Scholarlink

<p>The agroforestry sector faces significant challenges due to limited resources, labor shortages, and the increasing presence of new diseases and weed species in plantations. In the case of eucalyptus plantations, productivity is impacted by the proliferation of invasive plants, pests, and pathogens. Weeds aggressively compete for essential resources such as nutrients, light, space, and water, thus reducing growth and productivity. This study aimed to differentiate broadleaf and narrowleaf weeds in commercial Eucalyptus saligna plantations based on their spectral signatures and to identify the most informative wavelength bands for such discrimination. Spectral reflectance curves were acquired in the field using a FieldSpec<sup>®</sup> spectroradiometer in six plots of three forest stands in Rio Grande do Sul, Brazil. The Random Forest algorithm was applied for classification, and the importance of spectral bands was assessed using the Gini index. Two modeling approaches were evaluated: a hyperspectral approach using 501 predictor variables in the 400–900&#xa0;nm range, and a multispectral approach simulating the four bands of the commercial Parrot Sequoia<sup>®</sup> sensor. Broadleaf species predominated, accounting for 42 of the 52 weed species identified. Hyperspectral analysis revealed that the five most relevant wavelengths for discriminating weed groups were concentrated in the RedEdge region, while the multispectral classification emphasized the importance of the RedEdge and near-infrared bands. These results reinforce the potential of both approaches for accurate classification of weed leaf morphology, supporting the development of more efficient, site-specific weed management strategies based on spectral data. Practical implications include the possibility of using these remote sensing technologies and machine learning techniques for targeted and sustainable weed management in agricultural and forestry environments.</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

Use of hyperspectral sensor and random forest for broadleaf and narrowleaf weed differentiation

  • 期刊论文

  • 发表日期：

    2025年7月25日

查看全文
  • Pablo Fernandes,
  • Sally Deborah Pereira da Silva,
  • Roberta Aparecida Fantinel,
  • Norton Borges Junior,
  • Ivana Pires de Sousa-Baracho,
  • Mateus Sabadi Schuh,
  • Rudiney Soares Pereira,
  • Fernando Coelho Eugenio

摘要

The agroforestry sector faces significant challenges due to limited resources, labor shortages, and the increasing presence of new diseases and weed species in plantations. In the case of eucalyptus plantations, productivity is impacted by the proliferation of invasive plants, pests, and pathogens. Weeds aggressively compete for essential resources such as nutrients, light, space, and water, thus reducing growth and productivity. This study aimed to differentiate broadleaf and narrowleaf weeds in commercial Eucalyptus saligna plantations based on their spectral signatures and to identify the most informative wavelength bands for such discrimination. Spectral reflectance curves were acquired in the field using a FieldSpec® spectroradiometer in six plots of three forest stands in Rio Grande do Sul, Brazil. The Random Forest algorithm was applied for classification, and the importance of spectral bands was assessed using the Gini index. Two modeling approaches were evaluated: a hyperspectral approach using 501 predictor variables in the 400–900 nm range, and a multispectral approach simulating the four bands of the commercial Parrot Sequoia® sensor. Broadleaf species predominated, accounting for 42 of the 52 weed species identified. Hyperspectral analysis revealed that the five most relevant wavelengths for discriminating weed groups were concentrated in the RedEdge region, while the multispectral classification emphasized the importance of the RedEdge and near-infrared bands. These results reinforce the potential of both approaches for accurate classification of weed leaf morphology, supporting the development of more efficient, site-specific weed management strategies based on spectral data. Practical implications include the possibility of using these remote sensing technologies and machine learning techniques for targeted and sustainable weed management in agricultural and forestry environments.