Abstract
A recent class of hyperspectral anomaly detection methods that can be trainedonce on background datasets and then universally deployed -- without per-sceneretraining or parameter tuning -- has demonstrated remarkable efficiency androbustness. Building upon this paradigm, we focus on the integration ofspectral and spatial cues and introduce a novel "Rebellious Student" frameworkfor complementary feature learning. Unlike conventional teacher-studentparadigms driven by imitation, our method intentionally trains the spatialbranch to diverge from the spectral teacher, thereby learning complementaryspatial patterns that the teacher fails to capture. A two-stage learningstrategy is adopted: (1) a spectral enhancement network is first trained viareverse distillation to obtain robust background spectral representations; and(2) a spatial network -- the rebellious student -- is subsequently optimizedusing decorrelation losses that enforce feature orthogonality while maintainingreconstruction fidelity to avoid irrelevant noise. Once trained, the frameworkenhances both spectral and spatial background features, enabling parameter-freeand training-free anomaly detection when paired with conventional detectors.Extensive experiments on the HAD100 benchmark show substantial improvementsover several established baselines with minimal computational overhead,confirming the effectiveness and generality of the proposed complementarylearning paradigm. Our code is publicly available athttps://github.com/xjpp2016/FERS.