Performance of ResNet and Transformer-Based Models in the Classification of Steel Scrap

Abstract

The classification of steel scrap is of significant economic and environmental importance, particularly in the production of steel using electric arc furnaces (EAFs). EAFs primarily produce steel by utilizing scrap steel, offering an environmentally friendly alternative due to its lower energy consumption and reduced carbon emissions. This study analyzes the performance of different deep learning models in the classification of steel scrap. Detailed comparisons were made by training ResNet and Vision Transformer-based models on a dataset where different types of scrap were labeled. The results indicate that the ResNet34 model is the most successful model.