Annotated datasets for waste electrical and electronic equipment identification and battery detection and classification.

This paper describes the initial release of annotated datasets from the advanced robotics and artificial intelligence for critical raw materials recycling in the circular economy (XBAT+) project. Relatively small battery waste electrical and electronic equipment (WEEE) devices were sorted from cages of general WEEE during visits to recycling facilities and collection events. From 50 initially defined battery-WEEE categories, applying a frequency criterion reduced them to 15 categories comprising 421 battery-WEEE devices. Optical and x-ray imaging of battery-WEEE devices produced multiple images per device in some cases, resulting in 3045 static images. These were organised into high-quality, varying-quality, and RGB datasets, semi-automatically annotated, and split into training and test sets. Specifically, within each of the 15 categories, 20% of the images were allocated to the test sets, the rest to the training sets. Evaluation using YOLO26 and RF-DETR demonstrated promising mAP@50 and recall for supervised learning-based battery-WEEE identification and battery/nobattery detection, highlighting the potential of XBAT+ datasets to enable WEEE filtering and mitigate battery-based fire hazards in recycling facilities.