Study On Fast Classification Of Terahertz Spectrum For Traditional Chinese Medicine

In this paper,the practical technology of terahertz material detection will be researched;the common terahertz spectra of traditional Chinese medicine will be analysed;the performance of fast unsupervised learning algorithm in the field of traditional Chinese medicine detection will be discussed;a method of non-distorted rotation delay line will be given.The main contents of this paper are as follows:1.The calculation of the material parameters of the transmission terahertz time-domain spectroscopy system is deduced;the influence of reflection peak signal on frequency domain signal in terahertz time-domain pulse signal is analysed;the necessity of eliminating the reflection peak signal in time-domain signal is discussed.An introduction of the applications and disadvantages of supervised algorithm in THz detection is given.The advantages and the applications prospect of the non-supervised algorithm in THz detection are proposed.A detailed introduction of the principles of three kinds of hard clustering K-means,K-medoids and FCM algorithm is given.2.Detect the THz absorption coefficients of dozens of Traditional Chinese Medicine like Sanchi,Angelica Sinensis and Amur Corktree Bark.To amplify the characteristics,the paper proposes use multidimensional features instead of absorption coefficient as characteristic data.Present ACSUM method to improve the traditional clustering algorithms.Use Sanchi and its easily-confused Chinese Medicine Curcumae Rhizoma,Angelica Sinensis and its easily-confused Chinese Medicine Angelica Pubescens as samples to carry out experiments.Finally,in the experiment,the accuracy of sample clustering was 97.34%.3.The THz spectral analysis and recognition software of traditional Chinese medicine are written by Java Swing.The software has two basic modules including data processing and spectral recognition,and has four modules,including the spectrum calculation,the absorption coefficient calculation,the classification of traditional Chinese medicine and the clustering of traditional Chinese medicine.K-means,K-medoids and FCM are integrated in the spectral recognition module,and samples can be classified successfully,and a classification report will be produced.Finally,testing the software with two Chinese medicine,hou-po and Ligusticum chuanxiong Hort,achieves ideal goal.4.To deal with the drawback of slow scanning speed of the delay line using the translation stage in the laboratory,a rapid non-distorted rotation delay line is designed;the inherent defects of the special curved surface delay line and the advantage of the non-distorted delay line are discussed;the principle of the non-distorted and fast delay line is introduced;the optical path and the critical angle of the non-distorted delay line are deduced;the parameters of the delay line are optimized.