Images captured under insufficient lighting conditions face significant challenges, including notably reduced contrast and severe structural degradation. Most existing unsupervised learning-based approaches adopt a serialized two-stage processing strategy, which initially focuses on contrast enhancement followed by texture restoration. However, the cascaded processing exhibits inherent limitations that may introduce artifacts and compromise the texture details of the image during the contrast enhancement phase, thereby exacerbating the complexity and uncertainty inherent in the subsequent texture enhancement process. To address the above issue, I adopt a divide-and-conquer strategy and propose a novel parallel frequency-domain decoupling enhancement framework, which meticulously improves diverse degradation factors within separate frequency components and ultimately integrates them for comprehensive fine-tuning. Specifically, I design a cross-frequency domain fusion module to aggregate contextual information across different frequency domains, and propose a dynamic difference attention module to bolster the capabilities for structure restoration. Experiments on the LOL, MIT, and SICE datasets demonstrate that proposed method achieves better quantitative and qualitative results than the state-of-the-art algorithms.