Detection Of Transmission Lines’ Defects Based On Multi-source Image Feature Matching Of UAV

At this stage,although the UAV inspection of transmission lines has been widely used in the industry,it is difficult to automatically detect transmission line defects based on the images taken by the UAV inspection,which requires manual analysis,huge labor consumption,and low efficiency.Studying the UAV image transmission line defect detection technology based on deep learning target detection algorithm is of great significance for reducing the probability of multiple types of faults and further ensuring the safety and stability of the power grid.This article first establishes the detection process of transmission line defects,using the most popular deep learning Faster R-CNN algorithm as the defect identification detection algorithm,and describes the steps of the detection process.Constructed the transmission line defect data set,analyzed the characteristics of the four types of defects in the transmission line pole tower bird’s nest,the broken wire strands,the loose wire strands,and the glass insulator,and aimed at the insufficient samples of the four types of defect images in the existing UAV inspection photos Problem,using data enhancement technology to expand the UAV image samples,increased the number of four types of defect image samples,and constructed a data set.The image annotation of the four types of defect data sets of transmission lines was carried out.The Label Img annotation tool was used to annotate the UAV images of the four types of transmission lines.The four types of defects were classified according to the different performance characteristics of the four types of defects.Annotation methods have obtained different types of annotated image datasets.Proposed a deep learning based transmission line defect recognition algorithm,studied the deep learning based Faster R-CNN network target detection principle,established the Faster R-CNN network training process,by using four types of transmission line defects to label image data The network model is trained so that the Faster R-CNN network model has the ability to identify four types of transmission line defects.After training,a Faster R-CNN network model that can be applied to transmission line defect detection is derived.Finally,four types of defect identification experiments were carried out for transmission lines.The Faster R-CNN algorithm model derived after training with labeled image data sets was used to carry out self-detonation images of bird nests,broken wires,loose strands,and glass insulators taken by UAV inspection Automatic defect detection,by adjusting the training step size and labeling method,to verify the best training step size and four types of transmission line defect labeling methods that are most suitable for model detection.The experimental results show that after expanding the training data set with data augmentation technology,the recall rate of Faster R-CNN algorithm model for the detection of four types of transmission lines can all reach more than 80%,the accuracy rate can reach 96%,and the false detection rate below 3.5%,it indicates that the Faster R-CNN algorithm can meet the actual application requirements for the detection of four types of transmission line defects.