Research On Detection Technology Of Abnormal State Of Pantograph Based On Saliency And Gan Network

With the development of the railway system in our country,the safety and reliability of the railway system have been paid more and more attention.The catenary is a transmission line used by electrified railways to supply power to trains,and its operating status is of great significance to the safe operation of railways.Nowadays,the catenary suspension state detection and monitoring device(4C)is the main method of catenary state monitoring.It collects catenary images through a high-definition camera installed on the roof of the inspection vehicle,analyzes the images,and then realizes the detection of the various components of the catenary.Traditional manual investigation has high cost,low efficiency,and unstable results,which cannot meet the needs of informatization and intelligence of railway safety inspection.With the continuous development of deep learning and deep convolutional neural networks,huge technological innovations have been brought to the field of computer vision.Therefore,taking catenary insulators as the research object,research on insulator defect detection algorithms based on computer vision.First,The status detection of catenary insulators needs to locate the catenary insulators.In order to obtain the insulator location results without background information,this paper uses the Center Net network as the basis to improve it,and uses the anchorless detection network R-Center Net to locate the insulators.The network adopts an anchorless network structure to regress the key points of the image,which can effectively reduce the complexity of the network and the amount of calculation.At the same time,the angle output is added on the basis of the original network to realize the precise positioning of the catenary insulator.The test results show that the improved R-Center Net algorithm can achieve precise positioning of insulators,and obtain cleaner positioning results than traditional methods.In the detection of catenary insulator status,this thesis conducts research from two aspects: data and model.In terms of data,generative and non-generative defect samples are augmented,and Generative Adversarial Networks(GAN)are used to solve the problem of insufficient samples of catenary defects.In terms of model,combined with the characteristics of subtle differences in insulator defects,the intra-class distinction network WS-DAN network is used to realize the classification of catenary insulator defects.This model uses weak supervision and attention mechanisms to efficiently locate the local features of insulators,and then integrates global features to classify catenary insulator defects.The results show that this method can effectively detect insulator defects.