%0 Conference Proceedings
%4 sid.inpe.br/marte2/2019/09.05.13.26
%2 sid.inpe.br/marte2/2019/09.05.13.26.38
%@isbn 978-85-17-00097-3
%T Machine learning algorithms to land cover mapping with Landsat-8
%D 2019
%A Richetti, Jonathan,
%A Silva, Laíza Cavalcante de Albuquerque,
%A Becker, Willyan Ronaldo,
%A Paludo, Alex,
%A Comineti, Humberto João,
%A Johann, Jerry Adriani,
%@affiliation Universidade Estadual do Oeste do Paraná (UNIOESTE)
%@affiliation Universidade Estadual do Oeste do Paraná (UNIOESTE)
%@affiliation Universidade Estadual do Oeste do Paraná (UNIOESTE)
%@affiliation Universidade Estadual do Oeste do Paraná (UNIOESTE)
%@affiliation Universidade Estadual do Oeste do Paraná (UNIOESTE)
%@affiliation Universidade Estadual do Oeste do Paraná (UNIOESTE)
%@electronicmailaddress j_richetti@hotmail.com
%@electronicmailaddress laiza.cavalcante@hotmail.com
%@electronicmailaddress willyan.becker@outlook.com
%@electronicmailaddress paludo.alex@hotmail.com
%@electronicmailaddress humbertocomineti@gmail.com
%@electronicmailaddress jerry.johann@hotmail.com
%E Gherardi, Douglas Francisco Marcolino,
%E Sanches, Ieda DelArco,
%E Aragão, Luiz Eduardo Oliveira e Cruz de,
%B Simpósio Brasileiro de Sensoriamento Remoto, 19 (SBSR)
%C Santos
%8 14-17 abril 2019
%I Instituto Nacional de Pesquisas Espaciais (INPE)
%J São José dos Campos
%P 2061-2064
%S Anais
%1 Instituto Nacional de Pesquisas Espaciais (INPE)
%K data mining, classification, remote sensing, satellite image.
%X Data mining algorithms applied to satellite image can be used to land cover mapping. This brings agility to the process of mapping areas and the accuracy can be assessed. However, with many machine learning algorithms it is hard to assess the best one for a giving task. Therefore, this work aims to test different machine learning algorithms to classify land cover using high-resolution imagery. Four algorithms were tested: Bagged CART, Random Forest (RF), Neural Network, and Model Averaged Neural Network in the Landsat-8 tile path/row 223/078 from December 13, 2017. A sample of 42,676 pixels in eight different categories (city, exposed soil, soybean, corn, turnip, pasture, forest, and water) was used. From all pixels, 25,607 pixels (60%) were used as training set and 17,069 pixels (40%) were used as testing set. The results shown that RF algorithm performed better with overall accuracy of 97% and kappa of 0.946.
%9 Classificação e mineração de dados
%@language pt
%3 97284.pdf