We have developed novel techniques that utilize both spatial and spectral information for classifying pixels in hyperspectral images. Our methods demonstrated superiority against state-of-the-art algorithms on benchmark hyperspectral data sets with very few (like 10) training labels from each class.
The capability of identifying detailed classes of pixels is limited by spectral and spatial resolution of hyperspectral images.
Our method gives the best performance overall in accuracy even with a very small set of labeled pixels.
Especially, the gain in accuracy with respect to other state-of-the-art algorithms increases when the number of labeled pixels decreases, and therefore our method is more advantageous to be applied to problems with small training sets.
It is of great practical significance since expert annotations are often expensive and difficult to collect.
