Accuracy Assessment of Thematic Data Using Fuzzy Sets and Inter-Class Spectral Distances

摘要

The increasing use of thematic classifications of remotely sensed data requires robust methods of assessing the accuracy of these classifications. The use of fuzzy set theory in accuracy assessment provides the analyst with extensive information about the accuracy of a classification. Determining the fuzzy class membership values which are used to create this information is done by the ground reference data interpreters. This leads to the potential of significant effects due to inter-interpreter variation if multiple interpreters are used, and furthermore is not a quantitatively based method. This paper presents a new method of determining fuzzy class memberships which is based on inter-class spectral distances. This method has the advantages of reducing inter-interpreter variation, having a quantitative base, and being reproducible. A simulated data set is used to introduce and to help explain the technique. It is expected that this method of assigning fuzzy class membership values will result in fuzzy accuracy assessments that give a more representative estimate of the accuracy of a thematic classification than is currently available.