Steel Surface Defect Classification using Few-Shot Learning

A fundamental requirement of the manufacturing process is the detection of product defects, and deep learning-based models for quality control are already extensively used in this domain. While there is a drawback to this method which is requires a large dataset to function well, there is also an alternative that uses deep learning to identify defect types from a small sample of datasets, using a few-shot learning approach. In this paper we attempt to use the NEU dataset for few-shot learning, however due of the restricted diversity in the classes, the outcomes might not be ideal if we just use the dataset.