A Defect Classification Network Based on Deformation Dense Connection in Wire Rod Surface Image

Surface defects significantly impair the quality of wire rod products in steel manufacturing industry. In this paper, a defect classification network with deformation dense connection is proposed to integrate low-level texture features and high-level semantic features of defect image, in order to fully utilize the low-level texture feature. The deformation dense connection is composed of Dense Block(DB) and down-sampling cross-layer connection to enhance the utilization of low-level feature while avoiding the increasing of the depth of the network. Experimental results show that the proposed method achieved 97.02% classification accuracy. The performance of the proposed method is qualified in the application of quality control of wire rod production.