Comparison of Hyperspectral Unmixing Methods for Ship Detection on Airborne Hyperspectral Images

As marine traffic has increased, the importance of ship detection using remote sensing images has been emphasized. Especially, with a better performance for discrimination of target, the usage of hyperspectral data for marine surveillance has been increasing recently. In this study, we detected the vessels on airborne hyperspectral images and quantitatively analyzed the detection results. To obtain the airborne hyperspectral images and auxiliary data for the quantitative validation, the in-field airborne imaging experiment was carried out. In addition, four different end-member extraction techniques including N-FINDR, PPI, ICA, and VCA were applied for comparison of detection performance with hyperspectral unmixing methods. Detection results present significant differences by endmember extraction techniques. The N-FINDR and VCA techniques presented a total of 14 vessels, while the ICA technique detected seven vessels, and the PPI technique detected two vessels. The pixel-based probability of detection and false alarm ratio for all 14 ships were 98.83% and 4.30%, respectively. This study also addressed the important role of abundance fraction analysis for marine surveillance purpose.