Qualitative detection of amino acids in a mixture with terahertz spectroscopic imaging

Abstract

Terahertz time-domain spectroscopy (THz-TDS) is widely used in non-destructive testing and chemical identification due to the wide spectrum, low photon energy, distinctive absorption spectra, and higher intensity than other thermal radiators. Here, we propose a method to detect amino acids from a mixture at room temperature and in humid air. A THz-TDS spectrometer scans tablets of amino acid mixtures laid out on a polyethylene plate in transmission mode. The resulting spectroscopic imaging is then converted to probability maps (chemical maps), where the region containing the expected amino acid has higher pixel values, by a matrix manipulation with the matrix of absorption spectra. This approach is substantially resistant to air humidity and scattering caused by uneven grain sizes and is applicable to substances without strong absorption features. A classifier based on LeNet-5 is built to identify the subsection of the chemical map containing the expected substance, and the prediction reaches 100% accuracy for the testing dataset. The results of the classifier are compared with those of a support vector machine, whose accuracy is much lower due to its inferior ability to recognize complex patterns. The proposed method will enable us to qualitatively detect a component from an unknown mixture in a single measurement and will be useful for bio-molecule detection in real-life scenarios.