YOLO-HCS: a YOLO-based framework for small defect detection in industrial automated optical inspection
Machine Vision and Applications · 2026-11 · 37 (6)
DOI: 10.1007/s00138-026-01906-w ↗Kurz gefasst
Fachartikel in der Zeitschrift Machine Vision and Applications, online erschienen am 20.08.2026. Verfasst von Chun-Cheng Lin und 3 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 40 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Yang Ming Chiao Tung University.
Abstract (Originalsprache)
Connectors are used in electronic and industrial systems to establish electrical connections between components, and terminal surfaces play an important role in ensuring reliable current transmission. Accurate defect detection on connector terminals is therefore important for maintaining product quality. However, detecting small, irregular, and scale-varying defects remains challenging in automated optical inspection (AOI), particularly when defects appear on reflective metallic surfaces and occupy only a limited number of pixels. To address this challenge, this work presents YOLO-HCS, an application-specific YOLOv7-based defect detection framework for connector-terminal AOI. The framework integrates hierarchical positive sample selection (HPSS) to improve positive sample assignment for small and scale-varying defects, coordinate attention (CA) to enhance spatial localization of small targets, and SIoU loss to refine bounding-box regression through geometric constraints. Experimental results on two real-world AOI datasets show that YOLO-HCS achieves 85.3% and 92.2% mAP@0.5 on Dataset 1 and Dataset 2, respectively, outperforming YOLOv7 by 8.9 and 4.7% points. Under the stricter mAP@0.5:0.95 metric, YOLO-HCS achieves 41.9% and 54.4% on the two datasets, respectively, indicating improved localization performance under more demanding IoU thresholds. In terms of computational efficiency, YOLO-HCS requires 37.1 M parameters and 106.9 GFLOPs and maintains approximately 26–27 FPS under the 1280 × 1280 input setting. These results indicate that the proposed framework improves small-defect detection and localization while introducing only marginal computational overhead compared with YOLOv7. While the reported inference efficiency suggests potential compatibility with real-time AOI inspection under the studied setting, this work does not include long-term production-line deployment or operational cost analysis. Therefore, claims regarding deployment reliability, manual inspection reduction, and cost benefits should be interpreted as potential implications requiring further empirical validation.
Förderung
- National Yang Ming Chiao Tung University
Zitieren
APA
Lin, C., Weng, H., Lu, S., & Liu, H. (2026). YOLO-HCS: a YOLO-based framework for small defect detection in industrial automated optical inspection. Machine Vision and Applications, 37(6). https://doi.org/10.1007/s00138-026-01906-w
Vancouver
Lin C, Weng H, Lu S, Liu H. YOLO-HCS: a YOLO-based framework for small defect detection in industrial automated optical inspection. Machine Vision and Applications. 2026;37(6). doi:10.1007/s00138-026-01906-w
BibTeX
@article{Lin2026,
author = {Lin, Chun-Cheng and Weng, Hsiang-En and Lu, Shin-Hang and Liu, Heng-Yih},
title = {YOLO-HCS: a YOLO-based framework for small defect detection in industrial automated optical inspection},
journal = {Machine Vision and Applications},
year = {2026},
volume = {37},
number = {6},
doi = {10.1007/s00138-026-01906-w}
}
EUVIMED