOLOv8n model, identifying the acquired insulator image in combination with the model, and using precision (Precision) and mean average precision (mean average precision, mAP) as detection accuracy evaluation indicators; frames per second (frames per second, FPS) as a detection speed evaluation indicator; floating point operations (Gflops), parameter amount (parameter), model size (model size) as evaluation indicators of the lightweight degree of the detection model.

[0013] The present invention provides a high-voltage transmission line insulator defect detection method based on LightWeight-YOLOv8n, which is higher than the existing detection model in terms of accuracy and speed, has lower missed detection rate and false positive rate for small targets and occluded targets, has strong generalization and robustness, and can meet the requirements of transmission line inspection. It also makes the model lightweight, reduces the amount of model calculation and occupancy, and avoids wasting memory resources. It provides technical support for line inspection and plays an important role in the safe and stable operation of high-voltage transmission lines. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 is a flow chart of the method of the present invention; Figure 2It is the structure diagram of LightWeight-YOLOv8n of the present invention; Figure 3 This is a schematic diagram of the Labelme dataset annotation of the present invention; Figure 4 This is the MobileNetV3 backbone network diagram of the present invention; Figure 5 It is the structural diagram of CSPPC of the present invention; Figure 6 This is an example diagram of the detection results of the present invention. DETAILED DESCRIPTION

[0015] like Figure 1 As shown, a method for detecting defects in insulators of high-voltage transmission lines based on LightWeight-YOLOv8n comprises the following steps: S1. Use the high-definition camera of the drone to collect and organize the collected high-voltage transmission insulator data set; S2, perform image preprocessing operations on the sorted data set to expand the data set; S3, call Labelme to annotate and label the images one by one, and divide the final data set according to a certain ratio to obtain the high-voltage transmission line insulator data set; S4. Configure the basic environment of the algorithm and set up 5 types of online image enhancement projects before starting training to ensure the diversity and generalization ability of the data set.

[0016] S5, such as Figure 2 As shown, the MoblieNetV3 backbone network, SimAM attention mechanism, WIoU v3 border loss function, and SlideLoss classification loss function are combined with the YOLOv8n baseline model and reconstructed into the LightWeight-YOLOv8n model; S6. Put the obtained data set into the LightWeight-YOLOv8n model, and use the model to identify the acquired insulator image, and use precision (Precision) and mean average precision (mAP) as detection accuracy evaluation indicators; frames per second (FPS) as detection speed evaluation indicators; floating point operations (Gflops), parameter, and model size as evaluation indicators of the lightweight degree of the detection model.

[0017] S7. Use the trained model to accurately identify the image to be detected.

[0018] Specifically, step S1 includes: The S1.1 drone is embedded with a high-definition camera to conduct line inspection. The images of high-voltage transmission line insulators are captured to obtain a dataset of high-voltage transmission line insulators. When shooting, the light conditions are good, the background can be complex and diverse, and there are insulators of different categories and defects.

[0019] S1.2 Manually screen out high-resolution insulator images and discard blurry, unclear low-resolution images.

[0020] Specifically, step S2 includes: Perform preprocessing operations on the collected original data set to expand the number of data sets. The preprocessing methods are: histogram equalization, Gaussian filtering, median filtering, and canny operator. Put the preprocessed image into the original data set to obtain the expanded data set.

[0021] Specifically, step S3 includes: S3.1 As Figure 3 As shown in the figure, call Labelme to annotate the images one by one and label them. The labels are divided into four categories: glass insulator, porcelain insulator, composite insulator, and defect. At the same time, the final data set is divided into training set, validation set, and test set according to 6:3:1.

[0054] S3.2 converts the Labelme-annotated xml file into a txt file through the pycharm platform, and divides it into training set, validation set, and test set in the ratio of 6:3:1 to obtain a complete data set of high-voltage transmission line insulators.

[0022] Step S4 includes: The platform used for the test of the present invention is Pycharm11.0.15, the model framework is Pytorch1.12.0, the CPU is 12vCPU Intel(R) Xeon(R) Platinum 8352V CPU @ 2.10GHz, and the GPU is RTX4090, 24GB. The training is set to 300 rounds, and the SGD optimizer is selected.

