SSDC-MiniNet: A Tiny CNN Steel Surface Defect Classifier

SSDC-MiniNet: 소형 강철 표면 결함 분류 모델

Steel surface defect classification is important in producing an optimal product. However, worker errors while classifying the metal surface can lead to issues in production. So, recently researchers found a way to invest more in Industry 4.0 technology with methods such as Convolutional Neural Networks (CNNs). But models can be too large to be implemented in the factory. In response, we proposed a tiny model called SSDC-MiniNet with only 23,588 parameters and 0.105 GFLOPS. The model was trained on the North Eastern University (NEU) dataset, which consists of 1,800 images of six common metal surface defects. After testing the model, we achieved a competitive accuracy of 98.67% with only around 1% of the parameter count and flops.