I. Introduction
The CURRENT demand for novel security technology pushed the development of millimeter-wave and terahertz instrumentation during the past decades [1]–[3]. Many active and passive device concepts have been developed for locating concealed objects. Most of them have in common that the objects are identified by their optical properties in these frequency bands [4]–[6]. Metals, for instance, can be sensed by their high reflectivity utilizing millimeter-wave imaging [7]. Other concepts make use of the fact that some organic substances, such as explosives or drugs, have spectral fingerprints in the far-infrared [8]–[11]. Since these techniques are highly specific, limitations arise when applying them in real life situations. Databases of spectral material properties have to be collected, which should comprise dangerous materials as well as all possible cover materials. In the case of explosives, filling factors or the specific fabrication process can affect the spectra, which further expands the required databases. Finally, extreme challenges appear when concluding from the measured data onto the present materials [12]. Solutions of such inverse problems strongly diverge, in particular, in the presence of noise [13], which may result in enormous false positive rates.