Naderan S. Automatic building extraction from high-resolution satellite imagery using artificial neural networks

Object detection and extraction from high-resolution satellite imagery has becoming an important research topic in the field of photogrammetry and remote sensing. Geospatial data like buildings is one of the most critical feed of a GIS database. The dissertation is devoted to the development of methods and techniques for Automatic building extraction from high-resolution satellite imagery, aimed at enhancement of efficiency of object recognition and image segmentation. Analysis of satellite imagery is regarded in the consideration of an object-oriented approach and consists of the stage of object segmentation with subsequent recognition. The approach presented an image segmentation approach based on combination of image segmentation techniques and fuzzy c-means clustering and morphological image processing. The study proposes a neuro-fuzzy classifier that provides ranking of possible results by preferences and to increases recognition efficiency. In addition, the approach presents a genetic algorithm for learning purpose with using a Gaussian function as membership function. The experiment performed has shown the strength of the neuro-fuzzy classifier to recognize buildings from satellite imageries.