Natural numerical networks for Natura 2000 habitats classification by satellite images

Natural numerical networks are introduced as a new classification algorithm based on the solution of nonlinear partial differential equations forward-backward diffusion type complete graphs. The proposed natural network is applied to open important environmental and nature conservation task, automated identification protected habitats by using satellite images. In network, forward causes movement points in feature space toward each other. opposite effect, keeping away from other, caused backward diffusion. This yields desired classification. contains few parameters that optimized learning phase method. After optimizing topology graph, necessary for habitat performed. A relevancy map tool validating finding Natura 2000 appearances.