Examination of image data augmentation in CNN for the purpose of visual defect inspection process

Abstract

In recent years, deep learning has come to be used in various situations at production sites and its recognition and attention continue to rise. This is because the demand for automation and further efficiency has increased due to the background of the times such as labor shortages in that sites, and deep learning has become a powerful method to meet that demand. In this paper, we aim to improve the detection accuracy from the viewpoint of data augmentation in order to apply the existing object detection algorithm YOLOv3 for the industrial application of deep learning.