A unified deep learning Model for despeckling in multi-modal polarimetric SAR data

Polarimetric Synthetic Aperture Radar (PolSAR) captures detailed information about land cover, vegetation, and built environments. However, coherent image acquisition introduces speckle noise, which reduces clarity and affects analysis. Traditional filters can smooth noise but often blur spatial details, while many deep learning approaches work only with a fixed number of polarization channels. We propose a polarization-agnostic deep learning framework for speckle suppression that works with single-, dual-, and fullpolarimetric data. The model preserves structural details and polarization coherence, providing clear, consistent, and physically meaningful SAR imagery for diverse remote sensing applications.