Use of sampling polygons in supervisioned classifications of satellite images

Abstract

The territory dynamics has always been prominent in geographic and, more precisely, environmental research. Different methods and techniques can be used to perform these studies, among them, it is possible to highlight the application of remote sensing techniques with the use of orbital satellite images. This study aimed to analyze the data obtained through the relationship between the dimensions of the classified areas and the increasing number of sampling polygons, based on the supervised classification of satellite images using training areas or polygon clusters. Experiments were carried out with three classes for the years 1985 and 2018, namely water bodies, anthropic areas and "original" vegetation cover. The simulations carried out confirmed the hypothesis presented, i.e., that there would be a trend of stabilization in the data according to the increase of training areas. As a recommendation for future classifications, it is suggested to adopt at least fifty sample polygons per class, which best defines the region to be classified, since such areas of training should cover all the characteristics related to the classes to be adopted.

Keywords

• sensoriamento remoto. classificação supervisionada. áreas de treinamento. polígonos de amostragem. cluster de polígonos.