Machine learning based threat detection for dual modality X-ray transmission and coherent diffraction security screening systems

Abstract

The task of threat detection in X-ray based security screening applications not only demands a high probability of detection (Pd) but also low probability of false alarm (Pfa) to reduce operational cost and inconvenience. While material classification based on X-ray transmission-based features (i.e., material density and effective atomic number (Zeff)) have been demonstrated to yield high Pd but they suffers from relatively high Pfa due to the lack of material specificity especially in and around threat materials. It is also well known that X-ray coherent diffraction based material measurement are capable of providing more specific material information, at the molecular level, complementary to transmission based material information. In this work we report on the performance of joint classifiers, specifically support vector machines, that combine X-ray transmission with coherent diffraction features and quantify the improvement relative to transmission only system.