Thangka is a Tibetan traditional painting art known for its intricate textures and profound cultural significance. Typically displayed in low-light environments in museums to prevent light from affecting its quality, this also makes it difficult for visitors to capture details and is not conducive to digital preservation and transmission. Traditional image enhancement techniques, such as GAN networks that require paired images for training, are limited in their application to Thangka datasets. To address this issue, we propose an image enhancement method called BM3D-UGanNet, which can be trained without paired data. This method combines the BM3D algorithm with the UGanNet network structure, effectively enhancing the visual quality of Thangka images under low-light conditions while preserving their details and texture. This technology not only improves the viewing experience but also provides high-quality image resources for art research and education, holding significant academic and practical value. In addition, the low-light Thangka dataset we have collected provides valuable resources for future research.