An improved Vision Transformer-based method for classifying surface defects in hot-rolled strip steel

Abstract A new Vision Transformer(ViT) model is proposed for the classification of surface defects in hot rolled strip, optimizing poor learning ability original Transformer on smaller datasets. Firstly, each module ViT and its characteristics are analyzed; Secondly, inspired by deep VGGNet, multilayer fully connected layer VGGNet introduced into to increase capability; Finally, performing X-SDD hot-rolled steel strip defect dataset. The effect improved algorithm verified comparison experiments test results show that achieves better than terms accuracy, recall, F1 score, etc. Among them, accuracy set 5.64% higher ViT-Base 2.64% ViT-Huge; 4.68% 1.36% both respectively.