Automatic classification of Household waste using deep learning

Waste sorting serves as a highly effective approach to reduce waste at the source and promote resource recycling. It can significantly decrease the amount of waste needing final disposal while enhancing the rate of resource recycling. Additionally, this practice helps alleviate environmental issues such as soil and air pollution caused by the increasing volume of waste. This paper presents a deep-learning-based automatic waste-sorting device. It can automatically classify waste into the correct bins, avoiding mis-sorting or non-sorting. We used YOLOv12 as the deep learning model for object detection and integrated an object-relation module into YOLOv12 to strengthen the model’s ability to understand and utilize relationships between objects. This module was embedded in both the feature extraction and post-processing stages of YOLOv12’s detection process. Experiments showed that the smart waste-sorting device achieved a waste recognition rate of $\mathbf{9 7 \%}$, meeting real-world application requirements.