<p>To address the challenges of insufficient structural details, redundant background information, and underutilized cross-modal complementary information in infrared and visible image fusion, this paper proposes a novel cross-modal feature selection fusion network. By incorporating a multi-branch dynamic feature selection mechanism, cross-dimensional enhanced attention, and residual edge structure attention modules, this method significantly enhances structural edge and texture details and suppresses redundant information while maintaining global semantic consistency. Specifically, the model decomposes and fuses infrared and visible images, incorporating a multi-scale channel-space recalibration module (CDAA), a Residual Dynamic Attention Network module(RDANet), and a Multi-Scale Structural Enhancement Attention (MSSEA) to adaptively integrate low-frequency semantic information with high-frequency structural information. Experimental results on multiple public datasets (MSRS, TNO, and RoadScene) demonstrate that this method outperforms existing mainstream image fusion methods in terms of both objective evaluation metrics and subjective visual quality, effectively improving the clarity, contrast, and detail of the fused image.</p>