Scholarlink

<p>Precise identification of leaf diseases is crucial for the proper treatment of <i>Dendrobium officinale</i>. Based on the YOLOv5s model, an improved model (GBC-YOLOv5s) is proposed for detecting <i>Dendrobium</i> leaf diseases in complex environments. This model incorporates the GhostNet network (GhostConv and C3Ghost), the bidirectional feature pyramid network (BiFPN), and the CARAFE upsampling operator into the YOLOv5s model. The improvement shrinks the model size, enhances detection accuracy, and lowers computational complexity. An ablation experiment and confusion matrix analysis were conducted for the GBC-YOLOv5s model. Compared with the YOLOv5s model, its model size reduces by 42.3% and the mAP@0.5 increases by 2.4%, with the lower FLOPs (8.7G) and higher FPS (132). Meanwhile, the missed and false detections reduce, exhibiting the strong robustness in complex environments. Compared with other object detection models, the GBC-YOLOv5s model performed the best for accurately detecting <i>Dendrobium</i> leaf diseases while keeping the most efficient lightweight architecture. Given these advantages, the model can be used for identifying the <i>Dendrobium</i> leaf diseases in the complex environment and applied on the mobile or edge devices.</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

An Improved Lightweight Model for Detecting Leaf Diseases of Dendrobium Officinale Based on YOLOv5s

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • Jianian Li,
  • Zhu Yuan,
  • Zhengquan Liu,
  • Yang Luo,
  • Long Gao,
  • Jiaoli Fang,
  • Dejin Wang

摘要

Precise identification of leaf diseases is crucial for the proper treatment of Dendrobium officinale. Based on the YOLOv5s model, an improved model (GBC-YOLOv5s) is proposed for detecting Dendrobium leaf diseases in complex environments. This model incorporates the GhostNet network (GhostConv and C3Ghost), the bidirectional feature pyramid network (BiFPN), and the CARAFE upsampling operator into the YOLOv5s model. The improvement shrinks the model size, enhances detection accuracy, and lowers computational complexity. An ablation experiment and confusion matrix analysis were conducted for the GBC-YOLOv5s model. Compared with the YOLOv5s model, its model size reduces by 42.3% and the mAP@0.5 increases by 2.4%, with the lower FLOPs (8.7G) and higher FPS (132). Meanwhile, the missed and false detections reduce, exhibiting the strong robustness in complex environments. Compared with other object detection models, the GBC-YOLOv5s model performed the best for accurately detecting Dendrobium leaf diseases while keeping the most efficient lightweight architecture. Given these advantages, the model can be used for identifying the Dendrobium leaf diseases in the complex environment and applied on the mobile or edge devices.