| Using UAV + machine vision method to realize transmission line component defect inspection is a hot research field in the current electrical engineering application background.During the inspection of power lines,the existence of traditional algorithm for complex background is difficult to extract features,not both detection accuracy and model size,and the safety of the UAV flight.Therefore,this paper takes the accuracy and rapidity of insulator fault detection and the safety and stability of UAV flight as the research objects.An insulator defect detection model based on improved YOLOv5 s is proposed,and three-dimensional autonomous obstacle avoidance planning is studied.The main research contents of this paper are as follows:(1)Data sets of defective insulators with different backgrounds,sizes and materials are constructed.In view of the elongated structure characteristics of insulators,Kmeans++ algorithm is used to re-cluster prior boxes.(2)On the basis of YOLOv5 s network,on the premise of guarantee the insulator detecting precision,and designs the lightweight Ghost C3 and Ghost Conv module,reduce the model calculation,make the algorithm more easy to deploy.(3)A Bottleneck CSP structure is adopted,and a light-volume spatial and channel convolution attention mechanism is introduced to strengthen the characteristics of insulators and restrain the complex background characteristics.An improved Bi FPN structure is proposed,which implements multi-scale feature fusion and improves the detection ability of small targets.The new network structure obtained by the fusion of the above improvements and YOLOv5 s takes into account the detection accuracy and model size.(4)The electrical equipment inspection and obstacle avoidance system for rotor wing UAV is designed.An improved global path planning scheme based on fast random path search algorithm and a local tower optimization obstacle avoidance algorithm based on improved traditional artificial potential field method are proposed.Experimental results show that the improved algorithm in this paper has a m AP of92.3% on Insulator2022 data set,an increase of 3.6%,a reduction of 26.73% in the number of parameters,a reduction of 23.17% in floating-point arithmetic,and a reduction of 5.47% in the missed detection rate.In the open data set,the m AP of defective insulators reaches 99.5%,and the evaluation indexes are better than the mainstream algorithms of fain-RCNN,SSD,YOLOv3 and Yolov3-Tiny,as well as the insulator detection algorithms.Two improved 3D obstacle avoidance algorithms can effectively improve the robustness and safety of rotorcraft UAV line patrol. |
