ASSESSMENT OF EXTENSIONAL UNCERTAINTY MODELED BY RANDOM SETS ON SEGMENTED OBJECTS FROM REMOTE SENSING IMAGES

摘要

A newly developed random set model has been applied to model the extensional uncertainty of a wetland patch. The objective of this research is to explore the corresponding variables collected on the ground for validating the uncertain image objects and to report the quality of the random set modeling. The independent samples t-test and a correlation analysis have been used to identify the main variables, whereas the overall accuracy and Kappa coefficients quantify the quality of the random set model. The results show that significant correlations exist among covering function, Carex coverage and NDVI. This suggests that the covering function of the random set can be quantified and interpreted adequately by NDVI derived from satellite images and Carex coverage measured in the field. In addition, the core-set of the random set has an overall accuracy of 85 percent and a Kappa value equal 0.54, being higher than the median set and support set. We conclude that the random sets modeling of uncertainty allows us to perform an adequate accuracy analysis.