Research On Substance Detection Method Based On Terahertz Time-Domain Spectroscopy

With the development of ultrafast lasers,the terahertz radiation technology has gradually become the most active branch of scientific research.This technology obtains the optical parameters of the samples pending to be tested by measuring their terahertz spectrums.The study of the optical parameters helps to detect the composition of the substance.However,the application of this technology in the fields of food and medicine is still immature,and the accuracy of the test results needs improvement.In order to further explore the application of terahertz time-domain spectroscopy in substance detection and improve the accuracy of substance detection results,this thesis proposes to measure the terahertz spectra of tartaric acid,lactose monohydrate,talc,lotus root powder with the help of terahertz time-domain spectroscopy.The qualitative identification of substances and quantitative analysis methods of mixtures are studied.This thesis combines pattern recognition technology to achieve qualitative identification of substances and applies multiple regression algorithms to analyze the mixtures quantitatively.The research scope and findings are summarized below:1.This thesis does a detailed analysis of the generation and detection mechanism of terahertz waves,terahertz time-domain spectroscopy system,related machine learning algorithms and applicable scenarios.2.This thesis applies manual extraction of absorption peaks combined with support vector machine models to detect substances based on terahertz time-domain spectroscopy.Compared with the traditional hybrid model,principal component analysis combined with support vector machine,the proposed model has shorter training time and higher prediction accuracy,and the ability to identify multiple absorption peaks reaches 96%.Experimental results show that the model is more suitable for qualitative detection of substances compared with baselines.3.In order to improve the accuracy of substance detection,the paper proposes a material content regression model of gradient lifting regression algorithm combined with terahertz technology based on the manual extraction of absorption peaks and dimensionality reduction techniques.Compared with the three common regression models,least regression,partial least squares regression and support vector regression,terahertz time-domain spectroscopy combined with gradient lifting regression algorithm achieves the rms error less than 5%for quantitative analysis of the mixture and the error rate is reduced.The correlation coefficient of the prediction result is higher than 99.9%,and the fitting effect is the best.In addition,the new model is not only extremely accurate in the case of fewer data sets,but also causes little computation burden,like the three traditional models.As a result,the proposed model can be applied to food and drug detection based on terahertz time-domain spectroscopy technology in the future.