PPI selects endmembers by projecting n-dimensional data on a a straight line that is constrained to pass through the data mean, and selecting extrema.
SAM determines spectral uniqueness based on the angular distance between an endmember and a trial pixel relative to the origin (zero reflectance). That means that the SAM vectors for two PPI endmembers, one dark and one bright) could be identical. Has anyone explored the implications of this mismatch?
|