SpectraMorph: Structured Latent Learning for Self-Supervised Hyperspectral Super-Resolution

Hyperspectral sensors capture dense spectra per pixel but suffer from low spatial resolution due to photon splitting and long dwell times, causing blurred boundaries and mixed-pixel effects. Co-registered companion sensors such as multispectral, RGB, or panchromatic cameras provide high-resolution spatial detail, motivating hyperspectral super-resolution (HSR) through HSI-MSI fusion. Existing deep learning based methods achieve strong performance but rely on opaque regressors that lack interpretability and often fail when the MSI has very few bands. We propose SpectraMorph, a physics-guided fusion framework with a structured latent space. Instead of direct regression, SpectraMorph enforces an unmixing bottleneck : endmember signatures are extracted from the low-resolution HSI, and a compact multilayer perceptron predicts abundance-like maps from the MSI. Spectra are reconstructed by linear mixing, with training performed in a self-supervised manner via the MSI sensor’s spectral response function. SpectraMorph produces interpretable intermediates, trains in under a minute, and remains robust even with single-band MSI. Experiments on three benchmark datasets (Washington DC Mall, Pavia Center, and Botswana) show consistent improvements over state-of-the-art baselines. Code for SpectraMorph is available at https://github.com/ritikgshah/SpectraMorph.