SOME ACCURACY ASSESSMENT METHODS FOR FUZZY CLASSIFICATION OF REMOTE SENSING DATA- A COMPARATIVE STUDY

Abstract

Satellite images are important source for thematic data. With the advent of developed space technology, satellite data of better spatial as well as spectral resolution are available. Extraction of thematic information from satellite images is generally achieved through the application of a conventional classification, which allocates each pixel to a land cover class. Such approaches are inappropriate for mixed pixels, which contain two or more land cover classes, and a fuzzy classification approach is required. The accuracy of such classification may be indicated by the way in which the strength of class membership is partitioned between the classes and how closely this represents the partitioning of class membership on the ground. However, the accuracy of the representation provided by a fuzzy classification is difficult to evaluate. The objective of this thesis work is to evaluate some accuracy assessment measures for fuzzy classification and to carry out a comparative evaluation of those measures. Measures such as log entropy measure, Euclidean distance measure, cross entropy and information closeness measure, similarity measure and fuzzy set theory based measures have been used by different authors to evaluate the accuracy of fuzzy classification. However, most of the measures for accuracy analysis deal with normalized fuzzy set of data. To deal with truly fuzzy data, some novel measures, such as fuzzy correlation coefficient measure, class based distance measure and class based Kaufmann entropy measure have also been included in this study to evaluate the accuracy of fuzzy and other sub-pixel classifications. It has been observed that the accuracy indexes obtained by using the new measures provide better information regarding the accuracy of each individual class.