Published Paper Details:

Efficient food identification systems face challenges with the wide range of food categories and the high computational demands associated with extremely complex dishes. There is a need for a real-time solution that not only detects the food but also identifies its possible allergens and nutritional content, enabling efficient identification before distribution to those in need. Such a solution should ensure safe, accurate, and effective food identification in advance. This helps not only to reduce food waste but also to combat hunger. The proposed system's performance is evaluated using a confusion matrix, which helps assess model accuracy by displaying correctly and incorrectly classified data points and highlighting misclassification patterns. In our work, the ResNet50 model shows 85-90% accuracy with a loss of 0.3 on simple food images, but accuracy decreases with more complex food items. In contrast, InceptionV3, benefiting from multi-scale processing, achieves 88-92% accuracy with a loss of 0.25, demonstrating higher precision and recall with visually complex dishes. When combining ResNet50 and InceptionV3 through a stacking ensemble, performance significantly increases, reaching an overall accuracy of 93%, showing a substantial enhancement in classification accuracy across diverse food images.