Enhancing Data Through Augmentation to Address Few-class Scenarios of Few-shot Learning

Automatic prediction of defect types is a crucial aspect in ensuring the quality of production outputs. With the increasing demand for superior quality and complexity, maintaining efficiency and practicality has become paramount. Although deep neural networks are commonly utilized for fault diagnosis, they necessitate a significant amount of data. By employing few-shot learning methods, classification tasks can be performed with a limited number of image samples. However, utilizing a small number of classes (few-classes) in few-shot learning may lead to subpar model performance.