CALORIE ESTIMATION AND FOOD CLASSIFICATION USING DEEP LEARNING: A CULTURAL AND ARCHITECTURAL PERSPECTIVE

Abstract

Traditional food tracking method remain time consuming and culturally biased for different cuisines. However, the need for an appropriate dietary monitoring system is essential for managing chronic diseases. Currently, different deep learning models integrated automated solutions, but many of those models are mainly specific towards Western-centric datasets which constrains the generalizability across diverse cuisines.

This holistic review examines twelve empirical studies between 2020-2025 leveraging different models like YOLOv8, Mask R-CNN, ViT, and MobileNet which are applied for deep learning for food recognition and calorie estimation. Efficiency was evaluated on basis of the mAP, accuracy level, MAE, and real-time feasibility metrics, with diverse datasets across multi-regions of Indian, Koream, Malaysian, and Chinese cuisines.

Results shows that lightweight models maybe deployment ready (e.g. MobileNetV2, YOLOv8m) but they often lack segmentation accuracy. Vision Transformers exhibits higher accuracy but constrained by high computation cost. A balanced performance is provided by volume-based, API-driven, and regression-based calorie methods.

This study reveals the importance of culturally inclusive datasets, better multi-label classification of food items, and explainable AI to develop robust and globally scalable food recognition systems for health applications by considering multi-cuisines across Asian regions.