CLASSIFICAÇÃO DA VEGETAÇÃO DO PARQUE NACIONAL DA CHAPADA DAS MESAS, MARANHÃO, USANDO OBIA, MACHINE LEARNING E SOFTWARES LIVRES

Several studies have shown the importance of using machine learning algorithms for classifying targets in remote sensing images. When combined with Object-Based Image Analysis (OBIA) methods, they can yield excellent results, especially for high spatial resolution imagery. The present study aimed to classify the vegetation of the National Park of Chapada das Mesas (NPCM), using free, open access software. For the classification images from the Dove PlanetScope constellation were used, acquired at the dry and rainy seasons of the NPCM, together with elevation data from the ALOS World 3D model. The procedures were performed using the RSGISLib library of the Python language, the R language, and QGIS. The following procedures were carried: segmentation; classification; and filtering and validation. The Random Forests classification algorithm was chosen, and for accuracy analysis, the Kappa, global precision, and quantity and allocation disagreement indexes were computed from the confusion matrix, using the rsacc R package. Classification results had 71.49% overall accuracy. The use of tools and free software has shown promise for future similar studies.