Article Details

Abstract

To address the imbalance between detection accuracy and computational efficiency in lightweight object detection models for metal surface defect inspection, a structural optimization strategy based on YOLOv11n is investigated. Experiments are conducted on the NEU-DET dataset. Without altering the original multi-scale feature fusion framework, the C3k2-PConv module is introduced into the backbone and neck networks to reduce redundant computation and enhance feature representation efficiency. In addition, an LSCD-based detection head is adopted, and a Local Quality Estimation (LQE) mechanism is incorporated to optimize the localization-quality ranking of predicted bounding boxes, thereby improving the selection performance during the non-maximum suppression stage. Experimental results demonstrate that the proposed model achieves improved detection accuracy while significantly reducing the number of parameters and computational complexity. The mAP is higher than that of the baseline model, validating the effectiveness and practical applicability of the proposed method for lightweight metal surface defect detection tasks.

Keywords

NEU-DET; metal surface defect; Object Detection; YOLO; Partial Convolution (PConv)

References

1. [1] Luo Dongliang, Cai Yuxuan, Yang Zihao, etc A review of deep learning methods for industrial defect detection [J]. Chinese Science: Information Science, 2022, 52 (6): 1002-1039
2. [2] Lecun Y，Bottou L，Bengio Y，et al.Gradient-based Learning Applied to Document Recognition [J].Proceedings of the IEEE，1998，86（11）：2278-2324.
3. [3] Wu Yawei, Ming Bangming, He Jianfeng, etc Surface defect detection of lightweight steel based on YOLO-GR algorithm [J]. Combined Machine Tool and Automation Processing Technology, 2023 (11): 107-111115
4. [4] Li Sisi, Ge Huayong, Meng Yaqun, etc Steel surface defect detection algorithm based on AGD-YOLO [J]. Computer Measurement and Control, 2025, 33 (11): 142-148
5. [5] Li Heping, Chen Zhongju, Xu Haoran, etc Improving YOLO's algorithm for detecting surface defects in steel [J]. Modern Electronic Technology, 2024, 47 (13): 7-14
6. [6] Chen J R，Kao S H，Hao He，et al. Run, Don't Walk: Chasing Higher FLOPS for Faster Neural Networks [C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)，2023：12021-12031.
7. [7] Zhai Yongjie, Tian Jiming, Chen Penghui, etc Lightweight Phalaenopsis Seed Target Detection Algorithm Based on YOLOv8n Alolite [J]. Journal of Agricultural Engineering, 2025, 41 (4): 220-229
8. [8] Zhang Z，Li Y，Bai Y，et al. Convolutional Graph Neural Networks-based Research on Estimating Heavy Metal Concentrations in a Soil-rice System [J].Environmental science and pollution research international，2023，30（15）：44100-44111.
9. [9] Li X，Wang W H，Hu X L，et al. Generalized Focal Loss V2: Learning Reliable Localization Quality Estimation for Dense Object Detection [C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)，2021：11627-11636.
10. [10] Xu W，Wan Y，Zhao W N. ELA: efficient location attention for deep convolution neural networks [J/OL].Journal of Real-Time Image Processing，2025，22（4）：140.https://link.springer.com/article/10.1007/s11554-025-01719-6.
11. [11] Li C F，Xu A，Zhang Q B，et al. Steel Surface Defect Detection Method Based on Improved YOLOX [J].IEEE Access，2024，12：37643-37652.
12. [12] Lei Fuqiang, Ma Liuwen, Guan Peng, etc Research on Surface Defect Detection Algorithm for Steel Equipment Based on DFD-YOLOv11n [J]. Computer Engineering and Applications, 2026, 62 (2): 92-102
13. [13] He Pengfei, Chen Wanxin, Li Wei, etc MEP Net: Improved Model for Aluminum Metal Cover Defect Detection Based on YOLOv11n [J]. Packaging Engineering, 2025, 46 (21): 190-200
14. [14] Qiao Y F，Wei Y R，Wang K. Fast-YOLOv12: An Attention-guided Lightweight Network for Real-time Steel Surface Defect Detection [J/OL].Journal of Real-Time Image Processing，2025，23（1）：9.https://link.springer.com/article/10.1007/s11554-025-01807-7.

About the authors

Yang Ze (1993-), male, Han ethnicity, from Tonghua, Jilin, engineer, bachelor's degree, research interests include aviation maintenance; Xia Junnan (1990-), male, Han ethnicity, from Liaoyang, Liaoning, engineer, bachelor's degree, research interests include aircraft maintenance; Qi Qingguo (1971-), male, Han ethnicity, from Jilin, Jilin, engineer, bachelor's degree, research interests include aviation maintenance; Li Jinsuo (1997-), male, Han ethnicity, from Huludao, Liaoning, engineer, bachelor's degree, research interests include aviation maintenance; Corresponding author: Cui Liuqing (2001-), male, Korean ethnicity, from Hunchun, Jilin, assistant engineer, bachelor's degree, research interests include aviation maintenance.