Design Of Inspection System For Foreign Objects In Overhead Lines Based On Deep Learning

Overhead transmission line is the main way of power transmission in China.It is often wrapped by some foreign materials in the field environment,such as plastic film,balloon,kite,etc.If these foreign matters are not found and handled in time,it is easy to cause short circuit between lines,so it is necessary to conduct power line inspection.There are some disadvantages in common inspection methods,such as manual inspection method hsa low work efficiency and poor security;The inspection method of helicopter and UAV is expensive and difficult to operate;Robot patrol inspection can not cross obstacles on overhead lines or the efficiency is very low when crossing obstacles,which takes a lot of time.Aiming at the shortcomings of common inspection methods,such as poor operability,low efficiency,high cost and time-consuming,an inspection system for foreign objects in overhead lines based on deep learning is designed in this thesis,which is divided into inspection devices and ground control base station software.The inspection device is equipped with a pan-tilt camera and completes the inspection of overhead lines by hanging the wheel on the ground wire.The remote control operation of the inspection device is carried out on the control base station software,and the deep learning algorithm is embedded to prompt the foreign matters in the inspection screen,so as to assist the inspection staff to complete the inspection of overhead lines simply,conveniently and efficiently.The specific contents of this thesis are as follows:(1)According to the requirements of the project,the overall scheme of the system is designed,including the scheme design of the inspection device control system and the ground control base station software.In the scheme design,the wireless communication module,image-based foreign body detection algorithm and development language are selected.(2)Design of the control system for inspection device.Firstly,the overall scheme of the inspection device control system is designed.Secondly,the working principle analysis and program design of peripheral devices in the control system are carried out.(3)Based on the deep learning YOLOv3 algorithm,the design and model training of the foreign body detection algorithm have been completed.Firstly,the data set is constructed,including image normalization,data enhancement,etc.,and the image data is made into standard VOC format data set.According to the target category to be detected,the yolov3 algorithm is designed.The algorithm is trained with the data set,and the trained algorithm model is tested.The results show that the accuracy of the algorithm is more than 84%,and the detection rate is 25 fps.(4)Design of control software for ground base station.Firstly,the overall framework of the software is designed.Then the functional modules are developed in turn,including software interface adaptation,patrol video playback function,image data local saving and database saving,command sending and receiving,UDP video data sharing,export table generation,automatic obstacle crossing,memory patrol and visual algorithm embedding.Through indoor simulation on-site experiments,the inspection device and foreign body detection algorithm are tested.The test results show that the inspection device can autonomously cross the obstacles on the line.The foreign objects hanging on power lines can be automatically detected by the foreign objects detection algorithm at a speed of 25 fps and a detection accuracy rate higher than84%.The foreign objects detection algorithm is safe and efficient,and the inspection work of overhead lines is normalized by inspection device.The ground control base station software has friendly interface,simple operation,and more convenient data storage and query.Through the use of base station software,the inspection work is more simplified,and the overall efficiency of inspection is improved.