Steel Sheet Defect Detection Based on Deep Learning Method

Steel sheets have been widely used in the industrial field. With higher requirements for steel production, there is a growing need for factories to produce better quality steel sheets. Conventional steel sheet defect detection methods such as manual inspection are too laborious and inefficient. Therefore, in this paper, we manage to explore a possible solution for steel sheet defect detection and propose a novel image-based processing method. The image processing data enhancement method is used to extend the datasets for further training, then we use the transfer learning technique to train CNNs and extract features on the enhanced image set. A hierarchical model ensemble is applied to detect defects according to their locations. Experiments on enhanced datasets and real-world defect images achieve satisfying accuracy.