Research On Terahertz Spectrum Recognition Based On WGAN And Transfer Learning

At present,combined with advanced deep learning methods,research on terahertz spectrum recognition technology based on large-scale spectrum database has become the focus of terahertz application technology.Considering that due to the influence of experimental conditions and experimental equipment,the data in the terahertz spectrum database usually suffers from incomplete data categories and mismatched data sizes.To this end,this paper has conducted the following research on the incompleteness and imbalance of the data in the terahertz spectrum database:The data balance of each category is the basis of machine learning,and the data collected in the terahertz spectrum database is usually accompanied by the problem of data size mismatch.In response to this problem,this paper proposes an unbalanced terahertz spectrum recognition method based on WGAN.As a new method of generating data,WGAN uses the generated data under the condition of Nash equilibrium to supplement the data set and make it reach the category equilibrium.The generated data can effectively map the real data distribution,and the accuracy of identifying unbalanced spectral data can be improved by mixing the generated data with the real data.The use of terahertz spectroscopy to identify substances largely depends on the spectral accumulation of known substances in the database,but due to the limitation of experimental conditions,it is impossible to collect a complete terahertz spectrum database.Aiming at the recognition of the unaccumulated spectral data in the terahertz database,this paper proposes a terahertz spectral recognition technology based on transfer learning.By using the model to learn the characteristics of similar substances and transferring them to the identification of unknown substances,the purpose of obtaining the spectral data characteristics of unknown substances is achieved,so as to solve the problem of the identification of unincorporated substances.