Identifying pure urban image spectra using a learning urban image spectral archive (LUISA)

Abstract

In this study a learning urban image spectral archive (LUISA) has been developed, that overcomes the issue of an incomplete spectral library and can be used to derive scene-specific pure material spectra. It consists of a well described starting spectral library (LUISA-A) and a tool to derive scene-based pure surface material spectra (LUISA-T). The concept is based on a three-stage approach: (1) Comparing hyperspectral image spectra with LUISA-A spectra to identify scene-specific pure materials, (2) extracting unknown pure spectra based on spatial and spectral metrics and (3) provides the framework to implement new surface material spectra into LUISA-A. The spectral comparison is based on several similarity measures, followed by an object- and spectral-based ruleset to optimize and categorize potentially new pure spectra. The results show that the majority of pure surface materials could be identified using LUISA-A. Unknown spectra are composed of mixed pixels and real pure surface materials which could be distinguished by LUISA-T.