Determining uncertainties and their propagation in dynamic change detection based on classified remotely-sensed images

Abstract

This paper provides an approach to determine uncertainties and their propagation in dynamic change detection based on classified remotely-sensed images. First, the uncertainties of a classified image using maximum likelihood (ML) classification are determined. The probability vectors which are generated during the maximum likelihood classification are used as uncertainty indicators. Secondly, the uncertainty propagation of classified multi-date images is described using mathematical language for problem description. Based on this mathematical formulation, two techniques were used to calculate the uncertainty propagation. One is based on the product rule in probability theory and the other is based on a certainty factor model with probabilistic interpretation. Thirdly, a visualization technique, using 3-D and colour, was developed to present uncertainties.