Urban objects recognition feasibilities by airborne hyperspectral and multispectral remote sensing
Abstract
This paper explores the recognition uncertainty of urban objects by multiband imagery. The
purpose is to recognize the urban objects by their spectral signature, using an external spectral
library. Two Vis-NIR images were used for the study: a four bands Kompsat-2 multispectral
image and a 16 bands Ricola‘s airborne hyperspectral image, two supervised classifiers were
tested; a spectral based classifier, called the Spectral Angle Mapper (SAM), coupled to an external
spectral library and a machine learning based classifier called the Support Vector Machine
(SVM), in a second step the classification results obtained by the two classifiers were merged, the
goal was to take advantage of both techniques, to optimize the classification result. The classifiers
performance and the objects recognition feasibility were discussed for both images.