Research On Several Image Processing Technologies In AOI System Of Glass Substrate

The glass substrate is an important upstream product in the flat panel display manufacturing industry.With the widespread application of flat panel display technology in today's society,the market demand for glass substrates has greatly increased,so the defect detection technology of the glass substrate has also been paid more and more attention.The traditional manual visual inspection method for glass substrate defects has been unable to meet the requirements of contemporary large-scale production due to backward technology.The application of AOI(automatic optical inspection)and image processing technology in glass substrate defect detection has become more and more widely used.The pretreatment of defect images in automatic optical inspection of glass substrates,the extraction of feature parameters of different types of defect images,and the related image processing techniques in defect recognition and classification is mainly studied in this paper.In the study of the preprocessing algorithm for the defect image of the glass substrate,the filtering effect of the average filtering,Gaussian filtering and median filtering algorithm on the image noise was studied respectively.The combined comprehensive filtering method effectively filters the noise existing in the defective image of the glass substrate;in the image enhancement algorithm,the histogram equalization method is used to enhance the image contrast of the defect location area;in the image segmentation algorithm,the OTSU segmentation method is introduced,Segmentation of defective images is achieved.In the defect feature parameter extraction part of the glass substrate,the feature parameters were first analyzed for the defect binary image after image preprocessing,and geometric features,shape features,and image invariant moment features were selected as the classification basis for different defect types.When the defects in the region are labeled in the connected domain,the two-scan method is compared with the seed filling method.The convex hull detection algorithm of the Andrew scanning method and the minimum circumscribed rectangle labeling algorithm of the rotating shell method are used to obtain the minimum circumscribed rectangle related parameters of a single defect.Calculate the relevant characteristic parameter data of the defect image and normalize the data.In the classification and identification of different defect images,two classification methods of BP neural network and C-SVM support vector machine are used to realize the identification and classification of different defects.The three-layer network structure is used in the BP neural network,and the number of hidden layer nodes is determined through many experiments.The final recognition accuracy of the BP neural network reaches 87.6%.In the support vector machine,the defect type two-class classification method is adopted,and the radial basis kernel function is used to select the kernel function,and the parameter optimization of each SVM sub-classifier is performed,and the final SVM recognition accuracy rate reaches 92.4%.The results show that the support vector machine has a better advantage than the BP neural network in detecting defects on glass substrates.