Scholarlink

<p>Weeds are undesired plants that grow along with the main crop and strive for space, light, moisture, and nutrients; thereby causing huge losses in food crop yield (~ 45%) and economic losses (~ 80 thousand crores rupees per annum) in India. In the context of climate change, their impact is further amplified, as shifting weather patterns can enhance weed proliferation and resilience, making effective weed management crucial for ensuring sustainable agriculture and long-term food security. The study is based on creating a scalable framework for the automated detection of weeds using high-resolution smartphone images collected from the field. The layout of the experimental field is designed for common weed <i>Alternanthera sessilis</i> in rice crops, at the Research cum instructional farms of Indira Gandhi Krishi Vishwavidyalaya, Raipur, Chhattisgarh. Field images were collected during Kharif 2023 from <i>Alternanthera sessilis</i> treatments in three replicas designed for standing rice (IGKV-R1) crop. The research found that Deep Learning-(DL) based techniques like Convolutional Neural Networks are highly effective in weed detection in rice crops. The YOLOv8 architecture with custom classes (crop, weed, background), custom training dataset, and hyperparameter tuning was found to show good potential for weed detection (mAP = 0.60). Field-based weed index observations were found to agree with DL-based weed cover estimates (<i>R</i><sup>2</sup> = 0.946). The technology has been evaluated and proven on a modest level. It has a lot of potential to be scaled up for operational application.</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

Weed Detection Using Deep Learning in Rice Crops Under Field Conditions

  • 期刊论文

  • 发表日期：

    2025年8月13日

查看全文
  • Manju Tiwari,
  • Nitish Tiwari,
  • Prasun Kumar Gupta,
  • Shefali Agarwal

摘要

Weeds are undesired plants that grow along with the main crop and strive for space, light, moisture, and nutrients; thereby causing huge losses in food crop yield (~ 45%) and economic losses (~ 80 thousand crores rupees per annum) in India. In the context of climate change, their impact is further amplified, as shifting weather patterns can enhance weed proliferation and resilience, making effective weed management crucial for ensuring sustainable agriculture and long-term food security. The study is based on creating a scalable framework for the automated detection of weeds using high-resolution smartphone images collected from the field. The layout of the experimental field is designed for common weed Alternanthera sessilis in rice crops, at the Research cum instructional farms of Indira Gandhi Krishi Vishwavidyalaya, Raipur, Chhattisgarh. Field images were collected during Kharif 2023 from Alternanthera sessilis treatments in three replicas designed for standing rice (IGKV-R1) crop. The research found that Deep Learning-(DL) based techniques like Convolutional Neural Networks are highly effective in weed detection in rice crops. The YOLOv8 architecture with custom classes (crop, weed, background), custom training dataset, and hyperparameter tuning was found to show good potential for weed detection (mAP = 0.60). Field-based weed index observations were found to agree with DL-based weed cover estimates (R2 = 0.946). The technology has been evaluated and proven on a modest level. It has a lot of potential to be scaled up for operational application.