To deal with spectrum design for decoys under hyperspectral detection condition, the decoy spectrum design principles were analyzed in term of the feature significance. Different from other detection technologies, hyperspectral detection is a process of multi-dimensional data processing, and the target significance is embodied in feature space formed by dimensionality reduced data. Correspondingly, the proposed spectrum design method for decoys is as follows. The decoy is set in feature space, highlighting itself and shielding real target, and then the spectrum of the decoy is reduced by the relative relation between the decoy and other targets in feature space. The reduction is based on the linear relation between the feature space axis and the original data channels. Target detection results verified the effectiveness of the proposed decoy spectrum design method.