EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

CTVSeg-UNet: robust refinement for leaf vein phenotyping in curing tobacco

Yiming Ma, Hao Zhang, Wenjie Wang, Xiaohan Zhang, Mengjia Zhang, Hao Zhang, Zhaopeng Song, Junfeng Wu

Smart Agricultural Technology · 2026-12 · 15 · 102650

DOI: 10.1016/j.atech.2026.102650 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Smart Agricultural Technology, 2026-12. Verfasst von Yiming Ma und 7 weiteren Autor:innen. Die Arbeit nennt 38 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate leaf vein segmentation is essential for fine-scale plant phenotyping, as venation networks provide key structural information related to leaf morphology, transport function, and developmental or postharvest changes. However, vein extraction remains challenging due to the thin, branched, and low-contrast nature of venation structures, especially under dynamic agricultural conditions. Despite the widespread adoption of mIoU as the primary evaluation metric, its dominance by background pixels and insensitivity to false negatives often lead to overoptimistic assessment for thin, imbalanced targets like leaf veins. Flue-cured tobacco offers a representative scenario, as leaf traits change continuously during curing, increasing image diversity and segmentation difficulty. In this study, a dedicated acquisition system was designed for capturing tobacco leaf images during curing, and a diverse dataset was constructed by collecting samples from six cultivars across different production regions. Based on U-Net, the loss function was reweighted with a vein-to-background ratio of 40:1 to alleviate foreground-background imbalance. A collaborative upsampling mechanism combining bilinear interpolation and transposed convolution was designed to improve vein recovery across curing stages, and adversarial learning is incorporated to enhance mask realism and completeness. Mean pixel accuracy (mPA) was used as the primary metric in preference to mIoU. CTVSeg-UNet achieves 87.33% mPA, outperforming the baseline U-Net and existing segmentation models. Visual results confirmed robust segmentation of midribs and lateral veins across cultivars and curing stages. This study supports fine-scale phenotyping and intelligent curing regulation, and offers a transferable framework for segmenting thin biological structures under complex dynamic conditions.

Zitieren

APA

Ma, Y., Zhang, H., Wang, W., Zhang, X., Zhang, M., Zhang, H., Song, Z., & Wu, J. (2026). CTVSeg-UNet: robust refinement for leaf vein phenotyping in curing tobacco. Smart Agricultural Technology, 15, 102650. https://doi.org/10.1016/j.atech.2026.102650

Vancouver

Ma Y, Zhang H, Wang W, Zhang X, Zhang M, Zhang H, et al. CTVSeg-UNet: robust refinement for leaf vein phenotyping in curing tobacco. Smart Agricultural Technology. 2026;15:102650. doi:10.1016/j.atech.2026.102650

BibTeX
@article{Ma2026,
  author = {Ma, Yiming and Zhang, Hao and Wang, Wenjie and Zhang, Xiaohan and Zhang, Mengjia and Zhang, Hao and Song, Zhaopeng and Wu, Junfeng},
  title = {CTVSeg-UNet: robust refinement for leaf vein phenotyping in curing tobacco},
  journal = {Smart Agricultural Technology},
  year = {2026},
  volume = {15},
  pages = {102650},
  doi = {10.1016/j.atech.2026.102650}
}
Kostenlos starten →