Low-light enhancement aims to restore dark images to high-quality ones. It has broad applications in fields such as nighttime surveillance and autonomous driving. However, LLIE methods still struggle to restore details and extreme low-light scenes, and traditional augmentations like CutMix and CutOut introduce abrupt edges, disrupting illumination continuity, and damage structural and textural information, which can even degrade model performance. To address these challenges, this paper proposes FreqMix, a frequency-domain augmentation technique tailored for LLIE tasks. Our method mixes only the low-frequency components of two images to increase training diversity while preserving the high-frequency components critical for detail restoration. This Method helps retain the smooth illumination transitions and fine texture details of LLIE. Experimental results on the widely used LOLv1 and LOLv2-real datasets demonstrate an average PSNR improvement of 0.5 dB alongside significantly enhanced detail restoration, confirming its effectiveness as a data augmentation method for LLIE.