É possível detectar sintomas de nematoides em soja (Glycine max L.) utilizando sensoriamento remoto?
Carregando...
Data
2020-08-10
Autores
Santos, Letícia Bernabé
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
A utilização de técnicas de sensoriamento remoto teve expressivo aumento na agricultura nos últimos anos para avaliar diferentes estresses em culturas. Contudo, ainda são poucos trabalhos que avaliaram a viabilidade na identificação de nematoides. A atual amostragem de nematoides é intensiva em tempo e trabalho e não representa a variabilidade da infestação em todo o campo, dificultando a previsão precisa e o manejo. Portanto, foram desenvolvidos trabalhos visando verificar o potencial uso do sensoriamento aéreo na identificação de sintomas da ocorrência de nematoides na cultura da soja. No primeiro capítulo desta dissertação encontra-se a revisão de literatura contendo tópicos específicos relacionados ao tema central. No segundo, foi avaliada a variabilidade da ocorrência de nematoides em dois campos de produção da cultura da soja e o efeito da área de infestação em relação ao centro da reboleira (dentro, extremidade ou fora), assim como o comportamento multiespectral desses locais. Observou-se que as bandas do vermelho e NIR tiveram comportamento similar nas duas áreas avaliadas e foram capazes de se diferenciar em relação à localização na reboleira. No terceiro capítulo, o objetivo foi determinar a relação bivariada entre bandas espectrais individuais e índices de vegetação (IVs), todos relativos à condição da soja (plantas infectadas versus não infectadas). Utilizamos três algoritmos (Logistic Regression - LR, Random Forest – RF, Conditional Inference Tree – CIT) com três opções para entrada de dados: apenas bandas (modelo reduzido), apenas IVs e bandas mais IVs (modelo completo) para classificar as plantas. Os resultados demonstram a capacidade de se utilizar dados multiespectrais para distinguir plantas de soja infectadas por nematoides e não infectadas, utilizando sensoriamento aéreo combinado com aprendizado de máquina. No quarto capítulo, tem-se as considerações finais com os principais resultados, fatores limitantes, recomendações e os próximos passos da pesquisa.
Remote sensing techniques have increased significantly in agriculture in recent years to assess different crop stresses. However, there are still few studies that have evaluated the viability to identify nematodes symptoms. The current nematode sampling approach is time and labor intensive and does not represent the variability of the infestation across the field, complicating its accurate prediction and management. Therefore, this work was carried out to verify the potential use of aerial sensing to identify nematodes symptoms occurring in soybean. The first chapter of this dissertation is a literature review containing specific topics. The second chapter is a variability analysis of nematode occurrence in soybean fields and the effect of the infestation area in relation to the center of the hotspot (inside, border or outside) were evaluated, as well as the multispectral behavior of these locations. It was observed that the red and near-infrared bands had similar behavior in the two areas and were able to differentiate different infestation areas in relation to location in the hotspot. In the third chapter, the objective was to determine the bivariate relationship between individual spectral bands and vegetation indices (VIs), all related to the condition of the soybean (infected versus non-infected plants). We used three algorithms (logistic regression, random forest, and conditional inference tree) with three options for data entry: only bands (reduced model), only VIs and bands plus VIs (complete model) to classify the plants. The results demonstrate the ability of multispectral data to distinguish nematode-infected and non-infected soybean plants, using aerial sensing combined with machine learning. In the fourth chapter, there are the final considerations with the main results, limiting factors, recommendations and the next steps of the research.
Remote sensing techniques have increased significantly in agriculture in recent years to assess different crop stresses. However, there are still few studies that have evaluated the viability to identify nematodes symptoms. The current nematode sampling approach is time and labor intensive and does not represent the variability of the infestation across the field, complicating its accurate prediction and management. Therefore, this work was carried out to verify the potential use of aerial sensing to identify nematodes symptoms occurring in soybean. The first chapter of this dissertation is a literature review containing specific topics. The second chapter is a variability analysis of nematode occurrence in soybean fields and the effect of the infestation area in relation to the center of the hotspot (inside, border or outside) were evaluated, as well as the multispectral behavior of these locations. It was observed that the red and near-infrared bands had similar behavior in the two areas and were able to differentiate different infestation areas in relation to location in the hotspot. In the third chapter, the objective was to determine the bivariate relationship between individual spectral bands and vegetation indices (VIs), all related to the condition of the soybean (infected versus non-infected plants). We used three algorithms (logistic regression, random forest, and conditional inference tree) with three options for data entry: only bands (reduced model), only VIs and bands plus VIs (complete model) to classify the plants. The results demonstrate the ability of multispectral data to distinguish nematode-infected and non-infected soybean plants, using aerial sensing combined with machine learning. In the fourth chapter, there are the final considerations with the main results, limiting factors, recommendations and the next steps of the research.
Descrição
Palavras-chave
Detecção de doenças, Nematoides, Soja, Imagens multiespectrais, Aprendizado de máquina, Agricultura digital