🛡️ SafeBite AI – Street Food Hygiene Rating System

An AI-powered platform that analyzes street food stall images/videos to generate automated hygiene and food safety ratings.

Suitable for: Final Year Engineering Projects • Hackathons • IEEE Research Papers • Startup MVPs

📋 Table of Contents

Features
Tech Stack
Project Structure
Quick Start
Backend Setup
Frontend Setup
AI Model Integration
Model Training Pipeline
API Reference
Deployment
Architecture
Contributing

✨ Features

🔬 AI Hygiene Analysis

Object Detection (YOLOv8): Detects waste, gloves, masks, food items, cooking utensils, garbage bins, oil containers, smoke
Cleanliness Classification (CNN/ResNet50): Classifies stalls as Clean / Moderate / Unsafe
Hygiene Scoring Engine: 5 sub-scores + overall hygiene score (out of 10)
AI Recommendations: Natural language hygiene improvement suggestions

📤 Upload & Processing

Drag-and-drop image/video upload
Live webcam capture and analysis
Real-time CCTV analysis mode (WebSocket)
Supports: JPG, PNG, WebP, MP4, AVI

📊 Analytics Dashboard

Annotated image with bounding boxes
Hygiene metrics (Radar chart, Bar chart)
Historical trend analysis
Safety status: Safe / Moderate Risk / Unsafe

📄 Report Generation

Downloadable PDF inspection reports
Detected issues with timestamps
Annotated images embedded in PDF
AI-generated recommendations

🖥️ Admin Panel

All analyzed stalls with search/filter
Aggregate analytics
JSON/CSV export
Risk distribution charts

🚀 Advanced Features

Live webcam detection (WebSocket)
GPS/location tagging support
Vendor risk heatmap (Leaflet.js)
Multilingual-ready architecture
Municipality inspection mode

🛠️ Tech Stack

Layer	Technology	Purpose
Frontend	React 18 + Vite	SPA framework
Styling	TailwindCSS 3.4	UI styling
Animations	Framer Motion	Page/component animations
Charts	Recharts	Data visualization
State	Zustand	Global state management
Backend	FastAPI (Python 3.10+)	REST API server
Database	MongoDB + Motor	Async document storage
Computer Vision	OpenCV	Image preprocessing & annotation
Object Detection	YOLOv8 (Ultralytics)	Object detection
Classification	PyTorch + ResNet50	Cleanliness classification
PDF	ReportLab	Report generation
Real-time	WebSocket	Live camera analysis

📁 Project Structure

safebite-ai/
├── 📂 backend/                    # FastAPI backend
│   ├── main.py                    # Application entry point
│   ├── config.py                  # Settings & env config
│   ├── requirements.txt           # Python dependencies
│   ├── .env                       # Active environment (git-ignored)
│   ├── .env.example               # Environment template
│   ├── 📂 api/
│   │   └── 📂 routes/
│   │       ├── upload.py          # File upload endpoints
│   │       ├── analysis.py        # Analysis report endpoints
│   │       ├── reports.py         # PDF report endpoints
│   │       ├── admin.py           # Admin panel endpoints
│   │       └── realtime.py        # WebSocket live analysis
│   ├── 📂 db/
│   │   ├── database.py            # MongoDB + in-memory fallback
│   │   └── models.py              # Pydantic data models
│   ├── 📂 services/
│   │   ├── ai_pipeline.py         # Main AI orchestrator
│   │   ├── yolo_detector.py       # YOLOv8 object detection
│   │   ├── cleanliness_classifier.py  # CNN classification
│   │   ├── scoring_engine.py      # Hygiene scoring
│   │   ├── recommendation_engine.py   # AI recommendations
│   │   └── pdf_generator.py       # ReportLab PDF generator
│   └── 📂 utils/
│       └── image_processor.py     # OpenCV utilities
│
├── 📂 frontend/                   # React + Vite frontend
│   ├── index.html                 # HTML entry point
│   ├── package.json               # Node dependencies
│   ├── vite.config.js             # Vite configuration
│   ├── tailwind.config.js         # TailwindCSS config
│   ├── postcss.config.js          # PostCSS config
│   └── 📂 src/
│       ├── main.jsx               # React root
│       ├── App.jsx                # Router + layout
│       ├── index.css              # Global styles + design system
│       ├── 📂 api/
│       │   └── client.js          # Axios API client
│       ├── 📂 store/
│       │   └── useStore.js        # Zustand state store
│       ├── 📂 components/
│       │   ├── Navbar.jsx
│       │   ├── UploadZone.jsx
│       │   ├── AnalysisProgress.jsx
│       │   ├── HygieneScoreCard.jsx
│       │   ├── DetectionOverlay.jsx
│       │   ├── MetricsGrid.jsx
│       │   ├── RecommendationCard.jsx
│       │   ├── Charts.jsx
│       │   ├── ReportDownload.jsx
│       │   ├── AdminTable.jsx
│       │   ├── LiveCamera.jsx
│       │   └── ThemeToggle.jsx
│       └── 📂 pages/
│           ├── Home.jsx           # Landing page
│           ├── Upload.jsx         # Upload interface
│           ├── Analysis.jsx       # Processing status
│           ├── Dashboard.jsx      # Results dashboard
│           ├── Admin.jsx          # Admin panel
│           └── Reports.jsx        # Report history
│
├── 📂 ai_models/                  # AI model training
│   ├── train_yolo.py              # YOLOv8 training script
│   ├── train_classifier.py        # CNN training script
│   ├── dataset_structure.md       # Dataset documentation
│   └── 📂 data/                   # Training datasets (add your own)
│       ├── yolo_dataset/
│       └── classifier_dataset/
│
├── 📂 sample_data/                # Sample dataset structure
│   ├── README.md
│   ├── images/                    # Add sample images here
│   └── labels/                    # Sample YOLO labels
│
└── README.md                      # This file

