Redes neurais artificiais na predição do tempo de armazenamento de grãos de feijão

• Ano de defesa: 2018
• Autor(a) principal: Farias, Hiago Felipe Lopes de
• Orientador(a): Devilla, Ivano Alessandro
• Banca de defesa: Devilla, Ivano Alessandro, Melo, Francisco Ramos de, Resende, Osvaldo
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual de Goiás
• Programa de Pós-Graduação: Programa de Pós-Graduação Stricto sensu em Engenharia Agrícola
• Departamento: UEG ::Coordenação de Mestrado em Engenharia Agrícola
• País: Brasil
• Palavras-chave em Português: Escurecimento do feijão; Dureza do feijão; RNA; Multilayer perceptron; Feijão; Armazenamento
• Palavras-chave em Inglês: Dimming; Hardness; ANN; Multilayer perceptron
• Área do conhecimento CNPq: CIENCIAS AGRARIAS::ENGENHARIA AGRICOLA
• Link de acesso: http://www.bdtd.ueg.br/handle/tede/730

Bean is a widely cultivated crop in Brazil and the world. In the period of storage of grains, deterioration of the product occurs, which is gradual, irreversible and cumulative. Artificial Neural Networks (ANNs) have been used in a wide range of applications, such as: standard classification, recognition pattern, optimization, prediction and automatic control. In some cases, ANNs have performed better than the regression models. In the light of the above, this work aimed to evaluate the performance of artificial neural networks in predicting the storage time of bean grains as a function of color, tegument hardness and different temperatures. The grains were produced and stored by Embrapa Rice e Beans, located in the municipality of Santo Antônio de Goiás, harvest 2013/2014. Five groups of carioca bean cultivars with water content of 13% b.u. in the year 2014, the samples were stored in a Biochemical Oxygen Demand (BOD) type chamber, at temperatures (15, 21 and 37 ° C). Grain samples were collected at (36, 72, 108, 144 and 180) days of storage and staining and hardness evaluations of the tegument of the grains. The first evaluation was performed with the grains freshly harvested in the year 2014, identified as control samples. Data were normalized between -1 to 1, the trained networks were of the Multilayer Perceptron (MLP) type, after the training was selected the network that presented better performance to solve the problem. The best RNA had a success rate of 83.0% with training data and 91.2% with validation data, presented a correlation higher than 0.900 for training, validation and testing. Under the conditions in which this work was developed it can be concluded that RNAs can be used to estimate storage days as a function of color, hardness and temperature.