Visão computacional e outras técnicas de aprendizado de máquina aplicada ao processamento de frutas na produção de sucos

• Ano de defesa: 2022
• Autor(a) principal: Arroyo, Victória Hebling
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Engenharia Química
• Palavras-chave em Português: Visão computacional; Aprendizado de máquina; Redes neurais; Computer Vision; Machine Learning; Neural Networks; CNPQ::ENGENHARIAS; Engenharia química; Análise de imagem - Processamento eletrônico de dados; Suco de laranja - Indústria - Brasil; Frutas cítricas - Tecnologia pós-colheita
• Link de acesso: https://repositorio.ufu.br/handle/123456789/34562; http://doi.org/10.14393/ufu.di.2022.71

In the industry, evaluations such as texture, color and size of foods are performed manually, however, this form of inspection is subjective and tedious, thus, in order to generate consistency in the results, and other techniques have been developed, such as computer vision and machine learning. Thus, using these techniques, the objective of this dissertation is to develop an image acquisition system to evaluate Brix, juice volume and pH of oranges. Therefore, the methodology of this work consisted in photographing Pera Rio oranges, collecting information about these oranges and then configuring feedforward neural networks with backpropagation and convolutional networks. The results show that it is possible to analyze the volume of juice from the length of the oranges, width of the oranges and orange selection device. Furthermore, it was observed that the convolutional networks used to classify oranges for Brix and pH were not satisfactory, while the use of feedforward type neural network with input variables obtained from gray scale co-occurrence matrix texture analysis extracted from green color were satisfactory to classify pH and Brix.