Both approaches try to deal with non-linearities in the problem. However,
the linear spectral unmixing with higher moments tries to deal with the
non-linearities that arise due to the intra-class variability, while the
non-linear spectral unmixing tries to deal with non-linearities that arise
due to the nature of the physical processes involved. Ideally both approaches
should be combined. For a comprehensive review on both approaches see M.
Petrou, Chapter 4 in the book ``Information Processing for Mixed pixel
Classification in Remote Sensing'', Remote Sensing (code 3999), C H Chen,
editor, World Scientific, ISBN 981-02-3737-5, pp 69-83.
|