Robust Identification of Concealed Dangerous Substances by Spectral Correlation of Terahertz Transmission Images

Abstract

Terahertz (THz) images containing spectral information in each pixel are recorded in transmission mode using a fiber-coupled time-domain spectroscopy system. The images are acquired by mounting a sample holder on an x-y stage, which is stepped across the beam in the two transverse directions, while the transmitted THz waveform is captured. The materials under investigation consist of uncovered and hidden samples of an explosive (RDX) and simulants (lactose and tartaric acid). Spectral angle mapping is used to identify the materials in the THz images by comparing the spectrum in each pixel with a library of reference spectra for the different materials. We test the performance of several spectral characteristics derived from the measured transmission spectra. Robustness is studied by investigating the receiver-operating-characteristics (ROCs). The ROCs are used to find which of the spectral characteristics is most robust to different sample preparation conditions, without the need for extensive pre-treatment of the data, such as baseline correction. Simple theoretical considerations are used to support the experimental results.