Desempenho de diferentes abordagens classificadoras do uso e cobertura do solo urbano

Abstract

This study presents a comparative analysis of the performance of different classification approaches, specifically pixel-based and GEOBIA methods, in highly complex and variable urban environments. The dissertation is structured into chapters that will be turned into publications, covering both a bibliometric perspective on the use of remote sensing in urban land use and land cover classification and a practical evaluation of the classification methods. In the first chapter, through a bibliometric analysis and data visualization, we mapped the scientific production in the field of remote sensing applied to urban land classification, identifying publication patterns, leading authors, institutions, and prominent journals. The second chapter focused on the practical evaluation of classification techniques on high and medium-resolution data, using images from the Landsat 8, Sentinel-2, and WorldView-2 satellites, processed on the Google Earth Engine cloud computing platform. Cross-validation and holdout validation were applied to verify the accuracy of the classifications. Overall, this research provides a mapping of the evolution of scientific knowledge in the field, a quantitative evaluation of the classification approaches, and underscores the importance of combined techniques for efficient urban planning, contributing to the development of smart and sustainable cities.