Regresar

A Practical Computer Vision-Based Advanced Nutritional Analyzer for Adult Diets

Abstract:

This project tackles the challenge of subjectivity in the nutritional assessment of meals, specifically within the context of the Ecuadorian diet. It introduces a computer vision solution based on advanced convolutional neural networks, utilizing two adapted YOLO V8 models: one for detection and the other for segmentation. The system accurately identifies and analyzes the different food portions in served dishes, categorizing them into three classes: builders foods, regulators, and energy sources, ultimately determining whether the dish is well-balanced. This approach addresses the need for a precise estimate of meal proportions and nutritional composition, which is essential for promoting healthy diets and preventing issues related to nutritional imbalances. The project utilized refined datasets tailored to the Ecuadorian menu, resulting in a system that enhances the accuracy of nutritional assessments based on images of served meals. The findings highlight the effectiveness of the tool and recommend expanding the dataset to include a wider range of typical Ecuadorian foods, thus improving the precision and applicability of the system in various nutritional contexts.