This project describes an automated waste classification system which uses deep learning as well as image processing to classify waste into organic or recyclable. This system was trained on a dataset of images of waste and used convolutional neural networks. The system also used image various image processing techniques to enhance the input image quality and also improve the accuracy of the waste classification system. The system's output is a binary classification into organic or recyclable. This system has the potential to improve the current waste management systems and can also reduce the amount of waste that gets disposed in landfills.
Before running the project, ensure you have the following installed:
- Python 3.8+
- Jupyter Notebook
- NumPy
- OpenCV
- TensorFlow/PyTorch
- scikit-learn
- Matplotlib
This study proposed a deep learning approach and various transfer learning methods for waste classification using pre-trained models. The high accuracies are attributed to the maintenance of a fairly standard level of illumination on the waste being analyzed. The deep learning model consists of convolutional and fully connected layers and achieves a validation accuracy of 90.65%. The study also explores transfer learning approaches using pre-trained models such as VGG16, ResNet50, InceptionV3, and MobileNet. The results show that InceptionV3 performs the best with a validation accuracy of 93.20%. These transfer learning methods not only reduce the training time but also improve the generalization of the model, resulting in better performance on the test dataset.
The proposed deep learning approach and transfer learning methods are effective for waste classification, and the study demonstrates their potential for waste management.