Smart Diet Assistant: IoT Food Analysis

The Challenge

Manual calorie tracking and nutritional logging are tedious, error-prone, and a major barrier for individuals trying to maintain a healthy diet. The objective was to create an automated, user-friendly system that could analyze a meal from a single image, providing instant nutritional feedback.

My Solution

I engineered an end-to-end IoT-based dietary assessment system that combines hardware, computer vision, and deep learning. The system, powered by a Raspberry Pi, automates the entire process from image capture to nutritional breakdown, making dietary tracking seamless and intelligent.

Key Features & Technical Implementation:

• End-to-End IoT Pipeline: Designed a complete workflow where a user captures an image of their meal. The image is processed on-device by a Raspberry Pi, which communicates results via the lightweight MQTT protocol, ensuring efficient data exchange in an IoT environment.
• Precise Food Segmentation: Implemented Mask R-CNN, a state-of-the-art deep learning model, to perform instance segmentation. This allowed the system to not just identify different food items in a single image but also to isolate their exact shape and size, which is crucial for accurate portion estimation.
• Optimized On-Device AI: The core of the system is a CNN-based food recognition model. To enable deployment on the resource-constrained Raspberry Pi, I converted and optimized the model using TensorFlow Lite, significantly reducing its size and computational requirements without a major loss in accuracy.
• Nutritional Estimation: Beyond just classification, I developed regression models that take the segmented food items and their estimated portion sizes as input to accurately calculate total calories and macronutrient content (proteins, fats, carbohydrates).

This project showcases my ability to integrate hardware, optimized deep learning models, and cloud communication protocols to build a practical and innovative real-world application.