A Benchmark Dataset for Concealed Improvised Explosive Device Detection in X-ray Security Imaging

Threat detection in X-ray security screening is critical for preventing concealed threats in airports and other high-security venue where Improvised Explosive Devices (IEDs) remain among the most persistent and dangerous threats. The lack of a representative, and publicly available IED dataset has limited the development of machine-learning based automated threat detection systems. To address these issues, we propose an open access dataset, called IEDXray constructed for automated detection of IEDs. The dataset comprises 17,360 X-ray images captured under a strategic concealment protocol, covering scenarios ranging from isolated threats to heavily cluttered baggage environments. It includes diverse IED types-homemade explosives, batteries, and modified devices such as laptops, mobile phones, pagers, and walkie-talkies. To validate the dataset, we benchmark state-of-the-art detection models, including YOLOv10, Faster R-CNN, DETR, and GroundingDINO, establishing baseline results across multiple security-screening tasks. By reflecting real-world threat concealment, clutter, and variability, IEDXray provides the research community with a high-fidelity benchmark to advance automated explosive detection and improve x-ray security screening.