Inferência do teor de nitrogênio foliar em laranjeira-valência com imagens multiespectrais de alta resolução espacial

• Ano de defesa: 2019
• Autor(a) principal: Osco, Lucas Prado
• Orientador(a): Creste, José Eduardo
• Banca de defesa: Santos, Carlos Henrique dos , Watanabe, Fernanda Sayuri Yoshino , Mendes, Tatiana Sussel Gonçalves , Takata, William Hiroshi Suekane
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade do Oeste Paulista
• Programa de Pós-Graduação: Doutorado em Agronomia
• Departamento: Doutorado em Agronomia
• País: Brasil
• Palavras-chave em Português: Aeronaves Remotamente Pilotadas; Agricultura de Precisão; Classificação de Imagens; Índices Espectrais de Vegetação.
• Palavras-chave em Inglês: Remotely Piloted Aircraft; Precision Agriculture; Image Classification; Vegetation Spectral Indices.
• Área do conhecimento CNPq: CIENCIAS AGRARIAS::AGRONOMIA
• Link de acesso: http://bdtd.unoeste.br:8080/jspui/handle/jspui/1286

Nitrogen plays a fundamental role in the photosynthetic activity of plants, and its analysis is essential for the management of agricultural crops. One way of obtaining fast and inexpensive information on this nutrient is through multispectral image processing. Applications with images acquired by sensors on board of Remotely Pilot Aircrafts (RPA) have increased in Precision Agriculture. However, in citrus plants, in specific orange trees, little is known about the potential of multispectral images acquired with ARPs in the inference of leaf nitrogen content. Considering this context, the objective of this research was to evaluate the potential of multispectral images with high spatial resolution to infer the leaf nitrogen content in valencia-orange. As a method, we apply two approaches to infer the content of leaf nitrogen: (1) classification of images and; (2) spectral indices. In order to do this, we conducted an experiment in a valencia-orange orchard (Citrumelo Swingle rootstock) and collected leaves from different fields of the plantation. We conducted a flight with an eBee SenseFly ARP, equipped with the Parrot Sequoia camera, which records bands in the green, red, red-edge and near-infrared regions with a Ground Sample Distance (GSD) of 12 cm. In the first approach, we determined that the Spectral Angle Mappe (SAM) (r) classification algorithm had the best performance when classifying leaf nitrogen, with an overall accuracy of 85.7% and a kappa coefficient of 0.75. In the second approach, we determined that the spectral index Chlorophyll Vegetation Index (CVI) is the most accurate alternative (R² of 0.81 and Root Mean-Squared Error - RMSE) of 0.942 g.kg-1) among the indices tested to infer the leaf nitrogen content in valencia-orange. We conclude that it is possible to establish precise relationships between the foliar nitrogen content measured in the laboratory with the spectral response of the plant recorded in the image. Thus, multispectral images of high spatial resolution are efficient to discriminate levels of leaf nitrogen in valencia-orange. The approach investigated showed superior results in methods previously evaluated in literature.