Automated Meal Recognition Platform

Automated Meal Recognition Platform

Dietary self-reporting through manual food diaries suffers from recall bias and portion estimation error. This deep learning application enables automated nutritional logging by analyzing photographs of meals captured via smartphone cameras. The pipeline first segments the image into individual food items, then classifies each region into a hierarchical taxonomy of dishes and ingredients. A secondary density estimation network predicts serving volume from plate scale references and known geometric priors. The classified items and estimated masses are cross-referenced with a comprehensive nutritional database to compute macronutrient and micronutrient totals. The system is trained on a large-scale crowdsourced food image dataset and validated against doubly labeled water metabolic studies for caloric accuracy.

Components

Python 3.8+
TensorFlow/PyTorch
OpenCV
Flask/Heroku
Nutritionix API
SQLite Database