NDVI e índices do espectro visível como descritores de uma Random Forest para classificar a cobertura do solo no entorno da Represa de Itupararanga - SP por meio de imagens do satélite CBERS4A
Carregando...
Data
2023-02-15
Autores
Fernandes, Wanessa Monteiro
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
O sensoriamento remoto tem grande destaque na análise e monitoramento da superfície terrestre, especialmente devido a possibilidade de analisar grandes extensões de área, com destaque para os métodos computacionais que utilizam índices baseados nos canais do espectro eletromagnético. Visto isso, o objetivo deste estudo foi utilizar índices do espectro do visível e o NDVI (Normalized Difference Vegetation Index) obtidos a partir de imagens do satélite CBERS 4A em conjunto com a aplicação do classificador Random Forest para a análise de regiões no entorno da represa de Itupararanga, localizada na Região Metropolitana de Sorocaba (RMS). Para isso, utilizou-se os softawares QGis e Rstudio, nos quais foram realizados o pré-processamento das imagens, cálculo dos índices, coletas de imagens de treinamento das regiões características e implementação da Random Forest. As classes de cobertura do solo obtidas foram agricultura, área urbanizada, corpo hídrico, solo exposto, vegetação arbórea e vegetação rasteira, permitindo uma análise qualitativa e quantitativa pela Random Forest. Observou-se que o NDVI apresentou um melhor desempenho em relação aos outros índices utilizados na segmentação das diversas regiões de cobertura do solo, além da banda do infravermelho próximo (NIR), sendo os descritores mais relevantes para a classificação das regiões usando a Random Forest.
Remote sensing has great prominence in the analysis and monitoring of the Earth's surface, since through sensors it is possible to analyze large areas. One of the methods for this analysis are those that use indices based on the channels from the electromagnetic spectrum. This research aimed to study the use of visible spectrum indices and NDVI (Normalized Difference Vegetation Index) obtained from CBERS 4A satellite images and a Radom Forest as a classifier for the analysis of regions around the Itupararanga dam. For this, the softwares QGis and Rstudio were used, in which the pre-processing of the images, calculation of the indices, collection of training images of the characteristic regions and implementation of the Random Forest were carried out. The following land cover classes were considered in this study: agriculture, urbanized area, water, exposed soil, arboreal vegetation and undergrowth. In addition to the visual qualitative analysis of the index relevancies, the use of Random Forest allowed a quantitative evaluation. It was observed that the NDVI presents an excellent result in the segmentation of the different land cover regions and that, individually, the near infrared band (NIR) and the NDVI are the most relevant descriptors for the classification of the regions using Random Forest.
Remote sensing has great prominence in the analysis and monitoring of the Earth's surface, since through sensors it is possible to analyze large areas. One of the methods for this analysis are those that use indices based on the channels from the electromagnetic spectrum. This research aimed to study the use of visible spectrum indices and NDVI (Normalized Difference Vegetation Index) obtained from CBERS 4A satellite images and a Radom Forest as a classifier for the analysis of regions around the Itupararanga dam. For this, the softwares QGis and Rstudio were used, in which the pre-processing of the images, calculation of the indices, collection of training images of the characteristic regions and implementation of the Random Forest were carried out. The following land cover classes were considered in this study: agriculture, urbanized area, water, exposed soil, arboreal vegetation and undergrowth. In addition to the visual qualitative analysis of the index relevancies, the use of Random Forest allowed a quantitative evaluation. It was observed that the NDVI presents an excellent result in the segmentation of the different land cover regions and that, individually, the near infrared band (NIR) and the NDVI are the most relevant descriptors for the classification of the regions using Random Forest.
Descrição
Palavras-chave
CBERS4A, Geoprocessamento, Random Forest