A Developed Algorithm for Automating the Multiple Bands Multiple Endmember Selection of Hyperion data Applied on Central of Cairo, Egypt

Abstract

This study attempts to provide an answer regarding the utility of Hyperion imagery in mapping urban settings in developed countries. The authors present a novel method for extracting quantitative land cover information at the sub-pixel level from hyperspectral or Hyperion imagery. The proposed method is based on the multiple endmember spectral mixture (MESMA) proposed by Roberts et al. (1998b), but extends it to handle the high-dimensional pixels characterizing hyperspectral images. The proposed method utilizes a multiband multiple endmember spectral mixture analysis (Multiband MESMA) model that allows for both spectral bands and endmembers to vary on a per-pixel basis across a hyperspectral image. The goal is to select an optimal subset of spectral bands that maximizes spectral separability among a candidate set of endmembers for a given pixel, and accordingly to minimize spectral confusion among modeled endmembers and increase the accuracy and physical representativeness of derived fractions for that pixel.

The authors develop a tool to automate this method and test its utility in a case study using a Hyperion image of Central Cairo, Egypt. The EO-1 Hyperion hyperspectral sensor is the only source of hyperspectral data currently available for Cairo, unlike cities in Europe and North America, where multiple sources of such data generally exist. The study scene represents a very heterogeneous landscape and has an ecological footprint of a complex range of interrelated socioeconomic, environmental and urban dynamics. The results of this study show that Hyperion data, with its rich spectral information, can help address some of the limitations in automated mapping that are reported by previous studies. For this, proper bands and endmembers are selected and used within a multiple endmember, with a multiple-band SMA process to determine the best Root Mean Square Error (RMSE) and abundance percentages. This results in a better mapping of land cover extricated from hyperspectral imagery (Hyperion).

Keywords: Spectral Mixture Analysis, Hyperspectral Data, Hyperion Data, Cairo, Egypt