Classification for Glucose and Lactose Terahertz Spectrums Based on SVM and DNN Methods

Abstract

In recent decades, terahertz (THz) radiation has been widely applied in many chemical and biomedical areas. Due to its ability to resolve the absorption features of many compounds noninvasively, it is a promising technique for chemical recognition of substances such as drugs or explosives. A key challenge for THz technology is to be able to accurately classify spectral measurements acquired in unknown complicated environments, rather than those from ideal laboratory conditions. Support vector machine (SVM) and deep neural networks (DNNs) are powerful and widely adopted approaches for complex classification with a high accuracy. In this article, we explore and apply the SVM and DNN methods for classifying the frequency spectra of glucose and lactose. We measured 372 groups of independent signals under different conditions to provide a sufficient training set. The classification accuracies achieved were 99% for the SVM method and 89.6% for the DNN method. These high classification accuracies demonstrate great potential in chemical recognition.