Engenharia Florestal (Sede)
URI permanente desta comunidadehttps://arandu.ufrpe.br/handle/123456789/15
Siglas das Coleções:
APP - Artigo Publicado em Periódico
TAE - Trabalho Apresentado em Evento
TCC - Trabalho de Conclusão de Curso
Navegar
1 resultados
Resultados da Pesquisa
Item Uso de machine learning e sensoriamento remoto para a identificação da floresta tropical sazonalmente seca no Parque Nacional do Catimbau(2021-01-20) Monteiro Junior, José Jorge; Alba, Elisiane; El-Deir, Soraya Giovanetti; http://lattes.cnpq.br/3202139188457904; http://lattes.cnpq.br/1465154212352591; http://lattes.cnpq.br/0911037640720248The classification of seasonally dry tropical forests is one of the biggest challenges of environmental analysis by remote sensing, considering the forest physiognomic characteristics that are remotely similar to the characteristics of the exposed soil, generating sample errors in forest monitoring studies. The objective of this work was to use machine learning to understand the dynamics of land use and land cover in Catimbau National Park during periods of greater precipitation (wet) and less precipitation (dry) from LANDSAT imagery. The methodological treatment took place from the obtaining of LANDSAT data in 2019 for the wet period and dry period, the raw data were pre-processed in geographic information systems to (1st) select bands; (2nd) delimit the study area; (3rd) perform the atmospheric correction, and (4th) join the satellite bands (band set). A shapefile was created to train the machine learning algorithms containing samples of the classes found in the study area, these being the tree-shrub and shrub-herbaceous phytophysiognomies, anthropized areas, exposed soil, and other areas (i.e. clouds, water bodies, highways). In the R application, algorithms were used both for supervised classification (based on cross-validation, k-fold method, and Friedman and Nemenyi test) and for data spatialization using the aforementioned algorithms. With the described methods, it was possible to observe that the NDVI values promoted the idea that the shrub-herbaceous phytophysiognomy shows reflectance similar to the exposed soil in some areas in the dry period. In the wet period, the kNN algorithm performed better in-class differentiation and vegetation identification (Kappa = 0.9887). In the dry period, the kNN, SVM, and ANN algorithms did not show statistically significant differences regarding their performance, which are considered good classifiers for the period (Kappa = 0.9965; 0.9973; 0.9962, respectively). Therefore, the present study brought innovation in the use of Artificial Intelligence techniques to solve problems in the monitoring, management, and administration of seasonally dry tropical forests with remote data. Being an alternative method to identify, quickly and economically, changes in the forest structure.