Abstract: Prototype-based meta-learning has emerged as an effective approach for few-shot hyperspectral image classification (HSIC), but estimating deterministic prototypes directly from a limited ...