🍎 AI Food Spoilage Estimator & Shelf Life Predictor

An AI-powered ecological solution for reducing food waste through computer vision and thermal degradation modeling.

---

📄 Project Abstract

The AI Food Spoilage Estimator is a comprehensive software solution designed to tackle the global issue of food waste. By combining Computer Vision (CV) for food identification with Predictive Analytics (Machine Learning) for shelf life estimation, the system provides users with actionable insights into when their food items will become unsafe for consumption.

The system doesn't just "guess"; it utilizes established food science principles, including the Q10 Temperature Coefficient and USDA FoodKeeper averages, to provide scientifically grounded predictions tailored to specific storage environments (Refrigerated, Room Temperature, or Frozen).

---

⚠️ The Problem Statement

• Global Food Waste: Approximately 1.3 billion tons of food are wasted annually, with a significant portion occurring at the consumer level due to confusion over "Use By" vs. "Best Before" dates.
• Storage Complexity: Most consumers do not understand how small fluctuations in temperature (e.g., a fridge door being left open) or high humidity can drastically accelerate microbial growth and chemical degradation.
• Gap in Technology: While smart fridges exist, they are expensive. There is a lack of accessible, software-first tools that can instantly analyze a food item and provide a remaining shelf life estimate based on real-time environmental data.

---

🎯 Project Objectives

1. Instant Identification: Automate the classification of food items using state-of-the-art Deep Learning models.
2. Environmental Modeling: Account for temperature and humidity variables to provide a dynamic (not static) shelf life prediction.
3. Science-Driven Insights: Implement the Q10 temperature coefficient to model biological decay accurately.
4. Accessibility: Deliver the solution via a lightweight web interface accessible on both desktop and mobile devices.
5. Sustainability: Promote "Zero Hunger" (SDG 2) and "Responsible Consumption" (SDG 12) by preventing avoidable food spoilage.

---

✨ Key Features

• Hybrid AI Classification: Uses both Food-101 and ImageNet models for maximum accuracy.
• 15 Food Categories: Apple, Banana, Bread, Milk, Pasta, Pizza, Burger, Sushi, Meat, Fish, Egg, Vegetable, Fruit, Rice, Cake.
• Dynamic Shelf Life Prediction: RandomForest model predicts remaining days based on temperature, humidity, and storage type.
• Science-Based: Calculations based on USDA FoodKeeper data and Q10 temperature coefficients.
• Interactive Dashboard: Modern Glassmorphism UI for easy data input and visualization.
• Docker Ready: One-command deployment with docker-compose up.

---

🛠️ Technology Stack

• Backend: Python, Flask (Production API), Streamlit (Dashboard)
• AI/Vision: Hugging Face Inference API (Food-101 & EfficientNet-B0)
• ML/Analytics: Scikit-Learn (RandomForestRegressor), Pandas, NumPy
• Frontend: HTML5, CSS3 (Glassmorphism), JavaScript
• Deployment: Docker, Docker Compose, Python-dotenv

---

🔬 Scientific Methodology & "How It Works"

1. Image Classification (The "What")

The system uses a weighted ensemble/hybrid approach:

• Primary: Queries Kaludi/food-category-classification-v2.0 (Food-101 model) - Best for prepared dishes.
• Fallback: If confidence is low, queries google/efficientnet-b0 (ImageNet) - Best for raw ingredients.
• Result: Returns the highest-confidence match from either model.

2. The Q10 Spoilage Model (The "How Long")

Food degradation is a temperature-dependent chemical and biological process. The system implements the Q10 Temperature Coefficient:

$$Q_{10} = \left(\frac{R_2}{R_1}\right)^{\frac{10}{T_2 - T_1}}$$

• Logic: For many food items, the rate of spoilage increases by ~2.5x (Q10 factor) for every 10°C increase in temperature.
• Reference: We use 4°C (Standard Fridge Temp) as our baseline.

3. Humidity Correction

Microbial growth is accelerated by moisture. Our model applies a penalty/bonus factor where every 10% increase in humidity above 55% reduces the remaining shelf life by approximately 10%.

