DEEP LEARNING APPLICATIONS FOR BUILDING IDENTIFICATION IN THE RURAL AREA OF THE PANTANAL

Abstract

Environmental disasters have become increasingly intense and frequent, resulting in significant impacts on biodiversity, the environment, and the economy, often leading to loss of life. The environmental disaster can be potentiated by anthropic activities or events combined with natural events. In the Pantanal biome, one of the best-preserved biomes in Brazil, a large-scale environmental disaster occurred in 2020, resulting in the worst fires in recent history. Due to its characteristics and preservation, access in this region can be challenging, particularly with air transport and navigation. In addition, a good part of its extension comprises rural properties where the headquarters and houses are practically isolated due to the extension of the properties. Combining these factors, this study intends to detect buildings through artificial intelligence with deep learning techniques (object detection, instance segmentation, and semantic segmentation). This information has the potential to serve as a basis, a guide, or a source of data and information, especially in situations of an environmental disaster.

Keywords: environmental disasters, object detection, semantic segmentation, rescue, Pantanal, buildings.