Aplicação da espectroscopia do infravermelho próximo e sensor hiperespectral combinados com aprendizado de máquina para classificação da qualidade do arroz branco, parboilizado, preto e vermelho

• Ano de defesa: 2025
• Autor(a) principal: JULIANO LUCAS CARDOSO JESUS
• Orientador(a): Paulo Carteri Coradi
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Fundação Universidade Federal de Mato Grosso do Sul
• País: Brasil
• Palavras-chave em Português: Infravermelho próximo, sensor hiperespectral, aprendizado, qualidade do arroz.
• Link de acesso: https://repositorio.ufms.br/handle/123456789/11565

Rice is one of the most consumed essential cereals in the world. The quality of the cereal is assessed upon receipt and shipment of batches at storage and processing units, through physical classification and typing. However, the traditional physical classification method is performed visually, while the assessment is subjective and imprecise, and this is a time-consuming process. The objective of this study was to evaluate the use of near-infrared spectroscopy and hyperspectral sensor combined with machine learning to determine the physicochemical properties and quality classification of white, parboiled, black and red rice. At the Post-Harvest Laboratory (LAPOS) of the Federal University of Santa Marias, samples of white, parboiled, black and red rice were classified according to the Normative Instruction to determine physical defects. Samples with the percentage of defects of 2 kg each were produced and subsequently divided into 100 subsamples of 20 g each. Near-infrared spectroscopy, spectral variables, multivariate analysis and machine learning were evaluated. Pearson correlation analyses demonstrated significant interactions between the nutritional properties of rice. Principal component analysis discriminated the physicochemical characteristics of rice, determining distinct nutritional patterns, highlighting the high nutritional value of black rice in relation to the starchy profile of white rice. Hyperspectral signatures revealed differences for the types of rice and for each processed rice according to the physicochemical composition. The combination of near-infrared spectroscopy and hyperspectral sensor and machine learning algorithms achieved greater precision for all accuracy metrics, with the J48, SL, RF and SVM models achieving greater efficiency for classifying rice quality, and may constitute a new method for evaluating batches of rice grains in storage and processing units.