Multivariate and multiway analysis of hyperspectral and fluorescence landscape data

Publication Type:
Issue Date:
Full metadata record
Files in This Item:
Filename Description Size
Thumbnail01Front.pdf8.93 MB
Adobe PDF
Thumbnail02Whole.pdf116.15 MB
Adobe PDF
The majority of the methods that have been used for the analysis of hyperspectral images have focussed on classification of spectra within an image, and there have been few examples of using whole hyperspectral images (one image per sample) for classification purposes. In this project, feature vectors were created to capture the heterogeneity present in each hyperspectral image for subsequent classification. All of the feature vectors methods were compared to single-point spectra and literature methods. The MIQ (the median- interquartile superspectrum) achieved the best classification of a set of heterogeneous powder mixtures, classifying 100% of the external test set. Mid-infrared (MIR) hyperspectral images and LA-ICP-MS elemental maps were collected of 'real-world' samples of soil collected from various locations around Sydney. For both the MIR and LA-ICP-MS data, the MIQ feature vector achieved superior classification compared to the averages alone (used to represent single-point spectra). For the Sydney soil samples, the classification performance of the LA-ICP-MS elemental maps was superior to MIR reflection hyperspectral images. For the LA-ICP-MS data, using elemental ratios for calibration proved to be a more suitable strategy than using silicon as an internal standard. A major application of PARAFAC has been the analysis of fluorescent landscapes. In the other main area of this project, novel applications of PARAFAC were investigated. The first application explored was the use of PARAFAC and fluorescent landscapes for the characterisation of phytoplankton. Algal species were placed under different wavelengths of light and monitored over time to determine their response to these environments. PARAFAC was able to isolate components that were identified as p-carotene, chlorophyll a and b, and chlorophyll a degradation products. The advantage of using PARAFAC for the characterisation of phytoplankton is that it can simultaneously determine all pigment concentrations without the need for prior separation or extraction. Background interference due to highly coloured / patterned and fluorescence remains an issue with many techniques of fingerprint visualisation. The second application of PARAFAC in this project was to analyse fluorescent landscapes constructed by collecting series of images of fluorescent fingerprints on these traditionally difficult backgrounds. Whilst PARAFAC successfully isolated the fluorescence due to the fingerprint, there was typically only a small reduction in background interference. However, it is expected that PARAFAC would be more successful for samples where the background is also fluorescent, such as DFO- developed prints on fluorescent yellow paper.
Please use this identifier to cite or link to this item: