Applications and Progress of Image Recognition Techniques in Scrap Steel Classification and Grading: A Review

Abstract

With the growing demand for efficient and sustainable scrap steel recycling, intelligent classification and grading technologies based on image recognition have become essential for enhancing productivity and optimizing resource utilization. This review systematically examines recent advancements in scrap steel image recognition, emphasizing the application of deep learning techniques that outperform traditional machine learning methods in accuracy and efficiency. It highlights how convolutional neural networks and attention mechanisms enhance feature extraction, improve interpretability, and enable robust automatic classification in complex industrial environments. The review also explores the pivotal role of data acquisition strategies in ensuring model performance, underscoring the importance of data quality, diversity, and annotation in developing effective recognition systems. Furthermore, it analyzes the application of attention mechanisms in detail, demonstrating their ability to focus on salient image regions and enhance recognition accuracy. Finally, this review summarizes the key challenges in current research—such as limited domain-specific datasets, poor generalization across diverse scenarios, and constraints on real-time deployment—and outlines future research directions aimed at developing adaptive, interpretable, and scalable intelligent classification systems, providing valuable insights and references for advancing automated scrap steel processing.