Chemical Mapping and Concealed Explosive Imaging Using Terahertz Time-Domain Spectroscopy and Deep Learning

Terahertz time-domain spectroscopy enables non-invasive detection of chemicals through opaque materials, yet conventional methods often struggle with real-world sample variations and environmental interferences. To address these limitations, we developed a chemical imaging system that integrates THz-TDS with deep learning for accurate pharmaceutical and explosive classification. Utilizing plasmonic nanoantenna arrays for pulsed terahertz generation and detection in reflection mode, the system achieves a 96 dB dynamic range and a 4.5 THz bandwidth, supporting practical, stand-off operation. By analyzing time-domain terahertz pulses with deep neural networks, our approach effectively mitigates errors caused by environmental and geometric variations. Blind testing across eight chemicals demonstrated a 99.42% pixel-level accuracy. Furthermore, the system achieved 88.83% accuracy when imaging explosives concealed under an opaque cover, highlighting its robust generalization capability for practical security applications.