Vinodhini Shanmugapriya E, Shirly Edward A, Mercy Latha A
Terahertz (THz) spectroscopy and imaging have attracted increasing attention because they enable non-ionising and label-free probing of low-frequency molecular interactions and material properties. These characteristics make THz sensing particularly attractive for applications in biomedical diagnostics, agricultural inspection, and material characterisation. However, the broader adoption of THz technology remains constrained by challenges such as spectral congestion, environmental sensitivity, limited signal-to-noise ratio, and the complexity of hyperspectral datasets. Advancement in artificial intelligence (AI) provide promising approaches to address these limitations like automated feature extraction, improved spectral interpretation, and reliable analysis of high-dimensional THz measurements. This review examines the convergence of THz sensing and AI-driven modelling, highlighting machine-learning techniques that enhance signal processing, pattern recognition, and multimodal data analysis in THz-based non-destructive testing (NDT) systems. To illustrate these strategies, three representative examples based on publicly available datasets are discussed. These include regression-based estimation of fruit ripeness using sub-THz metamaterial sensing, multimodal classification of wood species using THz hyperspectral and X-ray datasets, and reconstruction of THz optical-constant spectra for anomaly detection. . Integration of sensing physics, AI-based modelling, and application-driven workflows,provides a structured perspective for reliable AI-enabled THz-NDT systems. This review further highlights the role of THz spectroscopy in Quantitative Non-Destructive Evaluation (QNDE).