Change Detection by Fusion/Contextual Classification based on a Hybrid DSmT Model and ICM with Constraints

The Dezert Smarandache Theory (DSmT) used for the fusion and the modeling of the classes sets of themes has shown its performances in the detection and the cartography of the changes. Moreover the contextual classification with the research for the optimal solution by an ICM (Iterated conditional mode) algorithm with constraints allows to take in account the parcellary aspect of the thematic classes, thus, the introduction of this contextual information in the fusion process has enabled us to better identify the topics of surface and the detection of the changes. The objective of this work is, in the first place, the integration in a fusion process using hybrid DSmT model, both, the contextual information obtained from a supervised ICM classification with constraints and the temporal information with the use of two images taken at two different dates. Secondly, we have proposed a new decision rule based on the DSmP transformation to overcome the inherent limitations of the decision rules thus use the maximum of generalized belief functions. The approach is evaluated on two LANDSAT ETM+ images, the results are promising.