Automatic identification of lines in vibrational-rotational spectra. Software based on a trainable neural network
A.D. Bykov, O.V. Naumenko, A.P. Shcherbakov
V.E. Zuev Institute of Atmospheric Optics of Siberian Branch of the Russian Academy of Science, Tomsk, Russia
Keywords: vibrational-rotational spectra, automatic identification, neural network, effective Hamiltonian, dipole moment
Abstract
This article presents an improved internet-accessible expert system, SLON, for analyzing high-resolution molecular spectra. It was developed in the Laboratory of Molecular Spectroscopy at Institute of Atmospheric Optics, Siberian Branch, Russian Academy of Sciences. The SLON expert system is based on a neural network model capable of making independent decisions when analyzing combination differences formed by groups of molecular transitions from different rotational sublevels of the ground state to the same excited vibrational-rotational state. The set of features by which the neural network distinguishes the correct variant of a combination difference from random realizations has been improved. Restrictions on the size of the analyzed spectrum have been removed. The format of the databases used is now universal and enables expanding the class of molecules under study. A modern multi-platform user interface allows this program to be compiled for Windows and Linux systems. The operating principles, operational experience, and prospects for the development of the created expert system are described.
|
