Steel Surface Defects Detection Based on Deep Learning

Abstract

Surface defects detection plays a significant role in quality enhancement in steel manufacturing. However, manual inspection of steel surface slows down the entire manufacturing process and is time consuming. Currently, many methods have been proposed for automatic defect detection on hot-rolled steel surfaces. These methods usually follow two steps: pre-processing and segmentation. The pre-processing step is intended to overcome the uneven illumination of images while the segmentation step generates a binary map to identify defects. This kind of method heavily depends on feature selection approaches, but the defect features are usually not easy to obtain. In this paper, we propose an automatic steel surface defects detection method based on deep learning. Two deep learning models for defect detection are evaluated. The experimental results show that the evaluated methods can detect steel surface defects more effectively and accurately than the traditional methods. This approach can be also applied to other industrial applications.