Terahertz spectral material recognition in online postal packages based on generative adversarial networks

Terahertz technology is the only technology that can achieve specific identification of hazardous chemicals, drugs, explosives, and other contraband inside mail packages without opening them. Therefore, overcoming the challenges of terahertz postal security inspection technology has broad market application prospects and extremely important social significance. In recent years, the combination of terahertz time-domain spectroscopy and deep learning has been widely applied in the field of material identification. However, in practical applications, based on the characteristic absorption spectra in the frequency range, the terahertz absorption spectra of amino acids vary with different packaging materials. Substance classification algorithms based on deep learning and machine learning show high accuracy in offline data models but lower accuracy during real-time online detection. Real-time detection of online amino acid samples based on terahertz time-domain spectroscopy technology should fundamentally solve these issues by increasing the training data, i.e., generating more data from the raw data. Generative adversarial networks (GANs) are a type of deep learning model that can learn the complex distribution of raw data. However, in the field of terahertz material identification, GANs have rarely been used to generate data to improve classifier performance. Therefore, this paper proposes a data augmentation method based on GANs. Then, a terahertz spectrum classification technique combining decision tree (DT), support vector machine (SVM), and convolutional neural network (CNN) is used to identify terahertz spectra within packages.