Environment Monitoring

Group Members Students: Hao Xizhe (SUSTech), Sun Quanen (UESTC), Wang Gengnan (BUPT), Hong Shiwei (SCU) Professor: Tan Wee Kek (NUS) Teaching Assistants: Wu Haiyang, Li Xin (NUS) Project Overview The Smart Refrigerator leverages AIoT (Artificial Intelligence of Things) principles to monitor and manage refrigerator contents. With this system, users can track food freshness, receive warnings on expired items, and access recommendations for food usage based on fridge contents. Why We Do This Project: In today’s fast-paced world, food waste has become a growing concern, and many households struggle with managing food storage effectively. Our project aims to address these challenges by: Reducing Food Waste: Proactively tracking expiration dates and providing timely alerts to prevent spoilage. Promoting Sustainability: Optimizing food usage with AI-powered recipe suggestions, encouraging users to consume available ingredients efficiently. Enhancing User Convenience: Automating inventory management to save time and effort, making food management smarter and simpler. With these goals, the Smart Refrigerator delivers a blend of convenience, sustainability, and innovation, ensuring smarter food management for households and businesses alike. 0. Poster Download 📄 Download the Project Poster 1. System Roadmap and Core Functionalities The system is built on a Raspberry Pi platform, integrated with sensors and AI-based predictions for real-time monitoring. Core Sensors and Functions: Environmental Monitoring: Tracks temperature, humidity, and door status using sensors. Food Inventory Management: Provides real-time updates on stored food, including weight and freshness predictions. Proactive Alerts: Sends warnings for expired items or abnormal environmental conditions. AI-Driven Suggestions: Recommends recipes based on available ingredients. Figure 1: System architecture with AI and IoT integration 2. Hardware Components The hardware setup is designed to ensure seamless integration and functionality: Sensors collect environmental and food-related data. Data is transmitted to the Raspberry Pi through ESP32 and Micro:bit modules. The Raspberry Pi acts as the central processing hub, analyzing data and interfacing with the AI and web-based components. Key Sensors and Modules: Environmental Sensors: DHT11: Measures temperature and humidity to ensure optimal storage conditions for food items. HC-SR04: Ultrasonic sensor for detecting door status and measuring distances. Passive Buzzer and RGB LED: Provide visual and auditory alerts for warnings, such as when the refrigerator door remains open. Weight Sensor (HX711): Collects weight data of stored food items with precision, enabling effective inventory management. Image Capture and Processing: Pi Camera: Captures images of food items for further processing and classification using AI models. Communication Modules: ESP32: Gathers sensor data (temperature, humidity, door status) and transmits it to the Raspberry Pi via MQTT protocol for real-time monitoring and warnings. Micro:bit Modules: Enable serial and radio communication for capturing, weighing, and managing inventory. 3. Artificial Intelligence Integration The Smart Refrigerator uses a Pre-trained YOLOv8 model to assess food expiration and predict freshness. The Raspberry Pi’s AI capabilities enable real-time predictions with high accuracy, evaluated through: Confusion Matrix and F1-Confidence Curve to track prediction performance. Figure 2: Confusion Matrix for YOLOv8 Model Figure 3: F1-Confidence Curve Showing Model Performance AI Model Workflow The AI integration follows a step-by-step roadmap to achieve accurate and real-time predictions: Step 0: Pre-trained AI Model on PC A YOLOv8 model is pre-trained on a labeled dataset of various food categories to detect and classify food items accurately. Step 1: Pi Camera Capture The Raspberry Pi utilizes the Pi Camera to capture real-time images of food items stored in the refrigerator. Step 2: AI Predictions on Raspberry Pi YOLOv8 is deployed on the Raspberry Pi for on-device inference. It detects food items, classifies them into predefined categories, and predicts the freshness. Step 3: Weight Information Collector The system correlates weight data collected via the HX711 sensor with the visual classification to ensure precise inventory management. Step 4: Update Data on Web Database The classified food data and weight measurements are stored in a cloud-based web database for further analysis and user access. Valid Predictions The YOLOv8 model effectively detects and classifies food items in real-time. The validation predictions show: Successful identification of common food items like apples, bananas, carrots, and fish with high confidence levels. Bounding boxes accurately localize each item in the images, enabling precise tracking. Figure 4: Validated YOLOv8 predictions for some food items. 4. Web Interface and Database Management The system features a web-based interface that allows users to interact with the Smart Refrigerator’s data. Features of the Web Interface Dynamic Dashboard: Displays real-time data on: Environmental conditions (temperature, humidity). Fridge door status (open/closed). Alerts for expiring or expired food items. Powered by HTML, CSS, and JavaScript for a responsive and interactive experience. Food Inventory Management: Lists stored food items along with weight, timestamp, and expiration status. Allows users to: Add or remove items from the inventory. View item-specific storage recommendations (e.g., temperature, shelf life). AI-Powered Recipe Suggestions: Leverages ChatGPT to provide recipe recommendations based on available ingredients. Recipes are dynamically generated using inventory data and a predefined recipe rule set. User-Friendly Alerts: Sends notifications for: Food nearing expiration. Abnormal fridge conditions (e.g., high temperature or humidity). Uses visual cues (color-coded indicators) for quick recognition of issues. Key Pages Index Page: Provides an overview of fridge contents, environmental conditions, and food expiration warnings. Figure 5: Web Interface - Index Page showing fridge contents and expiration alerts. Table List Page: Displays stored food data, including weight, names, and timestamps. Figure 6: Table list displaying detailed storage data. ChatGPT Recommendation Page: Generates recommendations for recipes or meals based on available fridge ingredients. Figure 7: Recommendations based on current ingredients. References Black Dashboard Template: Link Food Library: Link YOLOv8 Documentation: Link Aheleroff S, et al. IoT-enabled smart appliances under Industry 4.0. Advanced Engineering Informatics, 2020. Basa J J A, et al. Smart inventory management system. arXiv, 2019.

Jul 25, 2023