Increased accuracy multiband urban classification using a neuro-fuzzy classifier

This letter presents an improvement of an already proposed neural classifier, designed to exploit multiband data over urban environments. The original classifier, based on an Adaptive Resonance Theory (ART) network followed by a fuzzy clustering step, is here improved by directly using a neurofuzzy approach, the fuzzy ARTMAP neural network. We show that significant advantages in the classifications could be obtained by tuning the fuzzy ARTMAP learning parameters. Overall accuracy has increased on the same dataset of aerial and Synthetic Aperture Radar (SAR) images of the original work. Moreover, the proposed change in the original classifier structure reduces the implementation complexity and increases its capability to adapt to new inputs. To demonstrate the robustness of this new approach, we offer results on a multiband AIRSAR dataset (C-, P- and L-band images) over the urban area of Broni, northern Italy.