Scholarlink

<p>Hemp plant counting plays a crucial role in optimizing subsequent planning for the removal of hemp stamens. However, traditional object detection methods often lead to slow convergence during training and may fail to detect adjacent plants. To address these challenges, this paper proposes DN-DETR, a novel approach for hemp plant counting that integrates query denoising into DETR. This method effectively mitigates the issues mentioned above. Given the scarcity of publicly available hemp datasets, this study gathers images through UAV deployed over hemp fields and constructs a specialized dataset for hemp plant counting. The experimental results show that DN-DETR achieves the mAP of 94.3% and achieves the highest accuracy of 96.3% on test set, which is superior to YOLOv5, YOLOX, Faster R-CNN, DETR in a relatively fast time. Ablation experimental results show that each component in en-noising and de-noising module contributes to performance improvement. This research provides efficient technical support for hemp plant counting in hemp planting areas.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hemp Plant Counting from UAV Images via Detection Transformer by Introducing Query DeNoising

  • 期刊论文

  • 发表日期：

    2025年5月21日

查看全文
  • Qiufeng wu,
  • Zhuo miao,
  • Desheng sun

摘要

Hemp plant counting plays a crucial role in optimizing subsequent planning for the removal of hemp stamens. However, traditional object detection methods often lead to slow convergence during training and may fail to detect adjacent plants. To address these challenges, this paper proposes DN-DETR, a novel approach for hemp plant counting that integrates query denoising into DETR. This method effectively mitigates the issues mentioned above. Given the scarcity of publicly available hemp datasets, this study gathers images through UAV deployed over hemp fields and constructs a specialized dataset for hemp plant counting. The experimental results show that DN-DETR achieves the mAP of 94.3% and achieves the highest accuracy of 96.3% on test set, which is superior to YOLOv5, YOLOX, Faster R-CNN, DETR in a relatively fast time. Ablation experimental results show that each component in en-noising and de-noising module contributes to performance improvement. This research provides efficient technical support for hemp plant counting in hemp planting areas.