RT - Journal Article T1 - A band selection technique for optimized hyperspectral unmixing JF - kntu-jgit YR - 2018 JO - kntu-jgit VO - 6 IS - 3 UR - http://jgit.kntu.ac.ir/article-1-619-en.html SP - 93 EP - 114 K1 - Hyperspectral Images K1 - Unmixing K1 - Band selection K1 - Spectral Variability K1 - Similarity Measures. AB - Linear spectral mixture analysis (SMA) has been used extensively in remote sensing studies to estimate the sub pixel composition of spectral mixtures. The mathematical solution of the mixing problem is to resolve a set of linear equations using least squares approaches. The lack of ability to account for temporal and spatial variability between and among endmembers has been acknowledged as a major shortcome of conventional SMA approaches applying a linear mixture model using a set of fixed endmembers. Also, if endmembers are highly correlated, the matrix will become non-orthogonal, the inversion will be unstable and the inverse or estimated fractions will become highly sensitive to random errors (e.g., noise). In this paper, we present a new band selection method that comprises a band prioritization and a band de-correlation. The band prioritization will prioritizes all bands according to the reduced spectral variability of endmembers which will be used for unmixing. Bands are then selected on the basis of their associated priorities. Since the band prioritization does not consider as spectral correlation, a band de-correlation using the angles between bands are being applied to de-correlate prioritized bands. It is shown that the proposed band selection method effectively eliminates a great number of insignificant bands. Surprisingly, the experimental results on real and synthetic data sets show that with a proper band selection less than 0.2 of the total number of bands can achieve comparable performance using all bands. LA eng UL http://jgit.kntu.ac.ir/article-1-619-en.html M3 10.29252/jgit.6.3.101 ER -