[0023] Before starting training, set up 5 types of online image enhancement projects to ensure the diversity and generalization ability of the dataset. Online enhancement projects include Mosaic data enhancement (stopped at 290 rounds), Blur data enhancement (p=0.01, blur_limit=(3,7)), MedianBlur data enhancement (p=0.01, blur_limit=(3,7)), Togray processing (p=0.01), CLAHE data enhancement (p=0.01, clip_limit=(1,4.0), tile_grid_size=(8,8)). Online enhancement can improve the recognition accuracy of images, make the typical features of each category more distinct, and improve the quality of training.

[0024] Step S5 includes: S5.1 replaces the original YOLOv8n backbone with the MobileNetV3 lightweight network and adds the SimAM attention mechanism after the SPPF spatial pyramid module. Figure 2 As shown in the figure, in the CBS structure of the LightWeight-YOLOv8n model: C stands for Conv2d, B stands for BatchNorm2d, and S stands for SiLu activation function. The CBS module encapsulates the combined operation of convolution, batch normalization, and activation function. CBH structure: C stands for Conv2d, B stands for BatchNorm2d, and H stands for H-Swish. SPPF structure: first passes through the CBS module, then passes through 1, 2, and 3 BatchNorm2d splicing respectively, and finally passes through a CBS module.

[0025] like Figure 4 As shown in the figure, the MobileNetV3 model uses an inverted residual structure with a linear bottleneck. The inverted residual structure first performs 1×1 convolution to reduce the dimension (compression), then uses 3×3 convolution for feature extraction, and then uses 1×1 convolution to increase the dimension to obtain more feature maps (restoration), then performs residual connection, 3×3 convolution feature extraction, and finally uses 1×1 convolution to obtain more feature maps (upgraded dimension).

[0026] S5.2 As Figure 5 As shown, all C2f structures in the neck are replaced with lightweight CSPPC structures to reduce computational complexity: PConv reduces the number of floating point operations (FLOPs) by performing calculations on fewer channels. For example, if the partial rate is set to 1 / 4, the computation of PConv is only 1 / 16 of that of regular convolution. Reduce memory access: PConv reduces memory access compared to regular convolution, which is particularly beneficial for input / output (I / O)-constrained devices.

[0027] Maintain feature information flow: Although only a portion of the input channels are computed, the retained channels are still useful in the subsequent point-wise convolution (PWConv) layer, allowing feature information to flow in all channels.

[0028] S5.3WIoU v3 is used as a bounding box regression loss, which contains a dynamic non-monotonic mechanism and designs a reasonable gradient gain distribution. This strategy reduces the large or harmful gradients that appear in extreme samples. The loss method calculation pays more attention to samples of normal quality, thereby improving the generalization ability and overall performance of the network model.

[0029] According to the distance metric, the distance attention is constructed, and WIoU v1 with a two-layer attention mechanism is obtained. ; ; ; A non-monotonic focusing coefficient is constructed using β and applied to WIoU v1 to obtain WIoU v3 with dynamic non-monotonic FM. WIoU v3 achieves superior performance by using a judicious gradient gain allocation strategy with dynamic non-monotonic FM.

[0030] ; ; ; WIoUv3 uses a dynamic non-monotonic mechanism to evaluate the quality of anchor boxes, which makes the model pay more attention to anchoring boxes of ordinary quality and improves the model's ability to locate objects. For target detection tasks in drone aerial photography scenes, small objects with a high proportion increase the difficulty of detection. WIoU v3 can dynamically optimize the loss to weight small objects to improve the detection performance of the model.

[0031] like Figure 2As shown, the LightWeight-YOLOv8n model includes a Backbone module, a feature fusion Neck module, and a detection head Head module. The Backbone module is the part of this model used to extract image features, and includes a MobileNetV3 module, a SimAM attention mechanism module, a CSPPC lightweight module, and an SPPF module; SimAM is composed of a channel attention module (CBAM) and a spatial attention module (BAM); the feature fusion Neck is the part of this model used to connect the Backbone module and the detection head Head module, and includes a CBS module, an Upsample layer, and a CSPPC module; the Upsample layer is used for upsampling; the CBS module is composed of standard convolution, normalization, and SiLU activation functions; the CSPPC module is composed of a Dual new lightweight convolution layer and a Partial new lightweight convolution layer; the detection head Head module is the part of this model used to predict the target bounding box and category probability, and includes a WIoU v3 border loss function and a SlideLoss classification loss function.

