Fuzzy classification techniques in the urban area recognition

Abstract

Determination of scattered urban areas in a very heterogeneous environment can prove to be quite difficult using conventional classification techniques of remotely sensed images. On the other hand, fuzzy logic methods enable this difficulty to be overcome by assigning one pixel to more than one class according to a membership grade, determined using a pre-defined function. In this study, urban areas have been classified using fuzzy logic methods. The analysis was performed on a Landsat TM sub-scene (800/spl times/600 pixels) acquired over the province of Catanzaro (Calabria, Italy). The intrinsic characteristics of the ground coverage, as well as the rough topography, contribute to make this area a very heterogeneous one. The image was classified using a fuzzy parallelepiped classifier and membership values, associated to each pixel, were calculated. For each pixel, the classes, which contributed the most, were kept for the determination of the final pixel assignment. Global accuracy of fuzzy classification, estimated on mixed test area (chosen during a 2/sup nd/ ground truth campaign) reached a level of 0.75 areas were identified analysing the images that represent the combinations of Urban class with the other classes. The fuzzy classification results were compared to image classified using the traditional techniques, minimum distance and maximum likelihood. In terms of global accuracy, fuzzy technique appeared to be more accurate than conventional techniques.