Advanced spectral unmixing algorithm based on spectral information divergence

Spectral unmixing is a key issue of quantitative remote sensing. An advanced spectral unmixing algorithm based on per-pixel optimal endmembers selection named spectral information divergence-spectral mixed analysis (SID-SMA) was proposed. It determined the optimal endmembers subset using the criteria of SID and selected endmembers through two selection steps which could improve the precision of endmember selection and obtain small abundance estimation error. The results of simulated data from spectral library indicate that SID-SMA has better precision of endmember selection and abundance estimation. When the signal-to-noise ratio (SNR) is 100:1, the correct proportion of endmember selection arrives at 99.86% and total abundance error of 29 endmembers is less than 0.1 and the speed of SID-SMA is much faster.