Nutritional Analysi from Food Images

Abstract— Accurate dietary assessment is essential for managing nutrition-related diseases. However, traditional methods, such as food frequency questionnaires and recalls, are often time-consuming and biased. This project, "Nutritional Analysis from Food Images," offers an AI-driven solution to automate food identification and nutrition estimation from photographs. We aimed to address the limitations of conventional methods by developing a system that uses a pre-trained DenseNet-161 architecture along with data augmentation techniques and transfer learning to classify food items from the FOOD-101 dataset. Our model achieved a Top-1 accuracy of 93.27% and a Top-5 accuracy of 99.02%, outperforming several existing methods. The system takes a food photograph as input and provides the nutritional analysis along with a predicted list of ingredients. This makes it suitable for various and complex food types. Our findings show that AI-based visual analysis can greatly improve the efficiency, reliability, and accessibility of dietary tracking, offering a strong alternative to traditional nutrition assessment methods. Keywords— Nutritional Analysis, Food Image Classification, DenseNet-161, Dietary Assessment, Food Recognition, Portion Size Estimation, Ingredients Prediction, AI in Nutrition, Image Processing, Nutrition Tracking, Convolutional Neural Networks (CNN).