Morphological operators applied to X-band SAR for urban land use classification

This study provides an assessment of the potential for using contextual information with TerraSAR-X backscattering images in classifying urban land-use. Due to the lack of multi-frequency data, a contextual analysis was carried out to extract geometrical information of objects/classes within the images. Anisotropic morphological filters were applied to the backscattering image using a multi-scale approach. A range of different spatial domains were investigated by neural network pruning. The final map of land-use composed of seven different classes of interest was obtained using a Multi-Layer Perceptron neural network with an accuracy of 0.91 in terms of K-coefficient.