⚡ Quick Start

Prerequisites

Python 3.10 or higher
Node.js 18 or higher
MongoDB (optional — app works with in-memory fallback)
Git

1-Minute Setup

# Clone / navigate to the project
cd c:\Users\Varun\Documents\RVCE\food

# Terminal 1: Start Backend
cd backend
python -m venv venv
.\venv\Scripts\Activate.ps1
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

# Terminal 2: Start Frontend
cd frontend
npm install
npm run dev

Frontend: http://localhost:5173
Backend API: http://localhost:8000
API Docs (Swagger): http://localhost:8000/api/docs
API Docs (ReDoc): http://localhost:8000/api/redoc

🐍 Backend Setup

Environment Configuration

Copy .env.example to .env and configure:

# MongoDB (optional - falls back to in-memory storage)
MONGO_URI=mongodb://localhost:27017/safebite

# Security
SECRET_KEY=your-super-secret-key-change-in-production

# CORS
ALLOWED_ORIGINS=http://localhost:5173

# File storage
UPLOAD_DIR=uploads

# AI Model paths (optional - uses mock AI if not found)
MODEL_PATH=ai_models/weights/yolov8n.pt
CLASSIFIER_PATH=ai_models/weights/safebite_classifier.pt

Install Dependencies

cd backend
python -m venv venv
.\venv\Scripts\Activate.ps1
pip install -r requirements.txt

Note on torch/torchvision: If you have a CUDA GPU, install the CUDA version:
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118

Run the Server

# Development (auto-reload)
uvicorn main:app --reload --port 8000

# Production
uvicorn main:app --host 0.0.0.0 --port 8000 --workers 4

MongoDB Setup (Optional)

The app works without MongoDB using an in-memory store. To enable MongoDB:

Install MongoDB: https://www.mongodb.com/try/download/community
Start MongoDB service
Set MONGO_URI=mongodb://localhost:27017/safebite in .env

⚛️ Frontend Setup

Install Dependencies

cd frontend
npm install

Development Server

npm run dev
# Opens at http://localhost:5173
# API calls are proxied to http://localhost:8000

Production Build

npm run build
# Output in frontend/dist/

🤖 AI Model Integration

Current Status: Smart Mock Mode

By default, the application runs in Smart Mock Mode — the AI pipeline uses OpenCV-based analysis to generate realistic hygiene detections and scores without requiring trained model weights. This makes the app fully demo-ready out of the box.

Enabling Real YOLOv8 Model

Download pretrained YOLOv8 weights:

# Auto-downloads on first use
# Or manually download from: https://github.com/ultralytics/assets/releases

For custom-trained weights, set in backend/.env:

MODEL_PATH=ai_models/weights/your_trained_model.pt

Restart the backend server.

Enabling Real CNN Classifier

Train the classifier (see Model Training Pipeline below)
Set in backend/.env:

CLASSIFIER_PATH=ai_models/weights/safebite_classifier.pt

🏋️ Model Training Pipeline

1. Prepare Dataset

Follow the guide in ai_models/dataset_structure.md.

