The spectral mixture concept allowed a new approach on image classification procedures. This approach comprehends three processing stages: (a) endmembers detection, (b) elements location and (c) quantification. The aim of this work was to develop a comparative analysis of endmembers identification from an automatic process using images with different spectral resolutions: AVIRIS, Landsat ETM+ and ASTER. Endmembers detection techniques were developed to hyperspectral images but are being used to multispectral sensors. Thus it is important to evaluate limitations for the use of lower spectral resolution images. The selected area is a natural Savanna region (locally named Cerrado) in Niquelândia (GO), Brazil. It was used a hyperespectral image from the AVIRIS sensor that was spectrally resampled to Landsat/ETM+ and ASTER sensors. The AVIRIS sensor developed by NASA (National Aeronautics and Space Administration) was the first imaging sensor to measure the solar reflected spectrum from 400 nm to 2500 nm at 10 nm intervals. Endmembers' automatic identification from the ENVI software comprehends the following steps: a) spectral reduction by the Minimum Noise Fraction (MNF) transformation, b) spatial reduction by the Pixel Purity Index (PPI) and c) manual identification of the members using the N-dimensional visualizer. The comparison among the three sensors was done using PPI images. In spite of PPI image differences it can be seen similarities among existing targets. That similarity is due to vegetation targets predominance on these images. These endmembers were relative to photosynthetic and non-photosynthetic vegetations.
endmember; pixel purity index; image processing; remote sensing; hyperspectral
