Research On Overhead Transmission Line Component Identification Technology Based On Target Detection Algorithm

The power grid is an efficient and efficient energy transmission channel and optimization configuration platform,is a key link in the sustainable development of energy and electricity,plays an important pivotal role in the modern energy supply system,and is related to the national energy security.As the most basic component of the power grid,the integrity of the transmission line components is the guarantee of the stable operation of the power grid,so the transmission line should be inspected regularly.Although the use of unmanned aerial vehicles for line patrol inspection has greatly improved the efficiency of patrol inspection in recent years,the entire patrol process still requires the staff to focus on the search and identification of relevant components.This long time of focus is prone to visual fatigue,resulting in missed inspection.Therefore,if the UAV can quickly and accurately identify and mark the power components in the transmission line during patrol inspection,it will greatly improve the inspection efficiency of the staff.Therefore,the paper studies the identification and detection algorithms and embedded transplantation of the components(insulators,grading rings and spacers)that are prone to faults in transmission lines.The main research contents are as follows:1)Aiming at the problem that the current mainstream insulator defect detection algorithm model is large and the detection speed is slow.A lightweight Ghost-YOLOv5 insulator defect detection algorithm is proposed.The lightweight Ghost convolution is used to replace the general convolution,and the K-means algorithm for generating a priori frame is improved by the distance standard.The experimental results show that the average detection accuracy of the algorithm reaches 96.3%,the number of model parameters decreases by 41.1%,and the detection speed increases by 17.6%.The algorithm model is more lightweight and easy to deploy on embedded terminals.2)The grading ring and spacer in the transmission line components are small,and the size difference between them is too large,so that the common lightweight improvement will lead to the problem that the detection accuracy of small targets can not meet the requirements.Firstly,an image data set of overhead transmission line components including insulators,grading rings and spacers is constructed,and then a CBAM-Efficient-YOLOv5 transmission line component recognition algorithm based on improved attention mechanism is proposed.The algorithm uses Efficient Net lightweight network to replace the original backbone network for feature extraction,and improves its attention mechanism.CBAM mechanism with both channel and spatial attention is used to replace SE mechanism with only channel attention.The experimental results show that the accuracy of the algorithm for large,medium and small targets detection has been improved to more than 90%.3)Aiming at the problem that the algorithm model to be transplanted to embedded devices needs to be further lightened in practical applications.Firstly,the algorithm model is pruned by Slimming,and the number of parameters is halved on the basis of the original model.The Prune-CBAM-Efficient-YOLOv5 model with basically unchanged detection performance is obtained.Then Tensor RT was used to optimize the model structure and low precision processing,and finally successfully deployed in the embedded device Jetson TX2.The experimental results show that the detection accuracy of the compressed algorithm model is slightly reduced,and the detection speed reaches 30.3 frames/s,meeting the requirements of real-time target detection.