CFGFE: Coarse-Fine Grained Feature Enhancement Network for Pansharpening

Pansharpening seeks to enhance the spatial resolution of multispectral (MS) images by incorporating high-frequency spatial details from higher resolution panchromatic (PAN) images. However, most existing methods rely on single-granularity feature representations, which limits their capacity to comprehensively extract and fuse spectral and spatial information. We propose CFGFE, a coarse–fine grained feature enhancement network that addresses these limitations with a dual-branch architecture: a coarse-grained global modeling branch and a fine-grained frequency fusion branch. The former branch employs a wavelet-domain attention module to capture long-range dependencies and maintain global semantic consistency. In parallel, the latter branch decomposes both PAN and MS images into low- and high-frequency components and performs adaptive fusion via a frequency decomposition modulation module, enabling precise spectral–spatial integration. In addition, we introduce a differential feature coupling and enhancement module to jointly refine fine-grained frequency details and global semantic features, producing a more complete and accurate representation. Extensive experiments on multiple benchmark datasets demonstrate that CFGFE consistently surpasses state-of-the-art methods in both visual quality and quantitative metrics, confirming its effectiveness for high-quality pansharpening.