EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

FSCA-YOLO: A Recognition and Detection Algorithm for CT Images of PCB Drilling

Ziyang Li, Qingfeng Ma, Zhenbo Xu

IEEE Access · 2026 · 14 · 98876-98887

DOI: 10.1109/access.2026.3707539 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Ziyang Li und 2 weiteren Autor:innen. Die Arbeit nennt 35 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Open Project of Anhui Province Key Laboratory of Intelligent Computing and Applications “Research on PCB Defect Detection in CT Images Based on Deep Learning”.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

To address the issue of insufficient detection accuracy caused by the small size, low contrast, and interference from 3D artifacts of printed circuit board drill holes in Industrial Computed Tomography (CT) images, this paper proposes an FSCA-YOLO detection algorithm based on improved YOLO11n. First, the backbone network is replaced with FasterNet to establish a more efficient foundation for feature extraction. Second, the SegNext Attention mechanism is introduced at key stages to enhance the model’s ability to focus on critical features of the drill holes and suppress noise. Subsequently, construct the C3K2-AdditiveBlock module. This module incorporates a cross-attention sub-network, enabling the model to efficiently model the global contextual dependencies of features, thereby significantly enhancing the discrimination and representation of minute target features. Finally, an Auxiliary Head for small-target auxiliary detection is added, which leverages mid-level, high-resolution features to strengthen the recall rate for small objects. Experiments conducted on a self-built PCB drill hole CT defect dataset demonstrate that the FSCA-YOLO model achieves a mean Average Precision (mAP@50) of 98.1%, which is 4.1% higher than the baseline YOLO11n model, with improvements of 9.5% and 9.1% in precision and recall, respectively. mAP@50-95 also improved by 2.7%. The FPS of the final model is 93.3. Experimental results indicate that the proposed method significantly enhances the detection capability for 20– $150~\mu $ m drill holes on PCBs, as well as the recognition accuracy and robustness under complex interference conditions.

Förderung

  • Open Project of Anhui Province Key Laboratory of Intelligent Computing and Applications “Research on PCB Defect Detection in CT Images Based on Deep Learning”

Zitieren

APA

Li, Z., Ma, Q., & Xu, Z. (2026). FSCA-YOLO: A Recognition and Detection Algorithm for CT Images of PCB Drilling. IEEE Access, 14, 98876-98887. https://doi.org/10.1109/access.2026.3707539

Vancouver

Li Z, Ma Q, Xu Z. FSCA-YOLO: A Recognition and Detection Algorithm for CT Images of PCB Drilling. IEEE Access. 2026;14:98876-98887. doi:10.1109/access.2026.3707539

BibTeX
@article{Li2026,
  author = {Li, Ziyang and Ma, Qingfeng and Xu, Zhenbo},
  title = {FSCA-YOLO: A Recognition and Detection Algorithm for CT Images of PCB Drilling},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {98876--98887},
  doi = {10.1109/access.2026.3707539}
}
Kostenlos starten →