Article Details

Abstract

Addressing the issues of low detection precision, poor generalization, and insufficient real-time performance in industrial part surface defect detection, this paper proposes a Deep Learning detection method based on YOLOv8. The research utilizes YOLOv8m as the baseline model and conducts optimized training on the Northeastern University hot-rolled steel strip surface defect dataset by introducing data augmentation strategies (geometric transformations and color perturbations), label smoothing techniques, and a “Linear Warmup + Cosine Annealing” learning rate schedule. Experimental results demonstrate that the model achieves a mean Average Precision of 0.619 on the test set, with an overall recall rate of 0.94, representing a 2.1% precision improvement over YOLOv5m, while the inference speed is significantly superior to traditional models such as Faster R-CNN. The conclusion indicates that this method achieves high-precision detection while maintaining high real-time performance, effectively meeting the requirements for automated quality inspection in complex industrial scenarios.

Keywords

YOLOv8; Hot-rolled steel strip; Surface defect detection; Data augmentation

References

1. [1] Ma Jinlin, Cao Haojie, Ma Ziping, etc Improving the defect detection of YOLOv8 strip steel [J]. Computer Engineering and Applications, 2024, 60 (21): 183-193
2. [2] Li Shun, Yang Ying Defect detection of hot-rolled strip based on improved YOLOv5 [J]. Computer Simulation, 2024, 41 (5): 197-203
3. [3] Xu Yanwei, Li Jun, Dong Yuanfang, etc Overview of YOLO series object detection algorithms [J]. Computer Science and Exploration, 2024, 18 (9): 2221-2238
4. [4] Terven J，Córdova-Esparza D-M C，Romero-González J A R.A Comprehensive Review of YOLO Architectures in Computer Vision: From YOLOv1 to YOLOv8 and YOLO-NAS [J].Machine Learning and Knowledge Extraction，2023，5（4）：1680-1716.
5. [5] Safaldin M，Zaghden N，Mejdoub M. An Improved YOLOv8 to Detect Moving Objects [J].IEEE Access，2024，12：59782-59806.
6. [6] Dou Zhi, Gao Haoran, Liu Guoqi, etc Small sample steel plate defect detection algorithm for lightweight YOLOv8 [J]. Computer Engineering and Applications, 2024, 60 (9): 90-100
7. [7] Xiao Xuyuan Research on Real time Defect Detection Algorithm for Hot Rolled Strip Steel Driven by Deep Learning [D]. Chengdu: University of Electronic Science and Technology of China, 2025
8. [8] Chen Y L，Hung K C，Zhang J Y，et al. Improved YOLOv8n Models for Object Detection in Remote Sensing Images [J].IEEE Access，2025，13：97889-97900.
9. [9] Huang C Y，Cui J N，Li Y L，et al. Fusion YOLOv8s and Dynamic Convolution Algorithm for Steel Surface Defect Detection [J].Symmetry，2025，17（5）：701.
10. [10] Full text, Luo Hongtao Improving the Underwater Object Detection Algorithm of YOLOv8M [J]. Computer Technology and Development, 2025, 35 (1): 30-37
11. [11] Zhang Xiuquan Research on Rotation Object Detection Algorithm for Remote Sensing Images Based on YOLOv8L [D]. Taiyuan: North China University, 2024
12. [12] Long Y，Lin W S. Surface Defect Detection of Ultrathin Fiberboard Based on Improved YOLOv8x [J].Journal of Nondestructive Evaluation，2025，44（2）：58.
13. [13] Fajar P D，Moses L S，Dedy D P. Beyond Architecture: Hyperparameter Optimization for YOLOv8m in Multi-class Textile Defect Detection [J/OL].Results in Engineering，2025，28：108356.https://doi.org/10.1016/j.rineng.2025.108356.
14. [14] Su Weijiang, Sun Liuliu, Gao Yongqiang, etc Optimization of Infrastructure Violation Identification Method Based on Cross Entropy Loss Function [J]. Computer Simulation, 2025, 42 (3): 487-491
15. [15] Shi Shuo, Qin Jiajun, Yu Yang, etc Audiovisual Emotion Recognition Based on Improved ConvMixer and Dynamic Focus Loss [J]. Acta Sinica, 2024, 52 (8): 2824-2835
16. [16] Mei S Z，Liu C，Lyu X F. A Falls Recognition Framework Based on Faster R-CNN and Temporal Action Sequences that can be Deployed on Home Service Robots [J/OL].Measurement Science and Technology，2024，35（8）：085005.https://iopscience.iop.org/article/10.1088/1361-6501/ad44c9.
17. [17] Zhu Yu, Hao Xiaoli, Niu Baoning, etc Improvement of RetinaNet Object Detection Algorithm [J]. Computer Engineering and Design, 2024, 45 (11): 3271-3278
18. [18] Li J D，Tong C Y，Yuan H X，et al. A Complex Environmental Water-level Detection Method Based on Improved YOLOv5m [J].Sensors，2024，24（16）：5235.

About the authors

Liu Zhigang (June 1986-), male, Han ethnicity, from Weifang, Shandong, is an engineer and a master's student. His research interests include artificial intelligence, data annotation, and trustworthy data space; Zhao Xiaohui (May 1986-), female, Han ethnicity, from Yuncheng, Shanxi, engineer, master's student, research direction: artificial intelligence, digital economy.