@InProceedings{PaivaOliPasParMar:2020:EfSoGe,
               author = "Paiva, Roberto U. and Oliveira, Savio S. T. and Pascoal, Luiz M. 
                         L. and Parente, Leandro L. and Martins, Wellington S.",
          affiliation = "{Universidade Federal de Goi{\'a}s (UFG)} and {Universidade 
                         Federal de Goi{\'a}s (UFG)} and {Universidade Federal de 
                         Goi{\'a}s (UFG)} and {Universidade Federal de Goi{\'a}s (UFG)} 
                         and {Universidade Federal de Goi{\'a}s (UFG)}",
                title = "An Efficient Solution to Generate Meta-features for Classification 
                         with Remote Sensing Time Series",
            booktitle = "Anais...",
                 year = "2020",
               editor = "Carneiro, Tiago Garcia de Senna (UFOP) and Felgueiras, Carlos 
                         Alberto (INPE)",
                pages = "46--57",
         organization = "Simp{\'o}sio Brasileiro de Geoinform{\'a}tica, 21. (GEOINFO)",
            publisher = "Instituto Nacional de Pesquisas Espaciais (INPE)",
              address = "S{\~a}o Jos{\'e} dos Campos",
             abstract = "Over the last years, the volume of Earth observation (EO) data in- 
                         creased significantly due to the large number of satellites 
                         orbiting the planet. These data are being used by automatic 
                         classification approaches to generate land-use and land-cover 
                         (LULC) products for different landscapes around the world. Dynamic 
                         Time Warping (DTW) is a classical method used to measure the 
                         similarity between two time series. In this context, DTW-based 
                         algorithms are an efficient approach to handle EO time series. 
                         These algorithms can be used to generate meta-features (i.e., new 
                         features automatically derived from the orig- inal features) to 
                         improve the performance of classification models. However, these 
                         algorithms have a long processing time and depends on large 
                         computa- tional resources, making it difficult to use in large 
                         data volumes. Seeking to address this limitation, this work 
                         presents a full scalable parallel solution to optimize the 
                         construction of remote sensing meta-features. Additionally, a new 
                         classification strategy is presented, in which, the meta-features 
                         generated were used to train and evaluate a Random Forest model. 
                         Our results shows that both approaches leads to improvement in 
                         execution time and overall accuracy when compared to traditional 
                         methods.",
  conference-location = "On-line",
      conference-year = "30 nov. a 03 dez. 2020",
                 issn = "2179-4847",
             language = "en",
                  ibi = "8JMKD3MGPDW34P/43PLBGP",
                  url = "http://urlib.net/ibi/8JMKD3MGPDW34P/43PLBGP",
           targetfile = "p5.pdf",
                 type = "Dados espa{\c{c}}o-temporais",
        urlaccessdate = "13 maio 2024"
}