Authors:
S. Ajay Karthick, U. R. Sudarsananarayanan, I. I. Shrinaath, B. Deepa, Dilli Kasi Rao Kotha
Addresses:
Department of Computer Science and Engineering, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil Nadu, India. Department of Electronic and Communication Engineering, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil Nadu, India. Faculty of Engineering, Environment and Computing, Coventry University, Coventry, England, United Kingdom.
Defect detection on steel surfaces presents an important feature of Industry 4.0 and intelligent manufacturing, providing building safety, product quality assurance, and reducing production waste. Original quality control processes involve a lot of manual visual checking. Even the human visual approach requires too many resources, is highly personal, and does not allow one to keep up with the huge amount of data that has been generated by modern continuous steel rolling mills. Deep learning techniques, particularly efficient one-stage detectors such as those in the YOLO family, have been shown to cope with real-time defect detection tasks. At the same time, the practical execution of these algorithms, which are used to detect surface defects under the conditions of a chaotic industrial environment, faces some technical challenges. Particularly, when the problem is characterized by very low visual contrast, elongated and slender defects, very high inter-class similarity, and widespread texture noise, this results in relatively high false negative detection rates and poor localization capabilities of the conventional detectors. Therefore, to overcome the deficiencies of the current detectors, this paper proposes a new attention-enabled defect detection model based on a strong YOLOv8 baseline. The main contribution is that the architecture significantly enhances accurate localization and classification, while strictly maintaining the low interference needed by real-time feeds.
Keywords: Deep Learning; Intelligent Manufacturing; Quality Assurance; Low Visual Contrast; Texture Noise; Industrial Environment; Defect Detection; Steel Surfaces; Building Safety.
Received on: 17/07/2025, Revised on: 30/10/2025, Accepted on: 09/11/2025, Published on: 07/10/2026
DOI: 10.69888/FTSSM.2026.000815
FMDB Transactions on Sustainable Structures and Materials, 2026 Vol. 2 No. 2, Pages: 88-96