[22p-P01-1~5] 3.7 Optical measurement, instrumentation, and sensor (formerly 3.8)

[22p-P01-2] Analysis of spectra with wavelength fluctuation using neural network

〇Masaki Oba¹ (1.JAEA)

Keywords:neural network, spectrum, wavelength fluctuation

Training data with a slightly shifted wavelength was created and trained, and the analysis performance were investigated using the similarly wavelength-shifted data as the test data. As a result, when learning with training data without wavelength shift and analyzing test data with wavelength shift, there is a large difference of about 50 % between the true value and the analysis value of the content of each element. It was confirmed that the difference between the true value and the analysis value was improved to about 20 % when training was performed using training data with a wavelength shift.