Terahertz data combined with principal component analysis applied for visual classification of materials

A principal component analysis (PCA) method was used here to analyze three materials: glucose, intralipids, and water. The time-domain signal, frequency-domain information, refractive index, extinction coefficient, and dielectric function of these three substances were selected as original variables. We find that it is possible to analyze the similarity between different materials via the cluster distance in the principal component (PC) space and to analyze the difference between samples of the same material via the centroid distance, and the clustering effects of different original variables in PCA may be analyzed via the ratio of centroid distances. Because the thickness of solid-state matters and the concentration of liquid matters are related to the PC, the terahertz time-domain spectroscopy may be combined with PCA to perform visual classification on the materials, thus facilitating substance identification. Sample thickness and concentration may be deduced from the PC score of the materials.