Defect Diagnosis of Hot Rolling Steel Surfaces Based on Deep Learning

Abstract

Robust visual inspection systems have gotten more consideration in the area of product quality control in an effort to increase production effectiveness. The issue of hot-rolled steel strip surface examination utilising machine vision systems is the focus of the current study. A novel approach is suggested to diagnose (verify/classify) surface defects. The presented approach is a new simple construction supervised classifier “SC” based on Convolutional Neural Network (CNN) technique. The proposed approach has been applied and tested on a certified hot rolled steel surface defects dataset including healthy class and six common categories of steel flaws. An analysis of contrasts between SC and associated work approaches was performed. Results show that the proposed approach presents excellent performance in diagnosing flaws in the surface of hot rolled steel. The surface flaws dataset test results with 50% k fold for SC are 99.33% respectively.