Uso de séries temporais para o mapeamento da cafeicultura

• Ano de defesa: 2015
• Autor(a) principal: Souza, Carolina Gusmão
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: UNIVERSIDADE FEDERAL DE LAVRAS; DCF - Departamento de Ciências Florestais; UFLA; BRASIL
• Palavras-chave em Português: Café; Classificação de imagens; Sensoriamento remoto; Algoritmos de aprendizagem de máquina; Greenbown; Coffee; Image classification; Remote sensing; Machine learning algorithms; Greenbrown; CNPQ_NÃO_INFORMADO
• Link de acesso: http://repositorio.ufla.br/jspui/handle/1/9410

Coffee is one of the main agricultural activities, with great importance in Brazil and in the world, being the State of Minas the largest coffee producer in the country. Estimate the basic data from this culture correctly is a challenge, once obtaining such information have little detail and the sector is still missing accurate information. Geotechnologies has been promising to fill this gap, evaluating more correctly the dynamics of coffee. However, the mapping of these areas is still a difficult task, since these areas are too complex to map, presenting a high confusion among the targets. To meet this need, the goal this study was to propose a methodology for mapping of coffee, by multispectral and multi-temporal variables. The study was conducted in two distinct areas, which are located in the State of Minas Gerais, the first one in the South region and the second in the Midwest region of the State. Firstly, classifications was performed, using high-resolution satellite imagery RapidEye, testing different machine learning algorithms and the combination of different variables (spectral, geometrical and textural) in the classification process. The results showed that the Suport Vector Machine algorithm achieved the best results in the rankings for all areas, with overall accuracy of 88.33%. The textural variables when associated with spectral, improved a little accuracy, however, there was not significant difference when the ratings were compared. Although the results have been shown with good levels of accuracy, yet there was much confusion between classes. To overcome this gap, we proposed a new method for mapping using data as variables multi-temporal in the classification process. The results showed that using the multi-temporal variables, integrated the spectral variables, obtained overall accuracy of 93% and reduced significantly the confusion among the targets, making more precise classification process. The methodology proposed in this study was efficient to map coffee areas.