Abstract: Modeling discriminative spectral-spatial features is a key to improving hyperspectral image classification performance. However, existing methods cannot fully characterize the spatial ...