Paper Title
A CNN-based Defect Inspection for Smartphone Cover glass

Abstract
Due to the high transmittance, high strength, scratch resistance, and waterproof ability, glass becomes the first choice of smart phone cover material. For the purpose of ensuring the quality of smart phone cover glass(SCG), glass surface defect detection plays an important role on the production line. In this paper, a deep learning algorithm, convolution neural network(CNN), is used to find the SCG defect features by training classification models. Sliced from 25 raw defective SCG images taken by a 4K line-scan camera image-capture system, 336 images with 7 kinds of defect images and in defective images are sorted out as the dataset. The ability of different CNN models to identify various defects is compared. Additionally, image augmentation methods are introduced to increase the number of images. Different image processing methods are compared to improve the recognition ability as well. Finally, the accuracy of 98.15% is achieved, and the steps can be unified into a systematic strategy to apply to SCG defect detection. Compared with traditional automatic optical inspection (AOI) methods, time and labor costs on algorithm parameter adjustment during development and update a detect system can be significantly reduced. Keywords - Defect Detection, Deep Learning, Smart Phone Cover Glass, Image Augmentation, Image Processing