AI RESEARCH
Cosine-Normalized Attention for Hyperspectral Image Classification
arXiv CS.CV
•
ArXi:2604.01763v1 Announce Type: new Transformer-based methods have improved hyperspectral image classification (HSIC) by modeling long-range spatial-spectral dependencies; however, their attention mechanisms typically rely on dot-product similarity, which mixes feature magnitude and orientation and may be suboptimal for hyperspectral data. This work revisits attention scoring from a geometric perspective and