Techniques And Experiment Study Of Material Detection Based On Terahertz Time-Domain Spectroscopy

Terahertz Time Domain Spectroscopy Is Applied To The Detection Of Material,Due To The Thz Wave Characteristic O f Time Domain Spectrum Resolution Ability And Nondestructive Detection Ability,Thz Technology Will Bring A Broad Prospect For The Material Detection Technology.This paper takes protein,organic gas,tea and dangerous cargo as research objects,put these four materials on trial,meanwhile,carries on a systematic study in the THz optical parameters calculation,THz spectrum preprocessing,optimal fitting,object classification.The main research contents and innovation points summed up as follows:First,We Carried Out The Thz-TDS Experiment of Four K inds of Substances.According To The Detect Result O n Bovine Serum Albumin Samples At Different Temperatures,The Conclusion That The Thz Spectrum Experiment Result Could Reflect The Protein Molecule Conformational Analysis Problem Is Pointed Out.Experiments With Methanol Gas,Find O ut That The Organic Gas Showed The Absorption Peak At Thz Band And Acquires The Conclusion That The Thz-TDS Technology Could Be Used To Detect Small Molecule Organic Gases.Obtained The Thz Absorption Spectrums From Experiment On Tie Guanyin,Dragon Well Tea,Biluochun,Huangshan Maofeng,Pu'er Tea,Keemun Black Tea By Thz-TDS Experiment.Obtained The Thz Absorption Spectrums At The Range O f 0.1~0.25 thz By Thz-TDS Experiment With Three K inds O f Explosives,Namely: DNT,RDX,HMX.In The Thz Spectrum Preprocessing Stage,Compare And Analyze The Baseline Process Mode O f The First Derivative And Second Derivative.In The Meantime,Compare The Wavelet Denoising Method Which Based On Soft-Thresholding And Hard Thresholding.Combine the principal component analysis with fuzzy clustering method to fuzzy identifying the THz spectrum of solid macromolecular organic matter.The spectrum of BSA is adopted to deal with dimension reduction by principal component analysis and come to the conclusion that the score of dielectric loss angle tangent could reflect the relationship more between the BSA and the temperature than the score of the absorption coefficient,refractive index and dielectric spectrum inscriber.Use the score of dielectric loss angle tangent to build the standard model database.Take the fuzzy recognition method of Euclidean distance and nearest principle,discern the sample to be tested successfully.Use the BP neuralnetwork,GA-BP neuralnetwork and PSO-BP neuralnetwork to quantitative analysis for the small molecule gaseous organic matter.For the THz spectrum of methanol gas with different volume concentration,modeling with the three kinds of neuralnetwork,forecast the unknown concentration o f methanol gas.By a contrast between the model performance of the three kinds of neuralnetwork,shows that the the way to optimize the initial value and threshold value of BP neuralnetwork by particle swarm optimization and genetic algorithm has a much more convergence effect,and that the neural network value and actual value have much more relevant and the error of test data is smaller.However,compare with genetic algorithm of BP network,the particle swarm algorithm has smaller error of test data,more relevant but the rate of convergence is slower.The result provides a new way for spectrum detection of gaseous organic small molecule and detecting and concentration prediction.Qualitative analysis on the THz spectrum of the six kinds of tea by the improved support vector machine and the least squares support vector machine.Based on the successfully qualitative identification of the THz spectrum of the six kinds of tea by the traditional support vector machine and least squares support vector machine,To further reduce the identification error,take the particle swarm and genetic algorithm respectively to optimize parameters,combine the LS-SVM model and get the more precise result.To reduce the dimension on the THz spectrum of the primary tea by partial least squares,then combine the SVM and LS-SVM to discern the tea spectrum.The performance is better than the traditional support vector machine and least squares support vector machine.Integrate Laplace popular algorithms with support vector machine to classify a large number of unlabeled samples.Compare to the traditional support vector machine,the improved one has a higher degree of accuracy.The result provides new and effective ways to the classification of tea's THz spectrum,meanwhile it provides THz-TDS experiments new ways in the fields of food inspection and pharmaceutical quality control.