A new spectral unmixing algorithm based on spectral information divergence

Spectral unmixing is a common problem in hyperspectral remote sensing, and it is a key issue of quantitative remote sensing. This article proposed a spectral unmixing algorithm based on spectral information divergence (SID) named SID-SMA. It could improve the precision of abundance estimation through choosing optimal endmember subset used in unmixing. SID-SMA adopted the idea of iteration and added the process of negative endmembers removing which could obviously reduce the computation complexity and improve the speed. Through the results of simulated data from spectral library, it could be seen that the correct proportion of endmember selection by SID-SMA was very high, arriving at 99.86% when the signal-to-noise ratio (SNR) was 100:1. From the point of abundance estimation errors, the algorithm presented here had lower value than two other methods. Especially, when the SNR was 100, the error was less than 0.05.