MORPHO-SPECTRAL RECOGNITION OF DENSE URBAN OBJECTS BY HYPERSPECTRAL IMAGERY

Abstract. This paper presents a methodology for recognizing, identifying and classifying built objects in dense urban areas, using a morphospectral approach applied to VNIR/SWIR hyperspectral image (HySpex). This methodology contains several image processing steps: Principal Components Analysis and Laplacian enhancement, Feature Extraction of segmented build-up objects, and supervised classification from a morpho-spectral database (i.e. spectral and morphometric attributes). The Feature Extraction toolbox automatically generates a vector map of segmented buildings and an urban object-oriented morphometric database which is merged with an independent spectral database of urban objects. Each build-up object is spectrally identified and morphologically characterized thanks to the built-in morpho-spectral database.