Automating the Build-Up Process of Feature-Based Fuzzy Logic Models for the Identification of Urban Biotopes from Hyperspectral Remote Sensing Data

The automatic classification of urban biotopes from remote sensing (RS) data is not a task for a pixel-based classifier. All pixels of a biotope have to be taken into account during the classification of a biotope. Therefore, fuzzy logic models of biotope types are built with regard to the composition of the biotopes of different surface materials and their arrangement in the biotopes. The models consist of lists of numerical features and associated membership functions. The features have been calculated on RS data and are able to quantitatively assess characteristic differences between the biotopes of different types. There are not enough one-against-all features that separate one biotope type from all others. Therefore, the fully automated feature selection process aims at finding the set of features that separates two biotope types best with a pairwise maximum likelihood classification. The resulting list of features for the separation of two types is incorporated into the fuzzy logic models of these two types and serves there as one input variable. Thus, a model consists of n-1 input variables whereby n is the number of biotope types to distinguish. The application of the associated membership functions (Bayesian probability functions defined by means and covariances of training biotopes) on the input variables results in probability values between zero and one. The smallest of these probability values is taken as the crisp output value of a model and can be interpreted as a similarity value expressing the similarity of the classified biotope to the type of the model. The classification of the biotopes in the 14.5 km² test area with the developed models yields an overall accuracy of 87%.