<p>Infrared and visible image fusion (IVIF) serves as a critical enabling technology for target detection and remote sensing monitoring. However, existing methods often produce fused images with blurred semantics or degraded textures, primarily attributable to inadequate global context modeling and loss of fine-grained details. To overcome these limitations, we present PSFusion, which employs a progressive semantic-guided hierarchical fusion architecture for IVIF. The BaseSemanticEncoder extracts high-level semantic information through multi-level semantic feature distillation and dynamic sparse activation. This semantic information subsequently guides the ProgressiveDetailEncoder to gradually recover local texture details, forming a hierarchical feature representation system that evolves from global semantics to local structures. Furthermore, a global context module is incorporated to enhance cross-modal semantic consistency. The optimization process incorporates a compound loss function integrating perceptual loss with structural similarity constraints, coupled with a CosineAnnealingLR scheduler to accelerate convergence. Experiments on MSRS, TNO, and RoadScene demonstrate that PSFusion achieves state-of-the-art performance in both quantitative metrics and visual quality, demonstrating superior capability in maintaining structural integrity and textural fidelity, especially in complex scenarios. Ablation studies further substantiate the pivotal contributions of the semantic guidance paradigm and progressive fusion methodology.</p>