[0032] S5.4 SlideLoss is a method to solve the imbalance problem between simple and difficult samples through weight functions. Difficult samples are samples that are difficult for the model to correctly identify. Samples may become difficult to identify due to various reasons, such as occlusion, deformation, blur, illumination change, small size or low contrast with the background. In training, these samples will generate higher loss values ​​because it is difficult for the model to give accurate predictions for them. Simple samples, on the contrary, have high resolution and are easy to identify. Solving the problem of sample imbalance is the key to improving the generalization ability of the model. If the model has only seen simple samples for the most part, its performance may degrade when it encounters difficult samples in actual applications. Therefore, various strategies are used to solve this problem, such as resampling (oversampling difficult samples or undersampling simple samples), modifying the loss function (giving higher weights to difficult samples), or designing a new model structure to focus specifically on difficult samples. This patent uses a weight function such as Slide Loss to make the model pay more attention to those difficult samples during training, solving the imbalance problem between simple and difficult samples.

[0033] The formula of Slide Loss sliding loss function is as follows: ; Step S6 includes: After 300 rounds of training in S6.1, the trained best.pt file is obtained for image prediction.

[0034] S6.2 puts the verification set into the best.pt file, combines the model to identify the acquired insulator image, and uses precision (Precision) and mean average precision (mAP) as detection accuracy evaluation indicators; frames per second (FPS) as detection speed evaluation indicators; floating point operations (Gflops), parameter, and model size as evaluation indicators of the lightweight degree of the detection model.

[0035] The trained model is used to identify and locate insulator defect images collected on-site, allowing staff to repair them quickly and accurately.

[0036] CN117893789A improves the Faster RCNN baseline model to accurately identify transmission line insulators. The algorithm used in this model is not advanced, and it cannot detect multiple different types of insulators at the same time. It has weak generalization ability and no certain speed guarantee. CN113095253A uses the improved YOLOV3 algorithm to identify insulator defects, but the memory usage of this model is too large, which will lead to resource waste, and the detection speed is slow, which cannot meet the requirements of fast and real-time inspection of transmission line insulators. Compared with the solutions proposed by CN117893789A and CN113095253A, the present invention can provide faster detection speed, higher detection accuracy, and less model memory usage, providing a more feasible solution for drone inspection.

[0037] Multiple different target detection models were used to verify the same data set, and a comparison result table was obtained. Compared with other models, the model of the present invention has lower memory usage and greatly improves the speed without reducing the recognition accuracy. It has lower missed detection rate and false positive rate for small targets in complex backgrounds, has strong generalization and robustness, and can meet the requirements of power transmission line inspection. The detection results of the LightWeight-YOLOv8n model are shown in the figure below. Figure 6 shown.

[0038] The following is a comparative experiment table

[0039] In summary, the LightWeight-YOLOv8n model constructed in the present invention can quickly and accurately identify insulators and their defects in high-voltage transmission lines, provide technical support for line inspection, and play an important role in the safe and stable operation of the power grid.

[0040] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for detecting defects in insulators of high-voltage transmission lines based on LightWeight-YOLOv8n, characterized in that: The following steps are involved: S1. Use the high-definition camera of the drone to collect and organize the collected high-voltage transmission insulator data set; S2, perform image preprocessing operations on the sorted data set to expand the data set; S3, call Labelme to annotate and label the images one by one, and divide the final data set according to a certain ratio to obtain the high-voltage transmission line insulator data set; S4. Configure the algorithm basic environment and set up 5 types of online image enhancement projects before starting training to ensure the diversity and generalization ability of the data set; S5. Combine the MoblieNetV3 backbone network, SimAM attention mechanism, WIoU v3 border loss function, and SlideLoss classification loss function into the YOLOv8n baseline model and reconstruct it into the LightWeight-YOLOv8n model. S6. Put the obtained data set into the LightWeight-YOLOv8n model, combine the model to identify the acquired insulator image, and use the accuracy and average accuracy mean as the detection accuracy evaluation index; the number of frames transmitted per second is used as the detection speed evaluation index; the floating point operation amount, parameter amount, and model occupancy are used as the evaluation indexes of the lightweight degree of the detection model; S7. Use the trained model to accurately identify the image to be detected.

