Identificação e caracterização espectral da ferrugem (Austropuccinia psidii) do eucalipto por imagens de alta resolução obtidas de veículos aéreos não tripulados (vant) e em laboratório (espectroradiômetro)
Carregando...
Data
2018-05-15
Autores
Jim, André Stefanini [UNESP]
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
Plantios de eucalipto podem sofrer reduções na produção de até 48% devido à ferrugem das mirtáceas causada por Autropuccinia psidii. A identificação e quantificação da doença, ou fitopatometria, e o diagnóstico precoce de doenças é fator chave no manejo florestal. Em campo, o levantamento é realizado por amostragens, que alcançam apenas 5% da área total plantada e são realizadas por equipes de avaliadores por meio do uso de escalas diagramáticas. O uso do SIG na silvicultura tem sido bem-sucedido na detecção de diversos fatores considerados favoráveis às doenças. O termo Silvicultura de Precisão consiste na adoção de ferramentas de sensoriamento remoto embarcadas em distintas plataformas juntamente com sistemas computacionais para o processamento das múltiplas informações. Dentro do campo de sensores, uma plataforma alternativa para o diagnóstico é o emprego de VANTs (Veículos Aéreos Não Tripulados) com câmeras digitais de alta resolução espacial (pixels/cm²), apresentando diversas vantagens em relação aos dados advindos de satélites. Enquanto as informações espectrais coletadas por VANTs são misturas espectrais de dosséis, o espectroradiômetro pode coletar informações de objetos, ao nível de folha, sendo assim uma importante ferramenta para comparação de resultados de análises. Diante da necessidade de se criar uma metodologia para identificação e diagnóstico de doenças em florestas, o objetivo deste trabalho foi o de associar as informações espectrais aos procedimentos estatísticos multivariados, técnicas de regressão logística e bootstrap, de modo a obter modelos de classificação consistentes quanto à identificação e separação dos níveis de severidade para a ocorrência da ferrugem das mirtáceas em plantios jovens de eucalipto. Os resultados demonstraram que a regressão logística associada à função linear discriminante alcançaram as melhores taxas de acerto para dados de laboratório (espectroradiômetro), ao passo que a análise multivariada (por componentes principais) associada à função quadrática discriminante obtiveram as melhores taxas de acerto para dados de campo (VANT). O estudo permitiu concluir que as ferramentas empregadas foram adequadas para a diferenciação de plantas sadias de plantas infectadas por A. psiidi em plantios de Eucalyptus spp. Pôde-se concluir que o uso das imagens de VANT’s oferece uma análise preliminar ampla, possibilitando determinar áreas a receberem uma investigação mais detalhada, que eventualmente, pode ser via uso de espectroradiômetro. Quanto à caracterização do patógeno, observou-se que a técnica stepwise, coincidentemente, selecionou reflectâncias próximas à atividade das clorofilas a e b, excluindo a faixa espectral do verde.
Eucalyptus plantations may suffer reductions in yield of up to 48% due to myrtle rust caused by Autropuccinia psidii. The disease identification and quantification, or phytopatometry, and the early diagnosis are a key factor in forest management. In the field, the survey is carried out by samplings, which reach only 5% of the total planted area and are carried out by teams of raters through the use of diagrammatic scales. The GIS use in forestry has been used in detecting several factors considered favorable to diseases. The term Precision Forestry consists in the adoption of remote sensing tools embedded in different platforms with computational systems for the processing of the multiple information. In terms of sensors, an alternative platform for early diagnosis is the use of UAVs equipped with high spatial resolution digital cameras (pixels / cm²), presenting several advantages compared to the satellite's data. While the spectral information collected by VANTs are canopies spectral mixtures, the spectroradiometer can collect information from objects at the leaf level and thus is an important tool for comparing results analysis. The need to create a methodology for early identification and diagnosis of forest diseases, the aim of the present work was to associate spectral information with multivariate statistical procedures, logistic regression techniques, and bootstrap, in order to obtain consistent classification models, identification of severity levels of myrtle rust in young eucalyptus plantations. The results showed that the logistic regression associated with the linear discriminant function achieved the best hit rates for laboratory data (spectroradiometer), while the multivariate analysis (by main components) associated with the discriminant quadratic function obtained the best data set rates field (UAV). We could conclude that the tools used were adequate for the differentiation of healthy plants of plants infected by A. psidii in plantations of Eucalyptus spp. It was concluded that the use of VANT's images offers a preliminary extensive analysis, making it possible to determine areas to receive a more detailed investigation, which may be via the use of a spectroradiometer. As for the characterization of the pathogen, it was observed that the stepwise technique, coincidentally, selected reflectances close to the activity of chlorophyll a and b, excluding the spectral range of green.
Eucalyptus plantations may suffer reductions in yield of up to 48% due to myrtle rust caused by Autropuccinia psidii. The disease identification and quantification, or phytopatometry, and the early diagnosis are a key factor in forest management. In the field, the survey is carried out by samplings, which reach only 5% of the total planted area and are carried out by teams of raters through the use of diagrammatic scales. The GIS use in forestry has been used in detecting several factors considered favorable to diseases. The term Precision Forestry consists in the adoption of remote sensing tools embedded in different platforms with computational systems for the processing of the multiple information. In terms of sensors, an alternative platform for early diagnosis is the use of UAVs equipped with high spatial resolution digital cameras (pixels / cm²), presenting several advantages compared to the satellite's data. While the spectral information collected by VANTs are canopies spectral mixtures, the spectroradiometer can collect information from objects at the leaf level and thus is an important tool for comparing results analysis. The need to create a methodology for early identification and diagnosis of forest diseases, the aim of the present work was to associate spectral information with multivariate statistical procedures, logistic regression techniques, and bootstrap, in order to obtain consistent classification models, identification of severity levels of myrtle rust in young eucalyptus plantations. The results showed that the logistic regression associated with the linear discriminant function achieved the best hit rates for laboratory data (spectroradiometer), while the multivariate analysis (by main components) associated with the discriminant quadratic function obtained the best data set rates field (UAV). We could conclude that the tools used were adequate for the differentiation of healthy plants of plants infected by A. psidii in plantations of Eucalyptus spp. It was concluded that the use of VANT's images offers a preliminary extensive analysis, making it possible to determine areas to receive a more detailed investigation, which may be via the use of a spectroradiometer. As for the characterization of the pathogen, it was observed that the stepwise technique, coincidentally, selected reflectances close to the activity of chlorophyll a and b, excluding the spectral range of green.
Descrição
Palavras-chave
Assinatura espectral, Silvicultura de Precisão, Análise Multivariada, Modelos lineares generalizados, Bootstrap