Classificação de sementes de soja maduras e esverdeadas por meio de métodos ópticos

• Ano de defesa: 2018
• Autor(a) principal: Souza Sobrinho, Alcebíades Fogaça
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Lavras; Programa de Pós-Graduação em Engenharia de Sistemas e Automação; UFLA; brasil; Departamento de Engenharia
• Palavras-chave em Português: Sementes de soja esverdeada; Processamento de imagens; Iluminação por laser; Green soybean seeds; Image processing; Lighting by laser; Engenharia de Software
• Link de acesso: http://repositorio.ufla.br/jspui/handle/1/33435

Since the 1980's the Brazilian agriculture has been going through an unprecedented evolution. This is a result of successive efforts in research and field applications, which placed Brazil as a worldwide food supplier. However, the dynamism of agriculture constantly brings about new challenges that demand new solutions. The high occurrence of green soybean seeds is a problem, because of their low germination rate; even when they germinate, the seedlings are anomalous. Moreover, if they are processed into byproducts, the presence of chlorophyll in these grains makes them undesirable, since additional procedures are required to remove it. This work aimed at evaluating the differentiation of ripe and green soybean seeds under red laser, green laser, red LED lights and fluorescent light through image processing. We captured images (340x480 pixels) of ripe and green soybean seeds that were illuminated by red laser, green laser, red LED lights and fluorescent lights. Afterwards, we obtained the average grey scale v alues for each image, according to the Red, Green, Blue channels, and for images converted to an 8 -bit grey scale. The data were submitted to variance tests, Principal Component Analyses, Multiple Factor Analyses, and grey level calculations to classify the images. According to the results, illuminating the seeds with red laser, red LED lights and fluorescent lights was effective for image classification, reaching a 90% level of accuracy at Red channel.