MEASURING NUTRITION AND CALORIES FROM FOOD IMAGE CALCULATION IN DAY TO DAY LIFE

Abstract

Nutrition estimation from food images helps in understanding calorie and nutrient intake without using laboratory methods. Traditional chemical techniques are slow, costly, and destroy the food sample, whereas image-based methods are fast and non-destructive. This project uses RGB and Depth images along with a deep-learning model to accurately estimate calories, protein, fat, and carbohydrates. The system adopts the Swin Transformer for feature extraction and an Adaptive Feature Fusion and Enhancement (ADFE) module to combine RGB–D information effectively. Detailed-information enhancement and semantic information enhancement further improve feature quality. Experiments on the Nutrition5k dataset show that the model provides reliable nutrition prediction, making it useful for daily food analysis and health monitoring