Classification of spectral signatures in biological aerosols

In this thesis multivariate methods were used to evaluate pretreatment methods, such as normalization, as well as classification possibilities of data collected with Laser Induced Breakdown Spectroscopy (LIBS). The LIBS system that FOI is currently developing for the purpose of classifying biological airborne threats was used to collect data from ten different samples in a laboratory environment. Principal component analysis (PCA) shows that it is possible to observe differences between samples using the two types of data acquired from the LIBS system, i.e., 2D CCD camera images and 1D spectra extracted from the image. Further results using partial least squares discriminant analysis (PLS-DA) show that normalization of the data only has visual effects in the PCA score-plots and do not affect the models predictability. Results also show that cropping and binning the pixels in the image is possible to some extent without losing significant predictability.