Deep Model Based Phone Screen Detection And Classification

Screen is an important component in mobile phone devices.For screen manufacturers,it is urgent to develop an efficient,accurate and intelligent method to detect screen defects.With rapid development of machine vision and deep learning,many intelligent detection algorithms have been proposed.Although these algorithms have reached good performance for one special type of screen,they cannot be applied to another type of screen.Besides,these algorithms lack the ability of making a fine classification of the defects.For these reasons,we propose a defect detection system for mobile phone screens.The system mainly consists of a defect detection algorithm and a defect classification algorithm.The detection algorithm is superior in accuracy and generality,while the classification algorithm has a good performance in fine classification of screen defects.The defect detection algorithm we proposed is a combination of the convolutional neural networks(CNN)and GAN.First,we apply CNN to extract feature maps of screens,and take these feature maps as GAN's input to train the GAN model.Then,we use the discriminator of the pre-trained GAN as the final classifier.The classification results are regarded as the detection results of screen defects.In this paper,the defects detection problem is changed into a binary classification problem.The deep-learning model we obtained can effectively prevent interference from background noises and thus improve algorithm's generality.Methods that only detect defects of screens cannot meet manufacturer's requirements nowadays.Defects classification algorithm is also needed and has been gradually developed.In this paper,we put forward a method based on XGBOOST to solve the defects classification problem.We extract 8-dimensional vectors from the defects and use them to train the defects classification model and to get final classification results.Experimental results show that the system we proposed has good performance in defects detection and classification,which can bring great help to screen manufacturers.