Research On Object Detection Of Transmission Line Based On Deep Learning

With the construction and development of the society,it is becoming more and more important to maintain the stability of the power system.The electric power line inspection of unmanned aerial vehicle(UAV)is adopted by more and more power supply companies for its convenience,clear acquisition images,manpower and material resource economizing.Object detection is the foundation of the fault diagnosis of transmission components.This paper first analyzes the characteristics of aerial transmission line images.The type of image has the characteristics of complex background,unstable image quality,different viewing angles,different pixel sizes and unfixed positions in the image.These characteristics bring great challenges to object detection.In this paper,the recognition and fault detection techniques of five key components in aerial images are studied,which provides the theoretical basis for further automation of UAV.Firstly,current research of object detection is introduced,and the advantages and disadvantages of mainstream algorithms based on deep learning are analyzed.Secondly,this paper proposes a new sample preprocessing operation to solve the problem that SSD algorithm is easy to miss small object.The bilinear interpolation algorithm is used to compress the image,then the self-cropping algorithm is used to automatically crop the training sample image,and more pixel information about the target is retained,thereby improving the training effect of the deep learning model.In view of the insufficient use of high-level information of neural networks in SSD algorithm,this paper adopts a new framework structure based on FPN-SSD.Moreover,the original VGG network was replaced by the feature extraction network,and finally achieved very good results.The model improves the accuracy of object detection while ensuring faster detection speed,and the final detection accuracy is reached to 89.3percentage points.