Insulation Recognition And Damage Detection Of Catenary Based On Image Processing

Along with China's railway construction and operation of the cause of development,Railway transport not only carries the lifeblood of China's economy,but also the use of electrified railway,with environmental protection and efficient features.Especially in recent years,haze weather seriously affect our major cities,green travel has become a trend.This year's government work report proposed to fight the blue sky to defend the war,which is more reflected in the advantages of railway traffic green.But the safe operation of the railway,the need for traction power supply system to provide normal operation to provide protection,so traction power supply system is very important to the stable operation.As China's railway mileage is long,and mostly in the outdoor environment,so alone rely on manual inspection has been unable to meet the development needs of the railway.In recent years,image recognition technology,especially in-depth learning and rapid development of machine learning,for non-contact catenary fault detection provides a solid foundation.In this paper,the position of the insulator in the image is determined by the method of image processing and depth convolution network,and the algorithm of the local expansion of the image and the two value of dynamic threshold is used to detect the damage of the insulator.Due to the scattered research of catenary insulator,no sample insulators available,used to train classification algorithm or convolutional neural network,so this paper first on the 12000 inspection vehicle collected the original image data,marking and processing,and made into positive and negative samples of different scales set.Secondly,the traditional Adaboost cascade classifier and the SVM classifier are trained to identify and intercept the insulators in the original image using the model obtained after the training.Experiments show that the recognition rate of the insulator is only 52.7%,which can not meet the requirement of recognition rate.Based on this,we use the CaffeNet model in Caffe open source framework and optimize the model parameters.Then we use the model files obtained by training to classify the insulators.After testing,it was found that the recognition rate of the method was 99.8%at the time of testing,and this recognition rate could meet the requirements of insulator identification.The final use of this method will identify the insulator,cut from the original picture and save it,to detect the faulty insulator.Finally,the algorithm of template matching and the method of local expansion and dynamic threshold binarization are used to test the damage of insulator.Compared with the experimental results,it is found that the method of combining the local expansion of the image with the binarization of the dynamic threshold has a good effect and can distinguish the broken insulator well.This method can provide a kind of solution based on image processing for the damage identification of catenary insulator fault,and solve the problem of difficult to detect the damage of artificial insulator to a certain extent.