Fruit Weight Predicting by Using Hybrid Learning

Abstract

Modern lifestyle diseases are closely linked to diet, including weight control, healthy eating habits, and physical activity. This is particularly critical for individuals with poor blood sugar metabolism, who need to manage blood sugar-related conditions. A fast and intelligent algorithmic tool for quickly assessing carbohydrate content in food is crucial for effective management. Nutritionists often spend a lot of time on dietary assessments. With the widespread use of smartphones today, previous research has shown that assessing nutrients using a 45-degree plane photography method can introduce significant errors. Therefore, having an accurate tool for analyzing food carbohydrate content can help people understand their daily diet, explore nutritional patterns, and maintain a healthy diet. Food weight estimation is essential for carbohydrate analysis. Our research focuses on inferring food weight in fruits. We use a lightweight convolutional neural network model integrated with fast machine learning techniques, enabling real-time food weight prediction when taking smartphone photos. Through experiments, our model achieved an impressive 99.81% accuracy when tested with images of apples, bananas, and oranges, captured at different sizes, distances, and angles. However, dietary patterns are diverse and complex. Therefore, inferring various nutrients in diverse diets will require further research and development. This study lays the foundation for future exploration in this area.