Surface Recognition Of 27.5k V Electric Railway Rod Insulator Based On Image Processing

Insulator,as an important insulating component,is commonly used for electrified railway catenary.Its function is to suspend and keep the part of the suspension device electrically isolated from the insulating part in the railway system.Insulator not only bears the working voltage of itself and all kinds of over-voltage,but also undertakes the weight of the contact suspension and support structure and the mechanical load produced by the weather.Therefore,the insulation performance of the insulator has a great influence on the normal power supply of the railway catenary.Self-explosion,damage and other defects caused by the long-term fatigue wear and harsh environment lead to the decline of insulation performance,and then occur the phenomenon of discharge（flashover）.The power failure of substation caused by this phenomenon affects seriously the safe and stable operation of trains.That is why the insulation performance of the insulator must be guaranteed.Traditional manual identification has many disadvantages,such as heavy workload,poor real-time performance and poor accuracy,so how to quickly and accurately realize the accuracy and speed of insulator defect identification of electrified railway is extremely important.The image processing technology is used to realize the intelligent recognition of the defects of the 27.5k V electric railway rod insulator pictures collected by 4C device in this thesis.The main work of this thesis is as follows:First of all,responding to the requirements of the insulator image collection,this thesis studies and analyzes the actual situation of specific hardware devices of the electrified railway catenary suspension state detection monitoring devices（4C）,such as the sensors and the optical light source,and understands the imaging principle of the insulator,after according to the principle of insulator defect recognition,this thesis introduced the image processing algorithm process.Then,in the process of insulator image collection,the pepper and salt noise will be generated due to the photoelectric conversion process of industrial camera and the surrounding environment,so it is necessary to filter for the original insulator image collected.The amplitude of the salt and pepper noise obey random distribution approximately,and there are clean and noise point in the image to be denoised.For the median filtering algorithm can remove the false edge caused by the salt and pepper noise,the quick guide filtering algorithm is put forward,which replaces the mean filtering function f_meanof the quick oriented filtering algorithm with the median filtering function f_median.This improvement has function of keeping better edge details characteristics,which is conducive to extract the feature of follow-up insulator and improve the accuracy of the insulator surface defect recognition.Secondly,the characteristics of Canny edge detection algorithm and BP neural network algorithm are analyzed in this thesis.The insulator region is located by combining with improved Canny edge detection algorithm and Hough transform linear detection algorithm,which can avoid effectively the mislocation caused by missing part of the pixel points by Canny algorithm.After the image containing only insulator is obtained,an insulator recognition method based on improved genetic BP neural network algorithm（GA-BP）is proposed to solve the problems that the BP neural network algorithm is easy to converge to the local minimum value and the current insulator recognition accuracy is low.In the improved GA-BP algorithm,Powell-Beale conjugate gradient method is used to update continuously the threshold and weight of GA-BP algorithm,which improves the convergence speed of GA-BP network,achieves finally insulator identification and increases the accuracy of insulator identification.Finally,based on the regular shape of rod insulator umbrella skirt distribution,the thesis statistics the horizontal and vertical direction of the each insulator image pixels respectively,and calculates the length diameter ratio and umbrella skirt area of insulator of each insulator image.Then the thesis uses the conversion value of the camera calibration results,which can make it into the insulator length diameter ratio and umbrella skirt area as the insulator surface defect judgment standard.Finally,the thesis realizes the rod insulator defect status and defect position determination.The simulation experiment results show that the rod insulator surface defect identification system proposed in this thesis can identify accurately insulator damage and self-explosion defects,and improve effectively the accuracy of insulator surface defect identification,which can provide reliable basis for maintenance workers of the catenary.