Enhancing dietary management and nutritional analysis through deep learning: a multi-model approach for food classification and volume estimation

Nutrition and diet play pivotal roles in fostering a healthy lifestyle. Effective dietary management requires precise identification and analysis of consumed foods. However, manual tracking is susceptible to inefficiencies and human error. This research work proposes a novel scheme to mitigate these issues by developing a deep learning-based solution. This scheme employs Inception V3 for food classification with the benchmark datasets, Mask R-CNN for 2D area estimation, and MiDaS for single-view depth estimation. The proposed model secures an average classification accuracy of 98% and provides accurate nutrition information of food and volume. In addition, this research work addresses gaps in the current research by evaluating a range of deep learning models on a benchmark dataset of food items. It performs well and produces the good results on well-known datasets such as Food101 with 98% of accuracy, UEC-food256 with 97.25% of accuracy and Indian Food Images Dataset with 97.40% of accuracy. By integrating improved classification accuracy with volume estimation technique, the proposed scheme sets the foundation for precise and comprehensive dietary tracking, contributing to the fight against malnutrition and diet-related health issues globally.