| Terahertz Time Domain Spectroscopy(THz-TDS)is an important detection technique for obtaining material information and has been applied in many fields.Based on the unique properties of terahertz waves,the terahertz wave spectrum of the material contains extremely rich physical and chemical information.Therefore,terahertz spectroscopy has great application value.How to extract useful information from the THz spectrum? This requires an appropriate method to analyze the spectral signal.When analyzing terahertz spectra of substances,the most common methods are qualitative and quantitative analysis.However,due to system noise and water vapor absorption,if we do not pretreat the terahertz spectrum,eliminate these interferences.This can cause large errors in the qualitative and quantitative analysis results,and it may not be possible to extract feature information from the terahertz spectrum.In this paper,we introduce the wavelet analysis method to pretreat the measured terahertz spectral signal,reduce or even eliminate the influence of noise,and then use the processed spectrum for qualitative or quantitative analysis.The specific work content of the paper has two parts:First,the terahertz time-domain spectroscopy system was used to detect the terahertz spectra of potassium sorbate and maltose in the air environment(humidity 35%)and nitrogen atmosphere(humidity 3%),and the spectrum was analyzed.Then,taking the absorption spectrum in the nitrogen environment as a reference,the spectral data of potassium sorbate and maltose in air were denoised by wavelet transform,and the denoising effect was measured by relative deviation and root mean square error.In addition,the wavelet decomposition layer selection,threshold function and the best wavelet base were studied in the wavelet denoising process.Finally,the appropriate wavelet denoising method was found.Second,maltose was selected as a sample for quantitative analysis.Maltose and polyethylene were mixed and mixed according to a certain ratio to prepare maltose with different concentrations.Quantitative analysis of the maltose mixture was performed using SLR regression and PLS regression,respectively.Among them,SLR regression selected nine samples with concentrations of 10%-90% for modeling.Partial least-squares regression was used to model eight groups of samples with concentrations of 1%,2%,4%,5%,6%,7%,8%,and 9%.The results show that the correlation coefficients between the verification set and the prediction set of the SLR model at 1.10 THz are 0.991 and 0.987,respectively.The mean square error is 0.84% and 1.25%.The correlation coefficients of the calibration set and the prediction set at 1.59 THz was 0.894 and 0.886,respectively,and the mean square deviations were 1.02% and 1.83%,respectively.The correlation coefficients between the validation set and the prediction set of the PLS model are 0.997 and 0.993,the validation set RMSEC is 0.58%,and the prediction set RMSEP is 1.25%.Subsequently,the original absorption spectra of the eight groups of samples subjected to PLS regression were denoised by wavelet transform and analyzed by PLS regression after denoising.The results show that the accuracy of PLS model constructed after denoising by wavelet transform is improved.The correlation coefficients between the verification set and the prediction set of the PLS were 0.998 and 0.995,respectively.The correction set RMSEC was 0.46% and the RMSEP was 0.86%.The correlation coefficient is closer to 1,and RMSEC and RMSEP are getting smaller.The results show that the terahertz spectra of the maltose mixture are feasible for quantitative analysis combining Wavelet Transform. |
PDF Full Text Request