Salient target detection in hyperspectral images using spectral saliency

The researches on saliency detection for common images have been thoroughly explored in recent years, but little attention has been paid to hyperspectral images. However, salient region detection can contribute to finding significant targets in hyperspectral images. In this paper, spectral saliency is introduced to the spectral domain for detecting targets in hyperspectral images, which are computed by expanding the Itti's visual attention model. Firstly, spatial and spectral features are extracted from hyperspectral image. The conspicuity feature of the first three principal components, orientations, spectral angle and visible spectral band opponent are computed to sum into a saliency map. Finally, salient targets are detected by winner-take-all strategy finding the region with the highest value in the saliency map. Experimental results show that the proposed method have better effectiveness of target detection compared with the method based on RGB image and visible spectrum.