Analysis and Classification of Frequency-Domain Terahertz Reflection Spectra Using Supervised and Unsupervised Dimensionality Reduction Methods

Study of terahertz reflection spectra of optically thin RDX samples by terahertz imaging with spectral resolution

“…At the moment, there are many works on the topic of THz spectroscopy of explosives devoted to the study of samples that have undergone preliminary preparation aimed at minimizing scattering effects—the particles of the analyzed substance are crushed, mixed with particles that form a neutral isotropic matrix (for example, with a powder of polyethylene particles) and then pressed into tablet 10 – 13 At the same time, in the literature, one can find works devoted to spectral studies of explosive samples in the presence of significant scattering (including using terahertz time-domain spectroscopy—THz-TDS, for example, Refs. 14–16).…”

Aspects of Terahertz Reflection Spectroscopy

“…This is also evident from the perfect classification scores of the training data together with the lower scores of the test data. Particularly, we found that LDA fitted to the noise in the training data instead of the spectral characteristics Therefore, we applied regularized-LDA (RLDA) using a 10-fold stratified cross-validation to obtain the optimal regularization value β = 0.5 [198]. The RLDA algorithm returned almost perfect classification scores of the train data and a perfect generalization to the test data for all three classifiers (see Table 2.1).…”

“…Furthermore, for materials having a significantly higher refractive index than PE, e.g., PABA and theophylline, the reflection coefficient drops over the entire spectral range with decreasing concentration." [198] To accommodate the higher complexity of the data, we retained three features of both the PCA and LDA algorithms. Furthermore, the material concentrations were discarded from the class labels fed to the LDA algorithm as the principal scope was to identify the sample material.…”

“…Furthermore, the material concentrations were discarded from the class labels fed to the LDA algorithm as the principal scope was to identify the sample material. The classification scores are given in Table 2 [198].…”

“…: Classification scores of the PCA-and LDA-processed spectra of samples with various material concentrations, labeled according to their severity of hazard. Adapted from Publication A[198].…”