Scholarlink

<p>The controlled environment agriculture is a rapidly growing industry, where resource management is critical for operational efficiency and reduced costs. Skilled labor is important for stress management in plants grown in such environments. Thus, there is a need for technology and tools for automated crop stress management. This study introduces the development of an automated, plant health-based decision-support prototype system for crop stress monitoring and management applications in the controlled environment agriculture, which can potentially reduce labor costs. The system consists of an imaging sensor module, a control module, and a sprayer module. The imaging sensor module captures images of a group of plants, which is sent to the cloud for segmentation, mapping, and stress analysis. The control module processes the data and sends a signal to the sprayer module to trigger chemical application as needed. The images can be collected and processed in daily batches, with an automated actuation performed accordingly. We conducted a preliminary experiment on peas (<i>Pisum sativum</i> L.) by inducing nutrient-related stress (Dataset 1: 386 healthy and 126 stressed plants). We used YOLOv11 algorithm to generate individual masked images of plants and a neural network classifier as the decision model to trigger spraying if stress is detected. The segmentation model achieved high mean average precisions (mAP<sub>50−95</sub>) of bounding boxes and masks about 0.93 and 0.90, respectively, while the classification accuracy reached about 84% for Dataset 1. We also trained an additional classifier to explore broader applicability, achieving an accuracy of 91% using a image dataset of lettuce with multiple stress (e.g., chlorosis, tip burn, wilting) grown in a vertical farm (Dataset 2: 1162 healthy and 742 stressed plants). These results demonstrate a viable and adaptable approach for early, precise, and automated plant stress management, which can be integrated into controlled environment management to optimize operations and consequently enhancing crop profitability.</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

Development and evaluation of a smart crop stress monitoring and management system for applications in controlled environment

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月10日

查看全文
  • Toky F. Andriamihajasoa,
  • Lochana T. Marasinghe,
  • Rebecca J. McGee,
  • Sindhuja Sankaran

摘要

The controlled environment agriculture is a rapidly growing industry, where resource management is critical for operational efficiency and reduced costs. Skilled labor is important for stress management in plants grown in such environments. Thus, there is a need for technology and tools for automated crop stress management. This study introduces the development of an automated, plant health-based decision-support prototype system for crop stress monitoring and management applications in the controlled environment agriculture, which can potentially reduce labor costs. The system consists of an imaging sensor module, a control module, and a sprayer module. The imaging sensor module captures images of a group of plants, which is sent to the cloud for segmentation, mapping, and stress analysis. The control module processes the data and sends a signal to the sprayer module to trigger chemical application as needed. The images can be collected and processed in daily batches, with an automated actuation performed accordingly. We conducted a preliminary experiment on peas (Pisum sativum L.) by inducing nutrient-related stress (Dataset 1: 386 healthy and 126 stressed plants). We used YOLOv11 algorithm to generate individual masked images of plants and a neural network classifier as the decision model to trigger spraying if stress is detected. The segmentation model achieved high mean average precisions (mAP50−95) of bounding boxes and masks about 0.93 and 0.90, respectively, while the classification accuracy reached about 84% for Dataset 1. We also trained an additional classifier to explore broader applicability, achieving an accuracy of 91% using a image dataset of lettuce with multiple stress (e.g., chlorosis, tip burn, wilting) grown in a vertical farm (Dataset 2: 1162 healthy and 742 stressed plants). These results demonstrate a viable and adaptable approach for early, precise, and automated plant stress management, which can be integrated into controlled environment management to optimize operations and consequently enhancing crop profitability.