Multi-population evolution strategies for structural image analysis


To identify objects in aerial images, a special structural approach, based on a blackboard system, is used. The reference objects are described with generic models and a set of real-valued parameters. To adapt these parameters in an automatical way a closed-loop system is proposed using multi-population Evolution Strategies with a special form of migration. The result of the parameter optimization is demonstrated with an example of identifying bridges in aerial images. Applying this closed-loop system a reduction of computational effort was achieved.
Füger H, Stein G, Stilla U (1994) Multi-populations evolution strategies for structural image analysis. IEEE Conference on Evolutionary Computation (ICEC 94), Orlando, Vol I, 229-234
[ Stilla.de/pub ]