Article Details

Abstract

To address the issues of limited detection accuracy and insufficient feature representation capability under complex textured backgrounds in copper strip surface defect detection, a lightweight detection method based on the improved RT-DETR is proposed. By comparing different backbone networks, RT-DETR-R18, which has fewer parameters and lower computational cost, is selected as the base structure. In the backbone network, a Wide-Deep Dual-Branch (WDBB) module is introduced to enhance multi-scale feature extraction capability, and the global self-attention in the AIFI module is replaced with a deformable attention mechanism (DAttention) to guide the network to adaptively focus on high-value sampling areas, improving feature interaction and defect discrimination. Experimental results show that the improved model achieves precision, recall, and mAP@0.5 of 96.10%, 95.19%, and 97.60%, respectively, with overall performance better than that of the original model, demonstrating the effectiveness of this method in copper strip surface defect detection tasks.

Keywords

strip surface defect; RT-DETR; multi-branch structure; Attention Mechanism; detection method

References

1. [1] Gu Yingkui, Ye Biaobiao, Guo Mingjian, etc Rapid detection of appearance defects in biscuit packaging based on improved RT-DETR [J]. Food and Machinery, 2025, 41 (2): 234-241
2. [2] Zhang Jingwen, Sun Jian, Xu Hongwei, etc Glass insulator defect detection algorithm based on improved RT-DETR [J]. Electronic Measurement Technology, 2025, 48 (14): 96-105
3. [3] Wang Liang, Ouyang Basheng A USB-C connector appearance inspection model based on improved RT-DETR [J]. Automation and Instrumentation, 2025, 40 (1): 36-42
4. [4] Hu Jiwen, Zhang Guoliang, Shen Mingzhe, etc Improved RT-DETR model for surface defect detection of pine wood [J]. Journal of Agricultural Engineering, 2024, 40 (7): 210-218
5. [5] Liu Yameng, Zhao Youquan, Sun Zhentao, etc Constructing an Improved RT-DETR Algorithm for Detecting Circular Ripple Defects in Contact Lenses [J]. Journal of Electronic Measurement and Instrumentation, 2024, 38 (5): 1-9
6. [6] Zhou Dehui, Zhao Jun, Cheng Jinfeng Detection Algorithm for Minor Defects on Bearing Surface Based on RT-DETR [J]. Computer Applications, 2025, 45 (6): 1987-1997
7. [7] Zhao Y，Lyu W Y，Xu S L，et al. DETRs Beat YOLOs on Real-Time Object Detection [C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition，2024：16965-16974.
8. [8] He K M，Zhang X Y，Ren S Q，et al. Deep Residual Learning for Image Recognition [C]//2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR)，2016：770-778.
9. [9] Wan D H，Lu R S，Hu B T，et al. YOLO-MIF: Improved YOLOv8 With Multi-Information Fusion for Object Detection in Gray-Scale Images [J/OL].Advanced Engineering Informatics，2024，62：102709.https://doi.org/10.1016/j.aei.2024.102709.
10. [10] Xia Z F，Pan X R，Song S J，et al. Vision Transformer With Deformable Attention [C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition， 2022：4784-4793.
11. [11] Zhang Wei, Cai Yufan, Ye Lintao, etc Real time Transformer Small Object Detection Model Based on Feature Extraction Enhancement and Pyramid Structure [J]. Computer Science, 2025, 52 (S2): 375-385
12. [12] Wang C Y，Bochkovskiy A，Liao H Y M. YOLOv7: Trainable Bag-of-Freebies Sets New State-of-the-Art for Real-Time Object Detectors [C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition，2023：7464-7475.
13. [13] Varghese R J，Sambath M. YOLOv8: A Novel Object Detection Algorithm With Enhanced Performance and Robustness [C]//2024 International Conference on Advances in Data Engineering and Intelligent Computing Systems (ADICS)，2024：1-6.
14. [14] Khanam R，Hussain M. YOLOv11: An Overview of the Key Architectural Enhancements [PP/OL].（2024-10-23）.https://arxiv.org/abs/2410.17725.

Funding

Collaborative Education Project of the Ministry of Education (202101355003)

About the authors

Chen Long (1996-), male, Han ethnicity, from Hubei Xiaoren, is currently pursuing a master's degree in computer vision; Corresponding author: Li Xuebing (1970-), male, Han nationality, from Liuyang, Hunan, associate professor, master's student, research direction: computer vision, enterprise informatization.