A Novel Deep Learning Method for Detecting Defects in Mobile Phone Screen Surface Based on Machine Vision

Abstract

With the innovations in technology, the interest in the use of mobile devices is increasing day by day. Any defect that may occur during the production of smart mobile phones, which is among mobile devices, causes significant damage to both the manufacturer and the user. The careful detection of defects that may occur on the screen glass, which is one of the most striking defects among these defects, with the human eye significantly affects the workforce cost. Therefore, it is important to detect defects with the help of software. In recent years, many methods based on machine vision have been developed for the detection of any object or difference in the image. In this study, a new model structure called Yolo-MSD, based on machine vision and the Yolo-v3 deep learning model, which detects and classifies oil, scratch, and stain defect types on the glass on the touch screen surface used in the design of smart mobile phones, is proposed. The proposed model structure (Yolo-MSD) is obtained by reducing the number of blocks in the Darknet-53 network structure developed in Yolo-v3. As a result of the training, a success rate of 98.50% with the Yolo-v3 model and 98.72% with the Yolo-MSD model was achieved in detecting and classifying defect types. Therefore, it has been observed that the Yolo-MSD model structure is better than the Yolo-v3 model structure by making better feature extraction from the types of defects on the screen glass since it is both faster and has less complexity.