ON THE ESTIMATION OF SPATIAL-SPECTRAL MIXING WITH CLASSIFIER LIKELIHOOD FUNCTIONS

Conventional, hard classification algorithms that decide one class per pixel ignore the fact that many pixels in a remote sensing image repr esent a spatial average of spectral signatures from two or more surfac e categories. The mixing of signatures arises from the intrinsic, spat ially-mixed nature of most natural land cover categories, the physical continuum that may exist between discrete category labels, resampling for geometric rectification, and by the spatial integration defined b y the sensor's point spread function. By allowing for multiple classes per pixel, each with a relative membership likelihood, soft classific ation algorithms have the potential to ''unmix'' the pixel data into t he proportions of individual components. The potential and limitations of this approach are explored in this paper by empirical examples and analyses. A major conclusion is that the use of likelihood functions as estimators of mixing is valid for classes with high spectral signat ure separability, but problematic otherwise.