A tomographic workflow to enable deep learning for X-ray based foreign object detection

Mathé T. Zeegers, Tristan van Leeuwen, Daniël Maria Pelt, Sophia Bethany Coban, Robert van Liere, Kees Joost Batenburg

Abstract: Highlights•A tomographic workflow is proposed for deep learning based foreign object detection.•The workflow efficiently generates large annotated X-ray data required for training.•Real objects are scanned and 2D ground truth is generated with 3D CT reconstructions.•Results with trained networks show high detection accuracy with few training objects.•The workflow holds potential for applied industrial foreign object detection.