Scholarlink

<p>Stubble management is vital for the growth of Salix cheilophila, especially in arid regions where rapid detection is critical for smart stubble equipment. However, existing models are often computationally intensive, limiting their practicality. In this paper, we present and study YOLOv8n-VCAD, an enhanced method tailored for the detection of Salix cheilophila during the stubble period in the desert region of Shierliancheng, Inner Mongolia. By integrating VanillaNet as a lightweight backbone, we significantly reduce computational demands. A coordinate attention mechanism enhances feature extraction through location data, improving regression and positioning accuracy. Furthermore, we introduce an adaptive feature fusion pyramid network to bolster feature characterization and integration, leading to superior detection performance. We also replace traditional CIoU loss with DIoU loss to accelerate regression convergence. Experimental results indicate that YOLOv8n-VCAD achieves an accuracy of 95.4%, with floating-point operations at 7.4 G and parameters at 5.46&#xa0;M. Compared to the original YOLOv8, precision improves by 7.7% and recall by 1.0%, while computational complexity and parameters decrease by 16.8% and 7.9%, respectively. Overall, YOLOv8n-VCAD offers significant advancements in detecting Salix cheilophila, enabling rapid and accurate identification while reducing deployment costs and enhancing stubble system automation.</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

Enhanced YOLOv8 for efficient identification of salix cheilophila during stubble period

  • 期刊论文

  • 发表日期：

    2025年5月22日

查看全文
  • Haotian Ma,
  • Zhigang Liu,
  • Tianyou Song,
  • Zhifei Zhao,
  • Chenghui Pei,
  • Shuhan Wang

摘要

Stubble management is vital for the growth of Salix cheilophila, especially in arid regions where rapid detection is critical for smart stubble equipment. However, existing models are often computationally intensive, limiting their practicality. In this paper, we present and study YOLOv8n-VCAD, an enhanced method tailored for the detection of Salix cheilophila during the stubble period in the desert region of Shierliancheng, Inner Mongolia. By integrating VanillaNet as a lightweight backbone, we significantly reduce computational demands. A coordinate attention mechanism enhances feature extraction through location data, improving regression and positioning accuracy. Furthermore, we introduce an adaptive feature fusion pyramid network to bolster feature characterization and integration, leading to superior detection performance. We also replace traditional CIoU loss with DIoU loss to accelerate regression convergence. Experimental results indicate that YOLOv8n-VCAD achieves an accuracy of 95.4%, with floating-point operations at 7.4 G and parameters at 5.46 M. Compared to the original YOLOv8, precision improves by 7.7% and recall by 1.0%, while computational complexity and parameters decrease by 16.8% and 7.9%, respectively. Overall, YOLOv8n-VCAD offers significant advancements in detecting Salix cheilophila, enabling rapid and accurate identification while reducing deployment costs and enhancing stubble system automation.