EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

WheatNet: A lightweight network for wheat seedling detection and counting in complex fields

Yafeng Han, Binbin Liu, Xinfa Wang, Huiyuan Zhao, Xueyong Li, Guohong Gao, Liuping Chen, Xiaojuan Guo

Smart Agricultural Technology · 2026-12 · 15 · 102648

DOI: 10.1016/j.atech.2026.102648 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Smart Agricultural Technology, 2026-12. Verfasst von Yafeng Han und 7 weiteren Autor:innen. Die Arbeit nennt 46 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Henan Institute of Science and Technology.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Breeders use accurate wheat seedling counts to evaluate varietal adaptation and estimate yield potential. These counts also support irrigation, fertiliser management, and planting density adjustment. However, manual counting requires substantial time and labour and cannot provide the efficiency needed for large breeding trials. To address these limitations, this study develops the WheatSC dataset and proposes WheatNet, a detection network designed for dense wheat seedling scenes under field conditions. To handle the challenges posed by the small object size of wheat seedlings, slender leaf morphology, dense seedling distribution, and indistinct boundaries between adjacent plants in field images, WheatNet uses the Gated-Enhanced Vision-RWKV (GE-VRWKV) backbone to enhance multi-scale feature representations and introduces rotation-enhanced area attention (REAA) modules into the neck to improve the directional representation of leaf features during multi-scale fusion. In addition, the model adopts a joint Wise-IoU (WIoU) and normalised Wasserstein distance (NWD) bounding-box regression loss to improve small-object localisation stability. On the test set of the WheatSC dataset developed in this study, WheatNet achieved mAP@0.5 and mAP@0.5:0.95 values of 92.70% and 82.82%, respectively, exceeding the corresponding YOLO11n values by 3.98 and 4.86 percentage points. Its counting MAE was 0.972 seedlings per image. WheatNet contains 3.83 million parameters and achieved an inference speed of 156 FPS on the specified test platform.

Förderung

  • Henan Institute of Science and Technology

Zitieren

APA

Han, Y., Liu, B., Wang, X., Zhao, H., Li, X., Gao, G., Chen, L., & Guo, X. (2026). WheatNet: A lightweight network for wheat seedling detection and counting in complex fields. Smart Agricultural Technology, 15, 102648. https://doi.org/10.1016/j.atech.2026.102648

Vancouver

Han Y, Liu B, Wang X, Zhao H, Li X, Gao G, et al. WheatNet: A lightweight network for wheat seedling detection and counting in complex fields. Smart Agricultural Technology. 2026;15:102648. doi:10.1016/j.atech.2026.102648

BibTeX
@article{Han2026,
  author = {Han, Yafeng and Liu, Binbin and Wang, Xinfa and Zhao, Huiyuan and Li, Xueyong and Gao, Guohong and Chen, Liuping and Guo, Xiaojuan},
  title = {WheatNet: A lightweight network for wheat seedling detection and counting in complex fields},
  journal = {Smart Agricultural Technology},
  year = {2026},
  volume = {15},
  pages = {102648},
  doi = {10.1016/j.atech.2026.102648}
}
Kostenlos starten →