Close
Metadata

@InProceedings{SilvaMeirSilv:2020:UsPaLe,
               author = "Silva, Alexandre and Meireles, Sincler and Silva, Samira",
          affiliation = "{Universidade do Estado de Minas Gerais} and {Universidade do 
                         Estado de Minas Gerais} and {Universidade do Estado de Minas 
                         Gerais}",
                title = "Using Partial Least Squares in Butterfly Species Identification",
            booktitle = "Proceedings...",
                 year = "2020",
               editor = "Musse, Soraia Raupp and Cesar Junior, Roberto Marcondes and 
                         Pelechano, Nuria and Wang, Zhangyang (Atlas)",
         organization = "Conference on Graphics, Patterns and Images, 33. (SIBGRAPI)",
            publisher = "IEEE Computer Society",
              address = "Los Alamitos",
             keywords = "Butterfly Identification, Pattern Recognition, Partial Least 
                         Squares.",
             abstract = "Butterflies are important insects in nature, and along with moths 
                         constitute the Lepidoptera order. At the global level, the number 
                         of existing butterfly species is approximately 16,000. Therefore, 
                         the identification of their species in images by humans consists 
                         in a laborious task. In this paper, we propose a novel approach to 
                         recognize butterfly species in images by combining handcrafted 
                         descriptors and the Partial Last Squares (PLS) algorithm. A set of 
                         PLS models are trained using an one-against-all protocol. The test 
                         phase consists in presenting images to all classifiers and the one 
                         which provides the highest response value contains in the positive 
                         set the predicted class. The performance of the proposed approach 
                         is evaluated on the Leeds Butterfly dataset. Experiments were 
                         conducted using HOG and LBP descriptors, separately and combined. 
                         The approach using HOG singly reported an accuracy rate of 68.72%, 
                         while using only LBP resulted in an accuracy rate of 77.33\%. 
                         Combining both descriptors this value changes to 76.27%. The 
                         proposed approach achieves the best results in all three versions 
                         when compared to state-of-the-art approaches. Experiments have 
                         shown that describing images with LBP provides the highest 
                         accuracy values since it extracts texture information, what is an 
                         important characteristic to distinguish butterflies. However, 
                         information of color and shape, added by HOG, appears to make 
                         different species confused.",
  conference-location = "Virtual",
      conference-year = "Nov. 7-10, 2020",
             language = "en",
           targetfile = "example.pdf",
        urlaccessdate = "2020, Nov. 28"
}


Close