For YOLOv8:

ai_models/data/yolo_dataset/
├── images/train/   # ~2000 images
├── images/val/     # ~500 images
└── labels/train/   # YOLO format .txt files
    labels/val/

For CNN Classifier:

ai_models/data/classifier_dataset/
├── train/clean/       # ~1000 images
├── train/moderate/    # ~1000 images
├── train/unsafe/      # ~1000 images
└── val/clean/, moderate/, unsafe/

2. Train YOLOv8 Object Detector

cd ai_models
python train_yolo.py --epochs 50 --batch 16 --img 640

Options:

Flag	Default	Description
`--epochs`	50	Training epochs
`--batch`	16	Batch size
`--img`	640	Image size
`--model`	yolov8n.pt	Base model (n/s/m/l/x)
`--device`	auto	cuda/cpu

Best weights saved to: ai_models/weights/safebite_yolo_best.pt

3. Train CNN Cleanliness Classifier

cd ai_models
python train_classifier.py --head-epochs 5 --fine-tune-epochs 25 --batch 32 --model resnet50

Options:

Flag	Default	Description
`--head-epochs`	5	Warm-up epochs for classifier head
`--fine-tune-epochs`	25	Fine-tuning epochs after unfreezing the backbone
`--batch`	32	Batch size
`--lr`	0.0001	Learning rate
`--fine-tune-lr`	0.00001	Learning rate for full-model fine-tuning
`--patience`	5	Early stopping patience on validation macro-F1
`--min-delta`	0.001	Minimum F1 improvement needed to reset patience
`--label-smoothing`	0.05	Label smoothing for the loss
`--model`	resnet50	Architecture (resnet50/efficientnet_b0/mobilenet_v3)
`--freeze`	False	Freeze backbone layers
`--no-augment`	False	Disable data augmentation

Trained model saved to: ai_models/weights/safebite_classifier.pt

4. Update Backend Config

MODEL_PATH=ai_models/weights/safebite_yolo_best.pt
CLASSIFIER_PATH=ai_models/weights/safebite_classifier.pt

📡 API Reference

Base URL: `http://localhost:8000/api`

Upload

Method	Endpoint	Description
`POST`	`/upload`	Upload image/video for analysis
`GET`	`/upload/{id}/status`	Check processing status

Analysis

Method	Endpoint	Description
`GET`	`/analysis/{id}`	Get full analysis report
`GET`	`/analysis/history`	Paginated report history
`DELETE`	`/analysis/{id}`	Delete report

Reports

Method	Endpoint	Description
`GET`	`/reports/{id}/pdf`	Download PDF report
`GET`	`/reports/list`	List all reports

Admin

Method	Endpoint	Description
`GET`	`/admin/analytics`	Aggregate analytics
`GET`	`/admin/stalls`	All stall records
`GET`	`/admin/export`	Export all data as JSON

WebSocket

Protocol	Endpoint	Description
`WS`	`/ws/live`	Real-time camera frame analysis

Example: Upload and Get Report

import requests

# Upload image
with open("stall.jpg", "rb") as f:
    response = requests.post(
        "http://localhost:8000/api/upload",
        files={"file": f}
    )
    report_id = response.json()["report_id"]

# Poll status
import time
while True:
    status = requests.get(f"http://localhost:8000/api/upload/{report_id}/status").json()
    if status["status"] == "completed":
        break
    time.sleep(1)

# Get full report
report = requests.get(f"http://localhost:8000/api/analysis/{report_id}").json()
print(f"Overall Score: {report['scores']['overall']}/10")
print(f"Safety Status: {report['safety_status']}")

🚢 Deployment

Docker (Recommended)

# backend/Dockerfile
FROM python:3.11-slim
WORKDIR /app
COPY requirements.txt .
RUN pip install -r requirements.txt
COPY . .
CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000"]

# docker-compose.yml
version: "3.9"
services:
  backend:
    build: ./backend
    ports: ["8000:8000"]
    environment:
      - MONGO_URI=mongodb://mongo:27017/safebite
    depends_on: [mongo]

  frontend:
    build: ./frontend
    ports: ["80:80"]

  mongo:
    image: mongo:7
    volumes: ["mongo_data:/data/db"]

volumes:
  mongo_data:

Cloud Deployment

Platform	Service	Notes
Railway	Backend + MongoDB	1-click deploy
Render	Backend API	Free tier available
Vercel	Frontend	Optimal for React
Netlify	Frontend	Free tier
AWS EC2	Full stack	For production
Google Cloud Run	Containerized	Serverless

Environment Variables (Production)

MONGO_URI=mongodb+srv://user:pass@cluster.mongodb.net/safebite
SECRET_KEY=<generate-with-openssl-rand-hex-32>
ALLOWED_ORIGINS=https://your-frontend-domain.com

