Study on Abrasive Grains Detection based on Deep Learning

Abstract

The objective of this study is to detection abrasive grains using deep learning. We have used SSD for the object detection model to detect the abrasive grains, but in this study, we chose EfficientDet, which is an advanced form of SSD. This model consists of three stages. First, a backbone network based on EfficientNet. EfficientNet is a classification model using deep learning. Second, fusion multiple scale each feature maps that named BiFPN Layer. The third is the part that classifies and locates the abrasive grains. In this paper, we changed the aspect ratio of the default box to detect abrasive grains. Finally, we succeeded in determining the most suitable aspect ratio for the detection of abrasive grains.