Detecting Litter From Aerial Imagery Using the SODA Dataset

This study focuses on litter detection in the context of aerial imagery captured from low altitudes using consumer-grade UAVs. UAVs are equipped with high-resolution cameras; however, litter objects captured from aerial imagery are represented by a relatively small number of pixels compared to the total number of pixels in the image itself. This makes detecting litter from aerial imagery a small object detection problem. To assist in this study, the SODA dataset was created. The SODA dataset contains 377 aerial images captured using a UAV from different altitudes ranging between 5 meters and 30 meters. The images in the SODA dataset were collected in natural habitats environments and contain six different categories of litter. An evaluation of the SODA dataset was performed using YOLOv8, YOLOv5 and Faster R-CNN. The SODA dataset is made publicly available on GitHub.
¹
.