Anais do XX Simpósio Brasileiro de Sensoriamento Remoto

This paper aims to develop a classification of land use
and coverage, in order to distinguish planted and native
forest areas, aiming to use this information to build carbon
storage models in these areas. For this purpose, the
visible, NIR, SWIR and Red-Edges bands of the Sentinel-2A
satellite scene were used, in addition to the NDVI and SAVI
indices, and a GEographic Object-Based Image Analysis
- GEOBIA - was performed, as well as a classification
by decision tree. Spectral, spatial and textural attributes
were extracted from the segmentation performed, and later
samples were collected by dividing the region into a grid.
The classification was performed using the C5.0 algorithm,
with 100 trees generated. The results obtained were a mean
F1-Score of 0.956 and a Kappa index of 0.967. The achieved
classification presents a satisfactory quality, proving to be
promising in the differentiation of the proposed areas.