Enhancing Garbage Classification with Swin Transformer and Attention-Based Autoencoder: An Efficient Approach for Waste Management

Annually, an enormous quantity of waste is generated, and the majority of it is discarded, resulting in environmental degradation. Employing computer vision-based techniques to classify waste into distinct recycling categories may provide efficiency in waste material management and disposal. Hence, this study proposes a novel lightweight approach that utilizes the Swin Transformer and attention-based autoencoder for garbage image classification which consumes significantly fewer resources. The proposed method achieves a high accuracy rate (94.27%) in identifying different types of waste materials on the TrashNet dataset. Our experimental results demonstrate that it outperforms several existing methods for waste classification with very limited parameters.