An overhead transmission line conductor fitting defect identification method and system

Abstract

The application discloses an overhead transmission line conductor hardware defect identification method and system, which can not only significantly improve the detection speed, but also reduce the influence of shadows, illumination and complex backgrounds in the transmission conductor and hardware images on image defect identification, improve the detection accuracy of the algorithm, and has great reference for subsequent image analysis and fault identification. It comprises the following steps: collecting image data of the transmission line conductor and hardware in the detection area, and generating a data set; an improved YOLOv7 model is constructed, the original YOLOv7 backbone network of the YOLOv7 model is replaced by an EfficientNetV2 backbone network, and the original PANet module of the YOLOv7 model is replaced by a BiFPN module; the improved YOLOv7 model is used to train the data set to obtain a model weight file; and the model weight file is input into the improved YOLOv7 model to identify defects of the image of the transmission line conductor and hardware in the detection area.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method and system for identifying defects in conductor fittings of overhead transmission lines. Background Technology

[0002] High-voltage transmission lines and fittings are important components of the power system, and their operating conditions have a significant impact on the safe and stable operation of the power system.

[0003] Due to the complex weather conditions along the routes, the unique geographical distribution, and the wide coverage of transmission lines, they are susceptible to extreme weather events, resulting in faults such as wire jumpers, broken or scattered conductor strands, and damaged fittings, posing significant threats to the safe and reliable operation of the power grid. In recent years, accidents caused by faults in transmission line conductors and fittings have also shown an increasing trend due to severe weather. Therefore, actively exploring and researching fault detection technologies and preventative measures for transmission line conductors and fittings is of great significance for the safe and reliable operation of the power grid and the safety of people's property.

[0004] Existing methods for detecting defects in transmission lines and fittings can be mainly divided into two categories: traditional image-based methods and deep learning-based methods. Traditional image-based methods have low computational cost and fast processing speed, but their accuracy is significantly reduced due to the complex and variable operating environment of transmission lines and fittings. Images are easily affected by weather, lighting, background, shooting angle, and terrain, resulting in less than ideal performance. On the other hand, deep learning-based methods, while possessing strong anti-interference capabilities and high accuracy, suffer from slow processing speeds and cannot quickly identify conductor and fitting defects from a large number of transmission line inspection images, thus limiting their practicality. Therefore, how to quickly and accurately identify conductor and fitting defects from images has become a key focus of research in transmission line fault diagnosis. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention provides a method and system for identifying defects in overhead transmission line conductor fittings. This method not only significantly improves the detection speed but also reduces the impact of shadows, lighting, and complex backgrounds in images of transmission lines and fittings on defect identification, thereby improving the detection accuracy of the algorithm. It also provides valuable insights for subsequent image analysis and fault identification.

[0006] This invention is achieved through the following technical solution:

[0007] A method for identifying defects in conductor fittings of overhead transmission lines includes the following steps:

[0008] Collect image data of power line conductors and fittings within the detection area and generate a dataset;

[0009] An improved YOLOv7 model was constructed by replacing the original YOLOv7 backbone network with the EfficientNetV2 backbone network and replacing the original PANet module with the BiFPN module.

[0010] The improved YOLOv7 model was used to train the dataset, and the model weight file was obtained.

[0011] The model weight file is input into the improved YOLOv7 model to identify defects in images of power line conductors and fittings within the detection area.

[0012] Furthermore, the image data of the transmission line conductors and fittings within the acquisition and detection area includes:

[0013] Images of power transmission line conductors and fittings are acquired from any angle using image acquisition equipment mounted on a drone, the image size and aspect ratio are determined, and the images are scaled down proportionally.

[0014] Furthermore, after acquiring image data of the transmission line conductors and fittings within the detection area, the method further includes:

[0015] The acquired images are mapped to ground truth bounding boxes, and a labeling file containing location and species information is generated.

[0016] Furthermore, the generated dataset includes:

[0017] The acquired images were synchronously augmented, and the augmented data was divided into training set and dataset, and a training dataset in COCO format was created.

[0018] Furthermore, the synchronous augmentation of the acquired image includes flipping the image along the x-axis, y-axis, and origin, and then adding salt-and-pepper noise, Gaussian noise, increasing brightness, and decreasing brightness to the flipped image.

[0019] Furthermore, the EfficientNetV2 backbone network adopts a Fused-MBConv structure in the shallow network and an MBConv structure in the deep network;

[0020] The EfficientNetV2 backbone network incorporates a self-attention mechanism.

