🗑️ Garbage Classifier for Waste Management

AI-Powered Garbage Segmentation & Classification System

An intelligent waste classification system using Deep Learning for smart waste management and environmental sustainability

---

📋 Table of Contents

• About the Project
• Features
• Tech Stack
• Dataset
• Model Architecture
• Installation
• Usage
• Results
• Team
• Acknowledgments

---

🎯 About the Project

Waste management is a critical global issue, and improper segregation of garbage contributes to pollution, health hazards, and inefficient recycling. Manual waste sorting is error-prone and labor-intensive.

This project presents an intelligent Garbage Classifier for Waste Management using Deep Learning techniques. The system employs:

• YOLOv8 for detecting and segmenting garbage objects
• MobileNetV2 for refined classification into six categories
• Gradio for a user-friendly web interface with real-time predictions

The expected outcome is a reliable and scalable solution that can support smart cities, recycling plants, and households in effective waste segregation, thereby promoting environmental sustainability.

---

✨ Features

Feature	Description
🎯 Instance Segmentation	Precise pixel-level detection of garbage objects
📊 6-Class Classification	Biological, Cardboard, Glass, Metal, Paper, Plastic
🌐 Web Interface	User-friendly Gradio app for image uploads
📈 Visual Analytics	Pie charts showing garbage distribution
⚡ Real-time Processing	Fast inference on CPU/GPU
🔧 Adjustable Confidence	Configurable detection threshold

---

🛠️ Tech Stack

Category	Technologies
Deep Learning	PyTorch, YOLOv8 (Ultralytics), MobileNetV2
Computer Vision	OpenCV, PIL/Pillow
Web Interface	Gradio
Visualization	Matplotlib, NumPy
Data Source	Roboflow

---

📁 Dataset

Property	Details
Source	Roboflow Universe
Total Images	481
Annotation Format	YOLOv8 Segmentation
Image Size	640×640 (resized)
License	CC BY 4.0

Classes (6 Categories)

Class	Description	Color
🟢 Biological	Food waste, organic matter	Green
🟤 Cardboard	Boxes, packaging	Brown
🔵 Glass	Bottles, jars	Blue
⚪ Metal	Cans, foils	Gray
🟡 Paper	Documents, newspapers	Yellow
🔴 Plastic	Bottles, bags, containers	Red

---

🏗️ Model Architecture

┌─────────────────────────────────────────────────────────┐
│                    INPUT IMAGE                          │
└─────────────────────────┬───────────────────────────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────┐
│              YOLOv8 SEGMENTATION MODEL                  │
│  • Object Detection (Bounding Boxes)                    │
│  • Instance Segmentation (Pixel Masks)                  │
│  • Primary Classification                               │
└─────────────────────────┬───────────────────────────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────┐
│           OPTIONAL: MobileNetV2 CLASSIFIER              │
│  • Refined classification on cropped regions            │
│  • Higher accuracy for ambiguous cases                  │
└─────────────────────────┬───────────────────────────────┘
                          │
                          ▼
┌─────────────────────────────────────────────────────────┐
│                    GRADIO INTERFACE                     │
│  • Annotated image with masks & labels                  │
│  • Pie chart distribution                               │
│  • Detection summary                                    │
└─────────────────────────────────────────────────────────┘

---

🚀 Installation

Prerequisites

• Python 3.8 or higher
• pip package manager
• (Optional) CUDA-compatible GPU for faster inference

Setup

# Clone the repository
git clone https://github.com/YOUR_USERNAME/garbage-segmentation-app.git
cd garbage-segmentation-app

# Create virtual environment (recommended)
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Download Model Weights

Download the trained model weights and place in weights/ directory:

# Create weights directory
mkdir -p weights

# Download best.pt (link will be provided/you can train your own)
# Place the file as: weights/best.pt

Note: Model weights (~90MB) are not included in the repository due to size. See weights/README.md for download instructions.

---

💻 Usage

Launch Web Interface

python app.py

Then open your browser to http://127.0.0.1:7860

Command Line Options

# Create public shareable link
python app.py --share

# Custom port
python app.py --port 8080

# Custom host
python app.py --host 0.0.0.0

Python API

from models import GarbageSegmentor

# Initialize model
segmentor = GarbageSegmentor()

# Run inference
results = segmentor.predict("path/to/image.jpg")

# Get annotated image
annotated = segmentor.visualize(results)

# Get detection summary
summary = segmentor.get_detection_summary("path/to/image.jpg")
print(f"Total objects: {summary['total_detections']}")
print(f"Classes: {summary['class_counts']}")

---

📊 Results

Training Metrics

Metric	Value
Model	YOLOv8 Segmentation
Epochs	100+
Image Size	640×640
Batch Size	16

Confusion Matrix

Training Curves

Sample Detections

---

👥 Team

Minor Project | 5th Semester | BTech (Hons.) CSE - Artificial Intelligence
University Teaching Department (UTD), CSVTU Bhilai
Group Number: 09

Team Member	Contributions
Abhay Singh Sisoodiya	Model Development, Deployment (Gradio Web Interface)
Abhinav Anand	Data Collection (Roboflow Dataset), Documentation & Testing
Aditya Verma	Utilities (Preprocessing & Visualization), Deployment, Presentation
Anshul Yadav	Integration (Inference Pipeline)
Aman Banajre	Documentation & Testing (Report, README, Testing)
Harsh Kumar Chandrakar	Data Collection, Model Development, Integration

---

📁 Project Structure

garbage-segmentation-app/
├── app.py                  # Entry point for Gradio app
├── requirements.txt        # Python dependencies
├── .gitignore
├── LICENSE
├── README.md
│
├── models/                 # Core model code
│   ├── __init__.py
│   ├── yolo_segmentation.py
│   ├── mobilenet_classifier.py
│   ├── inference.py
│   ├── gradio_app.py
│   └── utils/
│       ├── preprocessing.py
│       └── visualization.py
│
├── weights/                # Model weights (download separately)
│   └── README.md
│
├── results/                # Training metrics & curves
│   ├── confusion_matrix.png
│   ├── results.png
│   └── ...
│
├── data/                   # Dataset documentation
│   └── README.md
│
├── docs/                   # Documentation
│   └── screenshots/
│
└── notebooks/              # Jupyter notebooks

---

🙏 Acknowledgments

• Ultralytics - YOLOv8 framework
• Roboflow - Dataset hosting and annotation tools
• Gradio - Web interface framework
• CSVTU Bhilai - University Teaching Department

---

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

---

🌱 Supporting UN SDG 11: Sustainable Cities and Communities

Made with ❤️ by Group 09 | UTD CSVTU Bhilai