Garbage Classification Using Computer Vision

Garbage classification is one of the initial steps in reducing the global garbage problem. Therefore, it is essential to segregate and sort the waste into recyclable and non-recyclable or in other categories which in turn reduces the pollution by reducing the overall dump in landfills and sending that waste to recycle plants. Convolutionary Neural Network (CNN) model and transfer learning model via various models are implemented to meet the purpose. Image-based dataset is trained, tested and validated to give provided results. CNN model is giving 64% accuracy while transfer learning via NasNetLarge performed better with about 88.07% accuracy. Further improvements can be made by expanding the dataset to include more labeled garbage images. The models could also be fine-tuned by unfreezing some higher layers and training the full network end-to-end on the garbage images. Experimenting with different CNN architectures and hyperparameters may also boost accuracy. In the future, combining multiple models into an ensemble could provide better results than any single model alone. With more computing resources, larger batch sizes and image resolutions could be used to take full advantage of the models capabilities. Overall, this is an important step in developing an automated garbage classification system, but accuracy can still be substantially improved with more data, model tweaking, and computational power.