Unmixing-Based PAN-Guided Fusion Network for Hyperspectral Imagery

The hyperspectral image (HSI) has been widely used in many applications due to its fruitful spectral information. However, the limitation of imaging sensors reduced spatial resolution that causes detail loss. One solution is fuse low (LR-HSI) and panchromatic (PAN) with inverse features get high-resolution (HR-HSI). Most existing fusion methods just focus on small ratios like 4 or 6, which might be impractical for some large ratios' HSI PAN pairs. Moreover, ill-posedness restoring information hundreds bands from only one band not solved effectively, especially under ratios. Therefore, a lightweight unmixing-based pan-guided network (Pgnet) proposed mitigate this improve performance significantly. Note process projected low-dimensional abundance subspace an extremely ratio 16. Furthermore, based linear nonlinear relationships between intensity abundance, interpretable inject (PDIN) designed details into feature efficiently. Comprehensive experiments simulated real datasets demonstrate superiority generality our method over several state-of-the-art (SOTA) qualitatively quantitatively (The codes pytorch paddle versions dataset could available at https://github.com/rs-lsl/Pgnet). (This improved version compared publication Tgrs modification deduction PDIN block.)