Real-Time Waste Detection Based on YOLOv8

In this work an intelligent waste detection system using YOLOV8 is proposed, Moreover, a comparison between YOLOv8, YOLOv5, YOLOv4, and YOLOv3, was performed. After training results shows that YOLOv8 tiny outperformed the previous methods named above. However, unlike the existing studies, In this paper using Yolov8 we can localize and detect two different classes for object detection task including, Landfill (open dump), and on ground waste, meaning from high and low distance used aerial images. The experiments are conducted on a collection of 998 images for landfills from SWAD dataset, and 772 images for on ground rubbish from UAVVaste dataset. The obtained results achieved a Precision of 85.9%, mAP@50 of 75.5%, and a Recall of 69.2%. This study demonstrated that our pretrained YOLOv8 detector significantly enhanced the precision of waste detection, taking into account different image background conditions.