Analysis of spectra using processed data as training data in the neural network

Oba, Masaki 

As a method of analyzing multi-element spectral data obtained by LIBS, etc., we are constructing an analysis system using a neural network. More learning data is expected to improve accuracy, but it takes time and effort to prepare many actual samples. Therefore, the spectral data of GdO, TiO and ZrO were mixed on the data by changing the ratio to create 462 types of processed learning data, and the data were learned. After that, we analyzed the content ratio between each element of 62 kinds of data of real samples obtained by microwave LIBS measurement and examined its characteristics. As a result, the content ratio was obtained with a difference of about 10% from the true value.