Deep Learning Based Smart Bin for Efficient Sorting of Recyclable, Non-Recyclable, and Compostable Materials

A significant amount of waste is produced every day from urban settings such as homes, offices, residential areas, and basically from our daily activities. Much of such waste remains unutilized due to inefficient sorting practices, which eventually result in environmental pollution. Efficient sorting of waste in these small settings simplifies the larger efforts of waste management and curbs environmental pollution to an extent. Recently, deep learning has attracted attention as a computational approach for solving waste classification challenges, and its application has been explored in other studies for improving the sorting methods. The novelty of the proposed method in this paper includes developing an automated smart bin to sort the waste into three sections: recyclable, non-recyclable, and compostable material, while facilitating compost production. Our smart bin uses the YOLOv8s model for detection, while the dataset contains diverse types of waste, from plastic to organic material. This paper explores different weights of the YOLO model: YOLOv8s and YOLOv8l provided 92.7% and 93.4% mAP@50, respectively, on the same dataset. While their accuracies are similar, YOLOv8s gives relatively good results with less processing time. Therefore, for real-time detection, YOLOv8s is best suited.