Wavelength Selection for Quantitative Analysis in Terahertz Spectroscopy Using a Genetic Algorithm

In the quantitative analysis of the components of mixtures based on terahertz time-domain spectroscopy (THz-TDS), wavelength selection of the spectra holds great importance for quantitative accuracy. The raw spectrum obtained in the experiments contains both high signal-to-noise ratio (SNR) and low SNR information, along with many types of noise, such as the scattering effect and random disturbances. Usually, the frequency band used for analysis is determined over multiple attempts to obtain the most satisfactory results. Currently, there is no common method for selecting the appropriate frequencies. In this paper, a wavelength selection method for the THz absorption spectra based on a genetic algorithm is presented, with the main focus on quantitative analysis by THz-TDS. Using this method, most of the useful data with high SNR were retained. The comparison between quantitative analysis errors obtained by the raw spectra and by the wavelength-selected spectra showed that this method could significantly decrease quantitative errors. After wavelength selection, the quantitative analysis errors of 12 mixture samples with widely differing concentrations are predominantly below 6% with a standard deviation of 0.0344.