Establishment And Application Of Drug Identification Platform Based On Terahertz Time-Domain Spectroscopy

THz-TDS(THz time-domain spectroscopy)uses the THz wave,an electromagnetic wave in the frequency range of 0.1～10 THz,which is between the infrared wave and the microwave.THz time-domain spectroscopy technology was originated in the 1980s.It was firstly used in telecom company AT&T,Bell Laboratory and IBM’s T.J.watson Laboratory.THz time-domain spectroscopy belongs to the coherent detection technology,and has been developing rapidly in the world in the 1990s and considered to be a promising new spectral measurement technology.Drug abuse is still an urgent problem to be solved in society,and drug testing is a long-term need for China’s social development.At pre sent,the detection of drugs is still lack of effective means.At this stag e,drug is often brought back to the laboratory and is detected by large-scale laboratory equipment.The portability and practicality is low.The vibrational rotational energy levels of a large number of drug molecules fall in the THz band,so the absorption spectrum of drugs in the THz band contains abundant physical and chemical information.Be cause of its low energy and strong penetrating power,THz can be used for nondestructive testing of drugs and is favored by people.The first part of work done in this paper is to use RBF(Radius Basis Function)neural network and SVM(support vector machine)neural network to train and recognize the THz fingerprint spectrum samples which has been processed for feature extraction by wavelet transform.There are 64 THz fingerprint spectrum samples of 9 kinds of drugs from different papers in total.The training and recognition has obtained good results,proved the feasibility of RBF neural network and SVM neural network in the identification of drug terahertz absorption spectrum in the aspect of application.The second part of work done in this paper is to use RBM(restricted Boltzmann machine)neural network and SVM(support vector machine)neural network to train and recognize the THz fingerprint spectrum samples which has been processed for feature extraction by binary encoding.There are 32 THz fingerprint spectrum samples of 3 kinds of drugs from different papers in total.The training and recognition has obtained good results,proved the feasibility of RBM neural network in the identification of drug terahertz absorption spectrum in the aspect of application.The third part of work done in this paper is,according to the RBF and SVM neural network trained in the first part work,a small program for terahertz drug fingerprint recognition is compiled in Matlab.It’s input is the terahertz absorption spectrum and the output is the drug types identified by the program.In addition,by Matlab and Java mixed programming,the identification process is released as a web service,deployed to the Tomcat server,so that users can access it through the network.