Avaliação da predição da qualidade de grãos de milho utilizando modelos de aprendizado de máquina nas etapas de transporte, secagem e armazenagem a partir do monitoramento de variáveis mensuráveis

• Ano de defesa: 2023
• Autor(a) principal: Rodrigues, Dágila Melo
• Orientador(a): Coradi, Paulo Carteri
• Banca de defesa: Teodoro, Paulo Eduardo, Costa, Douglas Romeu da, Zanatta, Fábio Luiz, Lutz, Éverton
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; Centro de Ciências Rurais
• Programa de Pós-Graduação: Programa de Pós-Graduação em Engenharia Agrícola
• Departamento: Engenharia Agrícola
• País: Brasil
• Palavras-chave em Português: Algoritmos preditivos; Redução de perdas e conservação de grãos; Tecnologias de pós-colheita
• Palavras-chave em Inglês: Predictive algorithms; Reduction of losses and conservation of grains; Postharvest technologies
• Área do conhecimento CNPq: CNPQ::CIENCIAS AGRARIAS::ENGENHARIA AGRICOLA
• Link de acesso: http://repositorio.ufsm.br/handle/1/30026

This work aimed to apply predictive models of Machine Learning to monitor the quality and early prediction of losses in the stages of transport, drying and storage of maize grains. This study was divided into three chapters. The first chapter addressed the review of the scientific literature on fundamentals and technological structures of post-harvest and its impacts on the quality of corn grains. The second chapter was a review of computational tools and artificial intelligence algorithms for early decision making to control the quality of corn grains in the post-harvest period. The third chapter was based on monitoring the quality of corn grains in the stages of transport, drying and storage, using Machine Learning models to predict quality and grain losses. The monitoring showed that the water content, the intergranular relative humidity alter the hygroscopic balance moisture of the grains, confident for the increase of the rhythm and consequently losses of dry matter during the transport. During drying, the air temperature caused thermal damage to the grains, increasing the electrical conductivity index. During storage, environmental conditions altered water content, causing a reduction in apparent specific mass, germination and crude protein, crude fiber and fat content of corn grains. In the transport stage, the model of artificial neural networks was the most indicated to predict the electrical conductivity, the apparent specific mass and the germination. The random forest model satisfactorily estimated the loss of dry matter. In the drying step, the models of artificial neural networks and random forest were the most suitable for predicting the variables. In storage, artificial neural networks and random forest were the most suitable to predict water content and germination, however, the decision tree model was the one that best predicted the results of apparent specific mass, electrical conductivity and starch.