Scholarlink

This paper presents a deep learning-based solution to the challenge of weed infestation in the context of modern agriculture. We leverage visual similarities of weed species through a hierarchical classification approach and explore real-time weed detection using state-of-the-art deep learning models. Our experiments using MobileNetV2 and NASNetMobile, optimized for edge applications, yield remarkable accuracies of 98.90 and 97.58%, respectively. Additionally, the paper explores real-time weed detection using the YOLOv7 and YOLOv8 framework, achieving a precision of 0.9860 and 0.9902, respectively. The deployment of these advanced technologies on the edge offers a transformative solution for weed management, ensuring precise and targeted weedicide application, thereby minimizing environmental harm and streamlining the weed management process for farmers.

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 Deep Learning Models for Real-Time Weed Classification on the Edge

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • K. S. Ramalakshmi,
  • Kingshuk Karmakar,
  • Krishna Bajaj,
  • Kumar Abhimanyu,
  • Gowri Srinivasa

摘要

This paper presents a deep learning-based solution to the challenge of weed infestation in the context of modern agriculture. We leverage visual similarities of weed species through a hierarchical classification approach and explore real-time weed detection using state-of-the-art deep learning models. Our experiments using MobileNetV2 and NASNetMobile, optimized for edge applications, yield remarkable accuracies of 98.90 and 97.58%, respectively. Additionally, the paper explores real-time weed detection using the YOLOv7 and YOLOv8 framework, achieving a precision of 0.9860 and 0.9902, respectively. The deployment of these advanced technologies on the edge offers a transformative solution for weed management, ensuring precise and targeted weedicide application, thereby minimizing environmental harm and streamlining the weed management process for farmers.