Publication

Spectrochimica Acta Part B: Atomic Spectroscopy 145, 1-7 (2018)
Signal preprocessing of deep-sea laser-induced plasma spectra for identification of pelletized hydrothermal deposits using Artificial Neural Networks

Author

Yoshino, S., Thornton, B., Takahashi, T., Takaya, Y. and Nozaki, T.

Abstract

This study investigates methods to analyze Laser-induced breakdown spectroscopy (LIBS) signals generated from water immersed deep-sea hydrothermal deposits irradiated by a long pulse (>100 ns) that are analyzed using Artificial Neural Networks (ANNs). ANNs require large amounts of training data to be effective. For this reason, we propose methods to preprocess full-field spectral signals into an appropriate form for ANNs artificially increase the amount of training data. The ANN was trained using a dataset of signals from immersed pelletized hydrothermal deposit samples that were preprocessed using the proposed method. The proposed method improved the accuracy of identification from 82.5% to 90.1% and significantly increased the speed of learning. The result shows that the ANN can be used to construct a generic method to identify hydrothermal deposits by long pulse underwater LIBS signals without the need for explicit peak detection.