A SYSTEMATIC REVIEW OF AI POWERED FOOD RECOGNITION SYSTEM

Abstract: More people face health issues tied to poor eating habits - conditions like obesity and heart problems keep rising. Because of this shift, better tools for tracking what we eat have become essential. Old ways of recording meals, such as handwritten logs or trying to remember everything eaten in a day, tend to be unreliable. Mistakes happen. Memories fade. Sticking with those methods over time? Rare. A new approach enters here - not magic, just smart engineering. This method uses artificial intelligence to turn meal photos into useful nutrition facts without guesswork. At its core sits a neural network model called ResNet-18, good at telling one dish apart from another. It learned from more than 100,000 food images labeled across 101 categories. Training happened using Food-101, a large-scale collection built for exactly this purpose. Speed matters; PyTorch handles number crunching while OpenCV prepares raw pictures for analysis. Cropping, color adjustments, noise cleanup - all done before classification kicks in. Once identified, each food links to stored nutrient values: calories, protein levels, fat amounts, carbs included. Results appear instantly through a live website powered by Streamlit. Snap a photo, get details moments later. No waiting. No spreadsheets. Starting off different, recent studies - especially those on visionlanguage systems and image-focused workout tools - show our method handles speed and accuracy without favoring one too much. Early tests suggest this automatic system cuts down the hassle of logging by hand, providing something sturdy and flexible enough to grow with personal wellness needs while supporting clearer food choices.
Though not perfect, it fits well where quick results meet reliable detection.

Keywords: Food Recognition, Deep Learning, ResNet-18, Nutritional Analysis, Computer Vision, PyTorch, Calorie Estimation, Food-101 Dataset