Research And Application Of Image Proceesing Technology For Recognition Of Overhead Powerline Porcelain Bottles Crack

Patrol of High Voltage Transmission Line is an important foundation to ensure the reliability of power transmission.Manual inspection has been the main way of transmission line inspection now.However,this method not only expends a mass of human and material resources,but also can not realize real-time supervision.It is a hot research field to use the visual system of the self-walking patrol robot in transmission line inspection for high voltage power line recently.The self-walking patrol robot for high voltage power line which integrates multiple inspection technologies can observe the power lines.The invention of the patrol robot breaks out a completely new concept and method for the high voltage line maintenance.The paper presents the research for damage inspection system in patrol robot under "863" project sponsored.The paper focuses on the recognition and orientation of porcelain bottles cracks in the vision system of patrol robot.The crack inspection of porcelain bottles is implemented by image processing technology.The simulation is completed by MATLAB and the inspection system is designed at last.The entire process is divided into five parts.1) System design.The hardwares include CCD,camera,control pan,image card and computer.The device can get the images of the object for the computer.2) Image preprocessing.The image of porcelain bottles need to take some image preprocessing such as grey scale transformation,de-noising,segmentation, morphologic processing.In the processing of the image segment,the method of OTSU and image threshold segmentation method which based on edge information are introduced and compared with fixed threshold and iterative method.3) Cracks orientation.Through the analysis of porcelain bottles images, characteristic region in the center of porcelain bottles are taken as the basis of location. The orientation of porcelain bottles are been completed by extracting the special shape of characteristic region,including the ratio between rectangular frame of characteristic region and length to width of porcelain bottles.Cracks in images are been located based on gray value of cracks in the characteristic region.4) Features extraction.The features are extracted from the image which is after focusing to distinguish the crack classifications which including transverse cracks, longitudinal cracks,block cracks,alligator cracks,non-distress cracks.Features including plane projection variance,vertical projection variance,HU invariant moment are selected in the paper.5) Images recognition.According to features are extracted from crack images, recognition will be completed by using the algorithm which was designed by combining genetic algorithm with BP and RBF network or using modified ART2 network.By analyzing and comparing the results of the simulation in the experimentation,modified ART2 network will be selected in the paper.