Target Detection Of Transmission Line Based On Convolutional Neural Network

Target detection is an important part of automation and is widely used in light industry、computer industry and other fields.Now,the theoretical research of deep learning and image processing is constantly developing.At the same time,due to the rapid development of GPU,the computing power of computers has doubled every year,effectively reducing the time of network training and algorithm running,and improving the accuracy and real-time performance of target detection.However,the detection system cannot be fully adapted in the complex working conditions such as the detection of transmission lines.Based on the complex background interference and small target traces of damaged equipment,the detection of transmission line related equipment still relies on manual surveys.An algorithm for automatic detection of damaged equipment in transmission lines is designed in this thesis.The algorithm in this thesis is based on YOLOV3 algorithm,which is aimed at damaged equipment on transmission lines.The algorithm improves the network structure for rust surface and lightning strikes.At the same time,a fractional defogging and low illumination enhancement algorithm based on dark channel is added.The algorithm can be embedded in the image acquisition function of aircraft and line-following robots.With the help of the line-following robots,The problems mentioned above can be positioned quickly to achieve the purpose of reducing economic losses and preventing further damage.The application of fractional calculus in images and the composition of convolutional neural networks are briefly introduced at the beginning of this article,then analyzes the principles and applicable places of commonly used neural network target detection algorithms.The detection accuracy and detection speed of YOLOV3 on target detection are mainly explained.Then an improved algorithm based on YOLOV3-tiny is proposed.The feed network is optimized for Darknet19,and the optimization of scale conversion is integrated.In this thesis,improved image processing technology is used to enhance picture contrast,and self-adaptive enhancement of night target and fog target definition is added.Aiming at the problem of small target area,the mode of using shallow features to fuse deep features,and adding edge sharpening based on fractional order to strengthen the edge and texture features of small targets can effectively prevent the feature loss of small targets in the deep network which effectively prevents the feature loss of small targets in the deep network.Finally,the collected pictures are input into the network for training.The experimental results show that the algorithm can be applied to all aspects of transmission line detection.In summary,the deep learning network and traditional image processing technologies have been improved and merged enhance the detection of damaged equipment on transmission lines.This article explores the automatic detection in the field of transmission lines.This research provides a set of algorithm support for the automatic detection of transmission lines by drones.