🏗️ Architecture

┌─────────────────────────────────────────────────────────┐
│                     Frontend (React)                     │
│  Upload → Analysis Progress → Dashboard → Admin Panel    │
└───────────────────────┬─────────────────────────────────┘
                        │ REST API / WebSocket
┌───────────────────────▼─────────────────────────────────┐
│                    Backend (FastAPI)                      │
│                                                          │
│  ┌─────────┐  ┌──────────────────────────────────────┐  │
│  │ Routers │  │         AI Pipeline                   │  │
│  │ upload  │→│  1. OpenCV Preprocessing               │  │
│  │ analysis│  │  2. YOLOv8 Object Detection           │  │
│  │ reports │  │  3. CNN Cleanliness Classification    │  │
│  │ admin   │  │  4. Scoring Engine (5 categories)     │  │
│  │ ws/live │  │  5. Recommendation Engine             │  │
│  └─────────┘  │  6. Image Annotation                  │  │
│               │  7. PDF Generation (ReportLab)        │  │
│               └──────────────────────────────────────┘  │
│                                                          │
│  ┌──────────────┐         ┌────────────────────────┐    │
│  │   MongoDB    │         │    File Storage         │    │
│  │ (or in-mem)  │         │  uploads/ (images, PDF) │    │
│  └──────────────┘         └────────────────────────┘    │
└──────────────────────────────────────────────────────────┘

📊 Hygiene Scoring

Category	Weight	Factors
Stall Cleanliness	25%	Dirty surfaces, overall image quality
Worker Hygiene	25%	Gloves, masks, bare-hand handling
Waste Management	20%	Waste proximity, garbage bin overflow
Oil Quality	15%	Oil container detection, dark coloration
Cooking Safety	15%	Smoke levels, utensil cleanliness

Score Range	Safety Status
7.0 – 10.0	✅ Safe
4.0 – 6.9	⚠️ Moderate Risk
0.0 – 3.9	🚨 Unsafe

🔬 AI Recommendations (Examples)

"Cooking area contains exposed waste near food preparation zone. Immediate removal required."
"Worker detected handling food without protective gloves. Hygiene violation observed."
"Cooking oil appears overused due to dark coloration. Oil replacement recommended."
"Excessive smoke detected near cooking area. Adequate ventilation improvement needed."
"Cooking surface shows signs of contamination. Immediate sanitization required."

🤝 Contributing

Fork the repository
Create feature branch: git checkout -b feature/amazing-feature
Commit changes: git commit -m 'Add amazing feature'
Push: git push origin feature/amazing-feature
Open Pull Request

📄 License

MIT License — Free to use for academic, research, and commercial purposes.

Built with ❤️ for a safer street food ecosystem

SafeBite AI — Protecting Public Health through Artificial Intelligence

Name		Name	Last commit message	Last commit date
ai_models		ai_models
backend		backend
frontend		frontend
sample_data		sample_data
.gitignore		.gitignore
README.md		README.md
sovle.js		sovle.js
test_upload.py		test_upload.py

Latest commit

History

Folders and files

Repository files navigation

🛡️ SafeBite AI – Street Food Hygiene Rating System

📋 Table of Contents

✨ Features

🔬 AI Hygiene Analysis

📤 Upload & Processing

📊 Analytics Dashboard

📄 Report Generation

🖥️ Admin Panel

🚀 Advanced Features

🛠️ Tech Stack

📁 Project Structure

⚡ Quick Start

Prerequisites

1-Minute Setup

🐍 Backend Setup

Environment Configuration

Install Dependencies

Run the Server

MongoDB Setup (Optional)

⚛️ Frontend Setup

Install Dependencies

Development Server

Production Build

🤖 AI Model Integration

Current Status: Smart Mock Mode

Enabling Real YOLOv8 Model

Enabling Real CNN Classifier

🏋️ Model Training Pipeline

1. Prepare Dataset

2. Train YOLOv8 Object Detector

3. Train CNN Cleanliness Classifier

4. Update Backend Config

📡 API Reference

Base URL: http://localhost:8000/api

Upload

Analysis

Reports

Admin

WebSocket

Example: Upload and Get Report

🚢 Deployment

Docker (Recommended)

Cloud Deployment

Environment Variables (Production)

🏗️ Architecture

📊 Hygiene Scoring

🔬 AI Recommendations (Examples)

🤝 Contributing

📄 License

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages

Base URL: `http://localhost:8000/api`