Prior unsupervised low-light image enhancement methods have exhibited commendable performance within indoor environments. However, adopting them in the wild scene whose low-light images consist of smooth low-light regions, extremely low-light regions, and strong lighting effects regions, none of them can restore these three regions well simultaneously. In addition, semantic information is often lost during the enhancement procedure for the challenge condition, profoundly impacting the subsequent image analysis in various instrumentation and measurement-related applications. To overcome these two problems, we propose a region-based unsupervised low-light image enhancement (RLLIE) in the wild with explicit domain supervision (EDS). Specifically, for the first problem, we propose to utilize EDS to convert unsupervised segmentation into a supervised one. Then with the segmentation regions, we restore each region separately with our proposed unsupervised local low-light image enhancement. Here the proposed segmentation method is the first unsupervised image segmentation method based on illumination conditions, our RLLIE is the first region-based unsupervised low-light image enhancement method based on regions with different illumination conditions and our RLLIE is also the first method to map images consisting of various illumination conditions regions into daytime. For the second problem, several region-based loss functions are proposed to establish the semantic consistency between each region and daytime separately. Comprehensive experiments conducted across both supervised and unsupervised datasets substantiate the efficacy of RLLIE: RLLIE improves peak signal-to-noise ratio (PSNR) from 14.73 dB (the optimal unsupervised baseline without severe semantic information lost) to 16.79 dB on the supervised dataset sRGB-SID and increases SPAQ from 73.54 (the finest baseline without significant semantic information lost) to 75.34 on the unsupervised dataset NU.