Utilização de redes neurais artificiais aplicadas na discriminação de padrões de doenças florestais

• Ano de defesa: 2015
• Autor(a) principal: Favan, João Ricardo [UNESP]
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual Paulista (Unesp)
• Palavras-chave em Português: Eucalipto - Doenças e pragas; Bacterioses; Imagens digitais; Topologia; Digital images
• Link de acesso: http://hdl.handle.net/11449/132129; http://www.athena.biblioteca.unesp.br/exlibris/bd/cathedra/20-10-2015/000853211.pdf

The forest sector has great importance for the Brazilian economy, whose, is raw materials supplier for reforestation of degraded areas as well as for industries like pulp and paper, coal, furniture, among others. However to achieve good results is necessary to use healthy forest seedlings. Bacterial stain of the eucalyptus and the stain caused by Cylindrocladium are two common diseases in nurseries and although the etiologic agents are different, their symptoms are similar, and may cause doubts in a diagnosis moment. The aim of this study was to develop an artificial neural network (ANN) to classify the two cited diseases using digital images. Images of both diseases were processed using a threshold technique to remove the image background, and their histograms were used to training a multilayer perceptron ANN with backpropagation algorithm. Ten essays with five different topologies ( 34, 128, 248, 368 and 511 neurons in the hidden layer) were carried out. All topologies reached on average 95% of correct classifications. The topology with 256 neurons in hidden layer was considered the most suitable one for this project