Bernard-Lunzer, John

Profile Picture

Filters

20131
Reset filters

Settings

Citations

Publication Search Results

Now showing 1 - 1 of 1
  • Thumbnail Image
    Publication
    Analysis of spectral variability in hyperspectral imagery with application to biodiversity estimation
    (2013) Bernard-Lunzer, John; Hunt, Shawn D.; College of Engineering; Velez-Reyes, Miguel; Jimenez, Luis; Department of Electrical and Computer Engineering; Rivera, Rosita
    This work presents a newly developed destriping algorithm for Hyperion hyperspectral imagery as well as methods to detect spectral homogeneity within hyperspectral images. Hyperion has been in operation for over a decade and provides a vast archive of hyperspectral data. The Hyperion data, however, suffers from several issues that degrade its quality. One of the most severe issues is strong vertical striping. Current available destriping algorithms for Hyperion do not provide satisfactory results, leaving behind newly generated stripes from erroneous stripe estimation. A new method called Local Median Removal is proposed to remove existing stripes introducing minimal new artifacts. The high spectral resolution of hyperspectral data can in theory, according to the spectral variation hypothesis, allow inference about the spatial heterogeneity of species of remotely sensed objects. A model of the spectra is proposed that shows that the mean squared amplitude of a group of spectra in an individual band is proportional to the variance of the spectra if all the spectra come from the same object. Statistical testing methods are proposed that will test a group of spectra, in which each band has been normalized by that bands mean squared amplitude, to determine if those spectra are all from a single object. The proposed methods were tested on synthetic and real-world data and show that they are able to detect class homogeneity even when the spectra tested come from two different objects with very similar spectra.