| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 3.39 MB | Adobe PDF |
Advisor(s)
Abstract(s)
As doenças relacionadas com a alimentação são um dos principais problemas de saúde pública, sendo a DM2 uma das principais doenças. Esta é uma doença crónica caracterizada por apresentar uma resistência à insulina, levando a um aumento dos níveis de glicose sanguínea, e que pode trazer problemas graves para os pacientes. Embora não exista uma cura, um estilo de vida saudável pode ser suficiente para a controlar. Desta forma o controlo da alimentação pode trazer francas melhorias à qualidade de vida destas pessoas, sendo que o desenvolvimento das tecnologias veio oferecer novos meios para tornar este processo mais rápido, simples e cómodo face aos métodos tradicionais. Assim, os objetivos do presente trabalho passam pelo desenvolvimento de um sistema, simples e intuitivo, que seja capaz de identificar uma refeição a partir de uma foto da mesma, e determinar os valores nutricionais dessa mesma refeição. Para tal, recorreu-se a métodos de machine learning para tentar desenvolver um sistema com estas características. O modelo proposto é baseado em RNC combinadas com métodos de transfer learning, sendo que para o desenvolvimento do mesmo recorreu-se a 5 classes da base de dados ETHZ Food-101. O sistema desenvolvido comtempla ainda um modelo para identificar e dar feedback da informação nutricional relativa às refeições da pessoa. Nos testes realizados ao modelo obtiveram-se bons resultados a nível de performance, de tal forma que pudesse considerar que o modelo desenvolvido é capaz de realizar previsões com elevada exatidão, mesmo com uma resolução reduzida, para as 5 classes usadas. Embora seja difícil comparar os resultados obtidos com outras abordagens presentes na literatura foi possível verificar que as RNC são o melhor método para este tipo de sistema.
Diet-related diseases are a major public health problem, with DM2 being one of the main ones. This is a chronic disease characterised by showing insulin resistance, leading to an increase in blood glucose levels, and which can bring serious problems for patients. Although there is no cure, a healthy lifestyle may be enough to control it. Thus, the development of technology has offered new ways to make this process faster, simpler, and more convenient than traditional methods. So, the objectives of this work are to develop a simple and intuitive system that can identify a meal from a photo of it and determine the nutritional values of that meal. For this purpose, machine learning methods were used to try to develop a system with these characteristics. The proposed model is based on the RNC combined with transfer learning methods, and for its development 5 classes from the ETHZ Food-101 database were used. The developed system also includes a model to identify and provide feedback on the nutritional information related to a person's meals. In the tests performed on the model, good performance results were obtained, such that it could be considered that the model developed can make highly accurate predictions, even with a reduced resolution, for the 5 classes used. Although it is difficult to compare the results obtained with other approaches present in the literature, it was possible to verify that the RNC are the best method for this type of system.
Diet-related diseases are a major public health problem, with DM2 being one of the main ones. This is a chronic disease characterised by showing insulin resistance, leading to an increase in blood glucose levels, and which can bring serious problems for patients. Although there is no cure, a healthy lifestyle may be enough to control it. Thus, the development of technology has offered new ways to make this process faster, simpler, and more convenient than traditional methods. So, the objectives of this work are to develop a simple and intuitive system that can identify a meal from a photo of it and determine the nutritional values of that meal. For this purpose, machine learning methods were used to try to develop a system with these characteristics. The proposed model is based on the RNC combined with transfer learning methods, and for its development 5 classes from the ETHZ Food-101 database were used. The developed system also includes a model to identify and provide feedback on the nutritional information related to a person's meals. In the tests performed on the model, good performance results were obtained, such that it could be considered that the model developed can make highly accurate predictions, even with a reduced resolution, for the 5 classes used. Although it is difficult to compare the results obtained with other approaches present in the literature, it was possible to verify that the RNC are the best method for this type of system.
Description
Keywords
Diabetes Tipo 2 Machine Learning Redes Neuronais Convolucionais Reconhecimento Alimentar Type 2 Diabetes Machine Learning Convolutional Neural Networks Food Recognition