Rough Sets based Measures for the Attribute Uncertainty of Classified Remotely Sensed Imagery

摘要

There are two fashions to measure the attribute uncertainty of classified remotely sensed imagery. The first fashion is based on the categorical scale, for example error matrix, kappa coefficient and etc. These measures of the first fashion can assess the uncertainty of each classified category but ignore the spatial distribution of uncertainty within each category. The second fashion is based on the pixel scale, for example probability vector, probability Shannon entropy and etc. Unlike those measures in the first fashion, measures in the second fashion can give the spatial distribution and variation of uncertainty but have no ability to provide the uncertainty index on the category scale. In this paper, two measures derived from rough sets theory are proposed to assess the uncertainty of classified data. These two measures not only consider the roughness within the each category but also offer the uncertainty index on the category scale.