BTWD: Bag of Tricks for Wheat Detection

Abstract

Accurate detection of wheat heads outdoors is a great challenge. Wheat color and shape distinctions, as well as overlaps and wind blurring in wheat photos, make it difficult to detect wheat heads. We propose a Bag of Tricks for Wheat Detection (BTWD), finding that a reasonable combination of some tricks will bring great improvement to the wheat detection results, and apply it on different networks such as YOLO v5x, YOLO v3, EfficientDet-D5, Faster R-CNN, etc. BTWD has greatly enhanced comparison with the original network without tricks. YOLO v5x with BTWD achieves 77.07% in average mAP, in comparison, only 70.78% without it on the Global Wheat Head Detection (GWHD) dataset.

Cite

Text

Wu et al. "BTWD: Bag of Tricks for Wheat Detection." European Conference on Computer Vision Workshops, 2020. doi:10.1007/978-3-030-65414-6_31

Markdown

[Wu et al. "BTWD: Bag of Tricks for Wheat Detection." European Conference on Computer Vision Workshops, 2020.](https://mlanthology.org/eccvw/2020/wu2020eccvw-btwd/) doi:10.1007/978-3-030-65414-6_31

BibTeX

@inproceedings{wu2020eccvw-btwd,
  title     = {{BTWD: Bag of Tricks for Wheat Detection}},
  author    = {Wu, Yifan and Hu, Ya-Han and Li, Lei},
  booktitle = {European Conference on Computer Vision Workshops},
  year      = {2020},
  pages     = {450-460},
  doi       = {10.1007/978-3-030-65414-6_31},
  url       = {https://mlanthology.org/eccvw/2020/wu2020eccvw-btwd/}
}