A Variational Latent-Space Framework for Uncertainty-Aware Spectral Image Emulation
The framework uses a Variational Autoencoder (VAE) to learn a probabilistic latent representation of hyperspectral data for emulation. It supports both spectrum-level and spatio-spectral emulation by learning how biophysical parameters map to this latent space. Results show that fully convolutional models perform better on real imagery due to their ability to preserve spatial coherence.