Transfer Learning for Neural Network Solution of an Inverse Problem in Optical Spectroscopy.

This study is devoted to inverse problems of optical spectroscopy that include qualitative and quantitative analysis of sample composition by its spectra. We consider the methods of optical absorption, IR absorption, and Raman spectroscopy for determining the concentrations of heavy metal ions in water. Since this problem has neither an analytical nor a direct numerical solution, practically the only way to solve it is by using approximation methods, including machine learning methods based on experimental data. Obtaining spectra is a laborious and expensive process, which makes it difficult to obtain a training sample of sufficient size. At the same time, experimental spectra are sensitive to impurities contained in water, including those of organic origin which are specific to each source. Therefore, it is not possible to create a universal training dataset with good representativity and providing a stable solution on any sample. Thus, in this paper, it is proposed to use the transfer learning approach to improve the quality of the neural network solution. The neural networks are pretrained on a basic dataset containing the spectra of solutions prepared in distilled water. Then, fine tuning and testing of the networks are carried out on specific datasets containing the spectra of solutions prepared in river water taken from different rivers. It is demonstrated that the transfer learning approach consistently provides the best results on the river water data. [ABSTRACT FROM AUTHOR]