Methodological And Experimental Research On Visual Defect Detection For Glass Cover Of Mobile Phone

Glass cover is one of the most important components of mobile phone,which has excellent mechanical and optical properties such as good light transmission,high strength,comfortable touch and so on.The manufacturing process of glass cover is very complex,and various defects are inevitable in the process of production and transportation.With the rapid development of smart phone industry,the design of glass cover is various,and the style of which is also updated quickly.Besides,the requirements of surface defect detection methods are also more and more rigid.This dissertation aims to study automatic optical detection technologies for glass cover of mobile phone,and the main contents are as follows:(1)The image acquisition scheme for 2D/3D glass cover of mobile phone is studied.Firstly,aiming at the problem of reflection interference caused by the 3D surface features of the detected object,the specular reflection light path on the surface is modeled and analyzed,and the image acquisition scheme of transforming the 3D surface into some planes is proposed.Then,due to the limited depth of field of imaging system,an all-in-focus image of 3D surface at varying distances in a scene cannot be captured.Therefore,the image acquired will get blurred due to the height difference between upper and lower surfaces.Faced with this problem,a multifocus image fusion method based on the combination of wavelet domain and spatial domain techniques is proposed.Finally,the imaging system scheme of glass cover is introduced,and an experimental platform is built up correspondingly.(2)The simulation algorithm of typical defects on glass cover based on structure and tex-ture generation models is studied.During a practical detection process,insufficient defect data,unbalanced defect types and the high cost of defect labeling can present problems.Therefore,it often takes considerable time and labor to collect actual samples to improve the accuracy of defect classification and recognition.To overcome the above problem,an artificial defect simulation algorithm is proposed,which includes three parts: external contour model,inter-nal structure generation model and texture generation model.By controlling and changing the parameters of the proposed algorithm,different types of artificial defects can be generated ran-domly.The experimental results confirm that the proposed algorithm can effectively simulate the characteristics of the actual defects of glass cover.(3)The defect detection algorithm of glass cover window area based on adaptive enhance-ment and feature cascade network is studied.Firstly,to ensure the generality and robustness of the detection algorithm,some preprocessing algorithms are introduced,such as pre-examination algorithm for sample location and ROI acquisition algorithm based on the shape template match-ing technology.Secondly,to accurately extract the defects with low contract,an adaptive image enhancement method combining Gaussian filter-based image difference technology and linear transformation is introduced.Next,to solve the problem of missing detection of some cluster defects,a clustering strategy of cluster defects is proposed.Based on the above algorithms,a cascade algorithm of BP neural network and CNN is proposed to achieve accurate and efficient detection of a variety of defects.It is a coarse-precise classification algorithm,which can ef-fectively solve the problems of defect standard comparison and accurate classification.Finally,the effectiveness and satisfactory performance of the algorithm are verified by experiments.(4)The defect detection algorithm based on Gabor convolutional layer and rotation region proposal network is studied.Due to the variety of scale and random direction of edge defects,the generality and robustness of traditional edge detection algorithms can not be guaranteed.To overcome the above problems,an intelligence-based object detection algorithm is proposed,which can realize accurate detection of multi-scale and arbitrary edge defects.The feature pyra-mid network(FPN)is utilized to perform the fusion of different scale feature maps,and the rotation region proposal network(R-RPN)is employed to generate proposals with different ra-tios,angles and scales.In addition,a Gabor filter layer is applied before the convolutional backbone to reduce background interference and enhance the characteristics of the irregular lin-ear edge defects.Finally,an end-to-end edge defect detection strategy is presented,and some experiments are performed to show the effectiveness of the proposed method.Finally,the research contents of this dissertation are summarized,and the future research direction is proposed.