Research On Detection And Classification Methods For Surface Defects Of Flat Panel Components

In the process of production,cutting and two processing of the screen,such as smart phone,because of the technical level,the workshop environment and some other environment and human factors,the mobile phone screen will inevitably have the quality of the production quality such as scratches,foreign objects and breakages.The emergence of these problems will not only affect the optical properties and impressions of the screen,but also affect the acceptance of products by consumers.Therefore,only through strict production inspection,will defective products be screened out,in order to meet the needs of enterprises,customers and consumers.The traditional production line mostly adopts the artificial detection method.This method has low detection efficiency,high demand for workers,and the result of detection can't be quantified.It has a certain influence on the production efficiency of the enterprise.The classification and quantity of defects and their location in glass panels determine the subsequent processing and processing of products.So,how to select the defective products accurately and efficiently in the transparent glass plate of mass production and classify the defects into a continuing problem.Based on the research of the characteristics of the white surface defects of glass bottles,a dark field lighting configuration composed of light source,zoom microscope and CCD is designed.Since the sag is the real damage that should be sent for further investigation,we should distinguish and remove the erroneous signals related to dust particles.The microscope and pattern recognition methods to classify the combination of dark field light scattering dust particles and digs.SDES for dark field image acquisition of optical sample.Then,each image is grayscale,texture and morphological analysis to extract the original feature data and compress it by principal component analysis.Based on the compressed feature data,support vector machine is used to build the classification model.The success rate of training set is 96.56%,and the success rate of prediction set is 93.90%.The classification results show that the method can be applied to identify actual optical samples from actual sunken and dusty particles.In this paper,the defects such as sags,scratches and dust are taken as the research objects.On the basis of consulting a large number of defects detection and the related literature of machine learning,the SVM classifier is used to study the surface defects detection and classification of the glass plate original.