EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Keypoint-Guided Deep Learning Framework for Non-Invasive Individual Authentication of Chinese Mitten Crabs (Eriocheir sinensis) in Food Traceability

Zhe Zhang, Shanshan Li, Na Luo, Helong Yu, Chuanheng Sun

Foods · 13.11.2026 · 15 (22) · 4075

DOI: 10.3390/foods15224075 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Foods, online erschienen am 13.11.2026. Verfasst von Zhe Zhang und 4 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Key Technology Research and Development Program of China, Scientific and technological innovation capacity building project of Beijing Academy of Agricul-tural and Forestry Sciences, the Key Research and Development Program of Jiangsu Province.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Reliable individual authentication can help prevent substitution of high-value aquatic products and improve food traceability. However, conventional traceability systems often rely on external labels, which may not ensure that recorded identities correspond to physical products. This study proposes a non-invasive, keypoint-guided deep learning framework (KG-CrabID) for individual authentication of Chinese mitten crabs (Eriocheir sinensis) using natural carapace features. A multi-source image dataset comprising 6830 images from 857 crab individuals was constructed using samples collected from four geographical origins under different imaging conditions. Keypoint-guided localization aligns the cephalothorax region and reduces interference from irrelevant anatomical structures and image backgrounds. A lightweight feature extraction network, CrabID-Net, incorporating multi-scale depthwise convolution and coordinate attention, was developed to learn discriminative visual representations. ArcFace metric learning was then applied to support both 1:1 verification and 1:N identification. In the evaluation, the proposed framework achieved a verification accuracy of 97.58% using a threshold optimized on the evaluation pairs and a Rank-1 identification accuracy of 96.32%. Additional experiments demonstrated the contributions of keypoint-guided localization and the proposed network components. A system was implemented to demonstrate individual registration, verification and identification. These findings support the potential use of cephalothorax-based visual authentication as a complementary tool for food traceability.

Förderung

  • National Key Technology Research and Development Program of China
  • Scientific and technological innovation capacity building project of Beijing Academy of Agricul-tural and Forestry Sciences
  • the Key Research and Development Program of Jiangsu Province

Zitieren

APA

Zhang, Z., Li, S., Luo, N., Yu, H., & Sun, C. (2026). Keypoint-Guided Deep Learning Framework for Non-Invasive Individual Authentication of Chinese Mitten Crabs (Eriocheir sinensis) in Food Traceability. Foods, 15(22), 4075. https://doi.org/10.3390/foods15224075

Vancouver

Zhang Z, Li S, Luo N, Yu H, Sun C. Keypoint-Guided Deep Learning Framework for Non-Invasive Individual Authentication of Chinese Mitten Crabs (Eriocheir sinensis) in Food Traceability. Foods. 2026;15(22):4075. doi:10.3390/foods15224075

BibTeX
@article{Zhang2026,
  author = {Zhang, Zhe and Li, Shanshan and Luo, Na and Yu, Helong and Sun, Chuanheng},
  title = {Keypoint-Guided Deep Learning Framework for Non-Invasive Individual Authentication of Chinese Mitten Crabs (Eriocheir sinensis) in Food Traceability},
  journal = {Foods},
  year = {2026},
  volume = {15},
  number = {22},
  pages = {4075},
  doi = {10.3390/foods15224075}
}
Kostenlos starten →