Analysis of hyper-spectral data derived from an imaging Fourier transform: A statistical perspective
Fourier transform spectrometers (FTS) using optical sensors are increasingly being used in various branches of science. Typically, a FTS generates a three-dimensional data cube with two spatial dimensions and one frequency/wavelength dimension. The number of frequency dimensions in such data cubes is generally very large, often in the hundreds, making data analytical procedures extremely complex. In the present report, the problem is viewed from a statistical perspective. A set of procedures based on the high degree of inter-channel correlation structure often present in such hyper-spectral data, has been identified and applied to an example data set of dimension 100 x 128 x 128 comprising 128 spectral bands. It is shown that in this case, the special eigen-structure of the correlation matrix has allowed the authors to extract just a few linear combinations of the channels (the significant principal vectors) that effectively contain almost all of the spectral information contained in the data set analyzed. This in turn, enables them to segment the objects in the given spatial frame using, in a parsimonious yet highly effective way, most of the information contained in the data set.
- Research Organization:
- Lawrence Livermore National Lab. (LLNL), Livermore, CA (United States)
- Sponsoring Organization:
- USDOE, Washington, DC (United States)
- DOE Contract Number:
- W-7405-ENG-48
- OSTI ID:
- 226081
- Report Number(s):
- UCRL-ID-123453; ON: DE96010037; TRN: AHC29610%%123
- Resource Relation:
- Other Information: PBD: 10 Jan 1996
- Country of Publication:
- United States
- Language:
- English
Similar Records
Tomographic spectral imaging: analysis of localized corrosion.
Quantitative Characterization of Hyper-Local Atmospheric Greenhouse Gas Sources