Deep Learning Waste Segregation

Uses a deep learning model trained on image data to automatically classify waste into different categories.

Comment

Waste Segregation using Deep Learning

PyTorch Python Visual Studio Code Jupyter Notebook

A computer vision-based project to classify and segregate waste using image recognition techniques.

Project Description

This project uses a deep learning model trained on image data to automatically classify waste into different categories.
The goal is to aid smart and sustainable waste management systems by helping machines identify recyclable and non-recyclable waste through image input.

Author

Ali Akbar Khan

Technologies Used

  • Python
  • Fastai
  • PyTorch
  • Gradio (for UI)
  • Jupyter Notebook

Model Details

  • Architecture: resnet18 pretrained CNN
  • Training: Fine-tuned for 10 epochs using Fastai's vision_learner
  • Metrics: Achieved high accuracy (close to 92%) on validation set

Dataset

The dataset contains labeled images of different types of waste, categorized into classes like:

  • Recyclable
  • Organic
  • Hazardous
  • Others (can be customized)

How to Run

1. Clone the repository

git clone https://github.com/aliiakbarkhan/deep-learning-waste-segregation.git
cd deep-learning-waste-segregation

2. Install dependencies

pip install -r requirements.txt

3. Run the notebook

Use Jupyter or VSCode to open waste-segregation.ipynb and run all cells.

4. Or launch Gradio app (if available)

python app.py

Sample Predictions

The model can take any image of waste and classify it into one of the predefined categories with high accuracy. Here's a snapshot of predictions from the Gradio interface.

File Structure

waste-segregation/
├── waste-segregation.ipynb     # Main notebook
├── model.pkl                   # Trained model (optional)
├── app.py                      # Gradio app (if created)
├── README.md                   # Project documentation
└── requirements.txt            # Dependencies

Built With

  • jupyter-notebook
Share this project:

Updates