Enhancing the interpretability of terahertz data through unsupervised classification
We present the applicability of hierarchical agglomerative cluster algorithms to terahertz (THz) spectroscopic analysis. We show the influence of different windowing and filtering methods in the spectral data preprocessing to enhance the clustering results. Two distance measures are compared. Classical Euclidean distance on the full frequency range and a distance working only on the minima of the spectra. We further show the adaptability of our clustering methods for THz hyper-spectral image classification and visualization.