Real-Time Material Identification Using a Fast and Simplified AI-Assisted Terahertz Spectrometer

Combining artificial intelligence (AI) with state-of-the-art spectroscopy has revolutionized data processing, significantly improving speed and accuracy. However, in the terahertz (THz) frequency range, AI-assisted techniques remain largely confined to research laboratories due to the complexity and cost of existing systems. Here, we introduce a compact and simplified multispectral THz spectrometer with a novel architecture, achieving performance comparable to conventional THz-TDS by leveraging AI for efficient data interpretation. Our compact system integrates a broadband fiber-coupled THz emitter and a custom-built rotating frequency-selective surface (FSS) chopper. Using synchronous detection by a fast intensity sensor, we capture multispectral data in a single rotation of the chopper wheel and analyze it with a deep neural network (DNN) model for rapid and reliable sample identification. We demonstrated real-time classification with over 98% accuracy within just 10 milliseconds of acquisition, even for materials lacking distinct THz fingerprints. This compact and cost-effective approach enables highly efficient THz spectroscopy outside laboratory settings, offering a scalable solution for industrial, biomedical, and security applications.