@InProceedings{Carvalho:2003:DeTiSe,
               author = "Carvalho, Luis Marcelo Tavares de",
          affiliation = "{Universidade Federal de Lavras}",
                title = "Declouding time series of Landsat data",
            booktitle = "Anais...",
                 year = "2003",
               editor = "Epiphanio, Jos{\'e} Carlos Neves and Fonseca, Leila Maria 
                         Garcia",
                pages = "2035 - 2042",
         organization = "Simp{\'o}sio Brasileiro de Sensoriamento Remoto, 11. (SBSR).",
            publisher = "INPE",
              address = "S{\~a}o Jos{\'e} dos Campos",
             keywords = "remote sensing, image processing, time series, cloud removal.",
             abstract = "Novel schemes based on multiresolution transforms were introduced 
                         to pre-process long time series of Landsat data. Particularly, 
                         removal of clouds and their shadows was tackled. We applied the 
                         product of wavelet scales to generate binary masks of corrupted 
                         observations, the robust smoother-cleaner wavelets to remove 
                         outliers in the data, and the wavelet shrinkage to estimate new 
                         values. Cloud contamination was simulated and the missing values 
                         were estimated using five methods: 1) mean value, 2) minimum 
                         value, 3) maximum value, 4) linear regression, and 5) the 
                         wavelet-based procedure. The product of wavelet scales not only 
                         identified clouded and shadowed pixels but also other anomalies 
                         like misregistration effects and changes of short duration (e.g., 
                         burn scars). The wavelet-based approach was more accurate for 
                         interpolating the missing values in clouded areas, whereas linear 
                         regression performed better in shadowed areas. The robust 
                         non-linear wavelet regression holds promise for effective time 
                         series analysis and has the potential to produce noise-reduced 
                         images at any point in the time series.",
  conference-location = "Belo Horizonte",
      conference-year = "5-10 abr. 2003",
                 isbn = "85-17-00017-X",
             language = "pt",
         organisation = "Instituto Nacional de Pesquisas Espaciais",
                  ibi = "ltid.inpe.br/sbsr/2002/11.14.16.26",
                  url = "http://urlib.net/ibi/ltid.inpe.br/sbsr/2002/11.14.16.26",
           targetfile = "15_198.pdf",
                 type = "Processamento de Imagens Digitais / Digital Image Processing",
        urlaccessdate = "10 maio 2024"
}