Cosine-Normalized Attention for Hyperspectral Image Classification

TL;DR

This paper introduces a cosine-normalized attention mechanism for hyperspectral image classification, emphasizing angular relationships to improve performance over traditional dot-product methods.

Contribution

It proposes a geometric, cosine-based attention formulation integrated into a Transformer, enhancing hyperspectral classification especially under limited supervision.

Findings

Consistently outperforms recent Transformer-based models on benchmark datasets.

Emphasizes angular relationships, reducing sensitivity to feature magnitude variations.

Provides a reliable inductive bias for hyperspectral data representation.

Abstract

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 introduces a cosine-normalized attention formulation that aligns similarity computation with the angular structure of hyperspectral signatures. By projecting query and key embeddings onto a unit hypersphere and applying a squared cosine similarity, the proposed method emphasizes angular relationships while reducing sensitivity to magnitude variations. The formulation is integrated into a spatial-spectral Transformer and evaluated under extremely limited supervision. Experiments on three benchmark datasets demonstrate that the proposed approach consistently achieves higher performance, outperforming several recent Transformer- and Mamba-based models despite using a lightweight backbone. In addition, a controlled analysis of multiple attention score functions shows that cosine-based scoring provides a reliable inductive bias for hyperspectral representation learning.