Detecção de mudanças em florestas e savanas utilizando análise estatística baseada em objetos geográficos

• Ano de defesa: 2015
• Autor(a) principal: Leite, Lucilia Rezende
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Lavras; Programa de Pós-Graduação em Engenharia Florestal; UFLA; brasil; Departamento de Engenharia Florestal
• Palavras-chave em Português: Savana brasileira; Floresta Amazônica; Sensoriamento remoto; Segmentação de imagens; Teste Qui-Quadrado; Distância de Mahalanobis; Brazilian savanna; Amazon Forest; Remote sensing; Image segmentation; Chi-square test; Mahalanobis distance; Silvicultura
• Link de acesso: http://repositorio.ufla.br/jspui/handle/1/10627

The aim of this study was to evaluate the change detection on the ground cover in areas of Forest and Brazilian Savanna, using images from Landsat 5 / TM and two methodologies based on iterative statistics object. The sensitivity of the methods in relation to the heterogeneity of the input data was evaluated using the reflectance data and vegetation indexes, and the use of different levels of confidence. The periods analyzed comprised the years 2000 to 2006 and 2006 to 2010. After the segmentation of the images were extracted the quantities descriptive average statistics and standard deviation of each object. The determination of change of objects was carried out iteratively based on Mahalanobis Distance and the chi-squared distribution. The results were validated using a previous visual detection and analyzed according to the ROC curve. Significant gains have been made in the shade of use and the bands 3 and 4 for both areas tested with 94.67% and 95.02% of correctly detected objects as change respectively in the areas of Forest and Savanna. The use of NDVI and different images proved unsatisfactory for the detection of changes in the areas tested.