[0021] Furthermore, the step of training the dataset using the improved YOLOv7 model to obtain the model weight file includes the following steps:

[0022] Based on the PyTorch deep learning framework, the Adamw optimizer is used to approximate the optimal solution, and then the SGD optimizer is used to adjust the improved YOLOv7 model.

[0023] Transfer learning was used to update the weights of the improved YOLOv7 model, generating a model file for defect detection of transmission line conductors and fittings. The learning rate was adjusted using a preheating and cosine annealing method, and the model weight file was generated after training.

[0024] Furthermore, after obtaining the model weight file, the method further includes evaluating the model weight file, wherein the evaluation metrics include accuracy and frame rate.

[0025] Furthermore, the specific steps for inputting the model weight file into the improved YOLOv7 model to perform defect identification on the images of transmission line conductors and fittings within the detection area are as follows:

[0026] Read the model weight file and input it into the improved YOLOv7 model for prediction and identification. Obtain the target location, type name, and confidence level of the defect, and mark it in the image. Defect identification is then complete.

[0027] A defect identification system for conductor fittings of overhead transmission lines includes:

[0028] The data acquisition module is used to collect image data of power transmission line conductors and fittings within the detection area and generate a dataset;

[0029] The model building module is used to build an improved YOLOv7 model. It replaces the original YOLOv7 backbone network with the EfficientNetV2 backbone network and replaces the original PANet module with the BiFPN module.

[0030] The training module is used to train the dataset using the improved YOLOv7 model and obtain the model weight file;

[0031] The identification module is used to input the model weight file into the improved YOLOv7 model to identify defects in images of transmission line conductors and fittings within the detection area.

[0032] Compared with the prior art, the present invention has the following beneficial technical effects:

[0033] This invention designs a method for identifying defects in overhead transmission line conductor fittings. It utilizes a drone equipped with a high-definition camera to acquire images of the transmission lines and fittings. Based on the YOLOv7 network structure, a lightweight self-attention network is proposed, combining EfficientNetV2 and CA modules to replace the YOLOv7 backbone network in the original YOLOv7 convolutional neural network model. This encodes the target location information, enabling faster and more accurate feature extraction. Simultaneously, the original PANet in the YOLOv7 convolutional neural network model is replaced with a BiFPN module, allowing for rapid and efficient fusion of extracted features, ensuring the output features contain both location and semantic information. This not only significantly improves detection speed but also reduces the impact of shadows, lighting, and complex backgrounds in the images of transmission lines and fittings on defect identification, enhancing the algorithm's detection accuracy. This method offers valuable insights for subsequent image analysis and fault identification. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1 This is a flowchart of a method for identifying defects in conductor fittings of overhead transmission lines according to the present invention;

[0036] Figure 2 This is a schematic diagram of the MBConv structure in the EfficientNetV2 network described in this invention;

[0037] Figure 3 This is a schematic diagram of the Fused-MBConv structure in the EfficientNetV2 network described in this invention;

[0038] Figure 4 This is a schematic diagram of the CA self-attention mechanism described in this invention;

[0039] Figure 5 This is a schematic diagram of the structure of PANet described in this invention;

[0040] Figure 6 This is a schematic diagram of the BiFPN structure described in this invention. Detailed Implementation

[0041] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0042] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0043] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0044] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection, an electrical connection, or a communication connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0045] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature being directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature being directly above or diagonally above the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0046] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0047] This invention provides a method for identifying defects in conductor fittings of overhead transmission lines, comprising the following steps:

[0048] Collect image data of power line conductors and fittings within the detection area and generate a dataset;

[0049] An improved YOLOv7 model was constructed by replacing the original YOLOv7 backbone network with the EfficientNetV2 backbone network and replacing the original PANet module with the BiFPN module.

[0050] The improved YOLOv7 model was used to train the dataset, and the model weight file was obtained.

[0051] The model weight file is input into the improved YOLOv7 model to identify defects in images of power line conductors and fittings within the detection area.

[0052] This invention designs a method for identifying defects in overhead transmission line conductor fittings. It utilizes a drone equipped with a high-definition camera to acquire images of the transmission lines and fittings. Based on the YOLOv7 network structure, a lightweight self-attention network is proposed, combining EfficientNetV2 and CA modules to replace the YOLOv7 backbone network in the original YOLOv7 convolutional neural network model. This encodes the target location information, enabling faster and more accurate feature extraction. Simultaneously, the original PANet in the YOLOv7 convolutional neural network model is replaced with a BiFPN module, allowing for rapid and efficient fusion of extracted features, ensuring that the output features contain both location and semantic information. Compared with existing algorithms for detecting defects in transmission lines and fittings, this invention, through lightweight network technology, effectively reduces computational complexity by extracting defect areas in the conductors and fittings, thereby effectively reducing accidents caused by conductor and fitting faults in transmission lines and providing a favorable guarantee for the safe operation of the lines. It features high accuracy and real-time detection. The defect identification method described in this invention is simple in principle, intuitive, and easy to implement, providing a new detection approach for the safe and stable operation of transmission lines.

