The　aim　of　pansharpening　is　to　fuse　low-resolution　multispectral　(MS)　images　from　various　sensors　with　high-resolution　panchromatic　(PAN)　data,　yielding　high-resolution　multispectral　(HRMS)　outputs.　Deep　learning–based　methods　have　developed　many　excellent　pansharpening　networks　with　modular　structures,　and　these　methods　often　outperform　traditional　approaches.　However,　fixed　modular　fusion　structures　tend　to　limit　the　full　utilization　of　cooperative　information　between　PAN　and　MS　images.　To　address　this　issue,　we　propose　a　dynamic　interaction　pansharpening　network　based　on　a　routing　decision.　Specifically,　we　design　a　dynamic　fusion-structure　space　that　achieves　interaction　between　spatial　and　spectral　information　by　optimizing　the　combination　to　construct　a　complete　path　space.　The　use　of　a　dynamic　decision-fusion　strategy　breaks　through　the　limitations　of　fixed　static　fusion　in　pansharpening　tasks.　Comprehensive　experiments—conducted　under　both　reduced-resolution　and　full-resolution　settings—on　the　WorldView-3　(WV3),　QuickBird　(QB),　and　GaoFen-2　(GF2)　datasets　demonstrate　the　superiority　of　our　approach.　Quantitative　comparisons　using　common　metrics　such　as　Spectral　Angle　Mapper　(SAM),　Error　Relative　Global　Dimension　Synthesis　(ERGAS),　and　Quality　with　No　Reference　(QNR)　consistently　show　that　our　method　achieves　state-of-the-art　performance.　For　example,　on　the　WV3　dataset,　PanRouter　reduces　SAM　from　5.7577　to　3.1229—a　relative　improvement　of　approximately　46%　compared　to　a　dense　fixed-fusion　baseline—while　incurring　only　a　minimal　increase　in　computational　overhead.　©　2025　Elsevier　Inc.