Calorie Sense: Food & Calorie detection

Accurate food detection and calorie estimation are of growing interest, especially with the growing trend towards health and nutrition. To tackle this, deep learning models can be used to recognize foods and estimate their caloric content using portion size, ingredients and nutritional information. The data set is pre-processed using image augmentation, normalization and feature extraction. A YOLOv11 object detection model is used to detect multiple food objects in an image, while a CNN based classifier refines the classification and calories estimation using the extracted features. A combination of CNN and YOLO model for better accuracy, VGG2 and YOLOv11 models for classification and object detection are compared in stratified k-fold cross-validation. These algorithms predict food items and estimate calories by analyzing features like food type, portion size, ingredient composition (if available), and serving size. The models are evaluated using the performance measure: Accuracy, Precision, Recall and F1-score are intercompared in the study. Experimental results showed that CNN with 91% accuracy and YOLOv11 with 95% in food recognition and the hybrid CNN-YOLO with 97% accuracies. The hybrid model increases the efficiency by incorporating complex features of food and improves approximation of calorie estimation. Our approach extends to automation of Dietary Assessment Systems providing real-time food recognition and calorie counting for people who are health conscious.