Visible-infrared person re-identification (VI-ReID) faces challenges in leveraging cross-modal semantic consistency within virtual-physical fusion systems. To address this, we propose PR-Net, a virtual surveillance-oriented framework that integrates spectral analysis with hierarchical feature fusion. The Fast Fourier Feature Transform (FFFT) aligns spectral distributions through frequency-domain correlation learning, establishing spectral-consistent representations for mixed reality rendering. The Super Token Aggregation (STA) captures both anatomical features and environmental context. Complemented by the Multi-Interaction Feature Extraction (MIFE) framework that fuses multiscale features through dual-channel interaction, our approach preserves identity discriminability for VR crowd simulation. Evaluations on SYSU-MM01 demonstrate 75.5% Rank-1 accuracy and 72.6% mAP under indoor search mode. This breakthrough provides a new paradigm for deploying ReID technologies in metaverse security systems requiring robust multimodal perception.