Research On Computer Vision-based Defect Detection Of High Voltage Transmission Line

In recent years,the State Grid Corporation of China has been committed to improving energy resource allocation capabilities and intelligence,ensuring power supply,and promoting the revolution in energy production and consumption.Therefore,the construction of smart grids has become a goal that power companies are constantly pursuing.Among them,the detection of important components of high-voltage transmission lines based on computer vision has gradually become one of the research hotspots in the construction of smart grids.This article mainly focuses on the automatic detection technology of two important components of transmission lines: insulators and dampers.Computer vision-based detection technology can accurately and efficiently automatically locate and detect important components of high-voltage transmission lines in aerial images,which is of great significance for liberating human resources and improving detection efficiency.The background of most aerial images of high-voltage lines includes complex terrain such as mountains,rivers,fields,etc.,which brings great challenges to the detection task.In the insulator detection,the orientation of each insulator in aerial images is different.The general object detection algorithm uses a regular rectangular bounding box without angle to locate the target.Therefore,the detection bounding box cannot tightly surround the insulator component,and some background noise.In addition,it is necessary to label a large amount of caps position information in order to detect missing insulator caps using deep learning methods.However,it is undoubtedly a huge task to label lots of insulator caps.In terms of detection of damper defects,damper is suspended on a high-voltage power line,which is small in most aerial images,making it difficult to detect the missing of the damper.Regarding the issues above,the research content and main work of this paper are as follows:(1)A cap-count guided weakly-supervised insulator cap missing detection method is proposed,which mainly includes three innovations: First,we represented the insulator target area by an inclined rectangular bounding box which solves the problem of multiangle insulators in aerial images.Secondly,the number of caps is used as the monitoring information.And the coordinates of each insulator cap are automatically generated based on the position information of the insulator string and the corresponding number of caps,reducing the reliance on data annotation.Finally,a detection model based on deep learning is constructed to optimize insulator strings and caps simultaneously.The experimental results show that the proposed detection method is better than other methods on the collected insulator data set.The accuracy rate and recall rate are 92.86% and 86.67%,respectively.It has greatly improved the efficiency of detection and detected insulator strings and caps at the same time.(2)Considering the characteristics of the pair of dampers in aerial images and the normal damper and the missing damper are regarded as two types of targets,we designed and implemented a method for detecting damper defects.The main work includes : for the problem that the damper has less defect data,we have generated a large number of damper defect data,and constructed a damper dataset.Secondly,for the small size of the damper,we proposed a detection method which can detect and classify dampers on three scales.And we use the complete damper pairs and the damper pairs without one as two categories for network training.Finally,through training the network model,the position,category and confidence information of the dampers are output.The experimental results proved the effectiveness of the proposed method.(3)We built an easy-to-use insulator and damper defect detection system.The system incorporates the two algorithm models proposed in this paper,as well as the function of the defect detection algorithm.