Classificação De Superfícies Impermeáveis Em Imagem Multiespectral Com Algoritmo De Machine Learning

Abstract

The urbanization process exposes the urban landscape to rapid and constant transformations. The change in land use and land cover patterns directly impacts the quality of life in cities. Therefore, monitoring the urban territorial composition becomes essential for urban management. To gain access to these data, studies have been applying remote sensing techniques combined with machine learning. Satellite images provide large-scale data with high temporal resolution, making it easier to detect changes in the landscape. Machine learning algorithms, on the other hand, provide classifications with greater accuracy compared to traditional methods. From this context and the available techniques, the study aims to evaluate the performance of the Support Vector Machine (SVM) algorithm in quantifying impervious areas in the urban perimeter of Presidente Prudente from a Planet image. The classification process was done using ArcGIS Pro software. The results demonstrate high performance for the SVM when applied in classification of impervious areas in urban territory. The accuracy of 94% shows that the method proposed in the work is useful as a tool for urban planning.