Uma abordagem multi-nível para classificação de imagens de moradias

Abstract

Agrarian settlements, for the most part, are made up of low-income families and many do not have sufficient financial conditions to own homes that guarantee a minimum of comfort. Policies for access to construction credit have difficulties in allocating necessary resources to meet the needs by region. Thus, the proposed work presents a model of Artificial Intelligence capable of classifying the Degree of Consolidation of residences, through their constructive characteristics. This model was implemented using 3 independent classification networks, where the first network analyzes and classifies the coarse details between the images, in order to avoid outliers classifications between the degrees. In the second stage, the characteristics and details of the houses are analyzed and classified, in order to more precisely classify their Degree of Consolidation. The results obtained demonstrate a considerably good rate of success in the classification of houses. In this way, these results can be used together with information from the housing locations to obtain a mapping of the regions that have the most houses with a low level of construction.