Research On Quadrotor Unmanned Aerial Vehicle Power Inspection Intelligent System

The electric power industry is an important industry for national production and daily life,playing a crucial role in fields such as economy,education,healthcare,and national defense.The stable operation of power transmission lines directly affects the stability of the power system,making it of great importance.Regular inspection of these lines is a vital means of ensuring safe operation.However,due to the vast and complex terrain,the wide distribution,and heavy workload of manual inspection tasks,it is difficult to quickly and accurately identify faults.With the rapid development of drone and AI technology,the field of power inspection has undergone gradual reform.In recent years,the technology of drone power inspection has become mature,but it is still mainly operated manually,which can lead to errors and difficulties in ensuring accurate inspection.Therefore,introducing AI into drone power inspection has practical significance.This article first analyzes the current situation of drone power inspection,identifies its shortcomings,and proposes corresponding strategies.A quadcopter drone power inspection intelligent system is built and summarized to lay the foundation for the deployment of specific algorithms in the following sections.Secondly,this article proposes an improvement strategy for the YOLOv5 image recognition algorithm for ground terminals,addressing the problems of large hardware resource consumption and slow recognition speed.Specifically,it improves the convolutional operation unit and residual network to solve the problem of slow detection in the original algorithm.Additionally,the original loss function cannot cope with extremely unbalanced training samples,so this article proposes using the Focal Loss loss function to replace the original loss function to solve the problem of unbalanced training samples.The model’s single image recognition time is 0.061 s,and the highest insulator recognition accuracy is 98.9%.Thirdly,this article proposes a lightweight improvement strategy for the YOLOv5 algorithm that can be loaded onto a micro-embedded platform.The YOLOv5 algorithm suffers from problems such as high hardware resource consumption and high computational heat generation,making it difficult to deploy on a micro-embedded platform.This article uses the Shuffle Net V2 network as the backbone network of the algorithm and replaces traditional convolution with depth-separable convolution to reduce the computational parameter volume and thereby reduce the computational burden on hardware resources.In addition,since this part of the algorithm is used for drone positioning and inspection of insulators,the Stem structure of the Pelee Net network is used to replace the original Focus structure,achieving specialized improvement for insulator features and further reducing CPU computational pressure.Furthermore,this article proposes the introduction of Transformer attention mechanism in the residual network to improve the algorithm’s ability to extract occluded targets.The improved algorithm can be deployed on a Raspberry Pi platform supporting Arm architecture,with a detection accuracy of up to 91.6%,an occluded insulator detection rate of 96%,and an average processing time of 0.97 s for a single image with a size of 640*640.Finally,this article proposes a path planning method for quadcopter drone power inspection.Taking a wind farm of the Shanxi branch of Huaneng New Renewables Co.,Ltd.as an example,it designs inspection routes for four common tower types,including singleloop suspension towers,single-loop straight towers,double-loop suspension towers,and double-loop straight towers.The intelligent inspection three-dimensional route is drawn with the latitude and longitude information of the poles.Additionally,the YOLOv5 lightweight image recognition algorithm is introduced to provide visual correction for the drone’s flight direction.Moreover,this article designs a collision avoidance mechanism for the drone,taking into account the flight path,flight altitude,and the presence of obstacles.