MSFFusion: Multi-Level Shallow Feature Fusion Network for Pansharpening

Pansharpening aims to reconstruct high-resolution multispectral (HRMS) images by integrating the spectral information of low-resolution multispectral (LRMS) images and the spatial details of panchromatic (PAN) images. Although deep learning-based methods have achieved significant progress, they often fail to sufficiently exploit shallow features during feature fusion. Since shallow features contain important edges, textures, and local structures, inadequate selection and integration of these features may weaken spatial detail reconstruction and affect spectral fidelity. To address this problem, we propose a multi-level shallow feature fusion network (MSFFusion) that effectively exploits shallow features and models cross-level and multi-scale feature interactions. Specifically, MSFFusion employs a multi-scale architecture to enhance feature interaction across resolutions. A memory feature supplementation fusion block (MFSFB) is introduced to progressively fuse shallow features with adaptive gating, enabling effective feature selection and spatial detail preservation. Furthermore, a multi-scale feature aggregation block (MFAB) is designed to promote cross-scale feature interaction, while a multi-level feature fusion block (MFFB) progressively integrates features from different levels, thereby improving fusion performance. Extensive experiments demonstrate that MSFFusion achieves better performance than several state-of-the-art (SOTA) methods in terms of both quantitative metrics and visual quality.