[0053] This invention also provides a defect identification system for overhead transmission line conductor fittings, used to implement the defect identification method described in this invention, comprising:

[0054] The data acquisition module is used to collect image data of power transmission line conductors and fittings within the detection area and generate a dataset;

[0055] The model building module is used to build an improved YOLOv7 model. It replaces the original YOLOv7 backbone network with the EfficientNetV2 backbone network and replaces the original PANet module with the BiFPN module.

[0056] The training module is used to train the dataset using the improved YOLOv7 model and obtain the model weight file;

[0057] The identification module is used to input the model weight file into the improved YOLOv7 model to identify defects in images of transmission line conductors and fittings within the detection area.

[0058] like Figure 1 As shown, the technical solution adopted in this invention is implemented according to the following steps:

[0059] Step 1: Acquire images of power transmission line conductors and fittings from any angle using image acquisition equipment installed on the drone. n Determine the image size and aspect ratio, and then scale the image down proportionally.

[0060] Step 2: Collect normal and defective images of transmission lines and fittings taken in Step 1, mark the ground true bounding box, and generate a labeling file containing location and type information;

[0061] Step 3: Simultaneously augment the images and annotation files from Step 2 to increase the number of training samples and improve the network's robustness against interference. Divide the augmented dataset into a training set and a dataset in an 8:2 ratio to create a training dataset in COCO format.

[0062] Step 4: Train the training dataset generated in Step 3 using the improved YOLOv7 model, obtain the model weight file, and evaluate it.

[0063] Step 5: Input the model weight file generated in Step 4 into the prediction network of the improved YOLOv7 model to perform defect detection on the images of power transmission lines and fittings taken in Step 1.

[0064] Furthermore, in step 1, the step of proportionally reducing the image is as follows:

[0065] Step 1.1: Acquire images of the power transmission line conductors and fittings from any angle using image acquisition equipment installed on the drone. n ;

[0066] Step 1.2, let I n The dimensions are W and H. Images taken by drones are typically 4000*3000 pixels with an aspect ratio of W / H. The scaled-up image is I. nThe dimensions are w and h, typically 800*600, with a width-to-height ratio of w / h, ensuring that W / H:w / h = 1.

[0067] Step 1.3, I n Divide the data into n disjoint blocks of n pixels, and calculate n according to formula (1-1);

[0068] n=W×H / (w×h) (1-1)

[0069] Step 1.4: Calculate the average value of the n pixels within each small block; this value is used as I. n 'Corresponding pixel value'.

[0070] Furthermore, in step 2, the steps for calibrating the ground truth bounding box are as follows:

[0071] Use the visual image annotation tool Labelme to annotate the image with bounding boxes. Specifically, use rectangles to mark the locations of defects in the image and add class names. Generate a JSON format information file.

[0072] Furthermore, in step 3, the specific steps for synchronously expanding the image and annotation file obtained in step 2 are as follows:

[0073] Step 3.1, for the image I after being reduced in step 1 n To expand, firstly, I... n Perform x-axis flip, y-axis flip, and origin flip on the flipped image. Then, add salt and pepper noise, Gaussian noise, increase brightness, and decrease brightness to the flipped image. After expansion, a single image becomes 20 images.

[0074] Step 3.2, expand the JSON file generated by the annotations in Step 2:

[0075] After the image flipping operation, the corresponding positional annotation information in the image should be symmetrically operated. For x-axis flipping and y-axis flipping, first calculate the distance d from the annotation point to the horizontal and vertical center lines. The value of d is positive if the annotation point is above the center line, and negative otherwise. Then update the symmetrical distance d' to the opposite of d. Finally, calculate the coordinate position of the new annotation point based on the center line distance d'. Iterate through each annotation point and generate a new JSON file.

[0076] To flip the image origin, first calculate the distances d and s from the labeled point to the horizontal and vertical midline, respectively. Then update the symmetrical distances d' and s' to the opposite of d and s, calculate the new labeled point coordinates, and iterate to generate a JSON file.

