DefectNET: multi-class fault detection on highly-imbalanced datasets

N. Anantrasirichai, David Bull

As a data-driven method, the performance of deep convolutional neural networks (CNN) relies heavily on training data. The prediction results of traditional networks give a bias tow…