---

📊 Supported Foods & Baselines

Category	Typical Fridge Life	Example Items	Scientific Logic
Apple	21 days	Whole apples, apple pie	High fiber, low oxidation rate
Banana	5 days	Fresh bananas	High Ethylene production
Bread	7 days	Loaves, toast, bagels	Moisture loss & retrogradation
Milk	7 days	Dairy products	Bacterial lactic acid production
Pasta	4 days	Spaghetti, lasagna, ramen	High water activity
Pizza	4 days	Leftover pizza	Fat oxidation & microbial risk
Burger	3 days	Hamburgers, hot dogs	Processed meat degradation
Sushi	1 day	Sushi, sashimi	Raw fish; high bacterial risk
Meat	4 days	Steak, pork, chicken	Protein degradation
Fish	2 days	Salmon, tuna, seafood	Rapid enzymatic breakdown
Egg	28 days	Eggs, omelettes	Natural protective shell coating
Vegetable	7 days	Salads, fresh veggies	Respiration & transpiration
Fruit	5 days	Berries, cut fruit	Cell wall breakdown
Rice	5 days	Cooked rice, risotto	Bacillus cereus risk
Cake	5 days	Cakes, desserts	High sugar stability

---

🚀 Quick Start & Installation

Option 1: Docker (Recommended)

1. Clone the repo and create .env:

   git clone https://github.com/yourusername/AI-Food-Spoilage-Estimator.git
   cd AI-Food-Spoilage-Estimator
   cp .env.example .env
   # Edit .env and add your Hugging Face token

2. Run with Docker:

   docker-compose up --build

3. Open http://localhost:5000

Option 2: Local Python

1. Install dependencies:

   pip install -r requirements.txt

2. Set environment variable:

   # Windows PowerShell
   $env:HUGGINGFACE_API_TOKEN = "hf_your_token_here"
   
   # Linux/Mac
   export HUGGINGFACE_API_TOKEN="hf_your_token_here"

3. Run the apps:

   • Flask (Main UI): python app_flask.py
   • Streamlit (Metrics): streamlit run app.py

---

⚙️ Environment Variables

Variable	Required	Description
HUGGINGFACE_API_TOKEN	Yes	Get free at huggingface.co/settings/tokens

---

📂 Project Structure

food-spoilage-estimator/
├── app_flask.py           # Production-ready Flask server
├── app.py                 # Interactive Streamlit dashboard
├── config.py              # Centralized configuration (Constants, Mappings)
├── Dockerfile             # Container definition
├── docker-compose.yml     # Multi-container orchestration
├── requirements.txt       # Python dependencies
├── .env.example           # Environment template
├── models/
│   ├── classifier.py      # Hybrid AI Vision logic (Hugging Face)
│   └── shelf_life_predictor.py  # RandomForest ML Model
├── utils/
│   └── data_loader.py     # Research-based dataset generator
├── templates/
│   └── index.html         # Modern Glassmorphism UI
└── data/                  # Generated training data (gitignored)

---

📈 Future Roadmap

• IoT Integration: Connect with real-time temperature/humidity sensors.
• Mobile App: Develop a Flutter cross-platform mobile application.
• Barcode Scanning: Fetch brand-specific data via OpenFoodFacts.
• Multi-Item Detection: Enable YOLOv8 for detecting multiple items in one fridge scan.

---

🤝 References & Context

Created for educational and research purposes in AI-driven sustainability.

• Food-101: Bossard et al., "Food-101 – Mining Discriminative Components", ECCV 2014
• Q10 Coefficient: Labuza, T.P. (1982). "Shelf Life Dating of Foods"
• USDA: FoodKeeper Application Data (2024)

📜 License

MIT License - Free for educational and commercial use.

---

Tip

Pro Tip for PPT: When presenting, highlight the Q10 Coefficient — it's the "Secret Sauce" that differentiates this project from simple image labelers by adding a layer of genuine food science.