[0077] Adding salt-and-pepper noise, Gaussian noise, increasing brightness, or decreasing brightness to an image does not change the positional information in the JSON file. Therefore, you only need to change the filename value in the JSON file to generate a new JSON file.

[0078] Further, in step 4, the improved YOLOv7 model is used to train the dataset generated in step 3, and the model weight file is obtained and evaluated. The specific steps are as follows:

[0079] Step 4.1: Replace the YOLOv7 backbone network.

[0080] The YOLOv7 backbone is replaced with the more lightweight EfficientNetV2 backbone. The network uses a NAS search function to explore the optimal match between input image size, network depth, and kernel size, maximizing model accuracy and inference speed. EfficientNetV2 uses a Fused-MBConv structure in shallow layers and an MBConv structure in deeper layers. This increased network depth avoids the problem of reduced receptive field caused by increasing kernel size. A schematic diagram of the MBConv structure is shown below. Figure 2 As shown, the Fused-MBConv structure is as follows: Figure 3 As shown.

[0081] Step 4.2: Optimize the EfficientNetV2 backbone network.

[0082] The convolution operation in convolutional neural networks fixes the kernel size, resulting in a fixed receptive field and an inability to focus on global image information. Adding a self-attention mechanism to the EfficientNetV2 backbone network improves the network's detection performance by obtaining global spatial information from the feature map through simple queries and assignments.

[0083] The CA self-attention mechanism embeds positional information into channel attention, capturing cross-channel information as well as orientation and positional awareness, thus enhancing the expressive power of image features. The CA module consists of two parts: coordinate information embedding and coordinate attention generation. First, to better preserve target positional information, a one-to-one feature encoding is used to encode each channel along the horizontal and vertical directions after the image is input into the CA structure. Therefore, the outputs of the c-th channel in the horizontal and vertical directions are shown in equations (4-1) and (4-2), respectively. Where z... c is the encoded value output by the c-th channel, where w and h are the width and height of the image.

[0084]

[0085]

[0086] Then, the feature maps from the two directions are concatenated, and the intermediate features are extracted. Finally, the feature maps are divided into two tensors along the horizontal and vertical directions, processed by convolution and activation functions, and then fused to generate coordinate attention. A schematic diagram of the CA self-attention mechanism is shown below. Figure 4 As shown, the CA module is introduced into the EfficientnetV2 backbone network to obtain a new backbone network.

[0087] Step 4.3, Improve the network feature fusion stage

[0088] YOLOv7 uses PANet as its feature fusion structure, which aggregates features from different levels through a bidirectional fusion network that operates from top to bottom and bottom to top. However, due to the complexity of PANet's structure, while achieving good results, it also increases computational cost. In this invention, to further improve the model's inference speed, a more efficient and faster BiFPN structure is used for feature fusion. Compared to PANet, BiFPN removes the top and bottom nodes of the intermediate layer, reducing computational cost, and adds skip connections, fusing more features. A schematic diagram of the PANet structure is shown below. Figure 5 As shown in the diagram, the structure of BiFPN is as follows: Figure 6 As shown.

[0089] Step 4.4, Training Process and Evaluation Indicators

[0090] The improved YOLOv7 model after steps 4.1, 4.2, and 4.3 was trained using the PyTorch deep learning framework. The images in the training dataset were all 800*600px in size. During training, the Adamw optimizer was first used to quickly approximate the optimal solution to reduce network training time and accelerate optimization. Then, the SGD optimizer was used to fine-tune the model to minimize the loss function. Simultaneously, to further accelerate training and improve accuracy, transfer learning was used to update the weights on the pre-trained model, generating a model file for detecting defects in power transmission lines and fittings. The learning rate was adjusted using a preheating and cosine annealing method, and the model weight file was generated during training.

[0091] The evaluation metrics include accuracy P and frame rate (FPS). Accuracy P is defined as shown in equation (4-3), where TP and FP refer to the number of correctly predicted and incorrectly predicted samples, respectively.

[0092]

[0093] FPS (Frames Per Second) refers to the number of images a network can predict per second. The higher the FPS, the faster the network's inference speed.

[0094] Furthermore, in step 5, the specific steps for defect detection of the transmission line and fitting images are as follows:

[0095] Using the improved YOLOv7 model from step 4, predictions are made on the downsized images of power transmission lines and fittings acquired by the UAV in step 1. First, the model weight file generated by training in step 4 is read; then, the image is input into the improved YOLOv7 model for prediction, returning the target location, category name, and confidence score; finally, a rectangle is drawn in the original image using the cv2.rectangle function to mark the target location, and the category and confidence score are placed in the upper left corner of the rectangle using the cv2.putText function, displayed, and saved.

