Automatic classification in Landsat images for the mapping of Otacílio Costa – SC

Abstract

This paper aimed to compare digital classification methods (supervised, unsupervised, object - oriented) in Landsat images in order to map the changes in land use and occupation for the years 2007 and 2017 for the municipality of Otacílio Costa - SC. For this purpose, images of the Landsat-5 TM sensor and the Landsat-8 OLI sensor were used. After the digital processing of the images, the classes of use and soil coverage were defined and the samples generated, divided into 60% of training and 40% of validation. Finally, the classification accuracy statistics for each method were calculated. The unsupervised methods were inefficient in all analyzed years, while the supervised ones were superior to the others. On the other hand, the object-oriented classification presented a classification considered excellent in 2007 and very good in 2017. The performance of the classification by the SVM method (Support Vector Machine) was excellent in 2007 and 2017, and it was considered the best evaluated method. From this, the mapping of the classes of use and coverage revealed a reduction of 4.8% of agricultural areas and 2.3% of urban areas and an increase of 1% for vegetation and 1.5% for water bodies