🌍 Waste Recognition Model for Environmental Sustainability

Overview

The Waste Recognition Model is a deep-learning-based solution designed to classify various types of waste materials with high accuracy. By integrating AI-powered waste classification, this project aims to enhance waste management efficiency, reduce environmental pollution, and promote sustainable waste disposal practices.

Key Features

• Accurate Waste Classification: Identifies waste materials across multiple categories.
• Disposal Recommendations: Suggests proper disposal methods to minimize environmental impact.
• Real-Time Processing: Provides quick and efficient waste recognition for seamless integration into waste management systems.
• User-Friendly Interface: Accessible via a web-based application.
• Versatile Deployment: Can be used in homes, offices, and industrial waste management systems.
• Environmental Impact: Reduces contamination and promotes sustainable waste disposal.

Waste Categories

The model is trained to classify waste into the following categories:

1. Plastic Waste - Bottles, wrappers, straws, bags.
2. Metal Waste - Aluminum, copper, iron, steel.
3. Organic Waste - Food waste, yard waste.
4. Glass Waste - Bottles, jars, containers.
5. Electronic Waste - Computers, phones, appliances.
6. Textile Waste - Clothing, fabric materials.
7. Medical Waste - Clinical and medical items.
8. Wood Waste - Construction wood, furniture.
9. Paper Waste - Newspapers, magazines, cardboard.
10. Shoes Waste - Footwear and related items.

Technologies Used

• Python
• Fastai
• Gradio
• Hugging Face
• HTML/CSS/JavaScript
• Tailwind CSS
• DaisyUI

🚀 Live Demo

Experience the model in action:

• Live Website
• Hugging Face Space

🖼️ Preview

Screenshots of the website

Hugging Face Model Preview

Model Performance

The model demonstrates high accuracy in waste classification with:

• Real-time waste identification
• Appropriate bin recommendations
• Proper disposal guidance

Final Model Accuracy:

epoch	train_loss	valid_loss	error_rate	accuracy	time
0	0.299003	0.159227	0.046041	0.953959	29:14

epoch	train_loss	valid_loss	error_rate	accuracy	time
0	0.491925	0.258303	0.073665	0.926335	03:01

Data Preparation

• Dataset Collection: Waste images are gathered from multiple sources, ensuring diverse and well-distributed data.
• Preprocessing: Images undergo resizing, normalization, and augmentation to enhance model generalization.
• Labeling: Data is labeled into predefined categories.
• Splitting: The dataset is divided into training, validation, and testing sets for model evaluation.

Model Training

• Architecture: The model is built using Fastai with a ResNet34 backbone for transfer learning.
• Optimization: Fine-tuned using techniques such as learning rate finder and gradual unfreezing.
• Training: The model undergoes multiple epochs of training to minimize loss and improve accuracy.
• Evaluation Metrics: Accuracy, loss, and confusion matrix are used to assess performance.

Inference

• Model Deployment: The trained model is deployed on Gradio and Hugging Face for easy accessibility.
• User Interaction: Users upload waste images, and the model predicts the waste category and provides disposal guidance.
• Real-Time Testing: The inference pipeline processes new images with minimal latency, making it efficient for practical applications.

Project Structure

WASTE-RECOGNIZER/
├── app/
│   ├── test_images/         # Sample images for validation
│   ├── app.py               # Main application file
│   ├── gradio.png           # Application preview image
│   └── README.md            # Application documentation
│
├── data/
│   ├── dataloaders/         # Utilities for data handling
│   │   └── README.md        # Documentation for dataloaders
│   └── README.md            # Data-related documentation
│
├── docs/
│   ├── images/              # Documentation images
│   ├── .nojekyll            # Jekyll configuration
│   └── index.html           # Main webpage content
│
├── models/
│   └── README.md            # Documentation for the model
│
├── notebook/
│   ├── DataPrep.ipynb       # Data preprocessing notebook
│   ├── DataTraining.ipynb   # Model training notebook
│   └── Inference.ipynb      # Inference and evaluation notebook
│
├── LICENSE                  # License file
├── README.md                # Project documentation
└── requirements.txt         # Dependencies

Setup Instructions

1️⃣ Clone the Repository

git clone https://github.com/AuvroIslam/wasteRecognizer.git
cd wasteRecognizer

2️⃣ Install Dependencies

pip install -r requirements.txt

3️⃣ Run the Model Locally

cd app
python app.py

4️⃣ Model Training (Optional)

cd notebook
jupyter notebook

Open and run DataPrep,dataTraining,Inference.ipynb to train and test the model.

Usage

• Upload an image of waste material.
• The model predicts the waste category.
• It provides disposal recommendations.

Testing

• Use images from test_images/ to validate model performance.
• Run inference using DataPrep,dataTraining,Inference.ipynb for real-world evaluation.

License

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

Contributions

We welcome contributions! Feel free to fork the repository and submit a pull request.

Contact for More Details

• 📧 Email: oitijya2002@gmail.com
• 🔗 LinkedIn: Oitijya Islam Auvro
• 💻 GitHub: AuvroIslam