Illicit Object Detection in X-ray Images Using Vision Transformers

Illicit object detection is a critical task performed at varioushigh-security locations, including airports, train stations, subways, andports. The continuous and tedious work of examining thousands of X-ray imagesper hour can be mentally taxing. Thus, Deep Neural Networks (DNNs) can be usedto automate the X-ray image analysis process, improve efficiency and alleviatethe security officers' inspection burden. The neural architectures typicallyutilized in relevant literature are Convolutional Neural Networks (CNNs), withVision Transformers (ViTs) rarely employed. In order to address this gap, thispaper conducts a comprehensive evaluation of relevant ViT architectures onillicit item detection in X-ray images. This study utilizes both Transformerand hybrid backbones, such as SWIN and NextViT, and detectors, such as DINO andRT-DETR. The results demonstrate the remarkable accuracy of the DINOTransformer detector in the low-data regime, the impressive real-timeperformance of YOLOv8, and the effectiveness of the hybrid NextViT backbone.