Target Detection And Defect Recognition Of Power Line UAV Inspection Image

In recent years,the rapid growth of national GDP and the continuous increase in electricity consumption in various fields have made the scale of power lines increasingly large.Since the power equipment is in an outdoor environment for a long time,long-term wind and rain will definitely affect its performance and even cause electric power accidents.Therefore,it is necessary to regularly inspect the equipment to ensure the normal operation of the system.With the rise of the drone industry and the gradual maturity of machine vision technology,it has become possible to replace drones with manual inspections.However,the environment in which electrical equipment is located is often complex and variable,which poses significant challenges for testing.In this practical context,this paper mainly studies the target detection and defect identification of insulators,towers,bolts,anti-vibration hammers and other objects in the UAV inspection power line image.The specific research contents are as follows:This paper first investigates the inspection of UAVs in power systems and related technologies at home and abroad,introduces various algorithms and their deficiencies,and then analyzes their characteristics from the pictures of drone inspections,respectively.Research on both detection and defect identification,in which defect recognition is based on target detection.In terms of target detection,this paper proposes four methods for generating potential regions,such as BING,Edgebox,Objectness and Selective search,for the characteristics of UAV patrol image resolution.Then put these potential regions into Faster R-CNN.Identification and positioning in neural networks such as SSD and YOLOv2.In order to enhance the robustness of the algorithm,a large number of transformations such as rotation and proportional contraction are performed on the data set.The experiment shows that the method has strong anti-interference ability,and even if there is a certain degree of blurry picture,the detection task can be completed well.In the aspect of defect identification,for the insulator and the anti-vibration hammer,the equal interval anomaly detection based on clustering and the equidistant anomaly detection based on image segmentation are proposed respectively.Experiments show that these methods can identify defects in objects under certain conditions,but rely too much on the quality of segmentation and clustering results.For more complex backgrounds,the effect of defect recognition is compromised.Therefore,this paper once again proposes an SSD network,which uses its strong anti-interference ability and fast detection speed to identify damaged insulators and shock-proof hammers in a complex background.