| Laser-induced breakdown spectroscopy (LIBS) has seen significant attention in recent years, in part because of several unique characteristics that distinguish it from other techniques for atomic emission spectroscopy. As a technology capable of fielded, portable deployment, it is possible to take analytical chemistry to the field, which may serve in a variety of applications such as industrial monitoring, geological surveys and hazard detection. The use of LIBS in a variety of material applications has been on the rise in recent years, however, in order for LIBS to successfully transition into the field, the sensor must be paired with appropriate algorithm for accurate and robust processing.;In this research dissertation, the result of testing two classification algorithms on eight LIBS datasets is reported. The results suggest that the standard cross validation techniques may not accurately estimate generalization performance and a proposed "Leave-One-Sample-Out (LOSO)" approach to experimental design for LIBS classifier validation may provide a more robust measure of performance. In another study focused on building a robust multi class classifier for LIBS, three modifications of the partial least square discriminant analysis (PLSDA) classifier were used to test six distinct LIBS datasets with different number of classes. The results show that the pairwise PLSDA classification scheme performed better than the traditional M-ary PLSDA classification scheme and the One-against-all PLSDA classification scheme especially on datasets with large number of classes.;The presence of contaminants in a LIBS spectral measurement can significantly degrade the generalization performance of classifier for LIBS. A proposed technique known as "Localized In-Sample Tunable Extreme-value Remover (LISTER)" is capable of removing these contaminants in a multivariate data, specifically LIBS spectral measurement. Removing the contaminated observations from the "contaminated" LIBS datasets is necessary to achieve robust validation for classifier for LIBS. The implementation of the recommendations and findings of this research work will lead to a reliable, methodical and automated process of using LIBS for elemental and material identification and ultimately enhance the possibility of LIBS becoming a fielded technology. |
