Abstract
Terahertz time-domain spectroscopy (THz-TDS) systems are widely used to obtain fingerprint spectra of many different biomedical substances, and thus the identification of different biological materials, medicines, or dangerous chemicals can be realized. However, the spectral data for the same substance obtained from different THz-TDS systems may have distinct differences because of differences in system errors and data processing methods, which leads to misclassification and errors in identification. To realize the exact and fast identification of substances, spectral standardization is the key issue. In this paper, we present detailed disposal methods and execution processes for the spectral standardization and substance identification, including feature extracting, database searching, and fingerprint spectrum matching of unknown substances. Here, we take twelve biomedical compounds including different biological materials, medicines, or dangerous chemicals as examples. These compounds were analyzed by two different THz-TDS systems, one of which is a commercial product and the other is our verification platform. The original spectra from two systems showed obvious differences in their curve shapes and amplitudes. After wavelet transform, cubic spline interpolation, and support vector machine (SVM) classification with an appropriate kernel function, the spectra from two systems can be standardized, and the recognition rate of qualitative identification can be up to 99.17%.
Highlights
Terahertz (THz) spectroscopy has a wide range of applications and has become an important research topic in recent years
It can be seen clearly that the spectra obtained from the two systems showed significant differences in their curve shapes and amplitudes, especially for amoxicillin, phenylalanine, p-toluylic acid, trinitrotoluene, and riboflavin. e main reason is that the USST system just gives out the original time-domain waveform, without any data processing, while the Gaojing system provides the smoothed data after a series of data preprocessing, whose details are unknown
E only misclassification was for phenylalanine. is could be attributed to the overfitting from the radial basis function (RBF) kernel function and the proximity of the absorption peak of phenylalanine (1.27 THz) to that of glutamic acid (1.25 THz). e proximity of these peaks was more problematic as these compounds only have one absorption peak in the 0.2–1.5 THz range
Summary
Terahertz (THz) spectroscopy has a wide range of applications and has become an important research topic in recent years. The vibrational and rotational frequencies of most polar molecules are located in the THz range, which can provide specific absorption responses (fingerprint spectra) for the identification of compounds [4, 5]. THz technology has been widely applied in the identification of many compounds and has been proven to have high recognition accuracy [7,8,9]. THz spectra of dinitrotoluene show large differences in the amplitudes and Scientific Programming frequency domain, which are caused by variations in system errors and data processing methods for different measuring systems [17, 18]. THz spectra of dinitrotoluene show large differences in the amplitudes and Scientific Programming frequency domain, which are caused by variations in system errors and data processing methods for different measuring systems [17, 18]. is may cause big errors in the database built and the substance identification. erefore, a standardization method for spectral acquisition and data processing is urgently needed
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
