An approach to urban surface features identification using Pushbroom Hyperspectral Imager

The hyperspectral data can identify the complicated urban surface features which can´t be identified using the multispectral data because of their lots of continuous bands in a narrow band width. The goal of this paper was to perform the urban surface features identification based on airborne Pushbroom Hyperspectral Imager (PHI) data. At first, some preprocessing of PHI data from original data, such as atmospheric correction and geometry calibration has been done. Then, the endmembers were located and identified using the n-d Visualizer on the pixels determined from pixel purify index (PPI) which was run on the Minimum Noise Fraction (MNF) transform result. The most crucial step is the matching between the target spectra and the reference spectra. In this paper, four matching algorithms were introduced and analyzed such as the spectral angle measure (SAM), the spectral correlation measure (SCM), the Euclidean distance measure(ED) and the spectral information divergence (SID).SAM and SID were used and compared in the target area. At last, the validation of the identification results was done and the error of the SID method was analyzed. The result indicated that water, vegetation and buildings can be identified exactly and the asphalt road can´t be distinguished from the buildings because their spectra are similar in PHI image. To be concluded, SID is a more effective algorithm for the spectral matching than SAM in that it can identify the minute difference of the surface features and there are too many isolated pixels in the SAM algorithm.