Research On Classification And Spectral Analysis Of Organic Molecules Based On Terahertz Spectroscopy Technology

Qualitative identification and structural analysis of organic molecules such as genetically modified materials and isomeric materials are widely used in the fields of agricultural security and fine chemicals.Terahertz(THz)waves have unique advantages in the field of non-destructive testing,as they can interact with organic molecules to produce "resonant" absorption,which can determine inter-and intra-molecular weak interactions,and are suitable for qualitative analysis and resolution of spectral absorption mechanisms.The traditional analysis method still has some problems,such as the difficulty of extracting high-dimensional nonlinear THz signal features under noise interference,the serious endpoint effect and modal confusion of THz measurement signals,the failure of experimental THz spectral feature "fingerprint" identification and the low reliability of theoretical calculation results for spectral analysis.In order to solve the above four typical problems,first of this thesis from the angle of the micro(the molecular level)in-depth studies three kinds of hydroxy benzoic acid isomers THz vibration spectrum absorption mechanism,then the THz spectrum of noise reduction and feature extraction methods,finally studies three kinds of transgenic soybeans and non-gmo soybean and three kinds of hydroxy benzoic acid isomers improve classification recognition method,In order to realize the molecular level characteristic "fingerprint" identification,intramolecular and intermolecular weak interaction region indication and fast nondestructive testing of organic molecules.The main research contents of the thesis are summarized as follows:(1)A fingerprinting scheme for the THz characteristics of hydroxybenzoic acid isomers by combining density general function theory with potential energy distribution analysis was proposed.The theoretical spectra of the three hydroxybenzoic acid tautomeric clusters were calculated using density functional theory(DFT)combined with B3 LYP hybridization generalization and D3 dispersion correction at the 6-311G++(d,p)group level.By decomposing the characteristic vibrational modes of the molecular cluster system through potential energy distribution analysis(PED),we obtained all the information on the vibrations of the experimental THz characteristic absorption peaks and the corresponding vibrational contribution values,and realized the effective fingerprinting of the THz experimental spectra of the three hydroxybenzoic acid isomers.(2)An energy decomposition analysis combined with visual molecular dynamics and interaction region indication is proposed for the THz vibrational spectral resolution of hydroxybenzoic acid isomers.Based on the quantum chemical calculations,the total interaction energy of the molecular cluster system was decomposed by the energy decomposition analysis combined with the classical molecular force field(EDA-FF)method to obtain all the bonding information and the corresponding energy contributions of the covalent and non-covalent bonding of the three hydroxybenzoic acid isomers.Visual molecular dynamics and interaction region indication(IRI)visualization were investigated to describe the covalent and non-covalent bonding interactions(weak interactions)by externally rendered atomic coloring and electron density gradient parametrization weighted by a certain ratio of electron density,and to realize the correct resolution of the theoretical calculation results to the experimental spectra and the indication of intermolecular weak interaction regions.(3)A classification model of hydroxybenzoic acid isomers based on variational modal decomposition-particle swarm optimization improved support vector machine is proposed.Different from the classical empirical modal decomposition(EMD)method,the recently proposed variational modal decomposition(VMD)method is used,which is in essence multiple adaptive Wiener filter groups,exhibiting better noise robustness and more suitable for processing nonlinear non-stationary short-time frequency THz signals.In order to further enhance the effect of highlighting small differences in different THz spectral data under background noise,particle swarm optimization combined with support vector machine algorithm(PSO-SVM)is investigated to obtain the global optimal solution by updating the particle swarm in multiple iterations to enhance the high signal-to-noise ratio and energy concentration of THz signals,and achieve the rapid nondestructive detection of three hydroxybenzoic acid isomers.(4)A THz spectral feature extraction method for transgenic soybean based on Floyd’s improved local linear embedding is proposed.The classical linear dimensionality reduction-based approach transforms the problem of noise reduction into a mathematical optimization problem by using nonlinear dimensionality reduction-improved Floyd local linear embedding means to realize the preprocessing of THz measurement signals.Based on this,the first few principal components after dimensionality reduction are used as target parameters for multiple linear regression analysis and partial least squares regression analysis to achieve robust extraction of high-dimensional nonlinear THz signals and accurate identification of transgenic and non-transgenic soybeans under background noise environment.