Applied Research On Deep Learning In Defect Detection Of Key Components On Transmission Towers

Manual inspection is the main traditional method of regular patrol for the high-voltage transmission towers.The workers need to check the tower with visual inspection.This method is not only inefficient,but also has safety issue to the workers.In recent years,the application of UAV in power industry is increased.This makes automatic inspection based on image processing technology possible.It is an interesting and challenging problem to detect the defects of the key components on the high-voltage tower in aerial images with complex background.In this paper,we are focused on automatic defect detection of insulator and pin in aerial image using Depth Learning,and the main work of this paper is as follows:1.Three kinds of data augmentation methods are developed.They are designed to solve the problem that the convolution neural network is hard to train due to lacking negative sample.The labeled insulator data sets are published online for other researchers;2.Based on the existing structure,a novel cascading object detecting framework is proposed.This structure is used for locating insulator and detecting defects.A variety of experiments were designed to verify the effectiveness of the structure.The defect detection accuracy of this structure on the published insulator data set is 91.25%;3.Pre-research on the detection of the pin defects in the aerial image is developed.A method combined locating and classifying is proposed.Aggregate channel features and Adaboost classifier is used to locate pin first,and then the defects of pins are detected by an 8-layer convolution neural network.