Multiscale pansharpening method combining spatial and frequency domain

Pansharpening is the process of creating high spatial resolution multispectral (HRMS) images by combining low spatial resolution multispectral (LRMS) images with high spatial resolution panchromatic (PAN) images. This study proposes a novel dual-domain multiscale network architecture to address the limitations of existing pansharpening methods, which predominantly focus on spatial-domain features while neglecting frequency-domain analysis. The framework comprises two complementary sub-networks, a spatial-domain feature fusion subnetwork and a frequency-domain feature fusion subnetwork, designed to integrate local spatial details and global frequency information. Specifically, the spatial-domain feature fusion sub-network incorporates an enhanced multiresolution analysis (MRA)-based module to refine the fusion of spatial information between PAN and LRMS images. Concurrently, the frequency-domain feature fusion sub-network introduces a Fourier transform-driven module to capture and combine frequency-domain characteristics. By this architecture leverages both high-frequency texture preservation and low-frequency spectral consistency, enabling the generation of high-quality pansharpening images with enhanced spatial and spectral fidelity.