Use of Image Endmember Libraries for Multi-Sensor, Multi-Scale, and Multi-Site Mapping of Urban Areas

Abstract

Building libraries of reference spectra for detailed mapping of urban areas at the level of building materials or plant species requires substantial effort. While in the last 15 years many approaches have been proposed to automatically extract pure material spectra from airborne hyperspectral imagery, the labeling of such spectra remains a tedious task. An interesting question, therefore, is to what extent the effort of building a library of reference spectra for a specific mapping task might be reduced by the re-use of image spectra collected from other imagery, covering multiple urban sites. In this chapter, we focus on methods for building multi-site libraries of reference spectra and for using these spectra in different urban mapping applications, and on the potential of generalized mapping models based on such libraries. We introduce the idea of a Generic Urban Library (GUL) offering users of remote sensing data the opportunity to share reference spectra, and to use spectra collected by others in their applications. Through specific case studies, we demonstrate the merits of sharing multi-site reference spectra using library-based mapping approaches. Building (more) generic urban libraries can be considered an important step in facilitating and fostering research on the transferability of urban mapping methods.