Scholarlink

<p>Effective pest prevention and control are crucial in agricultural production. Due to their diverse species and significant destructive potential, pest identification and localization must be accurate and timely. Traditional methods primarily rely on manual inspection, which can be inefficient for large-scale agriculture. To address this challenge, we have designed Pest-YOLOv8, a detection model based on YOLOv8 specifically for detecting small and diverse agricultural pests. It significantly improves the accuracy of pest classification and localization by incorporating the triple-coordinate attention (Triple-CA) module for more detailed feature capture. Additionally, it enhances the model’s ability to distinguish pests from the background in complex environments with dense weeds by utilizing the Wise-IoU loss function, thereby improving the model’s precise localization capability. Evaluated on the IP102 and Insects1201 datasets, Pest-YOLOv8 outperforms the basic YOLOv8, with improvements in mAP@50, mAP@75, and mAP@50:95 by 3.7%, 3.8%, and 2.8%, respectively, on IP102, and by 2.6%, 13%, and 7.3%, respectively, on Insects1201, despite a slight decrease in inference speed. This study introduces a rapid pest detection solution, providing technical support for agricultural applications.</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

Pest-YOLOv8: enhanced detection for small and complex agricultural pests using triple attention and wise-IoU

  • 期刊论文

  • 发表日期：

    2025年8月5日

查看全文
  • Bing Su,
  • Yueyang Zhu,
  • Yifeng Lin

摘要

Effective pest prevention and control are crucial in agricultural production. Due to their diverse species and significant destructive potential, pest identification and localization must be accurate and timely. Traditional methods primarily rely on manual inspection, which can be inefficient for large-scale agriculture. To address this challenge, we have designed Pest-YOLOv8, a detection model based on YOLOv8 specifically for detecting small and diverse agricultural pests. It significantly improves the accuracy of pest classification and localization by incorporating the triple-coordinate attention (Triple-CA) module for more detailed feature capture. Additionally, it enhances the model’s ability to distinguish pests from the background in complex environments with dense weeds by utilizing the Wise-IoU loss function, thereby improving the model’s precise localization capability. Evaluated on the IP102 and Insects1201 datasets, Pest-YOLOv8 outperforms the basic YOLOv8, with improvements in mAP@50, mAP@75, and mAP@50:95 by 3.7%, 3.8%, and 2.8%, respectively, on IP102, and by 2.6%, 13%, and 7.3%, respectively, on Insects1201, despite a slight decrease in inference speed. This study introduces a rapid pest detection solution, providing technical support for agricultural applications.