Register now After registration you will be able to apply for this opportunity online.
Multivariate analysis and visualization of hyperspectral data
In hyperspectral imaging a sample is raster-scanned through the focal point of a laser
beam and spectroscopic information is collected pixel by pixel. The resulting spectral
hypercubes contain thousands of individual spectra as well as the spatial coordinates
associated with each spectrum.
In hyperspectral imaging a sample is raster-scanned through the focal point of a laser beam and spectroscopic information is collected pixel by pixel. The resulting spectral hypercubes [Fig. 1(a)] contain thousands of individual spectra as well as the spatial coordinates associated with each spectrum. A univariate visual representation of such a dataset can be obtained simply by color-coding a specific spectral property (e.g., the intensity or position of a certain spectral band) and displaying this color for each pair of spatial coordinates in a map. However, potentially more information can be extracted from a spectral hypercube by employing multivariate methods using entire spectra, or large parts thereof, for image reconstruction [Fig. 1(b-e)]. This includes unsupervised learning methods such as dimension reduction techniques and cluster analysis. [2, 3]
The student will 1) familiarize him- or herself with a number of multivariate image reconstruction methods for the visualization and explorative analysis of hyperspectral data, and 2) investigate the applicability of said methods to pre-existing datasets obtained at the Photonics Laboratory from different samples of two-dimensional materials
using Raman- and photoluminescence spectroscopy.
References:
[1] https://hypers.readthedocs.io/en/latest/?badge=latest
[2] Gautam et al., EPJ Techniques and Instrumentation 2, 8 (2015)
[2] Miljkovi´c et al., Analyst 135, 8 (2010)
Prerequisites:
Experience or strong interest in data analysis (preferably in python).
In hyperspectral imaging a sample is raster-scanned through the focal point of a laser beam and spectroscopic information is collected pixel by pixel. The resulting spectral hypercubes [Fig. 1(a)] contain thousands of individual spectra as well as the spatial coordinates associated with each spectrum. A univariate visual representation of such a dataset can be obtained simply by color-coding a specific spectral property (e.g., the intensity or position of a certain spectral band) and displaying this color for each pair of spatial coordinates in a map. However, potentially more information can be extracted from a spectral hypercube by employing multivariate methods using entire spectra, or large parts thereof, for image reconstruction [Fig. 1(b-e)]. This includes unsupervised learning methods such as dimension reduction techniques and cluster analysis. [2, 3]
The student will 1) familiarize him- or herself with a number of multivariate image reconstruction methods for the visualization and explorative analysis of hyperspectral data, and 2) investigate the applicability of said methods to pre-existing datasets obtained at the Photonics Laboratory from different samples of two-dimensional materials
using Raman- and photoluminescence spectroscopy.
References:
[1] https://hypers.readthedocs.io/en/latest/?badge=latest
[2] Gautam et al., EPJ Techniques and Instrumentation 2, 8 (2015)
[2] Miljkovi´c et al., Analyst 135, 8 (2010)
Prerequisites:
Experience or strong interest in data analysis (preferably in python).
hyperspectral.pdf
