Research On Catenary Insulator Recognition And Pollution Detection Technology Based On Deep Learning

With the rapid development of science,industry and economy,electrified railway is also developing rapidly.In particular,China’s high-speed railway construction has reached a new milestone.China has a vast territory,and electrified railways extend in all directions,connecting almost all parts of the country together to form a huge network.At the same time,a transmission grid of the same scale is required for its power supply.This power supply network is called catenary,while catenary insulator functions as a key part of catenary.However,the working environment of catenary insulator changes with the change of deployment position,and it is exposed in the outdoor environment at all times,so it is easy to be contaminated by various kinds of dirt.Pollution flashover can be triggered on insulator under certain conditions,leading to the failure of insulation performance and affecting the safe operation of railways.At the current stage,the detection of insulator contamination is mainly by manual detection.The current detection method can no longer meet the current development needs of China’s railway network.According to the above problems,the following research has been carried out in this thesis:(1)The traditional computer vision method is greatly affected by light when recognizing contact line insulators,and it is even difficult to recognize under certain light,reflecting its limitations in use in certain weather.And its recognition accuracy in a complex background is very low,the thesis decides to use deep learning methods to identify catenary insulators in complex backgrounds.Considering that the entire system will be migrated to embedded devices and other small computing power-constrained devices,and the identification part should provide real-time information and real-time guidance for the following series of operations,This thesis combines the principles and characteristics of the Mobile Net-V1 deep separable convolutional neural network and the SSD one-stage detector,and builds the Mobile Net-V1-SSD model,which is a lightweight model with fewer parameters and faster speed.Since the accuracy of the lightweight model is slightly lower than that of the conventional network model,we try to perform feature fusion enhancement processing on the model in order to make an effective balance between real-time performance and accuracy,and complete related experiments.(2)According to the characteristics of the visible light image method,the background will interfere with the pollution detection to varying degrees during the pollution detection,this thesis uses the visible light image method,based on the phenomenon that the surface color of the insulator is the same and uniform,and if there is contamination on the surface of the insulator,the overall color of the surface will be changed.The insulator is converted from the RGB color space to the HSV color space,and then the color components are extracted.By controlling the color channel threshold,the pollution is identified.And complete the experiment according to actual work.(3)Based on the principle of high efficiency,low cost,applicability and as far as possible not to affect the normal operation of the railway,This thesis uses the visible light image method,According to the phenomenon that the surface color of the insulator is the same and uniform,and if there is contamination on the surface of the insulator that is different from the surface color,the overall color of the surface will be changed,the insulator is converted from the RGB color space to the HSV color space,and then the color components are extracted.By controlling the color channel threshold,the pollution is identified.And complete the experiment according to actual work.Experimental results show that: in terms of identification,insulators can be quickly and accurately identified under complex backgrounds and under different light;In terms of segmentation,the original pixels of the insulator can be quickly segmented from the background;in terms of contamination detection,the contamination on the insulator can be quickly detected from different angles.According to the analysis and discussion of experiment results,the method proposed by this thesis meets real-life requirements with good accuracy and real-time performance,which provides certain basis and ideas for the detection of catenary insulators and its contamination.