Optimization of libs measurement parameters via multivariate chemometrics for the classification purpose
| Authors | |
|---|---|
| Year of publication | 2016 |
| Type | Appeared in Conference without Proceedings |
| MU Faculty or unit | |
| Citation | |
| Description | The outputs of LIBS analysis are multivariate data sets with several thousand to tens of thousands variables in one spectrum. Such a comprehensive set of information contained in a single spectrum offers a challenge for processing all at once, quickly and efficiently. Multivariate analysis makes it possible by reducing large files of the complex, multivariate data to a smaller number of factors describing the differences between the samples. Chemometrics algorithms have already been applied on LIBS data for classification or quantification purposes. When focusing on classification, papers published in the past few years confirm the interest in multivariate classification approach. The most used multivariate classification method is principal component analysis (PCA). In these cases, however, only for dimension reduction, since it is an unsupervised technique and is suitable for classification of only simple systems, outlier detection or for preliminary view on the dependencies between the samples in the model. In this work we are primarily aimed on the classification of geo-samples using LIBS and chemometrics, namely PCA, which was already used for this kind of analysis, and support vector machines (SVM). The key part of classification, however, is the setting of appropriate measurement conditions, because these may differ from conditions used for quantification purposes. The standard procedure is to choose the measurement parameters on the base of signal-to-noise ratio of selected lines. Laser-induced plasma (LIP), however, is very complex in its evolution and emission. That is why we suggest to set the measurement parameters by using overall information which one can get from the spectra. This can be achieved by applying an algorithm, for example PCA. This is the main goal and novelty of presented work, since optimization of measurement parameters for the classification purpose especially has not been solved yet in any previous publication. |
| Related projects: |