Método baseado em análise de imagens speckle e aprendizagem profunda para identificação do defeito germinado na inspeção visual da qualidade de grãos de feijão

• Ano de defesa: 2023
• Autor(a) principal: Veiga Junior, Eli
• Orientador(a): Araújo, Sidnei Alves de
• Banca de defesa: Araújo, Sidnei Alves de , Deana, Alessandro Melo , Prates, Renato Araujo , Belan, Peterson Adriano
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Nove de Julho
• Programa de Pós-Graduação: Programa de Pós-Graduação em Informática e Gestão do Conhecimento
• Departamento: Informática
• País: Brasil
• Palavras-chave em Português: grão; feijão; germinado; visão computacional; speckle; inteligência artificial
• Palavras-chave em Inglês: grain; bean; germinated; computer vision; speckle; artificial intelligence
• Área do conhecimento CNPq: CIENCIA DA COMPUTACAO::SISTEMAS DE COMPUTACAO
• Link de acesso: http://bibliotecatede.uninove.br/handle/tede/3240

The automatic visual inspection of agricultural grains can bring a competitive differential for companies, because it allows a higher standardization of the results leading to the obtainment of products with higher quality and, therefore, with higher added value. Currently the visual inspection of bean grains is conducted manually in order to determine the Group, Class and Type of the product. The term "Group" is used to refer to the botanical species, while the "Class" is a classification of beans based on the color of the skin and the "Type" is related to the qualitative characteristics of the product, being determined according to the defects found in the sample, among which are: broken, careworn, burnt, moldy and sprouted, which in its advanced stage the grain presents visible radicle (sprout). The germinated defect in its initial phase is only detected by invasive methods, breaking the grain to get visualization. As an alternative, there are non-invasive methods in the literature based on speckle analysis that are already used in many areas, including agriculture, to analyze seed viability for planting. In this paper a method is presented that combines speckle image analysis and deep learning neural networks to identify the germinated defect in its early stage in bean grains, which would usually go unnoticed by human vision. First, the maps generated by Laser Spatial Speckle Contrast Analysis (LASCA) and by, Laser Temporal Speckle Contrast Analysis (LASTCA) were compared. Subsequently, 50×50 pixel windows were extracted from the LASTCA maps, which showed the best results, and submitted to a Convolutional Neural Network (CNN) that classified them to indicate the existence of the germinated defect in the analyzed grains. The proposed method was able to identify the defect with an accuracy of 92.33% and with high sensitivity (98.21%), demonstrating its applicability in the process of visual inspection of bean quality in agribusiness.