Análise de composição de bandas espectrais do satélite sentinel-2 para uso em classificação supervisionada

Abstract

Nowadays, it is increasingly necessary to know the characteristics of the area where one is working, because this information is crucial for decision making, allowing this area to be explored in a correct way and so that all its productive potential is used. Remote Sensing, along with supervised classification, has been widely used for the preparation of land use/land cover maps, providing products that allow the characteristics of the surveyed areas to be known. Thus, the present work aimed to elaborate such land use/land cover maps through supervised classification using different spectral band compositions obtained from a Sentinel-2 satellite scene, so that it would be possible to evaluate these classifications, since each composition enhances different information about the image used. The spectral bands that were found to have a spectral resolution greater than 10 meters underwent a process called downscaling, so that all bands were at the same spatial resolution. The maps generated were compared with images from Google Earth Pro to verify the quality of the classification generated for each composition through the Kappa Index and Overall Accuracy. Thus, it was verified, through this work, that the composition containing the Near Infrared, Red and Green bands was the one that presented the best Kappa index and Global Accuracy, thus consecutively a better result in the production of a land use/land cover map, thus allowing to know the characteristics of the study area in the best possible way.