An approach to model multitemporal knowledge in automatic classification process of remote sensors images
H. L. da C. Coutinho, Marco Aurélio C. Pacheco, Raul Queiroz Feitosa, Guilherme Lúcio Abelha Mota, Vanessa de Oliveira Campos · DOAJ (DOAJ: Directory of Open Access Journals) · 2005
The present work proposes a methodology to model multitemporal knowledge within an automatic image interpretation framework. The method uses state transition diagrams to express the possibilities of class changes within a given time span. The change possibilities are estimated upon data of the same region acquired at a previous date by using genetic algorithms. Classification results obtained from multitemporal reasoning are combined to pure spectral classification by using fuzzy logic techniques. The proposed approach was evaluated using LANDSAT 7 images of three consecutive years of an area in southwest Brazil showing encouraging results. The multitemporal knowledge always improved the classification performance in comparison to a pure spectral classification. In addition, the genetic algorithm was in all cases superior to the manual trial and error alternative for the parameter estimation of the multitemporal model.