A statistical evaluation of material discrimination approaches in cargo inspection systems

Abstract

In X-ray inspection systems employing dual-energy capabilities, material discrimination primarily centers on three methodologies based on curves: the $\alpha$ curve, the $R$ curve, and the $H-L$ curve. Despite extensive research in this field, there remains ambiguity regarding the comparative effectiveness of these methods under different conditions. This study aims to evaluate the accuracy of material discrimination using a standard cargo inspection system for four wedge-shaped materials. Multiple experiments were conducted, where experimental data were mapped onto the coordinate system for all three discrimination methods, and discrimination curves were generated using cubic interpolation. Regular residual analysis and statistical tests, including ANOVA and Tukey HSD, were performed to evaluate the effectiveness of the $R$, $\alpha$, and $H-L$ methods. No statistically significant differences in the performance of these methods were found. Unlike some previous studies where one method was favored over another, no inherent superiority of any particular method was observed in this research. Additionally, combining these methods yields no significant gains in accuracy or efficiency, and their simultaneous use is not recommended due to increased computational costs. The analysis of sample thickness effects revealed higher success rates for thicker samples, though no method outperformed the others across thickness ranges. These findings suggest that the choice of method should be guided by practical considerations rather than assumptions of inherent superiority.