EUVIMEDEUVIMED entdecken
Crossref

THzDPD-YOLO: a geometric morphology-aware enhanced defect detection method for polyethylene pipes

Enqi Shi, Huadong Song, Xinqian Li, Binghua Cao

Applied Optics · 20.08.2026 · 65 (24) · 8338

DOI: 10.1364/ao.599051 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Applied Optics, online erschienen am 17.08.2026. Verfasst von Enqi Shi und 3 weiteren Autor:innen. Die Arbeit nennt 26 Literaturangaben. Gefördert von: Science and Technology Program Project of the Zhejiang Provincial Administration for Market Regulation, The Research of National Engineering Research Center of Transducers.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Polyethylene (PE) pipes are critical to residential gas systems; however, defects such as voids and cracks at thermal fusion joints pose significant threats to public safety. Due to the irregular variations in defect size and morphology, combined with the blurred edges characteristic of terahertz TOFT imaging, traditional methods struggle to achieve high-precision defect detection in PE pipes. To address these issues, this paper proposes a detection method based on THzDPD-YOLO. First, a dual-frequency coupled context-aware (DFCCA) module is designed to effectively suppress background noise and enhance feature extraction capabilities. Second, a phenomenon-aware adaptive upsampling (PPGU) module is proposed, utilizing image gradient and variance as phenomenon prior information to improve the model’s defect localization ability. Finally, by integrating a dynamic head with deformable convolution (DCNv3), a deformable dynamic head (Dyhead-DCNv3) is reconstructed to enhance the model’s perception and detection accuracy regarding multi-scale and irregularly shaped defects. Experimental results demonstrate that the proposed method achieves an mAP@75 of 80.28% on the PE pipe defect terahertz TOFT imaging dataset, representing a 14.44% improvement over the original YOLOv11n. The results demonstrate the feasibility and potential of THzDPD-YOLO as an efficient computer-aided detection framework for PE pipe defects under THz-TOFT imaging conditions.

Förderung

  • Science and Technology Program Project of the Zhejiang Provincial Administration for Market Regulation
  • The Research of National Engineering Research Center of Transducers

Zitieren

APA

Shi, E., Song, H., Li, X., & Cao, B. (2026). THzDPD-YOLO: a geometric morphology-aware enhanced defect detection method for polyethylene pipes. Applied Optics, 65(24), 8338. https://doi.org/10.1364/ao.599051

Vancouver

Shi E, Song H, Li X, Cao B. THzDPD-YOLO: a geometric morphology-aware enhanced defect detection method for polyethylene pipes. Applied Optics. 2026;65(24):8338. doi:10.1364/ao.599051

BibTeX
@article{Shi2026,
  author = {Shi, Enqi and Song, Huadong and Li, Xinqian and Cao, Binghua},
  title = {THzDPD-YOLO: a geometric morphology-aware enhanced defect detection method for polyethylene pipes},
  journal = {Applied Optics},
  year = {2026},
  volume = {65},
  number = {24},
  pages = {8338},
  doi = {10.1364/ao.599051}
}
Kostenlos starten →