Physics-Informed Gaussian-Enforced Separated-Band Convolutional Conversion Network for Moving Object Satellite Image Conversion

Satellite imagery is crucial for monitoring global economic and ecological changes. However, integrating diverse datasets from different satellites is challenging due to differences in sensing technologies, lack of standardized calibration, and instrument hardware drift over time. Converting images from one sensor to another, like from WorldView-3 (WV) to SuperDove (SD), involves spectral band resampling and radiometric intensity scaling. A parametrized convolutional network approach has shown promise in non-linear conversion tasks across sensor domains, but it introduces artefacts when objects are in motion, due to temporal delays between multispectral band acquisitions. This results in spuriously blurred images of moving objects in the converted imagery. To resolve this, we propose an enhanced model, the Physics-Informed Gaussian-Enforced Separated-Band Convolutional Conversion Network (PIGESBCCN), which better accounts for spatial, spectral, and temporal correlations between bands.