A Method for Surface Detect Classification of Hot Rolled Strip Steel based on Xception

Abstract

Hot rolled strip steel is an important raw material for automobile, home appliance and other manufacturing industries, and the quality of its surface and plate shape has a vital impact on the products produced by end users. In actual industrial production, different measures should be taken for different kinds of strip steel surface defects. Therefore, it is of great significance to classify the surface defects of hot rolled strip accurately. An improved method based on Xception algorithm is presented. The algorithm can classify the hot rolled strip defects and is more suitable for the imbalance between categories to some extent. Compared with 91.18% of the original Xception algorithm, the classification accuracy of the improved algorithm reached 93.87% on the hot rolled strip defect dataset. The improved scheme solves the problem of unbalanced dataset samples to a certain extent and improves the classification accuracy of dataset significantly.