The Research On The Visual Detection Method Of Breakage Defect Of Insulators In Transmission Lines

The safe and stable operation of transmission lines has always been the focus of the power sector and users,and the status monitoring of transmission line equipment has the characteristics of large data volume and complex task requirements,which brings great challenges to the power industry.Among them,the insulator is the key equipment in the transmission line,due to the long-term exposure to the outdoor,it is very prone to failure,and is an important detection object of power inspection.In recent years,deep learning has become a key technology leading to the rapid development of artificial intelligence and has been important applications in the intelligent inspection of electric power,deep learning mainly relies on big data,through training to achieve effective feature extraction of data and complete a variety of tasks,such as object detection,image segmentation,defect detection,etc.This paper studies transmission line insulators’ breakage defect detection method and proposes a system scheme for that mission.The system first completes the real-time positioning and identification of insulators on the UAV side,stores the target image of the insulator,and then implements the insulator breakage defect detection on the server-side.The main work of this paper includes:1.Aiming at the problem of insufficient computing resources for UAVs,this paper studies a lightweight object detection model,and proposes an insulator object detection algorithm based on an improved CenterNet network.The CenterNet obtains the target position by directly recalling the center point and other information,avoiding the computational complexity issues caused by using anchor boxes,thereby improving the detection speed;and this paper adds the prediction of the target angle information in the CenterNet network to realize the insulator Orientation detection of objects,thereby reducing the background information redundancy in the bounding box.Experiments show that the method has good performance in precision and recall rate,and the network has the advantages of a small number of parameters and high detection speed.2.Aiming at the problem of few actual insulator defect samples,an unsupervised image reconstruction method based on autoencoder,memory autoencoder,and variational autoencoder is studied,and the method is verified in the insulator breakage defect detection experiment.Unsupervised defect detection methods enable the network model to reconstruct normal samples by training normal non-defective samples.This paper analyzes the principles of the above algorithms and summarizes their problems and shortcomings.3.An adversarial memory-variational autoencoder(AM-VAE)network based on image reconstruction is proposed for the task of breakage defect detection in insulators,which integrates memory encoding,variational autoencoder,and the generative adversarial network,and we designed different network modules.In the training stage,the network learns the image features of normal insulators without damage,obtains the normal insulator network model,and sets the error threshold;in the testing stage,the test image is reconstructed and output,and the reconstruction error value of the input and output images are calculated,and the error threshold can be determined.According to the reconstruction error,it is judged whether the insulator is damaged or not.At the same time,the defect location is further obtained by constructing the error heatmap of the input image and the reconstructed image.The experimental results show that the method can reconstruct the insulator image clearly,and excel in image quality metrics,detection accuracy,and recall rate.4.The software system for insulator breakage defect detection is designed and implemented,which provides strong support for the practical application of the algorithm proposed in this paper.