浙江农业科学 ›› 2025, Vol. 66 ›› Issue (5): 1158-1162.DOI: 10.16178/j.issn.0528-9017.20231115

• 经济作物 • 上一篇    下一篇

基于无人机载多光谱的烤烟叶片氮含量评估

金磊1(), 张驰2,*(), 邵小东3, 杜军1, 田晶晶1, 刘宇1   

  1. 1.安徽皖南烟叶有限责任公司,安徽 宣城 242000
    2.北京市农林科学院, 北京 100000
    3.云南省烟草公司 红河州公司,云南 弥勒 652300
  • 收稿日期:2023-11-17 出版日期:2025-05-11 发布日期:2025-05-20
  • 通讯作者: 张驰(1984—),男,四川成都人,副研究员,博士,研究方向为农业信息化,E-mail:zhangc@nercita.org.cn
  • 作者简介:金磊(1988—),男,安徽六安人,农艺师,硕士,研究方向为无人机农业信息化应用,E-mail:jinlei2004@163.com
  • 基金资助:
    安徽皖南烟叶有限责任公司科技项目(20220563004)

Assessment of nitrogen content in flue-cured tobacco leaves based on UAV-loaded multiple spectrum

JIN Lei1(), ZHANG Chi2,*(), SHAO Xiaodong3, DU Jun1, TIAN Jingjing1, LIU Yu1   

  1. 1. Anhui Wannan Tobacco Leaf Co., Ltd., Xuancheng 242000, Anhui
    2. Beijing Academy of Agriculture and Forestry Sciences, Beijing 100000
    3. Yunnan Tobacco Company Honghe Prefecture Company, Mile 652300, Yunnan
  • Received:2023-11-17 Online:2025-05-11 Published:2025-05-20

摘要:

该文研究无人机载多光谱遥感技术对大田烟叶氮素评估的可行性与有效性,为大面积集中连片种植的烟田提供一种高效、准确、无损的氮素营养诊断方式。2023年,在石屏县设置了不同处理的田间试验,并通过无人机载多光谱飞拍和田间采样获取小区多光谱图像数据和烟叶氮含量数据,利用多元线性回归(MLR)、偏最小二乘(PLSR)、随机森林(RF)等多种机器学习算法构建烟叶冠层多光谱特征与叶片氮含量之间的定量关系模型。结果表明,MLR构建的模型稳定性最高,RF构建的模型取得最高的相关度和最低的误差。本研究证实了多光谱遥感在大田烟叶氮素诊断方面的可行性,并取得了较好的结果,为更多烟草农学参数的遥感反演提供了参考。

关键词: 无人机, 烤烟, 多光谱遥感, 氮素诊断, 机器学习

Abstract:

To study the feasibility and effectiveness of multispectral remote sensing technology in field tobacco nitrogen assessment, and to provide an efficient, accurate and non-destructive nitrogen nutrition diagnosis method for large area concentrated tobacco fields, in 2023, field experiments with different treatments were set up in Shiping County. Data on plot multispectral images and tobacco leaf nitrogen content were collected through drone-based multispectral aerial photography and field sampling. Various machine learning algorithms, including multiple linear regression (MLR), partial least squares regression (PLSR), and random forest (RF), were used to construct quantitative relationship models between multispectral characteristics of the tobacco canopy and leaf nitrogen content. The results showed that the model built by MLR had the highest stability, while the model constructed by RF achieved the highest correlation and the lowest error. This study confirms the feasibility of using multispectral remote sensing for diagnosing nitrogen content in field tobacco leaves and has promising results, providing a reference for the remote sensing inversion of more agronomic parameters in tobacco cultivation.

Key words: UAV, flue-cured tobacco, multispectral remote sensing, nitrogen diagnosis, machine learning

中图分类号: