Visible-Near Infrared (VNIR) and Shortwave Infrared (SWIR) Spectral Variability of Urban Materials

The advent of relatively high spatial resolution hyperspectral imagery HSI provides a different perspective of the urban environment than lower spatial resolution hyperspectral data and either multispectral or panchromatic images. The objective of this thesis was to build and analyze a spectral library of urban materials and to understand how spectral variability affects the ability of classification algorithms to identify and discriminate various materials. The scope of the project was limited to non-vegetative impervious materials located on the Naval Postgraduate School campus. An airborne hyperspectral image, acquired September 30th 2011 was used for image-derived endmembers and a portable spectroradiometer was used to collect field spectra. Visual analysis of spectra was performed to assess intra- and inter-class variability and to identify spectral features and their causes. The spectral angle mapper SAM algorithm was used on the HSI data as a method to quantify intra-class spectral variability using a standard spectral angle. Classification maps were created with both SAM and mixture tuned matched filtering MTMF algorithms to determine how intra- and inter-class spectral variability affect the algorithm s ability to classify urban materials. The spatially complex nature of the urban environment negatively affected the performance of the SAM algorithm, but the ability to increase the spectral angle to account for materials with high spectral variability allowed improved interclass discrimination. The MTMF algorithm was better suited for intra-class discrimination of materials.