Smart Phone Screen Defect Detection Based On Deep Convolutional Neural Network

In recent years,with the rapid development of the network era,smart phones replace traditional computers into people's life in all aspects gradually.Smart phone not only satisfies people's communication requirements,but also meets people's entertainment needs.The screens of the smart phones become a window to interact with the information world for people.As living standards improve,the quality requirements of smart phone screens are increasing.In the manufacturing process of the smart phone screen,it is still inevitable to produce a smart phone screen with defects.How to improve the detection accuracy of smart phone screen defects and shorten the detection time is an urgent problem in the current production.At present,the main methods of smart phone screen defect detection are manual detection and machine vision detection.The use of workers to identify smart phone screen defects is not efficient,the cost of labor is high and continues to grow,and the individual's vision is still different,which makes it diff-icult to guarantee the quality of the detection.Machine vision inspection is a semi-automatic detection method.By studying the pixel distribution characteristics of images,digital image processing is used to extract features in a targeted manner,and then the types of screen defects are determined according to the characteristics of feature representation.Although the machine vision detection method can detect screen defects to a certain extent,the algorithm process is cumbersome,the detection accuracy is not enough,and the detection time is relatively long.In this paper,the existing smart phone screen defect detection methods were reviewed and improved.Aiming at six common defects of smart phone screens,the digital image processing method was used to preprocess data and expand data.The LabelImg annotation tool was used to label the image area and category the defects for each defective images.The deep convolutional neural network was used to automatically extract the features from the smart phone screen defect pictures.Then the extracted features were used to train the parameters of the Faster R-CNN algorithm until the system converges.The feasibility of the Faster R-CNN algorithm was verified on the standard dataset PASCAL VOC 2007.Experiments were carried out on the smart phone screen defect dataset using Faster R-CNN algorithm,and the superior performance of the algorithm on the smart phone screen defect dataset was obtained.The method used in this paper was compared with current mainstream smart phone screen detection method,which highlights the great advantages of the Faster R-CNN algorithm.To display the test result clearly,a front-end detection interface was written using PyQt.The algorithm was rewritten in a GPU version to observe the difference between GPU version algorithm and non-GPU version algorithm in training and testing time.The experimental results show that the GPU has excellent acceleration performance.The smart phone screen defect detection algorithm studied in this paper provides theoretical basis and technical solutions for industrial detection of smart phone screen defects,which has great application value and development prospect.