Detection Technology Of Power Line Insulator Bunch Drop In UAV Inspection Based On Deep Learning

Insulators,a vital component of power lines,provide electrical insulation as well as mechanical support and are the subject of inspection work.Insulators are subjected to internal pressure from mechanical and electrical loads during operation,as well as external attacks such as lightning strikes and icing,which can easily lead to failure and thus affect the sta ble operation of the power grid,due to long-term exposure to the complex and ever-changing open-air environment.The conventional manual inspection approach is progressively being replaced by the more easy,efficient,and safe UAV inspection method as science and technology advance.However,there are certain challenges with the real-time detection of insulator bunch drop in UAV inspection power lines due to the complex background and enormous volume of image data captured by UAV aerial photography.Therefore,a critical issue in the current power system inspection is how to accurately and rapidly realize the detection of insulator bunch drop in UAV inspection power lines.Aiming at the real-time and accurate detection of insulator string drop in power line UAV aerial photography,power line insulator is used as the detection object of this paper.Based on GPS positioning,deep learning and target detection technology,it is applied to the dynamic inspection process of UAV flight to realize the autonomous detection attitude control of UAV overhead power lines and the detection of insulator drop targets,in order to provide t echnical support for UAV inspection operations in overhead power lines.The main research contents are as follows:(1)UAV aerial photography insulator data enhancement processing.The aerial insulator data set based on multiple environments,multiple materials and different states is constructed,and the data set is enhanced by four enhancement methods: geometric transformation,color transformation,noise addition and special weather effects.(2)A lightweight insulator detection model based on YOLOv5 s a nd Mobile Netv3 is constructed.To address the problem of slow detection speed and low accuracy of the target detection model on the embedded platform,the feature extraction network of the YOLOv5 s target detection model is replaced with Mobile Netv3,a lightweight network incorporating the CBAM attention mechanism module,and the improved model detects insulator images with a resolution of 640×640 at 15frames/s in the CPU environment,and the the mean average accuracy reaches 94.25%.(3)Created an autonomous UAV flight system that is controlled by a Raspberry Pi.Given that most UAV flight control methods are currently manually controlled by ground personnel or use ground stations to make flight plans,the GPS-based UAV autonomous flight program is written and deployed to Raspberry Pi,and data communication is realized via a wired connection between Raspberry Pi and UAV flight control,and UAV flight control adjusts the UAV heading and flight speed based on the flight commands sent by Raspberry Pi.(4)Constructed an insulator drop string detection system.According to the overall structure of the UAV power line inspection system,the system was divided into three major modules,including insulator detection module based on the target detection model,contro l module,and ground monitoring module,and a simulated power line environment was built outdoors for insulator drop string detection experiments.The experimental results show that the inspection UAV can fly according to the predetermined inspection route,and accurately identify insulators and insulator bunch drop phenomenon at a detection speed of 4 frames per second.Therefore,the power line insulator and bunch drop detection system proposed in this paper can provide technical support for power line inspection.