2. According to the method for detecting defects in high-voltage transmission line insulators based on LightWeight-YOLOv8n in claim 1, it is characterized in that: In step S2, the image preprocessing operation includes histogram equalization, Gaussian filtering, median filtering, and canny operator preprocessing methods.

3. According to the method for detecting defects in high-voltage transmission line insulators based on LightWeight-YOLOv8n in claim 1, it is characterized in that: The step S3 is specifically as follows: S3.

1. Call Labelme to annotate the images one by one and label them. The labels are divided into four categories: glass insulators, porcelain insulators, composite insulators, and defects. The final data set is divided into training set, validation set, and test set according to 6:3:

1. S3.

2. Convert the Labelme-annotated xml file into a txt file through the pycharm platform, and divide it into training set, validation set, and test set in the ratio of 6:3:1 to obtain a complete data set of high-voltage transmission line insulators.

4. According to the method for detecting defects in high-voltage transmission line insulators based on LightWeight-YOLOv8n according to claim 1, it is characterized in that: The online image enhancement items in step S4 include: Mosaic data enhancement, Blur data enhancement, MedianBlur data enhancement, Togray data enhancement, and CLAHE data enhancement. The online enhancement function is used to improve the recognition accuracy of the image, make the typical features of each category more distinct, and improve the training quality.

5. According to the method for detecting defects in high-voltage transmission line insulators based on LightWeight-YOLOv8n in claim 1, it is characterized in that: The step S5 is specifically as follows: all modules in the YOLOv8n backbone network are replaced with the MobileNetV3 network, and the SimAM attention mechanism is inserted into the last layer of the backbone network to form a new MobileNetV3-SimAM backbone network module.

6. According to claim 5, a method for detecting defects in high-voltage transmission line insulators based on LightWeight-YOLOv8n is characterized in that: The step S5 comprises the following steps: S5.

1. Replace the original backbone network with the MobileNetV3 network and insert the SimAM attention mechanism after the MobileNetV3 network to form a lightweight new backbone network. S5.

2. Replace all C2f structures in the neck with lightweight CSPPC structures to improve the feature fusion capability of the model on the basis of lightweight; S5.

3. Replace the CIoU border loss function with the WIoU v3 border loss function, and replace the VFLoss classification loss function with the SlideLoss classification loss function.

7. The method for detecting defects in high-voltage transmission line insulators based on LightWeight-YOLOv8n according to claim 6, characterized in that: The LightWeight-YOLOv8n model includes a Backbone module, a feature fusion Neck module, and a detection head Head module. The Backbone module is the part of this model used to extract image features, and includes a MobileNetV3 module, a SimAM attention mechanism module, a CSPPC lightweight module, and an SPPF module; SimAM is composed of a channel attention module and a spatial attention module; the feature fusion Neck is the part of this model used to connect the Backbone module and the detection head Head module, and includes a CBS module, an Upsample layer, and a CSPPC module; the Upsample layer is used for upsampling; the CBS module is composed of standard convolution, normalization, and SiLU activation functions; the CSPPC module is composed of a Dual new lightweight convolution layer and a Partial new lightweight convolution layer; the detection head Head module is the part of this model used to predict the target bounding box and category probability, and includes a WIoU v3 border loss function and a SlideLoss classification loss function.

8. The method for detecting defects in high-voltage transmission line insulators based on LightWeight-YOLOv8n according to claim 1, characterized in that: The step S6 is specifically as follows: putting the obtained data set into the LightWeight-YOLOv8n model, combining the model to identify the acquired insulator image, and using accuracy and average accuracy mean as detection accuracy evaluation indicators; the number of frames transmitted per second is used as the detection speed evaluation indicator; the floating-point operation amount, parameter amount, and model occupancy are used as evaluation indicators for the lightweight degree of the detection model.

Citation Information

Cited By

  • Lightweight drainage pipeline defect detection method based on improved YOLOv8

    CN120355997A

  • Insulator icing detection and ice melting method based on deep learning

    CN120707520A

  • A Deep Learning-Based Method for Insulator Icing Detection and De-icing

    CN120707520B

  • Improved YOLOv8-based insulator defect detection method in severe weather

    CN120747630A

  • An improved YOLOv8-based insulator defect detection method under severe weather conditions

    CN120747630B