Evaluation of remote sensing images classifiers with uncertainty measures

Research output: Chapter in Book/Report/Conference proceedingChapter


Several methods exist to classify remote sensing images. Their appropriateness for a particular objective may be analysed by evaluating the classification accuracy with indices derived from the traditional confusion matrix, considering a sample of points within the regions used as testing sites. However, this analysis requires the existence of reference data corresponding to the ground truth for the used sample and its application is limited to a sample of points. This paper intends to show that, for classifiers from which uncertainty information may be obtained, the evaluation of the classifier performance can also be made with uncertainty indices. Two supervised classifiers are used, a Bayesian probabilistic classifier and a fuzzy classifier. Their performance is evaluated by using accuracy and uncertainty information and it is shown that similar conclusions may be obtained with both. Therefore, uncertainty indexes may be used, along with the possibility or probability distributions, as indicators of the classifiers performance, and may turn out to be very useful for operational applications.
Original languageUnknown
Title of host publicationSpatial Data Quality - From Process to Decisions
EditorsR Devillers, H Goodchild
Place of PublicationBoca Raton
PublisherCRC Press Taylor & Francis
ISBN (Print)9781439810125
Publication statusPublished - 1 Jan 2009

Cite this