Article Details

Abstract

In printing appearance inspection, complex image backgrounds easily lead to missed and false detections in anomaly detection models. To address this issue, this study proposes a product appearance defect detection method that combines YOLOv13 and unsupervised learning. The method uses YOLOv13 to extract the printing area and remove background interference, and then applies the EfficientAD model to the printing area images for defect detection. To verify the effectiveness of the method, this study constructs a self-built chip capacitor printing dataset and conducts comparative experiments. The experimental results show that after YOLOv13 is used to extract the printing area, the image-level area under the receiver operating characteristic curve (image_AUROC) increases by 25.4 percentage points and the image-level F1 score (image_F1Score) increases by 16.3 percentage points. Meanwhile, compared with the mainstream unsupervised anomaly detection models PatchCore and FastFlow, the image_AUROC of the EfficientAD model exceeds that of PatchCore and FastFlow.

Keywords

YOLOv13; unsupervised learning; chip capacitor printing; defect detection

References

1. [1] Diao Shuo, Kang Xiongbing, Yi Jinfeng Research on Automated Testing System for Multi functional Chip Capacitors [J]. China Testing, 2025, 51 (S1): 160-164
2. [2] Zhang Xiaoxue, Sun Bangyong, Tan Jiahai, etc Overview of Printing Defect Detection Methods [J]. Research on Printing and Digital Media Technology, 2025 (2): 1-10
3. [3] Lei M Q，Li S Q，Wu Y H，et al. YOLOv13: Real-Time Object Detection with Hypergraph-Enhanced Adaptive Visual Perception [PP/OL].arXiv:2506.17733 [cs.CV].https://doi.org/10.48550/arXiv.2506.17733.
4. [4] Batzner K，Heckler L，Konig R. EfficientAD: Accurate Visual Anomaly detection at Millisecond-level Latencies [C]//2024 IEEE/CVF Winter Conference on Applications of Computer Vision(WACV)，2024：128-138.
5. [5] He Zhijie, Xiao Wei, Liu Nanqing, etc A review of object detection techniques based on deep semi supervised learning [J]. Telecommunications Technology, 2025, 65 (3): 484-494
6. [6] Wang Wenpeng, Qin Yinchang, Shi Wenxuan A review of unsupervised deep learning methods for industrial defect detection [J]. Computer Applications, 2025, 45 (5): 1658-1670
7. [7] Liu Guixiong, Yan Yipu, Xing Xingao Research progress on unsupervised learning detection methods for visual surface defects [J]. China Testing, 2024, 50 (3): 1-12
8. [8] ROTH K，PEMULA L，ZEPEDA J，et al. Towards Total Recall in Industrial Anomaly Detection [C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition，2022：14298-14308.
9. [9] Yu J，Zheng Y，Wang X，et al. FastFlow: Unsupervised Anomaly Detection and Localization via 2D Normalizing Flows [PP/OL].arXiv.（2021-11-16）.https://doi.org/10.48550/arXiv.2111.07677.
10. [10] Huang Yi, Zhou Chun, Liu Xinjun, et al. Multi scale detection model for unmanned aerial vehicles in complex backgrounds based on YOLOv10 [J]. Optical Communication Research, 2025 (5): 45-52
11. [11] Liao Lida, Luo Xiao, Huang Bin, etc Defect detection of polycrystalline silicon solar cells in complex backgrounds [J]. Journal of Solar Energy, 2024, 45 (9): 295-303
12. [12] Qiu Yongfeng, Huang Xuan, Yuan Xinpan, etc Algorithm for detecting surface defects of steel under complex background [J]. Mechanical Design and Manufacturing, 2025 (10): 292-297

Funding

Capability Building of Embedded AI and Machine Vision Technology Research Platform (2022GDASZH-2022030604-08)

About the authors

Cai Haocong (1986-), male, Han ethnicity, from Jieyang, Guangdong, algorithm engineer, master's student, research interests include computer vision; Corresponding author: Yang Chenghu (1968-), male, Han ethnicity, from Huaihua, Hunan, professor-level senior engineer, master's student, research interests include artificial intelligence; Zeng Weikang (1997-), male, Han ethnicity, from Zhaoqing, Guangdong, software engineer, bachelor's degree, research interests include computer vision; Yuan Yong (1987-), male, Han ethnicity, from Jingzhou, Hubei, mechanical structure engineer, bachelor's degree, research interests include industrial robots; Lu Zhaocong (1999-), male, Han ethnicity, from Foshan, Guangdong, software engineer, bachelor's degree, research interests include computer vision.