Unlocking Terahertz technology with machine learning: A comprehensive review

Abstract

Terahertz (THz) technology has proven to be a powerful tool for investigating a wide range of physical, chemical, and biological phenomena, as extensively demonstrated by years of research and innovation. Today, THz time-domain spectroscopy has become a trusted technique for the characterization of the dielectric properties of materials, while advances in tomographic imaging have propelled THz waves from research laboratories to practical real-world applications. However, THz spectroscopy and imaging generate increasingly complex and high-dimensional data, posing challenges for traditional signal processing techniques, which often struggle to extract meaningful insights efficiently. Machine learning (ML) offers a transformative solution by extending classical signal processing capabilities to address broader challenges. The integration of ML into THz spectroscopy and imaging drives significant progress in areas such as data pre-processing, feature extraction, pattern recognition, and both qualitative and quantitative analyses. This review explores the application of these cutting-edge algorithms in THz technology from the past five years in non-destructive testing. It provides an up-to-date overview of the current technological landscape and challenges, providing insights and opportunities to propel further advancements in scientific research and commercial innovation.