Unhealthy dietary habits contribute to a rising global burden of non-communicable diseases, affecting over 2 billion individuals worldwide. Accurate dietary monitoring and personalized nutritional guidance are critical public-health interventions, yet traditional self-reported food journaling remains subjective, labour-intensive, and prone to substantial under-reporting bias. This paper presents NutriSnap-X, a novel AI-powered web application for automated food recognition and holistic nutrition analysis from user-captured food images. The system integrates a fine-tuned MobileNetV2 convolutional neural network achieving 92.6% top-1 classification accuracy across 101 food categories on the Food-101 benchmark dataset. Detected food items are linked to a curated USDA-sourced nutritional database to estimate macronutrient composition — calories, protein, carbohydrates, and fat — calibrated to per-image portion estimates. A multi-factor Health Score Algorithm then synthesizes this nutritional data with user profile parameters (age, weight, height, dietary goal, activity level) to produce a personalized health score and evidence-based dietary recommendations. Additional features include barcode-based packaged food lookup, a weekly analytics dashboard, and automated PDF nutritional report generation. The Flask-based web interface delivers end-to-end processing in under 530 milliseconds, enabling practical real-time deployment. Comparative evaluation demonstrates a weighted F1-score of 91.5%, outperforming existing open-source food AI systems while uniquely providing a complete personalized nutrition pipeline.
Ganesh et al. (Sun,) studied this question.