Abordagem colaborativa para segmentação de áreas degradadas e regeneradas

Abstract

Forests have huge environmental, social and economic importance. However, over the years they have been destroyed and transformed into pastures, agricultural fields, etc. Due to the popularity of remote sensing images, specifically the images acquired by the Landsat satellite series, several studies were proposed to find, monitor and analyze deforestation rates. However, researches indicate that images with higher spatial resolutions can overcome the results. Segmentation of remote sensing images is a crucial factor to support the inspection of deforested areas. Literature has different segmentation and pattern recognition techniques that have been used to solve this problem. Furthermore, these studies show the results of hybrid segmentation methods are better than single methods. Therefore, this research proposes a collaborative framework to segment and classify deforested areas. This approach analyzes four different segmentation methods (SAVI, AM techniques, multi-scale segmentation method and the marker-controlled watershed-based segmentation method) to suggest deforested areas in the State of São Paulo, Brazil. Results were compared with real environmental infractions found by the Military Environmental Police of the State of São Paulo manually and proved to be promising. Therefore, the proposal for a collaborative approach to suggest degraded areas becomes relevant as an approach to help the inspection of deforestation by police.