Classification of challenging Laser-Induced Breakdown Spectroscopy soil sample data - EMSLIBS contest

No Thumbnail Available
Date
2020-07-01
Authors
Vrábel, Jakub
Képeš, Erik
Duponchel, Ludovic
Motto-Ros, Vincent
Fabre, Cécile
Connemann, Sven
Schreckenberg, Frederik
Prasse, Paul
Riebe, Daniel
Junjuri, Rajendhar
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
We present results of the classification contest organized for the EMSLIBS 2019 conference. For this publication, we chose only the five best approaches and discussed their algorithm in detail. The main focus of the contest reflected both recent and long-term challenges of Laser-Induced Breakdown Spectroscopy (LIBS) data processing. The contest was designed with a purpose to raise a challenge in handling and processing a very large dataset, containing high-dimensional elemental spectra. For the contest, 138 samples were measured using a lab-based LIBS system. In total, the data set consisted of 70,000 spectra, separated into 12 classes according to their elemental composition. Due to its extensivity and complexity, the data set is unique within the LIBS community. The central idea was to simulate the so-called “out-of-sample” classification (i.e. according to similar elemental composition), implying various real-world applications. Even more, it reflects the current level of expertise in the LIBS community and the capability of the LIBS method itself.
Description
Keywords
Chemometrics, Classification benchmark, EMSLIBS contest, Laser-induced breakdown spectroscopy (LIBS), Machine learning
Citation
Spectrochimica Acta - Part B Atomic Spectroscopy. v.169