Research On Mobile Phone Screen Defect Detection Method Based On Deep Learning

Mobile phone screen has high hardness,high strength,scratch resistance,high transmission rate and excellent impact resistance,but in its production process,there will be some defects,affecting the quality and appearance of mobile phone.Therefore,it is necessary to adopt effective methods to detect the defects of the mobile phone screen,so as to improve the corresponding production process and improve the production quality of the mobile phone screen.In recent years,the development of deep learning has become more and more mature,mainly because of its excellent classification effect.This paper will take the defects of mobile phone screen as the research object,and detect the defects of mobile phone screen based on the method of deep learning.First of all,the characteristics of the defect of the mobile phone screen were analyzed,and the defect detection standards in this paper were defined,and the defects were determined to be divided into dust,point stab wound,line stab wound,scratch,hair silk,face stab wound,dust blob,stain,edge breakage defect,character defect and IR hole defect.The defect extraction algorithm is studied.Aiming at the universality of the algorithm for different types of mobile phone screens and the extraction accuracy of defects,this paper divides the defect extraction algorithm into four small modules.The information acquisition module is mainly used to solve the universality problem.Preliminary defect detection is mainly used to improve the efficiency of the program.The regional segmentation module mainly solves the problem of defect extraction accuracy.The defect extraction module adopts targeted image processing for each region,so as to accurately extract defects.Then,the defect classification algorithm is studied,and the multi-classifier method is adopted to classify the defects,which can improve the accuracy of classification.The classification items are: point-line-surface classification,point-defect classification,line-defect classification,face-defect classification,character-defect classification and IR hole defect classification.Six network structures of Inception-v3,Inception-v4,Resnet-v2-152,Inception-resnet-v2,Vgg16 and Vgg19 are studied,and the basic framework of defect classification algorithm is designed.Secondly,aiming at the imaging characteristics of defects,I participated in the discussion on the development of imaging hardware system,which solved the problem of high detection accuracy and low imaging contrast.Finally,the performance of six kinds of neural networks was tested for each classification item,and the network structure with the best performance was selected.The defect detection program of mobile phone screen was integrated,and the stability and accuracy of defect extraction and classification were tested with three different types of screens.The experimental results show that the defect detection program proposed in this paper has a miss rate of 0.87%,an overpass rate of 0.53%,an accuracy of 97.096%,and a detection time of 2.33s/p,which meets the basic requirements of industrial detection.