Adding probabilities to timed automata to improve landcover classification

The aim of this paper is to propose the use of a dynamic plot model to improve landcover classification on a sequence of images. This new approach consists in representing the plot as a dynamic system and in modeling its evolution (knowledge about crop cycles) using the timed automata formalism. In order to refine the results obtained by a traditional classifier, the observations given by a preliminary classification of images are combined with the expected states provided by simulation. The paper presents the general method, based on prediction and postdiction mechanisms, that have been adopted to improve the classification of a sequence of images. In a first experiment, plots are modelled using a timed automata formalism. The analysis of the results clearly demonstrates the interest of the method. A second experiment, in which plots are modelled using probabilistic timed automata, is described in a last section. The changes in the method are presented and the results are compared with the previous ones.