Research On Defect Detection Technology Of Glass Insulator Based On Machine Vision

In long-distance high-voltage and UHV transmission lines,glass insulators are an important insulation and support component.The defects such as bubbles,scratches,and stones in glass insulators constitute a huge factor for the safety of glass insulators and the convenience of maintenance of transmission lines.Therefore,it is necessary to detect various defects in glass insulators and ensure the yield rate.Based on machine vision,this thesis designs a set of algorithms for image acquisition,preprocessing,defect detection,feature extraction and classification of glass insulators.The main research contents are as follows:Firstly,this thesis studied the glass insulator lighting scheme,discussed the influence of light source type,shape,camera model,irradiation angle and other factors on imaging,and according to the appearance and material characteristics of the glass insulator,the LED backplane light source was selected,and the backlight illumination angle was also suitable.camera to acquire images.Secondly,this thesis studied the related content of glass insulator image preprocessing.Although the commonly used Gaussian low-pass filtering and median filtering can effectively denoise,they can also cause edge blur.This thesis is improved it based on this.First,each pixel in the image are classified into noise points and nonnoise points according to the sensitivity of the CCD sensor to noise,and then the filter window size and weighting coefficient are dynamically adjusted according to the number of noise points in the template.The measure results show that this method can save the edge information of glass insulators and defects in the image as much as possible when filtering Gaussian noise.After filtering,the sharpening technology is used to enhance the edge.Aiming at the characteristics of reflection on the surface of the glass insulator,this thesis designs an automatic spot detection and removal algorithm by estimating the maximum diffuse reflection angle,which effectively enhances the image.Then,from different perspectives,this thesis realizes the detection of defects in glass insulators based on morphology,edge,and MSB-FCN.Aiming at the fixation of traditional morphological structural elements,this thesis designs adaptive SV structural elements and corrosion operators to improve detection accuracy.The contours of the glass insulator will also be detected when detecting the edge of the defect.Because the contours of the glass insulators are arc-shaped and the curvature of the pixels on them are equal.Based on this feature,the ROI automatic matching algorithm is designed to remove the contours of glass insulators from the defect image.Finally,this thesis extracts the characteristics of the defect image and uses it as the basis for classification.In this thesis,geometric features and Hu invariant moments are extracted,a defect classification algorithm based on feature thresholds is designed according to geometric features,a BP neural network automatic classifier is designed according to geometric features and Hu invariant moments,and artificial light inspection is used as the standard to compare the the detection accuracy of two classification method.The measure results show that the BP neural network’s recognition rates for bubbles,scratches and stones have reached 85%,75%,and 72.5%respectively,which can realize the classification of glass insulator defect images.