The objective of the research is to develop robust approaches for terahertz time-domain spectroscopy (THz-TDS) in nondestructive evaluation and imaging. The focus is on implementing signal-processing and machine-learning strategies to overcome limitations in conventional analysis and to enable reliable inspection of stratified structures. This relies on an understanding of THz wave–matter interactions and the constraints of current deconvolution techniques. One thrust of the research is to investigate and test advanced methods for mill scale thickness measurement in the steel industry, including sparse deconvolution and neural-network–based regression and classification. Another thrust is to develop THz-based inspection methodologies for electronic components, using polarization-dependent responses and clustering algorithms to identify structural defects. A further thrust is to address cultural heritage applications, where convolutional neural networks with Co-teaching are employed to enhance the imaging of multilayer paper with iron gall ink.