Scholarlink

<p>Precise detection of stem bases in sugarcane plants is critical for developing intelligent harvesting. In this study, an improved YOLOv5m was proposed to detect stem bases of sugarcane plants. The model firstly introduced a CA attention mechanism module in the backbone network, thereby improving the specific extraction ability of sugarcane stem bases. In addition, a small target detection layer was added to the head to enhance multi-scale feature fusion, which enabled detection effect when dealing with small targets at the image edges. Compared with the existing models such as faster RCNN, YOLO7-tiny and YOLOv8n, the original YOLOv5m model obtained the preferred results. Moreover, the proposed improved YOLOv5m model achieved 96.9% of precision, 95.4% of recall, and 97.9% of mAP@50. When compared to the original YOLOv5m, these aspects boosted 9.1%, 9.3% and 9.3%, respectively. It took 0.166&#xa0;s for processing an image using the improved YOLOv5m model. Experimental results show that the improved YOLOv5m model achieved the most superior overall detection performance, which can provide a baseline for intelligent harvesting of sugarcane plants.</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

Stem Base Detection in Sugarcane Plants Using Improved YOLOv5m Model

  • 期刊论文

  • 发表日期：

    2025年4月17日

查看全文
  • Hongwei Li,
  • Jingyi Yang,
  • Jiuyang Song,
  • Minshuo Sun,
  • Wei Wei,
  • Shunsheng Zhang,
  • Tao Wu,
  • Yanzhou Li

摘要

Precise detection of stem bases in sugarcane plants is critical for developing intelligent harvesting. In this study, an improved YOLOv5m was proposed to detect stem bases of sugarcane plants. The model firstly introduced a CA attention mechanism module in the backbone network, thereby improving the specific extraction ability of sugarcane stem bases. In addition, a small target detection layer was added to the head to enhance multi-scale feature fusion, which enabled detection effect when dealing with small targets at the image edges. Compared with the existing models such as faster RCNN, YOLO7-tiny and YOLOv8n, the original YOLOv5m model obtained the preferred results. Moreover, the proposed improved YOLOv5m model achieved 96.9% of precision, 95.4% of recall, and 97.9% of mAP@50. When compared to the original YOLOv5m, these aspects boosted 9.1%, 9.3% and 9.3%, respectively. It took 0.166 s for processing an image using the improved YOLOv5m model. Experimental results show that the improved YOLOv5m model achieved the most superior overall detection performance, which can provide a baseline for intelligent harvesting of sugarcane plants.