Fechar
Metadados

@InProceedings{GirolamoNetoFonsKörtSoar:2018:MaBrSa,
               author = "Girolamo Neto, Cesare di and Fonseca, Leila Maria Garcia and 
                         K{\"o}rting, Thales Sehn and Soares, Anderson Reis",
          affiliation = "{Instituto Nacional de Pesquisas Espaciais (INPE)} and {Instituto 
                         Nacional de Pesquisas Espaciais (INPE)} and {Instituto Nacional de 
                         Pesquisas Espaciais (INPE)} and {Instituto Nacional de Pesquisas 
                         Espaciais (INPE)}",
                title = "Mapping brazilian savanna physiognomies using worldview-2 imagery 
                         and geographic object based image analysis",
            booktitle = "Proceedings...",
                 year = "2018",
         organization = "Geobia",
             keywords = "Data Mining, Random Forests, Image Classification, Very High 
                         Resolution Images, Cerrado, Carbon Storage, Remote Sensing.",
             abstract = "Brazilian Savanna, or just Cerrado, is considered one of the 25 
                         hotspots for biodiversity conservation priority in the world. 
                         Cerrado occurs on the central part of Brazil and has three major 
                         natural formations: Grasslands, Savannas and Forests. However, the 
                         challenge on mapping Cerrado relies on the division of these major 
                         formations into specific physiognomies. Distinguishing each of 
                         these physiognomies is an important task to better evaluate 
                         smaller ecosystems, access carbon storage with greater precision 
                         and improve the exactitude of greenhouse gases emissions. Thus, 
                         the aim of this work is to evaluate the potential of very high 
                         spatial resolution imagery in order to improve the classification 
                         of 8 Cerrado physiognomies: Rocky Grasslands, Open Grasslands, 
                         Shrub Grasslands, Shrub Savanna, Typical Savanna, Dense Savanna, 
                         Flooded Plains with Palmtrees and Evergreen Forest. A WorldView-2 
                         image was used for a protect area with over 30 thousand hectares 
                         of preserved Cerrado vegetation. Features such as surface 
                         reflectance, vegetation indices, tasseled cap transformation and 
                         spectral linear mixture models were used on the automatic 
                         classification. Random Forests algorithm was used with a 10-fold 
                         cross-validation. The Global Accuracy was of 67.7%. Values above 
                         70% of Users Accuracy were obtained for classes such as Rocky 
                         Grasslands, Open Grasslands, Typical Savanna and Evergreen Forest. 
                         On the other hand, Flooded Plains with Palmtrees were omitted from 
                         the classification. Omission errors were also noticed for the 
                         classes of Shrub Savanna and Dense Savanna; they were sometimes 
                         misclassified as Typical Savanna which has a similar vegetation 
                         structure and tree cover percentage. The use of very high 
                         resolution images provided advantages on distinguishing Cerrado 
                         physiognomies on an automatic classification procedure. The 
                         detection of some classes was very precise and, despite the 
                         obtained misclassifications, it is an advance to distinguish some 
                         physiognomies that lower spatial resolution sensors are, hardly 
                         never, capable of distinguishing.",
  conference-location = "Montpellier, France",
      conference-year = "18-22 june",
             language = "en",
           targetfile = "neto_mapping.pdf",
        urlaccessdate = "26 nov. 2020"
}


Fechar