Research And Implementation Of Defect Detection Algorithm For OLED Display

With the rapid development of display technology,more and more screen manufacturers begin to invest in mass production of OLED.In order to prevent undesirable OLED products from entering the market,it requires professional quality inspectors to detect the defects on the panel by human eyes.But detection by human eyes requires higher labor costs,and different quality inspectors may not have the same test results for the same panel.Therefore,in order to meet the demand of large-scale production in the factory,it has become an urgent task to study the corresponding automatic defect detection algorithm and let the machine vision replace the human eyes.There are point defects,line defects and surface defects(Mura defects)during the production of OLED.According to the characteristics of these three types of defects,this paper proposed various defect detection and classification methods respectively to improve the existing algorithms.The main research contents and contribution are as follows:(1)This paper proposed a neighborhood difference algorithm based on the periodicity of the panel image.This algorithm can effectivelysuppress the interference of the background texture and extract the point defects,which has solved the problem that the existing algorithm is insensitive to small defects.(2)This paper proposed a method of projection line based on the neighborhood difference algorithm.This method can realize the accurate extraction of line defects and effectively improve the detection accuracy of the existing algorithms.(3)In this paper,a background reconstruction method based on defect enhancement and surface fitting was proposed,which could let us get a "no defect" background map.Then,we calculated the residual difference between the background map and the enhanced image.Finally,the real defect was screened out according to the quantitative description method of Mura defects.This algorithm can extract weak Mura defects and achieve high accuracy under uneven background conditions.(4)In this paper,a classification method based on transfer learning was designed.This method can be used to complete the model with fewer training samples,and the experimental results showed that the method had high accuracy and could be applied to actual industrial production.These methods of this paper have been verified by a lot of actual data in the experiment,which showed a good performance.