AI-Based Estimation from Images of Food Portion Size and Calories for Healthcare Systems

초록

In the realm of nutrition science, it is well-recognized that individuals' dietary needs vary based on factors such as age, gender, and health status. This divergence in nutritional requirements is particularly critical for vulnerable groups, including newborns, the elderly, and individuals with diabetes, as their dietary choices can have profound implications for their health. Moreover, the dearth of Uzbek recipes in mainstream culinary literature, which predominantly focuses on Western cuisine, exacerbates the issue. To address these challenges, this study undertakes the ambitious task of constructing a comprehensive AI system, comprising both backend and frontend components, tailored to the nuances of Uzbek cuisine. The primary objectives encompass recipe classification, ingredient identification and localization, and the estimation of nutritional content and calorie counts for each dish. Convolutional Neural Networks (CNNs) are employed as the cornerstone of this computational solution, proficiently handling image-based tasks, including the recognition of diverse food items and portion size determination within Uzbek recipes. Food classification is executed using MobileNet, while the You-Only-Look-Once (YOLO) network plays a pivotal role in the dual functions of ingredient classification and localization within dishes. Upon rigorous training, testing, and system deployment, users can effortlessly capture images of food items through the smartphone application, facilitating the estimation of nutritional data and calorie counts. Ultimately, this vital information is presented to users via the smartphone interface, bridging the accessibility gap and enhancing comprehension of nutritional aspects within Uzbek cuisine.

키워드