To address the strong coupling between illumination and texture details in low-light images, we propose a dual-stream attention enhancement network named DFGNet, which integrates degradation awareness with cross-channel interaction. The proposed framework processes luminance (Y) and chrominance (CbCr) channels separately in the YCbCr color space. Through a Luminance Degradation-Aware Network and a Chrominance Degradation-Aware Network, illumination attenuation and noise degradation features are explicitly decoupled. A Cross-Channel Fusion Module is designed to establish semantic correlations between the two channels, enabling collaborative optimization of degradation features. Furthermore, a Degradation Feature-Guided Image Restoration Network is introduced. By incorporating gradient consistency loss constraints, the DFGIRN effectively suppresses noise while preserving structural edges and texture details. Experimental results on the LOL dataset demonstrate that the proposed method outperforms existing state-of-the-art approaches across standard evaluation metrics, achieving significant improvements in both visual quality and image fidelity for low-light enhancement.