The Kestrel Corporation visible-near IR band (525 to 1016 nm) airborne Fourier Transform Hyperspectral Imager was modified to include measurement of the polarization characteristics of several ground cover classes. The polarization contrast of typical terrestrial background and target objects was characterized. First, the t statistic was used as an index of class separation to determine whether polarized images were more useful for discriminating several cover classes than unpolarized images. Second, the information present in polarized images which is not present in unpolarized images was identified and described. This was done by regressing polarized and unpolarized images, generating images of predicted values for the polarized images using the regression coefficients, generating images of residuals by subtracting the actual values from the predicted values, and analyzing the statistical separation of cover classes in the residual images. A single polarized image was not more useful for identifying the cover classes than an unpolarized image. A residual image derived from a single polarized image and an unpolarized image provided a mean maximum statistical separation of t $EQ 18.3 for all cover class combinations. The sum of two orthogonal polarized images provided slightly greater separation, with a mean maximum separation of t $EQ 23.7.
展开▼
