Estimation Of Nutrition Components From Food

Abstract

This paper proposes a novel deep learning-based system for estimating nutritional components and ingredients from food using text input instead of images. Unlike traditional mobile applications that rely on food image analysis, our approach extracts semantically relevant information from user-provided textual descriptions to predict ingredients and nutritional values. The model is trained on a large, food-related corpus sourced from the internet. Experiments were conducted using an extended Food-101 dataset, consisting of 100 food categories with 1,000 samples per category. The system achieved a top-1 accuracy of 85%. It also demonstrates strong performance on subcontinental dishes. Designed as a mobile application, this system offers a non-intrusive, efficient tool for dietary analysis. It holds significant potential for applications in the healthcare industry, particularly in managing and preventing diet-related health issues.