EUVIMEDEUVIMED entdecken
Crossref

WheatSAM: A Two-Stage Wheat Head Automatic Segmentation Framework

Md Jaber Al Nahian, Tapotosh Ghosh, Farnaz Sheikhi, Ian McQuillan, Farhad Maleki

Proceedings of the Conference on Robots and Vision · 27.05.2025

DOI: 10.21428/d82e957c.12450d3b ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the Conference on Robots and Vision, online erschienen am 27.05.2025. Verfasst von Md Jaber Al Nahian und 4 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The Segment Anything Model (SAM) has demonstrated impressive zero-shot segmentation capability. However, it requires accurate prompts. In this paper, we benchmark SAM on wheat head segmentation for the first time, evaluating its performance using manually and automatically generated prompts. Our findings indicate that SAM provides good performance with high-quality manually annotated bounding boxes, and fine-tuning further enhances the segmentation performance. However, the performance of SAM significantly decreases when the bounding boxes are generated by YOLOV9 or Faster RCNN due to object detection errors. To mitigate this, we have proposed WheatSAM, a two-stage segmentation model in which SAM is fine-tuned with an error-aware module. The error-aware module synthetically adds controlled perturbations in bounding boxes during training to simulate real-world bounding box errors. By methodically regulating the error probability, shift, scaling, and overlap, we have achieved the Dice score of 89.73% with YOLO-generated bounding boxes, surpassing SAM’s zero-shot performance on YOLO-generated prompts, which was 89.44%. Our results demonstrate a basic trade-off between manual annotation and segmentation precision, illustrating the potential of error-aware training to bridge the gap between manually annotated prompts and automatic detection-based segmentation.

Zitieren

APA

Nahian, M. J. A., Ghosh, T., Sheikhi, F., McQuillan, I., & Maleki, F. (2025). WheatSAM: A Two-Stage Wheat Head Automatic Segmentation Framework. Proceedings of the Conference on Robots and Vision. https://doi.org/10.21428/d82e957c.12450d3b

Vancouver

Nahian MJA, Ghosh T, Sheikhi F, McQuillan I, Maleki F. WheatSAM: A Two-Stage Wheat Head Automatic Segmentation Framework. Proceedings of the Conference on Robots and Vision. 2025. doi:10.21428/d82e957c.12450d3b

BibTeX
@article{Nahian2025,
  author = {Nahian, Md Jaber Al and Ghosh, Tapotosh and Sheikhi, Farnaz and McQuillan, Ian and Maleki, Farhad},
  title = {WheatSAM: A Two-Stage Wheat Head Automatic Segmentation Framework},
  journal = {Proceedings of the Conference on Robots and Vision},
  year = {2025},
  doi = {10.21428/d82e957c.12450d3b}
}
Kostenlos starten →