| Power patrol inspection is a timely and accurate inspection of the operating status of power lines and ancillary equipment and the surrounding environmental conditions of power line corridors ensure the safe and reliable operation of transmission lines.At present,power patrol inspection is mainly completed by manual walking,power patrol inspection has high labor intensity,low efficiency,high cost,and there are relatively large hidden dangers to personnel safety.In this context,research on automated inspection technology based on computer vision has great significance.This thesis conducts research on the detection technology of tower defects in the process of power inspection.It specifically discusses and analyzes the detection and identification methods for the missing and loose bolts and other minor tower defects.In response to the above problems,the main contents of the thesis are as follows:(1)In this thesis,a bolt target detection data set is made according to the Pascal VOC data set format,and the YOLOv3 target detection algorithm is used to detect the bolts.(2)Aiming at the problem of poor performance of the classic YOLOv3 algorithm,this paper tries to optimize the multi-scale detection of YOLOv3 to improve the detection accuracy of the algorithm.The optimized detection accuracy has been increased from 0.86 m AP to0.90 m AP.In order to speed up the detection speed of the algorithm,this thesis uses Mobile Net V3 to replace the backbone network,and uses deep separable convolution to replace the convolution of the non-backbone network part.The final detection speed is increased from the original 18 FPS to 45 FPS.(3)This thesis uses data enhancement,transfer learning,and learning rate adjustment methods to solve the problem of poor model training caused by insufficient samples to a certain extent.In addition,this thesis also improves the Res Net50 network,which further improves the accuracy of the algorithm for identifying bolt defects. |
PDF Full Text Request