The challenges of high-frequency detail loss and insufficient cross-dimensional feature interaction in super-resolution tasks were addressed through a collaborative enhancement network integrating dynamic gating mechanisms with spectral constraints. Significant improvements were achieved through three key components: edge-preserving enhancement, frequency-domain information utilization, and channel-spatial feature fusion. Edge gradient features were extracted via parallel Sobel convolutions in an edge-aware upsampling module, where high-frequency texture representation was enhanced by a dynamic attention mechanism. A dynamic gating collaborative network was constructed with multi-scale gated convolutional units to adaptively regulate cross-channel feature interactions. Spatial-spectral joint encoding was implemented through dynamic weight allocation to strengthen local-global information fusion. A hybrid spectral constraint loss function was proposed, where L1 constraints were separately applied to real and imaginary components of frequency-domain FFT(fast Fourier transforms). A multi-dimensional supervision framework was established through joint optimization with an edge consistency term. State-of-the-art quantitative results across multiple scaling factors were demonstrated on five benchmark datasets (Set5, Set14, BSD100, Urban100, Manga109). Superior restoration of sharp edges and periodic textures is confirmed through visual comparisons with existing methods.