[0096] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0097] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that can be understood by those skilled in the art. The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A method for recognizing defects of an overhead power transmission line conductor fitting, characterized by, It comprises the following steps: Collecting image data of the conductor and fittings of the power transmission line in the detection area to generate a data set; An improved YOLOv7 model is constructed, the EfficientNetV2 backbone network is used to replace the original YOLOv7 backbone network of the YOLOv7 model, and the BiFPN module is used to replace the original PANet module of the YOLOv7 model; The improved YOLOv7 model is used to train the data set to obtain a model weight file; The model weight file is input into the improved YOLOv7 model to identify defects in the image of the conductor and fittings of the power transmission line in the detection area; The EfficientNetV2 backbone network uses Fused-MBConv structure in the shallow network and MBConv structure in the deep network; The self-attention mechanism is added to the EfficientNetV2 backbone network; The improved YOLOv7 model is used to train the data set to obtain a model weight file, which comprises the following steps: Based on the PyTorch deep learning framework, the Adamw optimizer is used to approach the optimal solution, and then the SGD optimizer is used to adjust the improved YOLOv7 model; The improved YOLOv7 model is used to train the data set to obtain a model weight file, which comprises the following steps:

2. The method for recognizing defects of overhead power line conductor fittings according to claim 1, characterized in that, The image acquisition device loaded on the unmanned aerial vehicle collects the image of the conductor and fittings of the power transmission line from any angle, determines the image size and aspect ratio, and proportionally reduces the image. After collecting the image data of the conductor and fittings of the power transmission line in the detection area, the following steps are further included:

3. The method for recognizing defects of overhead power line conductor fittings according to claim 1, characterized in that, Calibrate the collected image to the ground true box and generate a label file containing position and category information. The generated data set comprises:

4. The method for recognizing defects of overhead power line conductor fittings according to claim 1, characterized in that, The collected images are synchronously expanded, the expanded data is divided into a training set and a data set, and a COCO format training data set is prepared. The collected images are synchronously expanded, which includes x-axis flipping, y-axis flipping and origin flipping, and adding salt and pepper noise, Gaussian noise, enhancing brightness and reducing brightness to the flipped images.

5. The method for recognizing defects of overhead power line conductor fittings according to claim 4, characterized in that, After obtaining the model weight file, the model weight file is further evaluated, wherein the evaluation indicators include accuracy and frame rate.

6. The method for recognizing defects of overhead power line conductor fittings according to claim 1, characterized in that, After obtaining the model weight file, the model weight file is further evaluated, wherein the evaluation indicators include accuracy and frame rate.

7. The method for recognizing defects of overhead power line conductor fittings according to claim 1, characterized in that, The specific steps of inputting the model weight file into the improved YOLOv7 model to identify defects in the image of the conductor and fittings of the power transmission line in the detection area are as follows: Read the model weight file and input it into the improved YOLOv7 model for prediction and recognition to obtain the target position, category name and confidence of the defect, and calibrate it in the image, and the defect recognition is completed.

8. An overhead power transmission line conductor hardware defect recognition system, characterized by, It comprises: A data acquisition module for collecting image data of the conductor and fittings of the power transmission line in the detection area to generate a data set; The model construction module is configured to construct an improved YOLOv7 model, replace an original YOLOv7 backbone network of the YOLOv7 model with an EfficientNetV2 backbone network, and replace an original PANet module of the YOLOv7 model with a BiFPN module; The training module is configured to train a data set by using the improved YOLOv7 model to obtain a model weight file; The recognition module is configured to input the model weight file into the improved YOLOv7 model to perform defect recognition on an image of a power transmission line conductor and a fitting in a detection region; The EfficientNetV2 backbone network adopts a Fused-MBConv structure in a shallow network and adopts an MBConv structure in a deep network; The EfficientNetV2 backbone network is added with a self-attention mechanism; The improved YOLOv7 model is trained by using a data set to obtain a model weight file, and the method comprises the following steps: Based on a PyTorch deep learning framework, an Adamw optimizer is used to approach an optimal solution, and an SGD optimizer is used to adjust the improved YOLOv7 model; A model file for detecting defects of a power transmission line conductor and a fitting is generated by using transfer learning to update weights of the improved YOLOv7 model, and a preheating plus cosine annealing method is used to adjust a learning rate to train the model weight file.

Images