Mobile Phone Glass Cover Defect Detection Based On Deep Learning

As an important part of mobile phone screen,mobile phone glass cover has the function of protecting touch screen and ensuring light transmittance.In the production process,there will inevitably exist product defects and quality problems,including scratch,concave and convex points,ink stains,spots and so on.In order to improve the product yield,it is necessary to carry out effective quality inspection on the generated mobile phone glass cover.In order to solve the problem of mobile phone glass cover defect detection.Based on deep learning image processing theory,the main research of this paper are as follows.(1)In this paper,a two-sided telecentric linear array vision image acquisition system for mobile phone glass cover defect detection is built,and the imaging model and dynamic scanning imaging calibration method are studied.The geometric imaging model of the twosided telecentric linear array imaging system is established by the projection like transformation of the scanning direction of the linear array camera,and then combined with the affine imaging.The various distortion factors affecting the imaging effect are analyzed,including the distortion caused by the fluctuation of the scanning direction velocity and the distortion of the imaging lens.The internal and external parameters of the proposed imaging model are solved and the calibration process is given Real calibration experiments and simulation experiments verify the effectiveness of the proposed method.The experimental results show that this method can effectively calibrate the line and column directions of bilateral telecentric linear array imaging system.(2)Aiming at the problem of defect recognition in the defect detection of mobile phone glass cover plate,an improved deep learning network for defect recognition based on Yolo and SSD is proposed in this paper.Combined with the network structure and loss function characteristics of SSD and Yolo?v3,an improved lightweight network for mobile phone glass cover defect detection is proposed.Then this paper gives the experimental scheme and evaluation standard of defect recognition network,and the effectiveness of the recognition effect of the lightweight improved network on the defect recognition data set is verified by experiments.Through the actual product defect detection experiment of mobile phone glass cover samples,the accuracy and real-time detection speed of the defect detection algorithm are verified.(3)Aiming at the problem of defect classification in mobile phone glass cover defect detection,this paper develops a balanced simulation sample generation method based on Ga N network,and explores the mobile phone glass cover defect classification algorithm based on deep learning.In this paper,a method based on Generative AdversarialNetworks(GAN)for defect simulation is proposed,and the simulation results of concave and convex points,ink stains and scratches are evaluated through defect simulation experiments.Based on deep learning theory,the defect classification model of mobile phone glass cover plate is designed.Through the defect classification experiment,it is verified that the balanced sample generation using Gan network can effectively improve the classification ability of the defect classification network,and the deep learning classification network based on resnet50 can obtain the optimal defect classification accuracy.Experiments show that the classification effect of the defect classification method and the simulation effect of Gan network defects meet the requirements of this paper.