A datafield — Spectral preprocessing endmember extraction module of hyperspectral unmixing for marine oil spills

Abstract

Current methods that use hyperspectral remote sensing imagery in the detection and extraction of marine oil spills are quite popular. Because the mixed pixels exist broadly in the hyperspectral images, the automatic and accurate end member extraction technology plays most essential role in the hyperspectral unmixing. This paper proposed a data field-spectral preprocessing module prior to end member extraction. We compared it with other existing preprocessing algorithms by different image size, different dataset, different end member number, different noises, and combined them with different end member extraction algorithms. It is proved that the proposed algorithm provides accurate end member detection, including anomaly end member, so it has greater accuracy, stronger noise resistance and saves more time. Especially the proposed method is effective in marine oil spill monitoring.