Image Acquisition And Processing Technology Of TFT-LCD Array Substrate Defect Visual Detection System

Thin film transistor-liquid crystal display(TFT-LCD)has many advantages,such as small space occupation,convenient use,no radiation,easy integration,etc.It has been widely used in lightweight portable devices and large display devices.Because of its high degree of automation in manufacturing technology and suitable for large-scale industrialization,it has developed rapidly in the field of electronic display.In TFT-LCD automatic production process,in order to ensure product quality and production efficiency,automatic defect detection is particularly important.The automatic optical detection method based on machine vision has the advantages of high automation,less manual intervention and fast detection speed,which has important research significance.On the basis of studying the structure and detection principle of TFT-LCD array substrate defect visual inspection system,aiming at its characteristics of large field of view,high precision,short detection time and strong system timing,this paper first studies the key technology of image acquisition in the detection system;secondly,it studies the related algorithms of image processing;finally,it studies the data processing technology to improve the processing performance.This paper mainly studies the following points:1)In the image acquisition part of the system,in order to meet the imaging needs,this paper designs a composite imaging method which combines transmission imaging and reflection imaging,and uses multi-camera parallel acquisition technology to build the image acquisition system.Then a series of adjustment and calibration methods from light source to camera are proposed.These methods are used to adjust the image acquisition system so that it can acquire images clearly and accurately.2)In the image data processing of the detection system,in the pre-processing stage of the FPGA,the background is separated by one-dimensional Fourier transform method,the defect is segmented by dynamic double threshold method,and the effect of defect segmentation is verified by designing the detection rate experiment.In the stage of computer image processing,the defect region is marked by fast connected component analysis method,and then the defect feature is extracted.Finally,the shape feature of the defect is classified by C4.5 classification method.The classification effect of decision tree is validated by designing classification accuracy experiment.3)In the process of software implementation,a data parallel processing optimization method using ring buffer and OpenMP is proposed.The simulation and real-time detection experiments show that the method can improve the software performance of the detection system.