Detecção de plantas daninhas em pré-semeadura com base em dados espectrais de campo

• Ano de defesa: 2019
• Autor(a) principal: Pott, Luan Pierre
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; Brasil; Engenharia Agrícola; UFSM; Programa de Pós-Graduação em Engenharia Agrícola; Centro de Ciências Rurais
• Palavras-chave em Português: Manejo sítio-específico de plantas daninhas; Curvas espectrais; Bandas espectrais; Índices de vegetação; Site-specific weed management (SSWM); Spectral curves; Spectral bands; Vegetation indices; CNPQ::CIENCIAS AGRARIAS::ENGENHARIA AGRICOLA
• Link de acesso: http://repositorio.ufsm.br/handle/1/19101

According the concept of precision agriculture and new technologies for agriculture, there were several studies to improve tools at an extremely important stage in crop management, which is the identification and control of weeds. Therefore, the spatial variability of weed distribution is not being considered in deciding their management in most cases. In this sense, the objective of this study was: (i) the use of a hyperspectral sensor to identify more efficient spectral bands in distinguishing weeds from other targets (sandy soil, clay soil and plant residues) in pre-planting; (ii) elaborate vegetation indices to evaluate the accuracy of weed distinction and other targets. Two databases were used, the first from a field experiment conducted at the Federal University of Santa Maria as training data, and the second database was built with readings on-farm as validation data. The HandHeld 2 spectrometer, ASD®, with wavelengths of 325-1075nm, was used to perform spectral curves readings of weed species and other targets: clay soil, sandy soil, and residues. Subsequently, the wavelengths were grouped into spectral bands, as well as the calculation of vegetation indices for data analysis. The results showed that the data collected in the field experiment (training data) and in the farms (validation data) obtained similar spectral curves, where the red and near infrared spectral bands obtained higher accuracy compared to the other bands. The vegetation indices used increased the discrimination accuracy in relation to the isolated spectral bands. The work provides a valid tool for distinguishing weeds from other targets using proximal sensor pre-sowing of crops based on spectral curves.