COMPARAÇÃO ENTRE AS TÉCNICAS RANDOM FOREST E K-NEAREST NEIGHBORS PARA CLASSIFICAÇÃO DE IMAGEM DE SATÉLITE: ESTUDO DE CASO NA MICROBACIA DO RIO ALEGRE, RJ
COMPARAÇÃO ENTRE AS TÉCNICAS RANDOM FOREST E K-NEAREST NEIGHBORS PARA CLASSIFICAÇÃO DE IMAGEM DE SATÉLITE: ESTUDO DE CASO NA MICROBACIA DO RIO ALEGRE, RJ
-
Palavras-chave: análise de acurácia; monitoramento ambiental; uso e cobertura da terra.
-
Keywords: accuracy analysis; environmental monitoring; land use and land cover.
-
Abstract: This study aimed to classify land use and land cover in the Rio Alegre watershed (RJ, Brazil), using Random Forest (RF) and k-Nearest Neighbors (kNN) classification methods and to analyze the application of these algorithms. The analysis employed Landsat 8 satellite image to classify five land use and land cover classes: Water, Vegetation, Urban Area, Mosaic and Pasture. Data processing was performed using Quantum GIS software, version 3.34.11-Prizren. The results indicated that the kNN algorithm showed superior performance in identifying the classes, achieving an overall accuracy of 0.91, while the RF algorithm attained an accuracy of 0.87. Regarding the Kappa index, although kNN demonstrated higher overall accuracy, the RF algorithm exhibited a higher Kappa value. Greater uncertainty was observed in differentiating classes such as Urban Area and Mosaic, due to their similar spectral characteristics, particularly with the RF algorithm. Thus, it was concluded that while both algorithms are effective, kNN proved to be superior in this application. Finally, it is suggested that future studies employ images with different resolutions and compare them with other algorithms to enhance classification accuracy.
- Everaldo Zonta
- Vitória Duarte Miranda
- Gabriel Duarte Miranda
- Helena Saraiva Koenow Pinheiro